Download thesis - Curtin University
Download thesis - Curtin University
Download thesis - Curtin University
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Department of Spatial Sciences<br />
Improvement of Geodetic Parameter Estimation in Coastal Regions<br />
from Satellite Radar Altimetry<br />
Xiaoli Deng<br />
This <strong>thesis</strong> is presented for the Degree of<br />
Doctor of Philosophy<br />
of<br />
<strong>Curtin</strong> <strong>University</strong> of Technology<br />
August 2003
i<br />
Declaration<br />
This <strong>thesis</strong> contains no material which has been accepted for the award of any other degree<br />
or diploma in any university.<br />
To the best of my knowledge and belief this <strong>thesis</strong> contains no material previously published<br />
by any other person except where due acknowledgment has been made.<br />
Signature:<br />
………………………………………….<br />
Date:<br />
………………………...
ii<br />
ABSTRACT<br />
Altimeter-derived sea surface heights (SSH) are most probably in error in coastal<br />
regions due, in part, to the complex nature of echoes returned from rapidly varying<br />
coastal topographic surfaces (both land and sea). This dissertation presents improved<br />
altimeter-derived SSH results in coastal regions using the waveform retracking<br />
technique, which reprocesses the waveform data through a ‘coastal retracking<br />
system’.<br />
The system, based upon a systematic and comprehensive analysis of satellite radar<br />
altimeter waveforms in the Australian coast, provides an efficient means of<br />
improving altimeter-derived SSH data, not from a single retracker, but instead from<br />
several retrackers depending on the altimeter waveform characteristics. Central to the<br />
system is the use of two retracking techniques: the modified iterative least squares<br />
fitting and the threshold retracking algorithms. To overcome the ‘noise’ in<br />
waveforms caused by fading noise, the fitting algorithm has been developed to<br />
include a weighted iterative scheme. The retrackers adopted in the system include<br />
five fitting models and the threshold method with varying threshold levels. A<br />
waveform classification procedure has also been developed, which enables the<br />
waveform to be sorted and retracked by an appropriate retracker.<br />
Twenty-Hertz waveform data from the ERS-2 and POSEIDON satellites have been<br />
used to estimate a broad contaminated distance of ~10 km offshore the Australian<br />
coast. Two cycles of ERS-2 waveform data are, then, reprocessed using the coastal<br />
retracking system to obtain the improved SSH estimates. Using the AUSGeoid98<br />
geoid grid as a partly independent ground reference, the coastal retracking system<br />
has been able to reduce the contaminated distance to ~5 km. Improved ERS-1 SSH<br />
data from waveform retracking has also been obtained in another coastal region of<br />
Taiwan. Two altimeter-based gravity anomaly grids were created near Taiwan using<br />
the SSH data before and after retracking. Compared with ship-track gravity data,<br />
results show that the accuracy of the gravity anomalies has been improved from<br />
±13.9 mgal before retracking to ±9.9 mgal after retracking.
iii<br />
ACKNOWLEDGEMENTS<br />
I wish to extend my warmest and sincerest thanks to my supervisor, Professor Will<br />
Featherstone. Will has always been patient, helpful and supportive. I have learned a<br />
lot about the research methodology and geodesy from him. To him, I am indebted<br />
not just for his scientific supervision, but also for his understanding, help and<br />
friendship.<br />
I am most grateful to my associate supervisor, Professor Cheinway Hwang (National<br />
Chiao Tung <strong>University</strong>, Taiwan), for giving me the opportunity to work in his Space<br />
Geodesy Laboratory for a month, and for his help with many aspects of this research.<br />
I am grateful also to my other associate supervisors, Dr Jon Kirby and Professor<br />
Phillipa Berry (De Montfort <strong>University</strong>, UK), for their guidance during much of this<br />
research.<br />
My thanks go to all of the friendly team members, both staff and students, in the<br />
Western Australian Centre for Geodesy for their help and advice, particularly to<br />
people who worked with me in the lab: Troy Forward, Simon Holmes, Minghai Jia,<br />
Sten Claessens and Ireneusz Baran. To all the staff in the Department office, both<br />
past and present, I also extend my thanks.<br />
My thanks also go to Dr Michael Kuhn, Dr Mark Stewart and Dr Ruey-Gang Chang<br />
for their helpful comments towards this dissertation, and to Dr Graham Quartly<br />
(Southampton Oceanography Centre, UK), Dr Ronald Brooks (NASA, USA) and Dr<br />
George Hayne (NASA, USA) for providing some references and useful discussions.<br />
This research was funded by the International Postgraduate Research Scholarship<br />
(IPRS) and <strong>Curtin</strong> <strong>University</strong> Postgraduate Scholarship (CUPS). In-kind support was<br />
received from AVISO and ESA, by many of kindly providing POSEIDON and ERS-<br />
1/2 waveform data, respectively.<br />
Finally, I would like to thank my family for the constant love and support, in<br />
particular, my daughter, Lulu, with whom I shared happiness in the past two and half<br />
years. Without this love, patience and support, I am quite sure that this dissertation<br />
would never have been finished.
iv<br />
TABLE OF CONTENTS<br />
ABSTRACT.................................................................................................................ii<br />
ACKNOWLEDGEMENTS ........................................................................................iii<br />
TABLE OF CONTENTS............................................................................................ iv<br />
LIST OF FIGURES .................................................................................................... ix<br />
LIST OF TABLES ...................................................................................................xvii<br />
ACRONYMS AND ABBREVIATIONS ................................................................. xix<br />
1. INTRODUCTION ............................................................................................... 1<br />
1. 1 Applications of Satellite Radar Waveforms................................................. 1<br />
1.1.1 Description of the Satellite Radar Altimeter Waveforms .................... 1<br />
1.1.2 General Applications of Waveform Data............................................. 4<br />
1. 2 Problems in Coastal Areas ........................................................................... 6<br />
1.2.1 Topographic Effects on Waveforms .................................................... 7<br />
1.2.2 Topographic Effects on Geophysical Corrections ............................. 10<br />
1. 3 Justifications for This Study ...................................................................... 11<br />
1.3.1 Previous Work.................................................................................... 11<br />
1.3.2 Significance of the Improvement of Altimeter Data in Coastal<br />
Regions ............................................................................................................ 12<br />
1. 4 The Study Areas......................................................................................... 15<br />
1. 5 Thesis Outline ............................................................................................ 16<br />
2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES..... 18<br />
2. 1 Introduction................................................................................................ 18<br />
2. 2 General Information on Construction of Altimeter Measurements ........... 19<br />
2.2.1 Fundamental Radar Measurement Principles..................................... 19<br />
2.2.2 The Returned Waveform.................................................................... 21<br />
2.2.3 The Range and Sea Surface Height.................................................... 25<br />
2. 3 The Altimeter Waveform Data Used ......................................................... 26<br />
2.3.1 ERS-1 Waveform Data ...................................................................... 26<br />
2.3.2 ERS-2 Waveform Data ...................................................................... 27<br />
2.3.3 POSEIDON Waveform Data ............................................................. 28<br />
2. 4 External Input Data .................................................................................... 29<br />
2.4.1 The AUSGeoid98 Geoid Model of Australia..................................... 30
v<br />
2.4.2 Sea Surface Topography .................................................................... 31<br />
2.4.3 Marine Gravity Data around Taiwan ................................................. 34<br />
2.4.4 The Australian Digital Elevation Model (DEM) ............................... 35<br />
2. 5 Summary .................................................................................................... 36<br />
3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS............... 37<br />
3. 1 Introduction................................................................................................ 37<br />
3. 2 Convolutional Representation of Ocean-Returned Waveforms................. 38<br />
3.2.1 The Brown Model .............................................................................. 38<br />
3.2.2 Average Flat Surface Impulse Response Function ............................ 40<br />
3.2.3 Surface Elevation PTR of the Scattering Distribution ....................... 41<br />
3.2.4 Radar System PTR ............................................................................. 42<br />
3.2.5 Solution of the Radar Returns............................................................ 42<br />
3. 3 Modified Convolutional Representation of the Ocean Return Waveform. 43<br />
3.3.1 Effects of the Earth’s Curvature......................................................... 44<br />
3.3.2 Simplification of the Bessel Function................................................ 44<br />
3.3.3 Approximate Expression of the Non-Gaussian Scattering Surface ... 46<br />
3.3.4 Approximate Expression of the System PTR .................................... 49<br />
3.3.5 A Modified Solution of the Radar Returns Developed in This Study50<br />
3. 4 Ocean Waveform Retracking..................................................................... 55<br />
3.4.1 Least Squares Fitting Procedure ........................................................ 56<br />
3.4.2 Deconvolution Method ...................................................................... 58<br />
3. 5 Ice-Sheet Waveform Retracking................................................................ 60<br />
3.5.1 Fitting Algorithm - β-Parameter Retracking...................................... 62<br />
3.5.2 Off-Centre of Gravity (OCOG) Retracking Algorithm ..................... 65<br />
3.5.3 Threshold Retracking Algorithm ....................................................... 66<br />
3.5.4 Surface/Volume Scattering Retracking Algorithm ............................ 67<br />
3. 6 Land Waveform Retracking....................................................................... 69<br />
3. 7 Brief Introduction to Coastal Waveform Retracking................................. 71<br />
3. 8 Summary .................................................................................................... 72<br />
4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS .. 74<br />
4. 1 Introduction................................................................................................ 74<br />
4. 2 Background ................................................................................................ 75<br />
4.2.1 ERS-2 and POSEIDON Altimeter Footprints.................................... 76
vi<br />
4.2.2 ERS-2 and POSEIDON Altimeter Waveforms.................................. 76<br />
4. 3 Processing Methodology for Determination of Contamination Area ........ 78<br />
4.3.1 Threshold Retracking......................................................................... 78<br />
4.3.2 Mean Power of the Waveform ........................................................... 80<br />
4. 4 Data and Editing......................................................................................... 81<br />
4.4.1 ERS-2 ................................................................................................. 81<br />
4.4.2 POSEIDON........................................................................................ 83<br />
4. 5 Determination of Contamination Area....................................................... 84<br />
4.5.1 Using 50% of Threshold Retracking Points of ERS-2....................... 84<br />
4.5.2 Using the Mean Waveform and its Standard Deviation..................... 89<br />
4.5.3 Using Statistics of the 50% Retracking Point .................................... 94<br />
4. 6 Waveform Contamination Analysis........................................................... 97<br />
4.6.1 Waveform Contamination for Water-to-Land Ground Tracks .......... 97<br />
4.6.2 Waveform Contamination for Land-to-Water Ground Tracks ........ 101<br />
4.6.3 Contamination Caused by the Altimeter’s Operation ...................... 107<br />
4.6.4 Summary of Waveform Contamination in Coasts ........................... 109<br />
4. 7 An Experiment of the Land Effects on Waveforms Using the DEM Data....<br />
.................................................................................................................. 110<br />
4.7.1 Concept of Land Effects Using the DEM Model............................. 111<br />
4.7.2 A Preliminary Test of Land Effects Using the DEM Model ........... 114<br />
4. 8 Summary .................................................................................................. 116<br />
5. COASTAL WAVEFORM RETRACKING SYSTEM: DESIGN AND<br />
IMPLEMENTATION.............................................................................................. 119<br />
5. 1 Introduction.............................................................................................. 119<br />
5. 2 Coastal Waveform Retracking System Design........................................ 120<br />
5. 3 Extraction of the Geodetic Parameter: the Sea Surface Height ............... 122<br />
5. 4 Analysis of Waveform Shapes................................................................. 123<br />
5.4.1 Determination of the Peaks in the Waveform.................................. 123<br />
5.4.2 Waveform Classification.................................................................. 124<br />
5. 5 Fitting Functions ...................................................................................... 127<br />
5.5.1 The Ocean Model............................................................................. 127<br />
5.5.2 Five- and Nine-β-Parameter Models................................................ 131<br />
5.5.3 Comparison between the Ocean and Five-β-Parameter Models...... 131
vii<br />
5. 6 Iterative Nonlinear Fitting Procedure ...................................................... 136<br />
5.6.1 Solutions of the Waveform Retracking Fitting Function................. 137<br />
5.6.2 Linearisation of the Waveform Retracking Fitting Function........... 138<br />
5.6.3 Initial Estimates of the Unknown Model Parameters ...................... 139<br />
5.6.4 Data Weight and Outlier Detection Scheme .................................... 142<br />
5.6.5 Iterative Procedure ........................................................................... 144<br />
5. 7 Threshold Retracking............................................................................... 145<br />
5.7.1 Threshold Retracking Algorithm ..................................................... 145<br />
5.7.2 Selection of the Threshold Level ..................................................... 146<br />
5.7.3 Results of Selecting the Threshold Levels....................................... 149<br />
5. 8 An Assessment of Biases among Waveform Retracking Algorithms ..... 155<br />
5.8.1 Analysis Methods and Data ............................................................. 155<br />
5.8.2 Retracking Results and Analysis...................................................... 157<br />
5.8.3 Biases among Retracking Algorithms.............................................. 160<br />
5.8.4 Biases between the SSH Data before and after Retracking ............. 164<br />
5. 9 Summary .................................................................................................. 167<br />
6. WAVEFORM RETRACKING APPLICATION TO THE AUSTRALIAN<br />
COAST..................................................................................................................... 169<br />
6. 1 Introduction.............................................................................................. 169<br />
6. 2 Waveform Classification in Australian Coastal Regions......................... 170<br />
6.2.1 Typical Waveform Shapes ............................................................... 170<br />
6.2.2 Results of Waveform Categories ..................................................... 176<br />
6. 3 Waveform Retracking Results and Discussion........................................ 177<br />
6.3.1 Data Editing ..................................................................................... 177<br />
6.3.2 Waveform Retracking Using Different Algorithms......................... 177<br />
6.3.3 Waveform Retracking Using Different Fitting Functions ............... 179<br />
6. 4 Analysis of the Single-Track Retracked Results...................................... 182<br />
6. 5 Analysis of the SSH Data Before and After Retracking.......................... 185<br />
6. 6 Collinear Analysis.................................................................................... 187<br />
6.6.1 Methodology .................................................................................... 187<br />
6.6.2 Discussion of Results ....................................................................... 188<br />
6. 7 Evaluation of SSH Data Using Geoid Heights ........................................ 190<br />
6.7.1 Comparison with the AUSGeoid98 Geoid Model ........................... 190
viii<br />
6.7.2 Results of the Comparison between SSH Data and Geoid Heights. 191<br />
6. 8 Summary .................................................................................................. 193<br />
7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES NEAR<br />
TAIWAN FROM WAVEFORM RETRACKING .................................................. 196<br />
7. 1 Introduction.............................................................................................. 196<br />
7. 2 Data and Area........................................................................................... 198<br />
7. 3 Waveform Retracking .............................................................................. 200<br />
7.3.1 Retracking Scheme........................................................................... 201<br />
7.3.2 Examples of Waveform Retracking Results .................................... 202<br />
7. 4 The SSH Extracted from Waveforms ...................................................... 205<br />
7.4.1 Corrections ....................................................................................... 205<br />
7.4.2 The Weight Function ....................................................................... 211<br />
7.4.3 Data Filtering and Editing................................................................ 211<br />
7. 5 Marine Gravity Anomalies from Satellite Altimetry ............................... 212<br />
7.5.1 Along-track Deflections of the Vertical........................................... 216<br />
7.5.2 Averaging Gradients ........................................................................ 216<br />
7.5.3 Gravity Anomaly Computation........................................................ 219<br />
7.5.4 Comparisons between Marine and Ship Gravity Anomalies ........... 219<br />
7. 6 Summary .................................................................................................. 224<br />
8. CONCLUSIONS AND RECOMMENDATIONS .......................................... 226<br />
8. 1 Summary of Dissertation.......................................................................... 226<br />
8. 2 Specific Conclusions................................................................................ 228<br />
8.2.1 The Quantitative Estimation of the Contaminated Distance............ 228<br />
8.2.2 Practical Implementation of the Coastal Waveform Retracking System<br />
.......................................................................................................... 230<br />
8.2.3 Development and Validation of a Coastal Retracking System ........ 232<br />
8.2.4 Other Conclusions............................................................................ 235<br />
8. 3 Recommendations for Future Work......................................................... 235<br />
8.3.1 Extraction of Ocean Returns from the Mixed Waveform Shape ..... 236<br />
8.3.2 Improvement of the Correction Algorithms near Land.................... 236<br />
8.3.3 Improvement of the Altimeter Operation......................................... 237<br />
8.3.4 Geodetic Application ....................................................................... 237<br />
REFERENCES......................................................................................................... 238
ix<br />
LIST OF FIGURES<br />
Figure 1.1 Schematic altimeter mean return waveform over sea surfaces................... 3<br />
Figure 1.2 An observed ERS-2 waveform over deep oceans. (1 bin/gate = 0.45<br />
m and tracking gate = 32.5)....................................................................... 3<br />
Figure 1.3 Altimeter simultaneously illuminates land and ocean near the<br />
coastline..................................................................................................... 8<br />
Figure 1.4 Differences between the ERS-2 SSH and AUSGeiod98 geoid height<br />
along four collinear repeat tracks from cycles 42 and 43 at the<br />
Southern Australian coast, showing the metre-level uncertainty of<br />
SSHs closer to the coastline. ..................................................................... 9<br />
Figure 2.1 Basic measurement principle of the altimeter-derived SSH (From<br />
http://www.jason.oceanobs.com/html/portail/galerie<br />
/banque_img_welcome_uk.php3). .......................................................... 18<br />
Figure 2.2 A schematic geometrical description of the interaction of a pulse and<br />
the scattering surface, and the build up of a return waveform over<br />
the duration of a pulse. Note that, an average returned waveform<br />
from 50 typical individual returns is depicted, while a single typical<br />
return shows more noise. (PLF is the pulse-limited footprint). .............. 22<br />
Figure 2.3 Quasi-time-independent SST (contour interval is 10 cm) from Levitus<br />
et al. (1997). ............................................................................................ 33<br />
Figure 2.4 Coverage of ship-track gravity observations around Taiwan................... 34<br />
Figure 3.1 Values of d ( ξ = 0 ), showing that d
x<br />
Figure 3.6 Nine- β -parameter model, fitting a double-ramp function to two<br />
leading edges of the waveform samples.................................................. 64<br />
Figure 3.7 Schematic description of the OCOG method applied to an observed<br />
waveform. Using the OCOG method to calculate the amplitude (A)<br />
and width (W) of the rectangle, and the position of the 50% of<br />
amplitude interpolated at the leading edge as the estimate of<br />
expected tracking gate............................................................................. 66<br />
Figure 4.1 An example of the observed POSEIDON waveform over deep oceans<br />
(tracking gate = 31.5). ............................................................................. 77<br />
Figure 4.2 ERS-2 ground tracks (35-day repeat orbit, from March to April 1999)<br />
to a distance of 350 km from Australian shoreline (Lambert<br />
projection). .............................................................................................. 82<br />
Figure 4.3 The ocean-ice-mode flag in ERS-2 data supplied close to the<br />
Australian shoreline compared to the Wessel (2000) coastline,<br />
showing that most flags around Australia are on the land (Lambert<br />
projection). .............................................................................................. 82<br />
Figure 4.4 POSEIDON ground tracks (10-day repeat orbit, from January 1998 to<br />
January 1999) to a distance of 350 km from Australian shoreline<br />
(Lambert projection). .............................................................................. 83<br />
Figure 4.5 One cycle of ERS-2 contaminated waveforms (highlighted by circles)<br />
along ground tracks in the southeast coasts of Australia. ....................... 86<br />
Figure 4.6 Distributions of 50% retracking points plotted versus DS759.2 ocean<br />
depths for contaminated waveforms in areas of 105°≤λ
xi<br />
Figure 4.9 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for ERS-2 in five 10 km-wide bands from<br />
the Australian coastline........................................................................... 91<br />
Figure 4.10 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for POSEIDON cycle 197 in five 10 kmwide<br />
bands from the Australian coastline. .............................................. 92<br />
Figure 4.11 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for ERS-2 in five 2 km-wide bands from the<br />
Australian coastline................................................................................. 93<br />
Figure 4.12 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for POSEIDON cycle 197 in five 2 km-wide<br />
bands from the Australian coastline........................................................ 94<br />
Figure 4.13 Fifty percent retracking point frequency distributions for 5 cycles of<br />
Poseidon waveform data (left) and 1 cycle of ERS-2 waveform data<br />
(right). They are shown as the percentage of observations in each bin<br />
for five 2 km-wide bands around the Australian coastline. .................... 95<br />
Figure 4.14 The 20 Hz SSH profile of pass 21792 (ascending track, cycle 43 of<br />
ERS-2), approaching the eastern Australian shoreline from the<br />
Tasman Sea. The number beside the SSH samples corresponds to the<br />
waveform in Figure 4.15. ........................................................................ 98<br />
Figure 4.15 20 Hz waveforms of pass 21792 (ascending track, cycle 43 of ERS-<br />
2), approaching Australian eastern shoreline in the Tasman Sea. The<br />
number in the top left-hand corner is related to that in Figure 4.14........ 98<br />
Figure 4.16 The 20 Hz SSH profile of pass 21450 (ascending track, cycle 43 of<br />
ERS-2), approaching the western Australian shoreline from the<br />
Indian Ocean. The number beside the SSH samples corresponds to<br />
the waveform in Figure 4.17. ................................................................ 100<br />
Figure 4.17 The 20 Hz waveforms of pass 21450 (ascending track, cycle 43 of<br />
ERS-2), approaching Australian western shoreline in the Indian<br />
Ocean. The number is related to that in Figure 4.16............................. 100<br />
Figure 4.18 The 20 Hz SSH profile of pass 21636 (ascending track, cycle 43 of<br />
ERS-2), receding from the northwestern Australian shoreline to the<br />
Indian Ocean. ........................................................................................ 103
Figure 4.19 20 Hz waveforms of pass 21636 at the 1st second (~0-7 km alongtrack<br />
distance) from the shoreline (ascending track, cycle 43 of ERS-<br />
2), receding from the northwestern Australian shoreline to the Indian<br />
Ocean..................................................................................................... 103<br />
Figure 4.20 20 Hz waveforms of pass 21636 at the 2nd second (~7-14 km alongtrack<br />
distance) from the shoreline (ascending track, cycle 43 of ERS-<br />
2), receding from northwestern Australian shoreline to the Indian<br />
Ocean..................................................................................................... 104<br />
Figure 4.21 20 Hz waveforms of pass 21636 at the 3rd second (~14-21 km<br />
along-track distance) from the shoreline (ascending track, cycle 43<br />
of ERS-2), receding from the northwestern Australian shoreline to<br />
the Indian Ocean. .................................................................................. 104<br />
Figure 4.22 The 20 Hz SSH profile of pass 21886 (descending track, cycle 43 of<br />
ERS-2), receding from the southern Australian shoreline to the<br />
Southern Ocean. .................................................................................... 105<br />
Figure 4.23 20 Hz waveforms of pass 21886 at the 1st second (0-7 km alongtrack<br />
distance) from the shoreline (descending track, cycle 43 of<br />
ERS-2), receding from the southern Australian shoreline to the<br />
Southern Ocean. .................................................................................... 106<br />
Figure 4.24 20 Hz waveforms of pass 21886 at the 2nd second from the<br />
shoreline (ascending track, cycle 43 of ERS-2), receding Australian<br />
southern shoreline to the Southern Ocean............................................. 106<br />
Figure 4.25 20 Hz waveforms of pass 21886 at the 3rd second from the<br />
shoreline (ascending track, cycle 43 of ERS-2), receding Australian<br />
southern shoreline to the Southern Ocean............................................. 107<br />
Figure 4.26 A schematic geometrical description of the radar return from a point<br />
on the land surface at off-nadir angle ξ and associated colatitude<br />
angle φ where the land elevation above the mean sea level is H .<br />
The distance from the altimeter to the land surface is<br />
xii<br />
'<br />
R<br />
φ<br />
. ................... 113<br />
Figure 4.27 Retracking gate estimates corresponding to the land elevation. The<br />
search radius is 2.7 km (top) and 3.5 km (bottom). .............................. 115<br />
Figure 5.1 Block diagram of the coastal retracking system..................................... 121<br />
Figure 5.2 Typical ERS-2 waveform categories in Australian coasts ..................... 126
Figure 5.3 Variation of the leading edge’s slope with varying SWH (0 - 20 m),<br />
xiii<br />
modelling waveforms using the ocean model ( ξ = 0 °). ....................... 129<br />
Figure 5.4 Effect of the off-nadir angle (in degrees) on the amplitude, leading<br />
edge, and especially the slope of the trailing edge, modelling<br />
waveforms using the ocean model (SWH = 4 m). ................................ 130<br />
Figure 5.5 Variations of the trailing edge’s slope with varying parameter β 5 ,<br />
modelling waveforms using the five-β-parameter parameter function<br />
(linear trailing edge, SWH = 4 m, β<br />
21<br />
= 500 counts, β 5 = β<br />
51)........... 134<br />
Figure 5.6 Variations of the trailing edge’s slope with varying parameter β 5 ,<br />
modelling waveforms using the five-β-parameter parameter function<br />
(exponential decayed trailing edge, SWH = 4 m, β<br />
21<br />
= 500 counts,<br />
β 5 = β 51<br />
). ............................................................................................. 135<br />
Figure 5.7 Mean waveforms and the standard deviation in two 5 km-wide bands<br />
of 0-5 km (top) and 5-10 km (bottom) from the Australian coastline.. 148<br />
Figure 5.8 Ten 15º×15º coastal areas around Australia, showing the area number<br />
(Lambert projection). ............................................................................ 150<br />
Figure 5.9 Mean differences between the SSH data after retracking and<br />
AUSGeoid98 geoid heights in ten areas 5 km from the Australian<br />
coastline................................................................................................. 152<br />
Figure 5.10 Standard deviations of the mean differences in ten areas 5 km from<br />
the Australian coastline......................................................................... 152<br />
Figure 5.11 Mean differences between the SSH data after retracking and<br />
AUSGeoid98 geoid heights in ten areas 5-10 km from the Australian<br />
coastline................................................................................................. 154<br />
Figure 5.12 Standard deviations of the mean differences in ten areas 5-10 km<br />
from the Australian coastline. ............................................................... 154<br />
Figure 5.13 Geoid height and SSH profiles along ground tracks 21085 (top) and<br />
21364 (bottom) (ERS-2, cycle 42) before and after retracking using<br />
the ocean model, the five-parameter model and the 50% threshold<br />
level in coastal regions.......................................................................... 158<br />
Figure 5.14 Retracked and unretracked SSH profiles over open oceans, showing<br />
the biases among them. ......................................................................... 161
xiv<br />
Figure 5.15 Histogram of the differences between the ocean model and fiveparameter<br />
model.................................................................................... 162<br />
Figure 5.16 Histograms of the differences between the 50% threshold retracking<br />
technique and the ocean model as well as the five-parameter model. .. 162<br />
Figure 5.17 Histograms of the differences between unretracked SSH data and<br />
retracked SSH data using the ocean model and the five-parameter<br />
model..................................................................................................... 164<br />
Figure 5.18 Histogram of the differences between untracked SSH data and<br />
retracked SSH dada using the 50% threshold retracking technique...... 165<br />
Figure 6.1 Part of ERS-2 ground tracks off the northwest Australian coast,<br />
showing the orbit number of each ground track. The highlighted<br />
diamonds along tracks will be shown in detail waveform shapes later<br />
in this Section........................................................................................ 171<br />
Figure 6.2 Typical ocean waveforms from pass 21815 (cf. Figure 6.1). ................. 173<br />
Figure 6.3 Waveforms recorded when the ground track 21815 is approaching an<br />
island at latitude ~ -20.5º and leaving the island to water at ~ -20.9º.<br />
The data gap is due to the island (cf. Figure 6.1).................................. 173<br />
Figure 6.4 Waveforms leaving land to water (pass 21636 in Figure 6.1). The<br />
high-peaked specular responses are probably due to off-nadir<br />
brighter calm water. The waveform shifting (~ -22.25°
xv<br />
Figure 6.9 STD (positive values) of the collinear 20 Hz SSH differences of<br />
cycles 42 and 43 before and after retracking in six 5 km wide<br />
distance bands, showing improvement from SSH data after<br />
retracking............................................................................................... 189<br />
Figure 6.10 Mean differences of geoid heights (AUSGeoid98) and SSH data<br />
before and after retracking in six 5 km wide distance bands (cycle<br />
42). ........................................................................................................ 191<br />
Figure 6.11 STD (positive values) of the mean difference before and after<br />
retracking in six 5 km wide distance bands, showing improvement<br />
on SSH data after retracking (cycle 42). ............................................... 192<br />
Figure 6.12 Mean differences between geoid heights (AUSGeoid98) and SSH<br />
data before and after retracking in six 5 km wide distance bands<br />
(cycle 43)............................................................................................... 192<br />
Figure 6.13 STD (positive values) of the mean difference before and after<br />
retracking in six 5 km wide distance bands, showing improvement<br />
on SSH data after retracking (cycle 43). ............................................... 192<br />
Figure 7.1 Distribution of ERS-1 ground tracks ...................................................... 199<br />
Figure 7.2 The histogram of dh (the SSH difference between the five-parameter<br />
model and the ocean model (mean -0.7 cm; STD 13.5 cm; 3070<br />
observations). ........................................................................................ 202<br />
Figure 7.3 Three-ground tracks in the study area .................................................... 203<br />
Figure 7.4 SSHs from ground track A1 (a), D1 (b), and D2 (c) in Figure 7.3<br />
before and after retracking using the retrackers of the five-parameter<br />
and ocean models. ................................................................................. 204<br />
Figure 7.5 SSH profiles and parts of corrections along the track A1. ..................... 208<br />
Figure 7.6 SSH profiles and parts of corrections along the track D1. ..................... 209<br />
Figure 7.7 SSH profiles and parts of corrections along the track D2. ..................... 210<br />
Figure 7.8 Untracked SSH profiles, showing the filtered SSH data (dark curves)<br />
and unfiltered SSH data (light curves) along the tracks A1 (a), D1 (b)<br />
and D2 (c) in Figure 7.3. ....................................................................... 213<br />
Figure 7.9 Retracked SSH profiles, showing filtered SSH data (dark curves) and<br />
unfiltered SSH data (light curves) along the tracks A1 (a), D1 (b)<br />
and D2 (c) in Figure 7.3. ....................................................................... 214
xvi<br />
Figure 7.10 Flow diagram for the SSH extraction from waveforms and recovery<br />
of gravity anomalies.............................................................................. 220<br />
Figure 7.11 Gravity anomalies from altimeter data before retracking (units in<br />
mgal) ..................................................................................................... 221<br />
Figure 7.12 Gravity anomalies from altimeter data after retracking (units in mgal)<br />
............................................................................................................... 221<br />
Figure 7.13 Differences between altimeter-derived (before retracking) and shiptrack<br />
gravity anomalies ......................................................................... 223<br />
Figure 7.14 Differences between altimeter-derived (after retracking) and shiptrack<br />
gravity anomalies ......................................................................... 223
xvii<br />
LIST OF TABLES<br />
Table 2.1 Altimeter Operating Parameters................................................................. 20<br />
Table 2.2 Corrections applied to ERS-1 range measurements and their likely<br />
error......................................................................................................... 25<br />
Table 2.3 The repeat cycle characteristics of ERS-1. ................................................ 26<br />
Table 2.4 The reference ellipsoid characteristics of ERS-1/2 and T/P. ..................... 29<br />
Table 4.1 Contaminated 1 Hz ocean depths and 20 Hz distances from the nearest<br />
shoreline in different ............................................................................... 85<br />
Table 4.2 Differences between 50% retracking points and the tracking gate (bin<br />
31.5) for 5 cycles (January 1998 to January 1999) of Poseidon<br />
waveforms (each bin has the range distance of ~0.469 m). .................... 96<br />
Table 4.3 Differences between 50% retracking points and the tracker point (bin<br />
32.5) for 1 cycle (March to April 1999) of ERS-2 waveforms (each<br />
bin with the range width of ~0.454m)..................................................... 96<br />
Table 4.4 Descriptive statistics of the differences of retracking gate estimates<br />
before and after shifting rightward waveform samples in the range<br />
window (20 Hz waveforms from pass 21886 at the 2nd second from<br />
the coastline, 1 gate≈0.4542 m) ............................................................ 108<br />
Table 5.1 ασ² Values for ERS-1/2 (ocean mode) and TOPEX ............................... 132<br />
Table 5.2 Criteria of the absolute differences between two iterations (the ocean<br />
model for ERS-1/2) ............................................................................... 145<br />
Table 5.3 Along-track geoid and SSH gradients (track 21085)............................... 159<br />
Table 5.4 Along-track geoid and SSH gradients (track 21364)............................... 160<br />
Table 5.5 Biases between different retracking techniques from fitting a quadratic<br />
function to the ERS-2 SSH (cycle 42) differences over an ocean area<br />
(-30°S
xviii<br />
Table 5.8 Biases between the ERS-2 (cycle 43) SSHs before and after retracking<br />
over an ocean area (-30°S
xix<br />
ACRONYMS AND ABBREVIATIONS<br />
AGC<br />
Automatic gain control<br />
AGSO Australian Geological Survey Organisation, now Geoscience<br />
Australia (GA)<br />
ALT.WAP<br />
AVISO<br />
AWF<br />
DEM<br />
ECMWF<br />
EMB<br />
EODC<br />
ESA<br />
FFT<br />
GDR<br />
GPS<br />
IAG<br />
MLE<br />
OCOG<br />
PLF<br />
PTF<br />
PTR<br />
RMS<br />
SKB<br />
SSB<br />
SSH<br />
Altimeter waveform products<br />
Archiving, Validation and Interpretation of Satellite Oceanographic<br />
team<br />
AIDA waveform format<br />
Digital elevation model<br />
European Centre for Medium-Range Weather Forecasts<br />
Electromagnetic bias<br />
Earth Observation Data Centre, UK<br />
European Space Agency<br />
Fast Fourier transform<br />
Geophysical data record<br />
Global Positioning System<br />
International Association of Geodesy<br />
Maximum likelihood estimation<br />
Off centre of gravity method<br />
Pulse-limited footprint<br />
Probability density function<br />
Radar system point target response<br />
Root-mean-square<br />
Skewness bias<br />
Sea state bias<br />
Sea surface height
xx<br />
SST<br />
STD<br />
SWH<br />
T/P<br />
UK-PAF<br />
Sea surface topography<br />
Standard deviation<br />
Significant wave height<br />
TOPEX/POSEIDON<br />
United Kingdom Processing and Archiving Facility
Chapter 1. INTRODUCTION 1<br />
1. INTRODUCTION<br />
This dissertation is concerned with the intersection of two topics of geodetic satellite<br />
radar altimetry in coastal regions: satellite radar altimeter waveform retracking; and<br />
some applications of the waveform retracking. This Chapter will introduce both the<br />
background and the justification for the dissertation.<br />
1. 1 Applications of Satellite Radar Waveforms<br />
Radar altimeters have been flown on a number of satellites (GEOS-3, SEASAT,<br />
GEOSAT, TOPEX/POSEIDON (T/P), ERS-1 and 2, JASON-1, and ENVISAT) and<br />
been found to give useful information about the oceans, such as the large-scale<br />
movement of water and their total mass and volume (e.g., Marth et al., 1993; Chelton<br />
et al., 2001). The basic altimetric measurement is the vertical distance or range<br />
between the satellite and the at-nadir instantaneous sea surface. This range<br />
measurement, when combined with precise knowledge of the satellite orbit,<br />
propagation effects, and the Earth’s geopotential field, allows the determination of<br />
sea surface slopes and hence surface currents. The altimeter range measurements are<br />
determined from the time interval between the time that the radar pulse is transmitted<br />
and the time that the return reflected from the (assumed at-nadir) mean sea surface is<br />
received. Satellite radar altimeters transmit pulses at a certain frequency (e.g.,<br />
1,020 Hz for ERS-2 and 1,700 Hz for POSEIDON). The on-satellite processor<br />
averages the reflected radar returns to give the waveform that relates to the shape of<br />
the sea surface, which in turn contains information on currents, tides and ocean<br />
bottom features related to oceanography and solid Earth geophysics (e.g., Tokmakian<br />
et al., 1994)<br />
1.1.1 Description of the Satellite Radar Altimeter Waveforms<br />
Pulsed-limited radar altimeters produce a measurement of power as a function of<br />
time (e.g., Marth et al., 1993; Quartly and Srokosz, 2001; Chelton, 2001). From this<br />
temporal profile of received power, which is referred to as the altimeter waveform<br />
(described in detail in Section 2.2.2), the range to the at-nadir sea surface can be<br />
determined over oceans. In addition to the range, the waveform also indicates the
Chapter 1. INTRODUCTION 2<br />
reflectivity (estimated from the amount of power in the reflected pulse) and the largescale<br />
roughness of the scattering surface (estimated from the slope of the leading<br />
edge of the waveform). Of these, only the range will be considered in this<br />
dissertation as a geodetic parameter with a view to developing an improved range<br />
determination in coastal regions.<br />
The range is measured by the altimeter indirectly through a direct measurement of<br />
the two-way travel time between the transmission and reception of a radar pulse. The<br />
observed time delay can be converted into a range measurement as long as the<br />
propagation velocity is known. A determination of this velocity, with corrections for<br />
the instrument (e.g., the Doppler shift, antenna gain pattern, and automatic gain<br />
control (AGC) attenuation), the atmospheric refraction (e.g., wet and dry troposphere,<br />
the ionosphere), the sea-state bias, and geophysical adjustments (e.g., ocean tides and<br />
atmospheric pressure loading), is an essential part of radar altimetry (Chelton et al.,<br />
2001).<br />
The profile of the backscattered power (i.e., the waveform) is a function of range<br />
measured by the altimeter. It is the convolution of three terms based on specular<br />
reflection from the sea surface: the system’s point target response (PTR), the impulse<br />
response of the smooth spherical scattering surface, and the ocean-surface height<br />
distribution (e.g., Brown, 1977; Hayne, 1980; Rodriguez, 1988; Jensen, 1999). If the<br />
ocean waves on the sea surface are assumed to be linear, the corresponding statistics<br />
of surface elevation and slopes are Gaussian. This means that the ocean-surface<br />
height distribution is assumed symmetric about some mean value. Therefore, the<br />
altimeter waveform is an odd function relative to the midpoint (i.e., half-power point)<br />
on the leading edge of the waveform, and the range to the at-nadir sea surface<br />
corresponds to this midpoint of the leading edge. The Brown (1977) model is thus<br />
the general function used by the on-satellite data processor (Chelton et al., 2001).<br />
Figures 1.1 and 1.2 show a schematic altimeter waveform based on the above<br />
theoretical description and an observed ERS-2 waveform, respectively. As can be<br />
seen, the waveform consists mainly of three parts (Figure 1.1): the thermal noise,<br />
leading edge and the trailing edge (see Section 2.2.2 for the details), and recorded in<br />
64 range bins or gates (Figure 1.2). From Figures 1.1 and 1.2, there are differences
Chapter 1. INTRODUCTION 3<br />
between the modelled and actual waveforms. The observed waveform shows wraparound<br />
noise at the first and the last five bins, and noise along the trailing edge.<br />
Noise also affects the whole waveform. Due to these differences, it is hard to define a<br />
tracker point that is related to the mid-point on the leading edge of the waveform<br />
tracked by an altimeter. Therefore, it is replaced by a tracking gate, which is<br />
designed before the satellite is launched and located at bin 32.5 for ERS-1/2. The<br />
onboard tracking algorithm (e.g., Brown model) tries to centre the leading edge at the<br />
position of this gate. The ERS leading edge usually contains 3-4 range bins (Figure<br />
1.2).<br />
Returned Power<br />
Thermal Noise<br />
Leading Edge<br />
Trailing Edge<br />
Time or Bins<br />
Figure 1.1 Schematic altimeter mean return waveform over sea surfaces.<br />
1400<br />
1200<br />
1000<br />
Power (counts)<br />
800<br />
600<br />
400<br />
200<br />
0<br />
0 8 16 24 32 40 48 56 64<br />
Bins or Gates<br />
Figure 1.2 An observed ERS-2 waveform over deep oceans.<br />
(1 bin/gate = 0.45 m and tracking gate = 32.5).
Chapter 1. INTRODUCTION 4<br />
1.1.2 General Applications of Waveform Data<br />
The above-mentioned satellite radar altimetric tracking theory has allowed good<br />
estimates of geophysical and oceanographic parameters to be obtained over open<br />
ocean surfaces (e.g., Chelton et al., 2001; Hayne, 1980). The waveform shape based<br />
on the convolution form (e.g., Brown, 1977), especially the sharply rising leading<br />
edge of the waveform (see Figure 1.2), is the basis for the precise range estimation.<br />
Analysing the waveforms provides estimates of oceanographic parameters, which<br />
include the two-way travel time of the pulse (to give the range), the slope of the<br />
leading edge of the returned waveform (to give the significant wave height), and the<br />
backscattered power σ 0 (to relate empirically to the wind speed). These estimates<br />
are obtained by the on-satellite processing of the radar return based on the tracking<br />
algorithm of the Brown (1977) ocean statistics model (e.g., Tokmakian et al., 1994;<br />
Marth et al., 1993; Fernandes, 2003; Chelton et al., 2001). Thus, waveforms are the<br />
fundamental measurements of a radar altimeter.<br />
Although the waveform recorded in 20 Hz format is a basic measurement of the<br />
radar altimetry, the sea surface height (SSH) derived from the range and satellite<br />
altitude above an ellipsoid reference is generally used in geodetic and oceanographic<br />
investigations. Usually, the SSH is averaged to 1 Hz for a typical oceanographic<br />
application (e.g., ocean circulation) and 2 Hz (or a higher rate) for geodetic and<br />
geophysical applications (e.g., gravity anomalies) over an along-track distance per<br />
second (e.g., ~7 km for T/P). However, the 20 Hz waveform data have also found<br />
important applications both in oceanography and geophysics by applying a data postprocessing<br />
technique known as waveform retracking, which extracts the geophysical<br />
parameters from the waveform data using empirical, semi-empirical or physically<br />
based algorithms (e.g., Quartly and Srokosz, 2001; Chelton et al., 2001).<br />
Over oceans, the modified convolutional representation allows for non-linear wave<br />
parameters, such as skewness of the surface distribution, to be derived from the<br />
shape of the waveform (e.g., Barrick and Lipa, 1985; Hayne, 1980; Rodriguez, 1988;<br />
Tokmakian et al., 1994). The effects of atmospheric liquid water (both clouds and<br />
rain) on the waveform has been investigated using the waveform data over open<br />
oceans by Walsh et al. (1984), Quarlty et al. (1998), and Tournadre (1998).
Chapter 1. INTRODUCTION 5<br />
Satellite radar altimeters have principally been launched to study ocean variability<br />
and the marine geoid using measurements from repeated orbits, and map ocean<br />
gravity using measurements from geodetic missions (e.g., GEOSAT and ERS-1<br />
and 2). However, since Robin (1966, cited in Wingham et al., 1993) first recognised<br />
the potential of satellite altimetry for ice-sheet topography mapping, it has also been<br />
shown that many applications exist over non-ocean surfaces, such as digital terrain<br />
modelling and polar ice sheet topographic mapping in the field of ice dynamics<br />
(Berry et al., 1998; Bamber et al., 1997; Bamber, 1994; Brenner et al., 1997; Davis,<br />
1995; Partington et al., 1991; Remy et al., 1989; Remy and Minster, 1993; Frey and<br />
Brenner, 1990). Over non-ocean surfaces, the situation is more complex, less<br />
predictable, and the nature of return echoes varies widely as a function of the surface<br />
type. The data post processing involves range estimate refinement procedures (i.e.,<br />
waveform retracking) and a slope correction to provide the correct range for the<br />
corresponding ground location.<br />
Waveform retracking over non-ocean surfaces aims to determine the offset between<br />
the tracking gate derived from the on-satellite software and a known, fixed,<br />
instrument-independent position on the leading edge of the waveform and correct the<br />
satellite range measurement accordingly (e.g., Martin, et al., 1983; Davis, 1995;<br />
Bamber, 1994; Zwally, 1996). The slope-induced error correction aims to reduce the<br />
altimeter range between satellite and the closest point to that at nadir (e.g., Brenner et<br />
al., 1983; Cooper, 1989). Therefore, several non-linear range estimation algorithms<br />
have been developed mainly over ice since 1983, such as the β-parameter retracking<br />
algorithm (Martin et al., 1983), the threshold retracking method (Wingham et al.,<br />
1986), and the surface/volume scattering-retracking algorithm (Davis, 1993).<br />
These algorithms provide an important monitoring ‘tool’ for accurate estimates of<br />
growth or shrinkage of the ice sheets (e.g., the Greenland and Antarctic ice sheets).<br />
Wingham et al. (1998) indicate that the elevation of the Antarctic ice sheet fell by<br />
0.9±0.5 cm/year between 1992 and 1996. Information on the characteristics and<br />
changes of the firn (compacted snow), including surface roughness and volume<br />
scattering (e.g., Partington et al., 1989; Davis, 1993; Legresy and Remy, 1998), and<br />
information on the ice-sheet topography, including the surface melt streams and
Chapter 1. INTRODUCTION 6<br />
slopes (e.g., Phillips, 1998, cited in Zwally and Brenner, 2001) can also be derived<br />
from radar altimetry.<br />
Altimetry over land is even more complicated owing to rapidly varying topography.<br />
The echo waveform over land comprises reflections from an essentially static, nonhomogeneous<br />
surface. It contains various types of cover and each shows different<br />
characteristics. Some of the methods used over ice surfaces could be applied after<br />
proper modification. For example, Bamber et al. (1997) use a modified threshold<br />
method, with an optimal power threshold. Berry et al. (1998) develop an expert<br />
system to retrack waveforms over land from the ERS-1 mission by selecting one of<br />
the retrackers, which were designed to optimise the determination of the individual<br />
range correction. The ‘ice-mode’ operation of the ERS-1/2 enables the track to<br />
maintain lock over land on ~80% of surfaces (Berry, 2000a). Using a variety of<br />
different tracker algorithms, Berry (2000b) has determined land elevations, achieving<br />
a good repeatability between different passes. This has in turn revealed the<br />
inconsistencies in various digital elevation models (DEMs) obtained from<br />
conventional mapping techniques (e.g., Hilton et al., 2002).<br />
1. 2 Problems in Coastal Areas<br />
Modern satellite radar altimeters can measure the instantaneous SSH to a precision of<br />
approximately 5 cm in the open oceans (e.g., Shum et al., 1998; Chelton et al., 2001).<br />
However, in coastal regions, the waveform measurements are affected by the noisier<br />
radar returns from the (generally rougher) coastal sea states and simultaneous returns<br />
from the land (e.g., Brooks and Lockwood, 1990; Brooks et al., 1997; Nerem, 1995;<br />
Mantripp, 1996), and by less reliable geophysical, wet delay, orbit and instrument<br />
corrections (e.g., Shum, 1998; Andersen and Knudsen, 2000; Chelton et al., 2001;<br />
Fernandes et al., 2003; Quartly and Srokosz, 2001). Poorly modelled tides also affect<br />
the data owing to the shallow depth of the ocean and the irregular shape of the<br />
shoreline. This prevents the altimeter data from providing valuable information on<br />
the geoid shape, tides, the wind field, current characteristics, and dynamic the sea<br />
surface topography in coastal regions.
Chapter 1. INTRODUCTION 7<br />
1.2.1 Topographic Effects on Waveforms<br />
The shape of the radar return waveform can also be significantly affected in coastal<br />
regions by varying coastal topography, such as cliffs, shallow water tides, or the<br />
solid ground that is illuminated together with the sea surface. It is well known that<br />
the altimeter range may be estimated poorly by on-board tracking software in coastal<br />
regions (e.g., Nerem, 1995; Deng et al., 2002; Mantripp, 1996). As stated, oceanic<br />
waveform models are given under assumptions of scattering surface statistical<br />
homogeneity. However, this spatial uniformity of the topographic statistics and<br />
microwave properties of the surface cannot be met in coastal regions (Mantripp,<br />
1996; Quartly and Srokosz, 2001).<br />
In the proximity of the coastline, the altimeter simultaneously views both scattering<br />
from water and land surface (Figure 1.3). These two scattering surfaces may have<br />
different elevations, but the shortest range to the altimeter depends on the distance to<br />
the shoreline and the slope and reflectivity of the land (Brooks, et al., 1997; Brooks,<br />
2002). The nadir range between the altimeter and ocean surfaces may be the same as<br />
the slant range from land to the altimeter within the altimeter footprint. But as the<br />
satellite gets closer to the coastline, the slant range to the altimeter reflected from<br />
higher (or brighter) land may be shorter than that from the nadir oceans. Since the<br />
altimeter measures the shortest range to the reflecting surfaces, the land return within<br />
the footprint can contribute to the received power by the altimeter which then affects<br />
the returned waveform. The received power from different reflecting surfaces is<br />
dependent upon their respective backscattering coefficient and the angle at which<br />
they are being illuminated. These waveform features can be difficult to interpret<br />
using the Brown (1977) model. As a result, waveforms will be distorted by land<br />
return effects in coastal regions (Chapter 4).<br />
In addition to the land return effects, calm (usually brighter) water surfaces (both<br />
enclosed sea and inland water surfaces) near the coastline, such as bays and estuaries,<br />
make the returned power higher and narrower, thus contaminating the altimeter<br />
waveform as well. It will be shown in this research that land topography affects the<br />
waveform though the radar altimeter’s operation when crossing the coastline<br />
(Section 4.6). The altimeter changes its tracking mode when flying land-to-water or
Chapter 1. INTRODUCTION 8<br />
water-to-land, but data or signal reacquisition may take a second or more of time. As<br />
a result, land return effects on waveforms will last longer near the coastline. The<br />
effects of this performance on the waveform should also be considered when<br />
processing altimeter waveform data in coastal regions.<br />
Land<br />
Ocean<br />
Figure 1.3 Altimeter simultaneously illuminates land and ocean near the coastline<br />
The altimeter-derived SSHs are affected by the above factors. The difference<br />
between the SSH after applying all geophysical and environmental corrections and<br />
AUSGeoid98 geoid height along four ERS-2 collinear ground tracks from cycles 42<br />
and 43 is plotted in Figure 1.4. It shows that there is a large uncertainty in the<br />
altimeter-derived SSH data with respect to the geoid when approaching the coastline.<br />
The magnitude of the difference increases closer to the coastline. Since altimeterderived<br />
SSHs should provide the same geoid structure, but different noise<br />
components from the repeat observations, this large data scatter with respect to the<br />
geoid clearly shows the general problem that exists with the altimeter range data
Chapter 1.INTRODUCTION 9<br />
6<br />
5<br />
Cycle42<br />
Cycle43<br />
SSH -Geoid (m)<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0.06 3.07 6.37 9.42 11.62 14.25 16.76 19.30 22.04 24.89 27.81<br />
-1<br />
Distance from the coastline (km)<br />
15<br />
Cycle43<br />
Cycle42<br />
10<br />
SSH - Geoid (m)<br />
5<br />
0<br />
0.19 3.40 6.60 9.82 12.76 15.56 18.47 21.48 24.57 27.70<br />
-5<br />
-10<br />
Distance from the coastline (km)<br />
6<br />
5<br />
Cycle42<br />
Cycle43<br />
SSH - Geoid (m)<br />
4<br />
3<br />
2<br />
1<br />
0<br />
10.60 13.89 16.25 17.65 19.51 21.71 24.15 26.77 29.53<br />
Distance from the coastline (km)<br />
15<br />
10<br />
Cycle42<br />
Cycle43<br />
SSH - Geoid (m)<br />
5<br />
0<br />
0.07 3.17 6.38 9.64 12.38 15.22 18.16 21.20 24.30 27.45<br />
-5<br />
-10<br />
Distance from the coastline (km)<br />
Figure 1.4 Differences between the ERS-2 SSH and AUSGeiod98 geoid height along<br />
four collinear repeat tracks from cycles 42 and 43 at the Southern Australian coast,<br />
showing the metre-level uncertainty of SSHs closer to the coastline.
Chapter 1.INTRODUCTION 10<br />
from the on-satellite tracking algorithm in coastal regions. Therefore, waveforms can<br />
be contaminated by the land returns and sea states in coastal regions. This is the<br />
definition of the term contamination used in this dissertation.<br />
1.2.2 Topographic Effects on Geophysical Corrections<br />
In the vicinity of the coastline, land topography also affects the modelling of some of<br />
the geophysical corrections applied to the range measurement (e.g., Anderson and<br />
Knudsen, 2000; Shum et al., 1998; Fernandes et al., 2003). Of all corrections, the<br />
effects of wet tropospheric delay and the ocean tides on the altimeter data are most<br />
critical (e.g., Anzenhofer et al., 2000; Chelton, 2001). The result is that the accuracy<br />
of the altimeter-derived SSH will also degrade with the degradation of these<br />
geophysical corrections.<br />
For ERS-1/2 altimeters, the wet tropospheric corrections are derived from the actual<br />
measurements made by the onboard microwave radiometer (e.g., Anzenhofer et al.,<br />
2000; Fernandes et al., 2003). The corrections contained in the waveform data<br />
products are also calculated using an estimate of the total integrated water vapour<br />
and surface air temperature at the sub-satellite point using data supplied by the global<br />
forecast analysis in the European Centre for Medium-Range Weather Forecasts<br />
(ECMWF) (NRSC, 1995). However, the radiometer does not work correctly near the<br />
land and cannot switch on or off when the satellite flies across the coastline, causing<br />
some useful data to be unavailable and errors in the range measurements in coastal<br />
regions (e.g., Fernandes et al., 2003; Anzenhofer et al., 2000).<br />
Due to the spatial complexity of tides in shallow water, the global ocean tide models<br />
are still not accurate enough over shallow seas for detailed oceanographic studies.<br />
The error resulting from using global ocean tidal models can be at the decimetre<br />
level in coastal regions (Anzenhofer et al., 2000; Shum et al., 1998). For both<br />
ERS-1/2 altimeter waveform data and geophysical data records (GDRs), the ocean<br />
tide corrections are provided in a format of 1 Hz and based on the Schwiderski (1983,<br />
cited in NRSC, 1995) model rather than on the recently improved tidal models<br />
(NRSC, 1995). Since the ERS-1/2 satellite orbit is sun-synchronous and the choice of<br />
35 days for its main repeat period leads to severe aliasing of the most important tidal<br />
constituents, the use of the Schwiderski ocean tidal model is appropriate for the
Chapter 1.INTRODUCTION 11<br />
ERS-1/2 data. The problem is that the model cannot provide an adequate resolution<br />
for shallow water tidal constituents. This is because tides over shallow waters<br />
become highly complicated and vary rapidly in amplitude and phase. In these areas,<br />
tides are highly dependent on the bathymetry, the shape of the continental shelf, and<br />
the regional tidal regime (Andersen, 1999). In addition, shorter wavelength tidal<br />
features near sharply topographical changes are not represented in the current<br />
dynamical and empirical global tidal models (Tierney et al., 1998).<br />
Although relatively less research has been reported, the sea state bias (SSB)<br />
correction, which is a function of the significant wave height (SWH), is<br />
unfortunately affected by the land topography near the coastline. Land affecting the<br />
SWH and thus the SSB correction for GEOSAT altimeter has been reported by<br />
Fernandes et al. (2003). It will be shown in Chapter 7 that the erroneous SSB<br />
corrections have also been found for ERS-1 range measurements in the Taiwan Strait<br />
approaching the coastline.<br />
1. 3 Justifications for This Study<br />
1.3.1 Previous Work<br />
As mentioned above, improved determination of the SSH data through waveform<br />
retracking in coastal regions is essential because the coastal topography and the sea<br />
states contaminate the altimeter waveform (cf. Section 3.2.1). However, the potential<br />
of the altimeter waveform data in coastal regions has been the subject of<br />
comparatively little study. So far, considering the existing retracking algorithms used<br />
for coastal waveforms from the open literature, they are those developed for dealing<br />
with the waveform data over ice-sheet surfaces (cf., Anzenhofer et al., 2000; Brooks<br />
et al., 1997).<br />
One earlier study by Brooks and Lockwood (1990) indicates that the presence of land<br />
within a radar altimeter’s footprint affects the altimeter’s tracking performance and<br />
thus contaminates the altimeter waveform data. They analysed GEOSAT waveform<br />
data over the Tuamotu Archipelago in the southern Pacific Ocean to determine the<br />
contaminated distance (1.0-6.2 km) from the islands. Their research pointed out that<br />
the island can locally affect the SSH by more than ±4 m. An edit criterion based on a
Chapter 1.INTRODUCTION 12<br />
combination of waveform data, automatic gain control (AGC) and SWH was<br />
suggested, but waveform retracking was not involved. Brooks et al. (1997) retracked<br />
TOPEX waveforms along 18 satellite ground passes in Pacific Rim coastal zones<br />
using the threshold retracking technique (cf. Sections 3.5.2 and 3.7). Anzenhofer et al.<br />
(2000) present a retracking system in coastal regions, which consists of only a single<br />
retracker and a fitting algorithm is employed. The five-β-parameter fitting function<br />
with either a linear trailing edge or exponential decay trailing edge (cf. Section 3.5.1)<br />
is used by Anzenhofer et al. (ibid.) to retrack ERS-1 waveforms.<br />
Using only a single retracker, however, limits the precision of the recovered SSH<br />
data because of various levels of complexity of the waveform data in coastal regions<br />
(see Sections 5.8.2 and 6.4). Therefore, developing an alternative approach to<br />
improving the precision of recovered SSH data in these areas is necessary, which<br />
must be able to cope with diverse waveforms and estimate precise SSHs from these<br />
reprocessed data. This is one of the justifications for this dissertation.<br />
1.3.2 Significance of the Improvement of Altimeter Data in Coastal Regions<br />
Over open oceans, satellite altimetry is one of the most effective techniques that<br />
provides global and accurate homogeneous coverage of the shape of the sea surface.<br />
The shape of the sea surface is the only physical variable measured from space that is<br />
directly and simply connected to large scale movement of water and the total mass<br />
and volume of the ocean. In coastal regions, although there are generally other data<br />
sources available, such as tide gauges, airborne and shipborne gravity data sets,<br />
altimetry is still a significant technique that is able to provide reliable and uniform<br />
surface information. Since the launch of the SEASAT satellite, there has been a<br />
demand for combining the uniform altimetric coverage of data with other data<br />
sources in order to investigate the oceanic and climatic environment. As stated,<br />
however, the precision of the altimeter data is degraded because of the coastal<br />
topography (both land and sea).<br />
As new satellite missions, such as GFO-1, ENVISAT, JASON-1, and CRYOSAT<br />
will contribute more to the existing altimeter data sets of SEASAT, GEOSAT,<br />
ERS-1/2 and T/P, applications of altimetry in coastal regions will continue to grow.<br />
The coastal areas represent eight percent of the ocean surface. Most importantly,
Chapter 1.INTRODUCTION 13<br />
economic and engineering activities are mainly concentrated in these areas, such as<br />
fisheries and offshore oil drilling (Menard et al., 2000). Altimeter data are also<br />
important for understanding coastal tides, interactions between coastal currents and<br />
open ocean circulation, mesoscale and coastal variability, geoid modelling, and<br />
marine gravity anomaly determination. However, this could only be achieved by<br />
specialised improvement of altimeter measurements (both the data themselves and<br />
the corrections) in coastal regions and accurate data sampling through merging, in an<br />
optimal way, different altimeter data sets such as T/P and ERS-1/2, or JASON-1 and<br />
ENVISAT in the future.<br />
After declassification of the GEOSAT geodetic mission, cross-track resolution of the<br />
satellite altimeter profiles has been improved considerably. In particular, the crosstrack<br />
resolution is improved further because of other satellite altimeter missions,<br />
especially the ERS geodetic missions. Consequently, the cross-track resolution has a<br />
trend exceeding the along-track resolution (~7 km). ERS and GEOSAT geodetic<br />
missions, for instance, have cross track spacings of ~8 km and ~5 km at the equator,<br />
respectively (Chelton et al., 2001). Thus, it becomes necessary to keep and improve<br />
20 Hz data for refining the along-track resolution not only in open oceans, but also in<br />
coastal regions.<br />
When compared ERS-1 with GEOSAT altimeters, the ERS-1 range is poorly tracked<br />
by the altimeter. This is because ERS-1 was not a geodetic satellite and had its<br />
geodetic mission added at the end of its life. The prime use of its data was to monitor<br />
the trailing edge of the waveform and then to monitor the air-sea interactions. The<br />
number of the gates along the leading edge is usually 3-4 sample bins. This makes<br />
difficult for the on-satellite tracking algorithm to precisely define the waveform<br />
shape and thus the range to the surface. Instead, GEOSAT has twice the number of<br />
gates along the leading edge to estimate the range more precisely (Fairhead et al.,<br />
2001). Fairhead et al. (2001) conclude that the ERS-1 ranges have been poorly<br />
estimated by the on-satellite tracking algorithm.<br />
In addition, ocean surface waves are a fundamental limitation to the recovery of the<br />
gravity field from altimeter SSH profiles. For instance, to recover the gravity field at<br />
an accuracy of 1 mgal at 20-km full wavelength requires the ocean surface height
Chapter 1.INTRODUCTION 14<br />
change over a 10-km horizontal distance to a precision of 1 cm (Sandwell, 2003;<br />
Fairhead et al., 2001). However, standard on-board data processing of altimeter data<br />
provides a SSH precision of only ~5 cm at this length scale (Chelton et al., 2001;<br />
Sandwell, 2003). The results from Fairhead et al. (2001) show that reprocessing<br />
ERS-1 waveforms significantly reduces the along-track noise particularly in the 50-<br />
10 km wavelength range, thus improving spatial resolution. This also suggests the<br />
improvement of geodetic parameter estimation from waveform retracking is<br />
necessary over both coastal regions and open oceans.<br />
The present theory of altimetry, developed to describe scattering from the ocean<br />
surface or to retrack the return waveform from the non-ocean surface, does not deal<br />
properly with the waveforms in coastal areas. Data post-processing over topographic<br />
surfaces of ice and land have been attended to since 1983 (e.g., Martin et al., 1983;<br />
Partington et al., 1989; Wingham et al., 1986; Davis, 1995; Zwally, 1996; Berry,<br />
2000a), but there is still a comparable lack of post-processing in the coastal areas<br />
(see Section 1.3.1 and Chapter 3). Therefore, in order to fully realise the potential of<br />
the altimetry in coastal areas, it is necessary to improve the altimeter measurements<br />
and obtain a better understanding of the SSH estimates derived from satellite radar<br />
altimetry. Accurate measurements in coastal regions would quantify one important<br />
source of the SSH and thus enable better understanding of the various geophysical<br />
and oceanographical features in coastal regions. This requires more sophisticated<br />
correction models and data processing techniques than in the open oceans and on<br />
land.<br />
Finally, the International Association of Geodesy (IAG) has created a special study<br />
group (3.186) for studying altimetry data processing for gravity, geoid and sea<br />
surface topography determination (http://space.cv.nctu.edu.tw/IAG/main.html). The<br />
subject of this study has been listed as one of the research problems in the group. The<br />
Australian Research Council International Researcher Exchange Scheme (grant<br />
number: X00001267) also supported a project related to this study. Therefore, the<br />
results will contribute significantly not only to altimeter applications, but also to the<br />
altimetric data processing theory and practice. Essentially, satellite altimeter tracking<br />
will realise its potential over the whole Earth’s surface that is covered by the
Chapter 1.INTRODUCTION 15<br />
altimeter orbits, rather than ignoring eight percent of the Earth’s surface in coastal<br />
regions.<br />
1. 4 The Study Areas<br />
The study areas used to demonstrate the retracking system are the Australian coastal<br />
regions and the Taiwan Strait. They are associated with different coastal<br />
characteristics.<br />
As well as for geographical convenience to the author, Australia is chosen for this<br />
study for the following reasons. Firstly, it is surrounded by part of three large,<br />
interconnected ocean basins of the Southern Hemisphere: the Pacific, Indian and<br />
Southern Oceans. Secondly, the Australian coastline (including Tasmania), is<br />
approximately 36,700 kilometres in length (Zann, 1990, cited in Resource<br />
Assessment Commission, 1993), which is one of the longest coastlines in the world<br />
for a single country. Finally, Australia’s coastal zone exhibits the following broad<br />
range of topographical features: coastal plains, hills and mountains, beaches, cliffs,<br />
estuaries, oceans, coral reefs, and islands. Thus, it is assumed that the performance of<br />
altimetry around Australia is reasonably representative of what can be expected in<br />
other parts of the world.<br />
Although the Australian coast provides an almost perfect area for the investigation of<br />
waveform retracking, it is not an area for the application of recovery of gravity<br />
anomalies. In 1992, the Australian Geological Survey Organisation (AGSO, now<br />
Geoscience Australia) released a gravity database that includes denser ship-track<br />
gravity observations around Australia bound by 108°E ≤ λ ≤ 162°E and<br />
8°S ≤ ϕ ≤48°S. However, these AGSO marine gravity data were not provided in a<br />
format such that the crossover adjustment can be conducted to reduce some serious<br />
biases among overlapping tracks (cf. Featherstone, 2003). They were also not<br />
crossover adjusted before supply (ibid.). Therefore, they cannot offer any definitive<br />
indication on the improvements of SSH data through the accuracy of the satellite<br />
altimeter-derived gravity anomalies in the Australian coastal region.<br />
Instead, another study area chosen is the narrow Taiwan Strait (roughly ~110 km<br />
wide), which links the East China Sea to the South China Sea. This area is chosen
Chapter 1.INTRODUCTION 16<br />
because firstly there are some on-going research activities, detailed and reliable shiptrack<br />
gravity anomalies, and bathymetric data available in the area (e.g., Hwang and<br />
Wang, 2002). Secondly, it is a shallow water region with high tidal amplitudes,<br />
which allows the evaluation of the retracking system in a different coastal area.<br />
Thirdly, ~2.5 years of ERS-1 waveform data including all phases provide denser<br />
measurement coverage over the area. This will benefit the recovery of the gravity<br />
field there.<br />
1. 5 Thesis Outline<br />
Chapters 1 and 2 introduce the background of this dissertation and the geodetic<br />
principles of satellite altimetry. The altimeter data products, waveform data sets and<br />
some external input data sources used in this study will be summarised and described<br />
briefly in Chapter 2. In Chapter 3, the existing algorithms developed to retrack<br />
altimeter waveforms over ocean surfaces and non-ocean surfaces will be reviewed<br />
and discussed in the context of coastal retracking. The ocean model, which will be<br />
used in this study, will be deduced, and a detailed qualitative analysis of the model<br />
will be carried out.<br />
Chapter 4 quantifies the contaminated distance around Australian coast using 20 Hz<br />
waveform data of ERS-2 (one cycle, 35-day repeated orbit) and POSEIDON (five<br />
cycles, 10-day repeated orbit). Several examples from ERS-2 waveforms along<br />
satellite ground tracks near the coastline are analysed in detail to understand how<br />
waveforms are contaminated and what the characteristics of contaminated<br />
waveforms are at the Australian coast (Section 4.6). A preliminary test that estimates<br />
the land effects on waveforms in the vicinity of land will be given in Section 4.7.<br />
Chapter 5 develops a coastal retracking system that is based upon a detailed preanalysis<br />
of the coastal waveforms given in Chapter 4. The retracking algorithms used<br />
will be the fitting and threshold methods. Issues involving the fitting algorithms are<br />
fitting functions of an ocean and β-parameter models, the least squares iterative<br />
procedure, linearisation of the parameters, determination of the initial estimates of<br />
parameters, and the weight scheme. A method will be introduced that uses iterative<br />
weights to detect the outliers in the waveforms ensuring the effective convergence of<br />
the fitting procedure. For the threshold retracking algorithm, a method will be
Chapter 1.INTRODUCTION 17<br />
described that selects an appropriate threshold level for coastal waveforms. Finally, a<br />
coastal retracking system will be developed, which makes it possible to improve the<br />
SSH precision from reprocessing the altimeter waveforms, correcting the range<br />
measurements, and extracting the precise SSH data from corrected ranges.<br />
In Chapter 6, the coastal retracking system developed in Chapter 5 is applied to the<br />
Australian coast (0-30 km from the coastline) using two cycles (42 and 43, March to<br />
May 1999) of ERS-2 20 Hz waveform data. Results of waveform classification and<br />
waveform retracking will be presented and discussed in Sections 6.2 and 6.3,<br />
respectively. The internal validation of the quality of the retracked altimeter data will<br />
be conducted by analysis of the single track SSH data in Section 6.4, comparisons<br />
between SSH data before and after retracking in Section 6.5, and a collinear analysis<br />
in Section 6.6. Using an external (partly) independent reference of the AUSGeoid98<br />
geoid model, the results of retracking and comparisons with the unretracked and<br />
external control data will be discussed in Sections 6.4 and 6.7 to demonstrate the<br />
improvements from waveform retracking.<br />
Chapter 7 investigates the improved accuracy of the altimeter-derived coastal marine<br />
gravity field from SSH data after waveform retracking in the coastal region of the<br />
Taiwan Strait. Using ~2.5 years of ERS-1 20 Hz waveform data (
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 18<br />
2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES<br />
2. 1 Introduction<br />
Satellite altimetry employs a nadir-pointing, high-resolution radar to measure the<br />
range to the ocean’s surface with an accuracy of a few centimetres. The range to the<br />
at-nadir sea surface is measured by the altimeter directly through a measurement of<br />
the time difference between the transmission of a radar pulse and the reception of the<br />
echo. If the altitude of the satellite above a reference ellipsoid is known, the SSH<br />
above the reference ellipsoid (Figure 2.1), which is the measurement expected from<br />
the satellite altimetry and corresponds to the geophysical characteristics of the sea<br />
surfaces, can be calculated by subtracting the range from the satellite altitude. In<br />
order to precisely estimate the SSH, it is necessary to know the principle of the<br />
altimeter’s operation, estimates of the range and then the SSH, and the error sources<br />
for the determination of the SSH. These will be described next.<br />
Figure 2.1 Basic measurement principle of the altimeter-derived SSH (From<br />
http://www.jason.oceanobs.com/html/portail/galerie<br />
/banque_img_welcome_uk.php3).
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 19<br />
In addition, Section 2. 3 will briefly summarise the waveform data sets used in this<br />
study. Some external input data sources will be described in Section 2. 4.<br />
2. 2 General Information on Construction of Altimeter Measurements<br />
The altimeter range measurement allows the determination of the SSH related to a<br />
reference ellipsoid and hence oceanographic, geophysical and geodetic applications,<br />
such as the determination of gravity anomalies, mesoscale eddies, Rossby waves and<br />
the surface currents under an assumption of geostrophy (Chelton et al., 2001). Unlike<br />
synthetic aperture radar, radar altimeters do not form an image observation, but make<br />
point measurements at or close to the nadir of the satellite as it follows its orbit. The<br />
coverage of the ground tracks is built up over time as the orbit arcs cover the Earth’s<br />
surface. In addition, the shape of the return pulse has been found to give information<br />
on SWH, while the backscattered power has been used to estimate the surface wind<br />
speed (ibid.).<br />
2.2.1 Fundamental Radar Measurement Principles<br />
The altimeter-observed time delay can be converted into a range to the at-nadir sea<br />
surface as long as the propagation velocity is known as<br />
ct<br />
R = (2.1)<br />
2<br />
where R is the range, c ≈ 3×10 8 m s -1 is the free-space speed of light and t is the<br />
two-way travel time of the radar pulse. The range resolution is given by<br />
c<br />
∆ R ≥<br />
τ<br />
(2.2)<br />
2<br />
where τ is the pulse length. The resolution of one discrete measurement of range<br />
depends on the altimeter’s operating parameters (Table 2.1). For instance, the<br />
resolution is approximately 0.45 m for the ERS-1/2 altimeter operating in an ‘ocean<br />
mode’ with a pulse length τ = 3.03 ns. In order to measure dynamic ocean signals,
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 20<br />
Table 2.1 Altimeter Operating Parameters<br />
Altimeter<br />
Launch<br />
date<br />
Altitude<br />
(km)<br />
Inclination<br />
(deg)<br />
Beamwidth<br />
(deg)<br />
Frequency<br />
(GHz)<br />
PRF<br />
(Hz)<br />
No.<br />
of<br />
gate<br />
s<br />
Tracking<br />
gate<br />
Sampling time<br />
(ns)<br />
Waveform<br />
frequency<br />
(Hz)<br />
No of<br />
waveforms<br />
in average<br />
Seasat 27/06/1978 800 108 1.6 13.5 1020 60 30.5 3.125 10 100<br />
Geosat 12/03/1985 800 108 2.1 13.5 1020 60 30.5 3.125 10 100<br />
ERS-1 17/07/1991 784 98 1.3 13.8 1020 64 32.5 3.03 (ocean mode) 20 50<br />
12.12 (ice mode)<br />
ERS-2 21/04/1995 784 98 1.3 13.8 1020 64 32.5 3.03 (ocean mode) 20 50<br />
12.12 (ice mode)<br />
TOPEX (Ku) 10/08/1992 1334 66 1.0 13.6 4500 128 32.5 3.125 10 2×228<br />
TOPEX (C) 10/08/1992 1334 66 2.7 5.3 1200 128 35.5 3.125 5 4×60<br />
POSEIDON 10/08/1992 1334 66 1.1 13.65 1700 60 29.5 3.125 20 86<br />
GFO 10/02/1998 800 108 1.6 13.5 1020 128 32.5 3.125 10 100<br />
Jason-1 (Ku) 07/12/2001 1334 66 1.3 13.6 1800 104 32.5 3.125 20 90<br />
Jason-1 (C) 07/12/2001 1334 66 3.4 5.3 300 104 32.5 3.125 20 15<br />
Envisat (Ku) 01/03/2002 784 98 1.3 13.6 1800 128 3.125 18 100<br />
Envisat (S) 01/03/2002 784 98 5.6 3.2 450 64 6.25 18 25
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 21<br />
the range and timing precisions are required to a few centimetres (e.g., ~4 cm) and<br />
less than 0.2 ns, respectively (Mantripp, 1996; Chelton et al., 2001). This is achieved<br />
by averaging the estimates from several individual pulses (cf. the rightmost column<br />
in Table 2.1).<br />
2.2.2 The Returned Waveform<br />
A schematic description of the interaction of a pulse-limited radar altimeter with a<br />
diffuse, horizontal and planar sea surface is illustrated in Figure 2.2. Initially<br />
( 0 t < t0<br />
< ), a pulse of electromagnetic energy is transmitted from an on-satellite<br />
altimeter antenna, propagating in a spherical wavefront. The area of pulse that can be<br />
received on ground is defined by antenna beam-width ( θ<br />
A<br />
). When the wavefront<br />
encounters the nearest crests of those ocean waves directly beneath the satellite (i.e.,<br />
at nadir) at t = t0<br />
, it illuminates a point and a reflected signal begins to return to the<br />
altimeter. As time and the pulse progress, the wavefront reaches the surface at points<br />
further from the nadir point. This increases the area of interaction between pulse and<br />
surface, and the illuminated point spreads out rapidly to form a disc (assuming a<br />
locally flat surface). Properly reflected facets within this disc scatter energy back to<br />
the altimeter during the period of t<br />
0<br />
< t < t1<br />
(Figure 2.2). Once the rear of the pulse<br />
reaches the lowest trough at nadir, the region of interaction between the surface and<br />
pulse forms an annular ring of increasing diameter, narrowing width, and the<br />
constant mean area. The backscattered energy reaches its maximum at the point of<br />
transition to an annular ring (i.e., t = t ). Thereafter ( t > t<br />
1<br />
1<br />
), the backscattered energy<br />
begins to decay owing to the finite antenna beam width and the fewer proper<br />
reflected facets available at larger off-nadir angles. The returned power is recorded<br />
over the duration of the pulse, building up a returned waveform (see the bottom in<br />
Figure 2.2) with a rapidly rising leading edge and long decay of the trailing edge.<br />
In order to reduce the fading noise arising from coherency among individual<br />
waveforms, returned waveforms are actually an average of a number of pulses, 50 in<br />
the case of ERS-1/2 (Mantripp, 1996; Quartly et al., 2001), forming theoretically a<br />
average return altimeter waveform over ocean surface shown in Figure 1.1.
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 22<br />
SIDE VIEW<br />
transmitted<br />
pulse<br />
R<br />
θ A<br />
PLF<br />
PLAN VIEW<br />
(Illuminated area)_<br />
t 0<br />
t 1<br />
t 2<br />
RETURN WAVEFORM<br />
leading edge<br />
trailing edge<br />
average return<br />
typical return<br />
t 0 t 1 t 2<br />
time<br />
Figure 2.2 A schematic geometrical description of the interaction of a pulse and the<br />
scattering surface, and the build up of a return waveform over the duration of a pulse.<br />
Note that, an average returned waveform from 50 typical individual returns is<br />
depicted, while a single typical return shows more noise. (PLF is the pulse-limited<br />
footprint).<br />
This averaged returned waveform (Figure 1.1) is a time series of the mean returned<br />
power recorded by a satellite altimeter, and is referred to as the convolution of three<br />
terms in the time domain (e.g., Brown, 1977; Hayne, 1980; Chelton et al., 2001).<br />
Over open oceans, it can be described by the Brown (1977) model (see Chapter 3). A<br />
modelled or ideal average return altimeter waveform over ocean surface is shown in<br />
Figure 1.1. It consists mainly of three parts:<br />
(a)<br />
The thermal noise contains the thermal noise power generated by the altimeter<br />
prior to the first return of a signal from the scattering surfaces. Its effect is to<br />
add a constant power level to the return waveform. It is to be noted that the
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 23<br />
(b)<br />
(c)<br />
observed waveform of ERS-1/2 does not show the thermal noise because of the<br />
operating feature of the altimeter (see Section 4.4).<br />
The leading edge contains the return power from the scattering surfaces within<br />
the pulse-limited footprint (see below for explanation), which involves in the<br />
information about the SWH and range between the satellite altimeter and the<br />
mean sea surface at the nadir.<br />
The trailing edge consists of the return power from the scattering surface<br />
outside the pulse-limited footprint, which can be approximated by a straight<br />
line whose slope depends on the altimeter antenna pattern and the off-nadir<br />
angle (see Chapter 5).<br />
To date, all spaceborne radar altimeters use the pulse-limited (i.e., only the earliest<br />
returns reflected from nadir are recorded) operation to measure the range to the<br />
surface. The pulse-limited operation itself defines one of the diameters of the<br />
measurement footprint (Martin et al., 1983; Quartly et al., 2001). The footprint is the<br />
area of the reflecting surface which is illuminated by the altimeter, and its size<br />
depends on the satellite altitude, the width of the range window and the features of<br />
the reflecting surface (see Section 4.2.1). The pulse-limited footprint is the maximum<br />
circular area from which backscatted power can be simultaneously received (Brooks<br />
et al., 1978). The spreading of the pulse over the pulse-limited footprint corresponds<br />
to the risetime of the leading edge of the return waveform, while the range to the<br />
nadir sea surface measured by the altimeter corresponds to the midpoint at the<br />
leading edge of the waveform.<br />
In practice, a pulse compression technique (Marth et al., 1993; Chelton et al., 2001)<br />
is used to obtain a high signal-to-noise ratio and an acceptable demand on the<br />
satellite’s power system. The radar altimeter emits long-duration chirps, which are<br />
linearly frequency-modulated pulses, then the received return signal is dechirped and<br />
mixed with a deramping chirp. This mixed signal is then filtered and digitised in the<br />
frequency domain where there is a one-to-one correspondence between frequency<br />
and two-way travel time. Individual waveform samples result finally from a fast<br />
Fourier transform (FFT) procedure. The resultant signal is a function of the<br />
difference frequency, and its properties depend not only on the particular part of the<br />
nadir scattering surface, but also the characteristics of the filter frequency used in the
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 24<br />
chirp generation and the deramping processing. In this case, the power spectral<br />
density estimates at discrete frequencies (64 bins for the ERS and POSEIDON) are<br />
obtained and have a similar shape to that shown in Figure 1.1, with only the<br />
difference for the two domains that the abscissa is frequency or two-way travel time<br />
(Chelton et al., 2001). The further details on the actual implementation can be found<br />
in Marth et al. (1993), Mantripp (1996), and Chelton et al. (2001).<br />
Radar altimeters operate as closed-loop regulators in both the time-delay (range<br />
tracking) and radiometric (AGC) domains (Marth et al., 1993; Jensen, 1999). Thus,<br />
the on-satellite radar processor attempts to hold the reference position of the<br />
altimeter waveform at a fixed position and with fixed amplitude so that successive<br />
waveforms may be integrated or averaged without any loss of measurement<br />
resolution. This is conducted by adjusting the timing of the data sampling and by the<br />
attenuators within the receiver (ibid.). As a result, the processor must constantly<br />
measure the varied range to the surface and reflectivity of the scattering surface. It<br />
will be found in Section 5.4.1 (Figure 5.3) that the waveform’s leading edges are<br />
centred on the midpoint, and this remains true regardless of the surface roughness. It<br />
is this stable behaviour that makes the AGC and range tracking algorithm work.<br />
The range to the surface can vary from several metres over ocean surfaces to several<br />
kilometres over non-ocean surfaces within the diameter of the footprint (Mantripp,<br />
1996; Chelton et al., 2001). However, in the vertical dimension the altimeter can<br />
receive the returns only within a ‘range window’ specified by the width of the<br />
frequency spectrum. For instance, it is ~28 m and ~112 m for ERS-1/2 in the ocean<br />
mode (1 range bin equals ~0.45 m) and ice mode (1 range bin equals ~1.82 m),<br />
respectively, and ~30 m for POSEIDON over oceans (1 range bin equals ~0.47 m).<br />
The onboard tracker continuously adjusts the range window to keep the leading edge<br />
of the waveform at a specified position at the centre of the range window. This<br />
position is known as the ‘tracking gate’ (or tracking point), which is designed prior to<br />
the launch of the satellite and is the range bins 32.5 for ERS-1/2 and 31.5 for<br />
POSEIDON. When the surface elevation varies too rapidly for the tracker to respond,<br />
the altimeter is said to have ‘lost lock’. Usually, the altimeter cannot track surfaces<br />
where the mean slope is greater than about one degree (Barrick and Lipa, 1985;<br />
Mantripp, 1996).
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 25<br />
2.2.3 The Range and Sea Surface Height<br />
If the altimeter tracker and AGC circuitry worked perfectly, the position of the<br />
tracking gate would always be centred halfway up the return waveform leading edge<br />
at the tracking gate. Therefore, the tracking gate would itself be an accurate measure<br />
of the two-way travel time. In order to turn this measurement into the range from the<br />
electromagnetic mean sea surface, and, ultimately, a surface elevation (e.g., SSH),<br />
several corrections need to be applied to the data during ground processing. These<br />
include the instrumental and satellite corrections (e.g., software processing delays<br />
and Doppler shift), propagation corrections (e.g., dry and wet tropospheric delay and<br />
ionospheric delay), and geophysical corrections (e.g., tides and inverse barometer<br />
effect). Table 2.2 lists some corrections applied and the correction error to ERS-1<br />
range measurements (cf. Cudlip et al., 1994).<br />
Table 2.2 Corrections applied to ERS-1 range measurements and their likely error.<br />
Error Source Correction magnitude Error<br />
Doppler shift ±33 cm ±0.3 cm<br />
Centre of gravity offset 0-90 cm ±0.1 cm<br />
Ionospheric delay 0-100 cm ±5 cm<br />
Tropospheric delay-dry 1.7-2.5 m ±1 cm<br />
Tropospheric delay-wet 0-50 cm ±5 cm<br />
Ocean Tide ±50 cm a ±10 cm<br />
Earth Tide ±30 cm ±1 cm<br />
Ocean Loading Tide ±10 cm
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 26<br />
measurements, the SSH can be estimated by subtracting the corrected range from the<br />
altitude.<br />
2. 3 The Altimeter Waveform Data Used<br />
Three types of waveform data products from ERS-1, ERS-2, and POSEIDON have<br />
been used in this study. They are provided in the specific data format themselves by<br />
the European Space Agency (ESA) for ERS-1/2 and the Archiving, Validation and<br />
Interpretation of Satellite Oceanographic (AVISO) team for POSEIDON.<br />
2.3.1 ERS-1 Waveform Data<br />
The European Remote Sensing satellite ERS-1 was launched on 17 July 1991 by<br />
ESA. The ERS-1 measurements effectively consist of a combination of several<br />
different missions (Table 2.3).<br />
Phase<br />
Table 2.3 The repeat cycle characteristics of ERS-1.<br />
Repeat cycle<br />
(days)<br />
Number of orbits<br />
per cycle<br />
Ground-track<br />
spacing at the<br />
equator (km)<br />
ice 3 43 931<br />
Multi-discipline 35 501 80<br />
Geodetic 168 2411 16<br />
The ERS-1 Ku-band radar altimeter (RA) is one of the remote sensors carried onboard<br />
the satellite. It emits 1020 pulses per second and the on-board processor<br />
averages the returns in groups of 50 to produce 20 Hz waveforms (e.g., Quartly et al.,<br />
2001). The ERS-1 radar altimeter waveform is provided in a set of power signals<br />
with respect to time at 64 sample bins (cf. Table 2.1 and Figure 1.2). It is noted that<br />
ERS-1 geodetic mission contains two cycles of measurements and its second cycle of<br />
orbit drifts away 8 km from the first cycle of orbit on the equator, thus making a<br />
denser data coverage for geodetic applications.<br />
ERS-1 altimeter waveform products (ALT.WAP) are supplied by the United<br />
Kingdom Processing and Archiving Facility (UK-PAF) on behalf of ESA. The
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 27<br />
ALT.WAP product contains all the information telemetered in the altimeter source<br />
packets together with the corrections, calibration and orbit data required to further<br />
process the data into the GDRs. The data are structured following the CEOS format,<br />
which are unformatted, binary, direct access and with fixed record lengths (NRSC,<br />
1995).<br />
Because of the common format of the CEOS for all on-board remote sensing data<br />
(Mansley, 1996, cited in Anzenhofer et al., 2000), the original ALT.WAP data in the<br />
CEOS format are compressed to remove the un-used information for the altimeter,<br />
generating compressed waveform products in the AIDA waveform format (AWF).<br />
The intermediate ERS-1 waveform written in AWF contains only relevant<br />
information to produce the altimeter GDRs, leading a data reduction from 1 GByte to<br />
about 550 MByte for each 3-day data set (cf., Anzenhofer et al., 2000). They are<br />
stored on the CD-ROM and contain 1 Hz of geophysical and environmental<br />
0<br />
corrections, and 20 Hz of observing time, range, location, orbit altitude, σ (radar<br />
backscattering cross section per unit area), SWH, and waveform data in 64 sample<br />
bins. The exact details of the AWF format can be found in, e.g., Anzenhofer et al.<br />
(2000).<br />
Nearly 2.5 years of ERS-1 waveform data are used in this research, which include<br />
the data from all three phases. The data are supplied by the Ohio State <strong>University</strong>,<br />
USA, but they were originally provided by UK-PAF. Unfortunately, the waveform<br />
data used in this study are not provided in a format such that the orbit number can be<br />
obtained to identify individual ground tracks. This presents a problem for averaging<br />
repeated SSH to reduce the time variability and data noise when computing gravity<br />
fields described later in Section 7.5.2. Therefore, a case-specific averaging method<br />
will be developed to circumvent this problem.<br />
2.3.2 ERS-2 Waveform Data<br />
The ERS-2 satellite was launched on 21 April 1995 into a high inclination orbit<br />
(98.5°), which provides a near-global coverage. Its mission consists of a combination<br />
of the multi-discipline and tandem phases with both 35-day orbits. ERS-2 operates in<br />
two tracking modes: ocean mode and ice mode. The on-satellite tracking algorithm
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 28<br />
used for the ocean mode is known as the Sub-optimal Maximum Likelihood<br />
Estimator (SMLE), while the off-centre of gravity (OCOG) algorithm is used by the<br />
ice mode (Scott et al., 1994).<br />
The ERS-2 waveform data used in this study are taken from CD-ROMs supplied by<br />
De Montfort <strong>University</strong>, U.K., which are the ALT.WAP products supplied originally<br />
by UK-PAF on behalf of ESA. The original data were read, compressed and stored<br />
on the CD-ROM at De Montfort <strong>University</strong>. Each ERS-2 waveform record consists<br />
of 1 Hz and 20 Hz data. Geophysical and environmental corrections are supplied at 1<br />
Hz, and observations, such as the range, waveform longitude and latitude, altitude,<br />
σ<br />
0<br />
, and 64 waveform samples, are supplied at 20 Hz. Two cycles of waveform data<br />
(35-day repeat orbit, March to May 1999) are used in the study.<br />
A problem found in the ERS-2 waveform data edited by De Montfort <strong>University</strong> is<br />
that most of the Doppler range corrections exceed the minimum (-50 cm) and<br />
maximum (50 cm) acceptable values (cf., NRSC, 1995). Therefore, this correction is<br />
not applied to the altimeter range measurement. As stated, the returned signal is<br />
analysed in the frequency domain. Thus, any change in the frequency of the returns<br />
causes an error in the estimation of the range to the at-nadir sea surface. The Doppler<br />
shift changes the relative velocity between the altimeter and the sea surface, thus<br />
causing an error in the range estimate. The range correction for the Doppler shift is<br />
typically ±33 cm for ERS data (cf., Mantripp, 1996). Without applying this<br />
correction will cause the error to the range measurement, which is ~±0.3 cm from<br />
Table 2.1. However, comparing with the required range accuracy of a few<br />
centimetres (e.g., ~4 cm), the effect of this correction is small and can be neglected.<br />
2.3.3 POSEIDON Waveform Data<br />
The POSEIDON altimeter is a solid-state radar altimeter carried by the<br />
TOPEX/POSEIDON (T/P) satellite, which was launched on 10 August 1992 into a<br />
66-degree inclination orbit. It emits pulses at 1700 Hz and averages the returns in<br />
groups of 86 every 53 ms. The waveform is sampled in 64 bins of 3.125 ns width,<br />
with the tracking gate between bin 31 and 32. Waveform samples at the first two and<br />
last two (aliased) bins have been removed from the data products provided by the
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 29<br />
AVISO team. Since TOPEX and POSEIDON altimeters share an antenna on board<br />
the satellite, the POSEIDON altimeter is operated for about 10 percent of the mission<br />
time. The POSEIDON data used in this study comprise five cycles (January 1998 to<br />
January 1999) of 20 Hz waveform data (ten-day repeat orbit) taken from the CD-<br />
ROMs supplied by AVISO.<br />
Table 2.4 The reference ellipsoid characteristics of ERS-1/2 and T/P.<br />
Items ERS-1/2 T/P<br />
Equatorial radius (m) 6378137 6378136.3<br />
Flattening 1/298.257223563 1/298.257<br />
The POSEIDON waveform product is structured as the classical T/P GDRs in terms<br />
of cycle, pass files, headers and format. Each file is a fixed-length unformatted<br />
record and contains a header. All file headers are ASCII and all other data are VAX<br />
binary integers (AVISO/Altimeter, 1998). It consists of ten-day repeat cycles of data,<br />
which were separated into ascending and descending passes. A POSEIDON<br />
waveform record includes 1 Hz elementary data records of a pass-file and 20 Hz data<br />
of the waveform measurements (ibid.). The data are referenced to different ellipsoids<br />
(Table 2.4), so it is necessary to convert the data to a consistent reference (e.g.,<br />
GRS80) if they are used together.<br />
2. 4 External Input Data<br />
The external input data sets used in this research for providing necessary information<br />
for data editing and the result valuation include:<br />
(1) GSHHS (0.2 km resolution) shoreline model (Wessel and Smith, 1996)<br />
– provide information of the coastline’s location;<br />
(2) The DS759.2 (5′×5′ resolution) ocean depth model (Dunbar, 2000) and the<br />
Australian bathymetric model (30" resolution) (Buchanan, 1991) – identify<br />
whether radar returns reflect from land or water, and for result analysis;<br />
(3) Australian DEM (9"×9" resolution, version 2) (Hutchinson et al., 2001)<br />
– provide land elevations above local mean sea level;
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 30<br />
(4) A water/land grid (0.5′×0.5′ resolution) derived from the Generic Mapping<br />
Tools (GMT) high-resolution shoreline data (Wessel, 2000) – combined with<br />
the bathymetric models to flag data accurately;<br />
(5) AUSGeoid98 geoid grid (2′×2′ resolution) (Featherstone et al., 2001) – provide<br />
“ground truth” for validating the algorithms;<br />
(6) Sea surface topography (SST) model (Levitus et al., 2002) – provide the SST<br />
correction for estimating the gravity anomalies; and<br />
(7) Marine gravity data around Taiwan (Hwang and Wang, 2002) – provide the<br />
ground reference for the comparison between the gravity anomalies derived<br />
from SSHs before and after retracking.<br />
Some of these data will be described in detail below.<br />
2.4.1 The AUSGeoid98 Geoid Model of Australia<br />
The principal objective of the retracking system is to estimate the tracking error<br />
caused by the coastal waveform contamination and correct the altimeter range<br />
measurements accordingly. The analysis of the system discussed in Chapters 5, 6 and<br />
7 has arisen because of the necessity of knowing how good the retracking system is.<br />
To assess the retracking system, objective and independent ground references must<br />
be used. In this study, the AUSGeoid98 gravimetric geoid model of Australia<br />
(Featherstone et al., 2001) will be used as one of these independent ground<br />
references. The altimeter-derived SSH before and after retracking will be compared<br />
statistically with the geoid heights interpolated at the observation location from<br />
AUSGeoid98 (Chapter 5 and 6). This model is chosen just because it is the most<br />
recent and precise model readily available around Australia coastal regions at the<br />
time of this study. It is acknowledged that other reference sources of local geoid<br />
models (e.g., AUSGeoid91 and AUSGeoid93 models) or tide gauge data exist.<br />
However, the AUSGeoid98 includes denser and more types of data than other global<br />
or local models and uses arguably improved computational techniques (cf.,<br />
Featherstone et al., 2001), thus precisely representing the geoid features in Australia.<br />
The AUSGeoid98 geoid model was computed using data from the EGM96 global<br />
geopotential model, the 1996 release of the Australian gravity database (both over
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 31<br />
land and in marine areas), a national-wide digital elevation model (DEM), and<br />
satellite altimeter-derived marine gravity anomalies. The geoid height has a<br />
resolution of two minutes in longitude and latitude with respect to the GRS80<br />
ellipsoid. Comparisons of AUSGeoid98 with GPS and Australian Height Datum<br />
(AHD) heights across the continent indicate an RMS agreement of ±0.364 m (ibid.).<br />
It is also acknowledged that the altimeter-derived gravity anomalies, the global two<br />
minute grid of Sandwell et al. (1995), were used in the computation of the<br />
AUSGeoid98 model. However, the ship-track gravity data are also combined with<br />
the altimeter-derived gravity data using the ‘draping’ technique (cf., Kirby and<br />
Forsberg, 1997; Featherstone et al., 2001). This technique involves least squares<br />
collocation, as well as a 200 km high-pass filter. The satellite gravity grid is first<br />
carried out by this filter and the difference between the filtered altimeter grid and the<br />
ship-track gravity anomalies is calculated. Next, a difference grid is generated using<br />
least squares collocation. Then, this grid is added back to the filtered altimeter grid<br />
where no ship-track gravity data existed. Thus, because of the use of the ‘draping’<br />
technique, the ship-track gravity data contribute mainly to the geoid signal in areas<br />
which are close to the Australian coast. Therefore, the AUSGeoid98 can still be a<br />
quasi-independent reference for the purpose of assessing the SSH data before and<br />
after retracking.<br />
The problem that must be noted is that many ship-track gravity data used in the<br />
AUSGeoid98 model have not been cross-over analysed and adjusted (Featherstone et<br />
al., 2001; Featherstone, 2002). Some serious biases among overlapping tracks exist.<br />
Therefore, these ship-track gravity data should be used with some caution. This is<br />
also the reason that the marine gravity data around Taiwan are used in this <strong>thesis</strong> (cf.<br />
Section 2.4.3).<br />
2.4.2 Sea Surface Topography<br />
In order to estimate the gravity anomalies and geoid height from altimeter-derived<br />
SSH, both the time-dependent and time-independent (or explicitly quasi-timeindependent)<br />
sea surface topography (SST) must be removed (Hwang et al., 2002b;<br />
Chelton et al., 2001). The SST is the dynamic sea surface height associated with<br />
geostrophic surface currents (Chelton et al., 2001). The long-term mean SST from a
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 32<br />
reference surface can be computed from a compilation of ship observations of<br />
temperature and salinity profiles (ibid.). For instance, a 1°×1° mean SST grid related<br />
to 3000 dB computed from 100 years of ship observations of temperature and salinity<br />
profiles and smoothed by Levitus and Boyer (1994, cited in Chelton et al., 2001) can<br />
be found in Chelton et al. (Figure 52, p.88, 2001). Another Levitus SST version<br />
(1997) is provided by NOAA with the monthly averages on a 1°×1° grid for 12<br />
months (cf, Hwang et al., 2002b), which is related to a reference sea surface of<br />
1000 m depth level (Levitus et al., 2002). In this study, the quasi-time-independent<br />
SST values adopted is from a 1°×1° grid averaged from these 12 month SST values<br />
by Hwang et al. (2002b). Figure 2.3 shows the averaged Levitus (1997) SST, which<br />
appears to be the similar profile to the version of Levitus (1994) SST, but the<br />
different values due to the different references (cf, Chelton et al., Figure 52, p.88,<br />
2001).<br />
This SST grid (Figure 2.3) will be used to interpolate the required SST values at the<br />
altimeter data point around the Taiwan Strait in Chapter 7. It is applied because of<br />
the existence of the Kuroshio Current passing east of Taiwan and flowing along the<br />
eastern boundary of the East China Sea near the research area (118°E ≤ λ ≤ 123°E<br />
and 22°N ≤ ϕ ≤ 27°N). The Kuroshio Current is a strong western boundary surface<br />
current (Le Traon and Morrow, 2001). It causes the larger SSH gradients (>1 mm/km<br />
from Figure 2.3), so that the effect of SST cannot be neglected in the area around<br />
Taiwan (Hwang, 2003).<br />
However, this SST grid will not be applied to the altimeter data at the Australian<br />
coast in this study. Firstly, from the SST data released by Levitus (1997) to create<br />
this quasi-time-independent mean SST (Figure 2.3), there were no data sources near<br />
the Australian coast. The SST values at this area are based on an interpolated<br />
procedure from a 1°×1° grid (Hwang, 2003), thus causing additional errors to the<br />
altimeter-derived SSH there. Secondly, the ocean currents near the Australian coast,<br />
such as the Leeuwin Current, are weak when compared with the Kuroshio Current<br />
(Le Traon and Morrow, 2001; Hwang, 2003), and do not cause large SSH gradients<br />
(~0.2 mm/km). Thus, the SST effects on the SSH can be ignored at this area.
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 33<br />
Figure 2.3 Quasi-time-independent SST (contour interval is 10 cm) from Levitus et<br />
al. (1997).<br />
Finally, it is known that the SSH can be expressed as (Chelton et al., 2001)<br />
SSH h + h +<br />
=<br />
g d<br />
C + ε<br />
(2.3)<br />
where h g is the geoid height related to the reference ellipsoid, h d is the SST height, C<br />
is the range correction, and ε is the measurement error. However, it must be noted<br />
that the absolute SST values cannot be obtained from an existing SST model,<br />
because all existing SST versions are related to some reference surface (cf, Chelton<br />
et al., 2001; Levitus et al., 1997). From Equation (2.3), if the altimeter-derived SSH<br />
is used as an absolute measurement relative to the reference ellipsoid, nonuniqueness<br />
will be caused by applying the SSTs from different references.<br />
Due to these reasons, the SST was not applied to the altimeter-derived SSH data at<br />
the Australian coast. The comparison between the SSH and the AUSGeoid98 geoid<br />
height focuses on the observation of their profile variation rather than their difference.<br />
The standard deviation of the difference between them will be as an indicator of the<br />
SSH accuracy rather than the mean difference of them (Chapters 5 and 6).
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 34<br />
2.4.3 Marine Gravity Data around Taiwan<br />
To assess the accuracy of gravity anomalies derived from retracked satellite altimeter<br />
waveform data, the ship-track gravity data provided by Hwang and Wang (2002a)<br />
around Taiwan are used in this study (Figure 2.4). These marine gravity data contain<br />
observations conducted from more than 50 marine geophysical cruises during the<br />
period from 1965 to 1996. They are obtained from the gravity databases at Oxford<br />
<strong>University</strong>, the National Geophysical Data Centre data set of the National Oceanic<br />
and Atmospheric Administration (NOAA), and several overseas universities and<br />
agencies (Hsu et al., 1998; Hwang and Wang, 2002a).<br />
Figure 2.4 Coverage of ship-track gravity observations around Taiwan<br />
Some basic data cleaning procedures had been applied to these ship-borne gravity<br />
data, such as removing observations during turning the ship and crossover<br />
adjustment (c.f. Hsu et al., 1998; Hwang and Wang, 2002a). Most gravity data near<br />
Taiwan were provided by five geophysical cruises conducted in 1996 and GPS<br />
navigation was used to obtain high-quality positions and Eötvös corrections. Other
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 35<br />
ship-borne gravity data show lower precision because the Transit navigation system<br />
was used. Therefore, the five cruises were used as control data and others were<br />
adjusted with respect to them. The mean difference and standard deviation of the<br />
crossover points of total shipboard gravimetric data are 6.3 mgal and ±11.2 mgal (see<br />
Hsu et al., 1998). The free-air gravity anomalies were recomputed on the GRS80<br />
ellipsoid.<br />
Through the use of GPS positioning techniques and these careful data adjustments,<br />
the gravity data around Taiwan form a ‘ground control’ on the accuracy of the<br />
satellite altimeter-derived gravity anomalies. Gravity anomalies at 397 points along<br />
the ship tracks are used in comparisons (Chapter 7), which are taken in the area<br />
118°E ≤ λ ≤ 122°E and 23°N ≤ ϕ ≤ 26°N west of Taiwan to be consistent with the<br />
altimeter data (Figure 7.1).<br />
2.4.4 The Australian Digital Elevation Model (DEM)<br />
The shape of the waveforms in coastal regions have been found to differ from those<br />
recorded over open oceans from the previous studies (e.g., Brooks, 1997) and this<br />
study (Chapters 4 and 6). One of the causes of this waveform contamination is that<br />
the waveform may be mixed with the land and ocean returns when the altimeter<br />
simultaneously illuminates land and water. The DEM can provide information about<br />
the terrain elevation and topography on land within the altimeter footprint near the<br />
coastal zone, and thus be used for analysis of the land effects on the waveforms<br />
(Section 4.7).<br />
A DEM is a representation of the terrain using averaged elevation information. The<br />
recent upgraded (version 2) 9 second Australian DEM is a grid of elevation points<br />
covering the whole of Australia with a grid spacing of 9 seconds in longitude and<br />
latitude (approximately 250 m) in the Geocentric Datum of Australia 1994 (GDA94)<br />
horizontal reference frame (Hutchinson et al., 2001). The source data used to create<br />
the 9 second Australian DEM consists of:<br />
(1) revised data sets (spot heights, linear watercourse features, Australian coastline,<br />
and coastal inlets) from Australian GEODATA digital topography products
Chapter 2. SATELLITE ALTIMETER MEASUREMENTS AND DATA SOURCES 36<br />
(including TOPO-250K Relief theme, Drainage layer of TOPO-250K, and<br />
COAST-100K Coastline and State Borders);<br />
(2) trigonometric data points (converted to the GDA94 coordinate system) from<br />
the National Geodetic Data Base;<br />
(3) radar altimeter point elevation data (300 observations) for Lake Eyre; and<br />
(4) additional data sets (spot heights, stream line data, sink point data, cliff line<br />
data and associated contour line data) from digital 1:100 000 scale mapping.<br />
They are supplied by National Mapping Division of Geoscience Australia<br />
(formerly AUSLIG) (ibid.).<br />
Errors in the DEM depend mainly on the resolution of the DEM and the roughness of<br />
the terrain (i.e., the slope). Theoretical estimates and tests of the 9 second DEM<br />
against trigonometric data distributed evenly across the continent indicate that the<br />
standard elevation error of the DEM varies from ~±7.5 m to ~±20 m for most of the<br />
continent (ibid.). Errors are larger in highland areas with steep and complex terrain,<br />
where the largest errors can exceed 200 m (ibid.).<br />
2. 5 Summary<br />
This Chapter has introduced the basic fundamental working principle of a radar<br />
altimeter, the altimeter range measurements and the required sea surface information<br />
(i.e., SSH), and data products. Some external input data sources used in this study,<br />
such as the AUSGeoid98 geoid model, the Australian 9" DEM and the ship-track<br />
gravity around Taiwan, were also described in this Chapter.
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 37<br />
3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS<br />
3. 1 Introduction<br />
Radar altimeter return waveforms are the basic measurement for observing<br />
geophysical parameters of the Earth surfaces, in particular, the ocean. They give the<br />
range between the satellite and the mean sea surface at nadir (via pulse transit time),<br />
the SWH (via return waveform shape characteristics), and the roughness of the sea<br />
surface (via received energy, from which wind speed is empirically inferred).<br />
Retracking is a procedure of waveform data post-processing to improve parameter<br />
estimates over those given by GDRs. These parameters contain range corrections due<br />
to the estimation algorithm and the limited computational time on-board the satellite,<br />
and unmodelled ocean surface effects, such as the surface skewness. Waveform<br />
retracking has different emphases for different applications depending on the<br />
reflecting surfaces.<br />
Over ocean surfaces, altimeter retracking algorithms aim to reduce the following<br />
errors: short-wavelength random noise, and more importantly, potential longwavelength<br />
errors caused by SWH biases, range biases, the altimeter antennamispointing<br />
angle, or unmodelled non-Gaussian ocean surface parameters, such as<br />
the surface skewness and kurtosis (e.g., Hayne and Hancock III, 1990; Hayne et al.,<br />
1994; Hayne, 1980; Rodriguez, 1988; Rodriguez and Martin, 1994b). Over nonocean<br />
surfaces, the on-board satellite altimeter tracker cannot follow the sharp<br />
change of the surface topography (i.e., land/ice), causing an obvious departure<br />
between the midpoint of the leading edge of the waveform and the predesigned<br />
altimeter tracking gate. This leads to an error in the telemetered range measurements<br />
made by the altimeter. Waveform retracking, therefore, aims to produce accurate<br />
surface elevation measurements over ice sheets and land (e.g., Martin et al., 1983;<br />
Ridley and Partington, 1988; Davis, 1993a; Berry et al., 1998). The algorithms<br />
developed to retrack altimeter waveforms over ocean surfaces and non-ocean<br />
surfaces will be discussed in Sections 3.4, 3.5 and 3.6.<br />
In coastal regions, waveforms are contaminated by the land topography (see<br />
Chapters 1, 2 and 4), which exhibits a variety of terrains including the coastline,
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 38<br />
mountains, islands, lowlands, cliffs, reefs, and estuaries, as well as the sea states,<br />
such as standing and very calm water. To the author’s knowledge, all previously<br />
published retracking efforts involve retracking coastal waveforms using existing<br />
algorithms developed for ice sheets (e.g., Anzenhofer et al., 2000; Brook et al., 1997).<br />
A description of these methods will be given in Section 3.7.<br />
3. 2 Convolutional Representation of Ocean-Returned Waveforms<br />
The shape of the radar return waveform represents the observations measured by a<br />
satellite radar altimeter as a function of two-way travel time, which is related to the<br />
scattering characteristics of the reflecting surface. Moor and Williams (1957, cited in<br />
Brown, 1977) first proposed the convolutional representation of the return waveform.<br />
They demonstrated that for a ‘rough’ (i.e., incoherent) scattering surface, the average<br />
return power (as a function of delay time) can be expressed as a convolution of the<br />
transmitted power waveform envelope with a quantity involving<br />
0<br />
σ (radar<br />
backscattering cross section per unit scattering area), the antenna gain, and the range<br />
from the altimeter to any point on the Earth’s surface. Furthermore, when<br />
considering the vertical distribution of the surface height and radar receiver effects,<br />
the average return power (as a function of the delay time) is a convolution of three<br />
terms: the smooth spherical surface, the probability density function of the height of<br />
the specular points, and the radar system’s point target response (e.g., Brown, 1977;<br />
Hayne, 1980; Rodriguez, 1988).<br />
3.2.1 The Brown Model<br />
The mathematical model used to derive the average altimeter returned waveform<br />
from incoherent surface scattering has been based upon physical optics theory,<br />
whereby the surface is treated as a set of specular facets with a given height and<br />
slope probability density distribution. The time-series of the mean returned power<br />
waveform P (t)<br />
measured by a satellite altimeter is expressed as a convolution of<br />
three terms in the time domain as (Brown, 1977)<br />
P( t)<br />
= P ( t)<br />
∗ q ( t)<br />
∗ P ( t)<br />
(3.1)<br />
fs<br />
s<br />
PTR
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 39<br />
where t is the time measured at the satellite receiver such that t = 0 corresponds to<br />
the range to the mean sea level at nadir, P fs<br />
(t)<br />
is the average impulse response from<br />
a flat surface, q s<br />
(t)<br />
is the surface elevation probability density function (PDF) of<br />
specular points within the altimeter footprint, and P PTR<br />
(t)<br />
is the radar system point<br />
target response (PTR). Each of the three terms in Equation (3.1) will be presented<br />
separately later in Sections 3.2.2, 3.2.3 and 3.2.4.<br />
In order to give an analytical representation of the convolution in Equation (3.1),<br />
Brown (1977) first presented a simplified expression of the terms for near-normal<br />
incidence under the following assumptions, which are common to all satellite radar<br />
altimeter systems. The basic assumptions that are inherent in the convolution model<br />
of near-normal incidence rough surface backscatter (Barrick, 1972; Brown, 1977;<br />
Fedor et al., 1979) are as follows:<br />
(1) The root-mean-square (RMS) elevation variations are greater than the radar<br />
wavelength;<br />
(2) The ocean surface is statistically homogeneous and stationary over the number<br />
of pulses needed to determine a mean waveform;<br />
(3) The radar cross section is constant over the illuminated area;<br />
(4) There is no backscatter of the electromagnetic energy from beneath the surface;<br />
(5) The illuminated surface always contains a statistically significant number of<br />
specular points;<br />
(6) The surface elevation and surface slope density distributions are statistically<br />
independent; and<br />
(7) Multiple scattering among various parts of the surface is neglected.<br />
These assumptions are generally valid for an altimeter operating over open ocean<br />
surfaces. However, they are not generally the case over non-ocean surfaces, and<br />
hence an alternative approach must be sought (e.g., Mantripp, 1996). In coastal<br />
regions, areas of land can be simultaneously illuminated with water, thus causing<br />
land returns to the altimeter. Therefore, it is not always the case that the above<br />
assumptions are valid in the coastal zone (ibid.).
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 40<br />
3.2.2 Average Flat Surface Impulse Response Function<br />
Using a Gaussian approximation of the antenna gain, Brown (1977) presents P fs<br />
(t)<br />
over the illuminated area of the Earth’s surface as<br />
where<br />
P fs<br />
1/ 2<br />
( t)<br />
= Aexp(<br />
−α ' t)<br />
I ( β ' t ) U ( )<br />
(3.2)<br />
0<br />
t<br />
4c<br />
α'<br />
= cos(2ξ<br />
)<br />
(3.3)<br />
γ h<br />
a<br />
4<br />
β ' =<br />
γ<br />
a<br />
⎛ c<br />
⎜<br />
⎝ h<br />
⎟<br />
⎠<br />
⎞<br />
1/ 2<br />
sin(2ξ<br />
)<br />
(3.4)<br />
sin 2 θ<br />
γ<br />
a<br />
=<br />
ln 4<br />
(3.5)<br />
2<br />
In Equation (3.2), U (t)<br />
is a unit step function, I ( t<br />
1/ ')<br />
is the modified Bessel<br />
0<br />
β<br />
function (see Section 3.3.2), A is an amplitude scaling term containing the off-nadir<br />
pointing angle ξ and several other constants, such as the radar wavelength, the ocean<br />
reflectivity and the radar antenna gain (Brown, 1977). In Equations (3.3) and (3.4),<br />
h is the satellite altitude above the reference ellipsoid surface, and γ<br />
a<br />
is an antenna<br />
beamwidth parameter defined by Brown (1977). In Equation (3.5), θ is the antenna<br />
beamwidth.<br />
The derivation of Equations (3.3) and (3.4) assumes that the altimeter antenna pattern<br />
is approximately a Gaussian function, otherwise the closed form of Equation (3.2)<br />
cannot be derived from an integration of P fs<br />
(t)<br />
(see Brown, 1977, Equation 2).<br />
According to Brown (1977), this approximation is generally valid out to the point on<br />
the antenna pattern for which there is no appreciable contribution to the<br />
backscattered power.<br />
In Equation (3.2), A can be divided into two parts as<br />
A = A0<br />
Aξ<br />
, where 0<br />
A is an<br />
amplitude scaling term involving only constants of the radar parameters and the<br />
ocean reflectivity, and A ξ<br />
is a term related to the off-nadir pointing angle ξ . They<br />
can be written as
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 41<br />
where G<br />
0<br />
, λ and<br />
A<br />
0<br />
2 2 0<br />
0 λ cσ<br />
( 0)<br />
2 3<br />
( 4π<br />
) L h<br />
G<br />
= (3.6)<br />
4<br />
p<br />
⎛ 4 ⎞<br />
= ⎜ −<br />
2<br />
A<br />
ξ<br />
exp sin ξ ⎟<br />
(3.7)<br />
⎝ γ ⎠<br />
L<br />
p<br />
are radar system parameters (Brown, 1977).<br />
3.2.3 Surface Elevation PTR of the Scattering Distribution<br />
The instantaneous illumination pattern of the altimeter depends on the specific wave<br />
height field at time t within the footprint. Because of the random nature of the sea<br />
surface elevation distribution, it is more instructive to consider the PTR of specular<br />
points for the determination of the waveform. As a first-order approximation, the<br />
distribution of the sea surface elevation is Gaussian (Chelton et al., 2001). The scale<br />
of the wave-height distribution is described by the SWH, which is defined to be the<br />
average crest-to-trough height of the 1/3 highest waves. Brown (1977) uses a<br />
Gaussian specular point PDF to describe the statistics of the linear surface elevations.<br />
It is given as<br />
q s<br />
( ζ )<br />
2<br />
1 ⎛ ζ ⎞<br />
exp<br />
⎜ −<br />
⎟<br />
1 / 2<br />
(3.8)<br />
(2π<br />
) σ<br />
s ⎝ 2σ<br />
⎠<br />
= s<br />
where ζ is the surface elevation (positive upward) above the mean level of the<br />
specular points, σ s<br />
is the RMS surface elevation of the specular points related to<br />
SWH by SWH = 4(2 / c)<br />
σ . The PDF in Equation (3.8) is expressed in the space<br />
s<br />
domain in distance units (metres), while the ocean surface elevation density function<br />
q s<br />
(t) is written in the altimeter’s time domain in Equation (3.1). To coincide with<br />
other two terms in Equation (3.1), a variable<br />
t = −2ζ / c must be used to convert<br />
q s(ζ ) from the space domain to the two-way range time domain in time units<br />
(nanoseconds). The relevant standard deviation can be written as σ = 2σ c .<br />
t s<br />
/
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 42<br />
3.2.4 Radar System PTR<br />
The radar system PTR is primarily the transmitted radar pulse shape. For an idealised,<br />
linear, frequency-modulated altimeter radar pulse, the radar system PTR is given by<br />
(Ulaby et al., 1982, cited in Rodriguez, 1988)<br />
P PTR<br />
[(<br />
at / 2)( T − t )]<br />
2<br />
sin<br />
( t)<br />
− T ≤ t ≤ T<br />
(3.9)<br />
=<br />
2<br />
( at / 2)<br />
where T is the radar transmitted pulse duration (or the compressed pulse length if<br />
pulse compression is employed) and a is a constant which depends on the radar<br />
bandwidth. Thus, P PTR<br />
(t)<br />
is a symmetric expression of the impulse response about<br />
t = 0 . However, using the PTR expressed by Equation (3.9), it is not possible, in<br />
general, to obtain the analytic function of Equation (3.1). Since the width of the PTR<br />
of the short pulse radar altimeters is of the order of 20 ns (e.g., Brown, 1977),<br />
Equation (3.9) can be simplified to a Gaussian function as<br />
P<br />
PTR<br />
2<br />
⎡ ⎤<br />
⎢<br />
1 ⎛ ⎞<br />
⎜<br />
t<br />
( t)<br />
≈ η − ⎟ ⎥<br />
p<br />
PT<br />
exp<br />
⎢ 2<br />
(3.10)<br />
⎥<br />
⎣ ⎝σ<br />
p ⎠ ⎦<br />
where η<br />
P<br />
is the pulse compression ratio, P T<br />
is the peak transmitted power, and<br />
is a measure of the pulse width<br />
σ<br />
P<br />
σ<br />
P<br />
= 0. 425T<br />
(3.11)<br />
in which 0.425 is an approximation of the compressed pulse shape by a Gaussian<br />
function (Marth et al., 1993).<br />
3.2.5 Solution of the Radar Returns<br />
Using Equations (3.2), (3.8) and (3.10), Brown (1977) shows that the convolution of<br />
Equation (3.1) can be reduced to the following expression<br />
P(<br />
t)<br />
≈ P<br />
fs<br />
( t)<br />
∞ ∞<br />
∫∫<br />
0 −∞<br />
⎛ c ⎞<br />
⎜ ⎟P<br />
⎝ 2 ⎠<br />
PTR<br />
⎛ c<br />
( t −τ<br />
) qs⎜<br />
⎝ 2<br />
( τ − ˆ τ ) dτ<br />
d ˆ τ<br />
⎟ ⎠<br />
⎞<br />
(3.12)
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 43<br />
1 ⎡ ⎛ t ⎞ ⎤<br />
≈ ηP<br />
PT<br />
Pfs(<br />
t)<br />
σ<br />
P ⎢erf<br />
⎜ ⎟ + 1<br />
2<br />
⎥<br />
⎣ ⎝ 2σ<br />
⎠ ⎦<br />
where erf (x)<br />
is the error function, and σ is the composite risetime given by<br />
2 2 ⎛ 2<br />
= σ p<br />
+ ⎜ s<br />
2<br />
σ ⎞<br />
⎟<br />
⎝ c σ<br />
⎠<br />
(3.13)<br />
In Equation (3.12), the first term, P fs<br />
(t)<br />
, includes the effects of the antenna<br />
beamwidth and the off-nadir pointing angle. The flat surface emphasises that an<br />
incoherent surface scattering process is assumed. The second term, P PTR<br />
(t)<br />
, contains<br />
the effects of the receiver bandwidth on the transmitted pulse. The third term, q s<br />
(t)<br />
,<br />
contains the sea-state effects. It is assumed that waves on the sea surface are linear<br />
(i.e., an incoherent surface) and therefore the corresponding statistics of surface<br />
elevations and slopes can be expressed by a Gaussian function.<br />
Equation (3.12) is the fundamental model of the returned waveform over ocean<br />
surfaces, named the Brown model. It is important because it clearly shows that the<br />
simplified expression of the altimeter return waveform can be applied to extract<br />
geophysical parameters over oceans from the observed return waveform. However, it<br />
should be noted that some effects are neglected in the Brown model, such as the<br />
curvature of the Earth’s surface, approximation of the Bessel function, and the<br />
nonlinearity of the actual surface waves. These remain an open issue, which has been<br />
addressed by other researchers, such as Rodriguez (1988), Hayne (1980), Challenor<br />
and Srokosz (1989), and Rodriguez and Martin (1994a), and will be summarised in<br />
the following Sections.<br />
3. 3 Modified Convolutional Representation of the Ocean Return Waveform<br />
The Brown model has been modified, e.g., by Hayne (1980) and Rodriguez (1988),<br />
to present general expressions for the radar mean return waveform from non-<br />
Gaussian surfaces. This has allowed more parameters to be estimated from the radar<br />
return, such as the skewness of the sea surface (e.g., Lipa and Barrick, 1981;<br />
Rodriguez, 1988; Rodriguez and Martin, 1994b; Challenor and Srokosz, 1989). This<br />
Section will summarise the modifications made to the Brown model by these authors.
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 44<br />
Then, an analytic expression for the radar ocean return will be derived from some of<br />
these modified expressions for the purposes of this dissertation.<br />
3.3.1 Effects of the Earth’s Curvature<br />
The effects of the Earth’s curvature are neglected in Equations (3.3) and (3.4), but<br />
the off-nadir angle for a given pulse delay time t is slightly smaller than that in the<br />
flat-Earth approximation. Considering its effect, Rodriguez (1988) shows that the<br />
corrected functions α and β related to Equations (3.3) and (3.4) are<br />
4c<br />
1<br />
α = cos(2ξ<br />
)<br />
(3.14)<br />
γ h (1 + h / R)<br />
4 ⎛<br />
β = ⎜<br />
γ ⎝<br />
c<br />
h<br />
1 ⎞<br />
⎟<br />
(1 + h / R)<br />
⎠<br />
1/ 2<br />
sin(2ξ<br />
)<br />
(3.15)<br />
where R ≈ 6371005 m is the spherical radius of the Earth (e.g., GRS80; Morita,<br />
1980). The flat-Earth approximation in Equations (3.3) and (3.4) is not a realistic<br />
case. For the ERS-2 and TOPEX orbit heights of about 785 km and 1336 km,<br />
respectively, this flat-Earth approximation leads to a 5.8% and 9.1% reduction of the<br />
illuminated area of the flat-Earth surface. Thus, the average radar impulse response<br />
from a smooth spherical surface is given by (Rodriguez, 1988)<br />
P s<br />
1/ 2<br />
( t)<br />
= Aexp(<br />
−α t)<br />
I ( β t ) U ( )<br />
(3.16)<br />
0<br />
t<br />
Equation (3.16) includes effects due to the Earth’s curvature which, in general,<br />
cannot be ignored (Hayne and Hancock III, 1990).<br />
3.3.2 Simplification of the Bessel Function<br />
The convolution in Equation (3.1) does not allow for easy analytical integration<br />
because of the presence of the modified Bessel function (Equation 3.2). It can be<br />
seen from Equations (3.2) and (3.4) that the off-nadir angle ξ and return time t are<br />
two variables in the Bessel function. The off-nadir angle is ideally equal to zero, but<br />
in practice the antenna does not always point along the nadir and the maximum<br />
possible pointing error is 0.5 degrees (e.g., Challenor and Srokosz, 1989; Brown,
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 45<br />
1977). In addition, the return time is also a small value compared to the leading edge<br />
rise time (Rodriguez, 1988). For such small off-nadir angles and the case of shortpulse<br />
radar altimeters, Hayne (1980) expands I 0<br />
in the small-argument (assumed<br />
convergent) series expansion (Jeffrey, 1995):<br />
n<br />
2<br />
2<br />
1<br />
0<br />
( ) ∑ ∞ ⎛ z ⎞ ⎛ ⎞<br />
I z =<br />
⎜ ⎜ ⎟<br />
= 0 4<br />
⎟<br />
(3.17)<br />
n ⎝ ⎠ ⎝ n!<br />
⎠<br />
which allows for the term-by-term integration in the convolution model in Equation<br />
(3.1) (Hayne, 1980). Rodriguez (1988) expands the Bessel function presented in the<br />
smooth surface impulse response using Equation (3.17), but keeps only terms of<br />
order<br />
2<br />
β t and lower. Then, ( 1/ 2<br />
)<br />
I β is replaced by exp( 2 t / 4)<br />
0<br />
t<br />
β to give a simpler<br />
expression for analytical integration. Thus, considering the curvature of the Earth and<br />
the above approximation for I<br />
0<br />
, Equation (3.2) can be expressed as the average radar<br />
impulse response from a smooth sphere (Rodriguez, 1988):<br />
P s<br />
2<br />
⎡ ⎛ β ⎞ ⎤<br />
( t)<br />
≈ Aexp⎢−<br />
⎜α − t⎥U<br />
( t)<br />
⎣ 4<br />
⎟<br />
(3.18)<br />
⎝ ⎠ ⎦<br />
The error caused by this approximation was analysed by Rodriguez (1988) based<br />
upon various off-nadir angles and return times, for SEASAT, GEOSAT and TOPEX.<br />
The results show that the error is less than one percent for all times when the offnadir<br />
angles is approximately less than half the altimeter beamwidth (see Table 2.1)<br />
(Rodriguez, 1988).<br />
The antenna off-nadir angle is also called the off-nadir pointing error, which<br />
represents the displacement of the boresight axis of the altimeter antenna from the<br />
nadir. Because of peak power limitations and the high operating altitude, spaceborne<br />
radar altimeters generally use high gain antennas. A direct consequence of the high<br />
gain requirement is a narrow beamwidth, typically less than 3° (Brown, 1977). With<br />
current satellite attitude control systems, the beamwidth is 1.34° for ERS-1/2 and<br />
1.1° for TOPEX (Ku-band, see Table 2.1). Thus, the off-nadir angle approximated as<br />
half the antenna beamwidth cannot be larger than 1°. In addition, it is highly unlikely<br />
that the pointing error could exceed 1° without causing loss of tracking (cf. Barrick
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 46<br />
and Lipa, 1985). Therefore, Equation (3.18) is a reasonable approximation based on<br />
the typical magnitude of the off-nadir angle.<br />
3.3.3 Approximate Expression of the Non-Gaussian Scattering Surface<br />
As stated, the shape of the returned waveform is determined by the distribution of<br />
specular surface scatterers rather than by the actual sea surface height distribution<br />
within the footprint. The differences between the distributions of the scatterers and<br />
the sea surface height lead to biases between the altimeter-measured scattering<br />
surface and the actual mean sea surface, which must be corrected for the altimeter<br />
range measurements (Chelton et al., 2001; Rodriguez, 1988). The range to the sea<br />
surface is estimated from the half-power point of the leading edge of the waveform,<br />
which corresponds to the return from the median height of the specular scatterers<br />
referred to as the electromagnetic (EM) sea level (ibid.).<br />
The difference between the true mean sea level and the mean EM surface in the<br />
footprint is called the sea state bias (SSB), which comprises the electromagnetic bias<br />
(EMB) and the skewness bias (SKB). Owing to the non-Gaussian nature of the sea<br />
surface, the wave troughs are brighter than the peaks for radar wavelengths, arising<br />
in a greater backscattered power per unit surface area from wave troughs than from<br />
wave crests. The result is that the mean EM surface measured by altimeters is biased<br />
lower than the true mean sea surface toward wave troughs. This effect is known as<br />
the EMB. From previous studies (cf. Chelton et al., 2001), the EMB has been<br />
observed to increase monotonically with increasing the SWH. The SKB is the error<br />
in tracking the mean height of the specular scatterers, which is the height difference<br />
between the mean scattering surface and the median scattering surface (ibid.). The<br />
SKB is generally proportional to the SWH, but it is not related in any simple manner<br />
to the SWH or any other geophysical quantity that can be inferred from altimeter<br />
data.<br />
The specular point PDF q s<br />
(t)<br />
in Equation (3.1) represents the probability density<br />
distribution of the specular scatterers within the altimeter footprint. It thus includes<br />
the effects of the SSB. It is evident that the SSB in the surface elevation distribution,<br />
through its effect on the specular point PDF, affects the ability of the on-satellite
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 47<br />
tracking algorithm to determine the exact range to the at-nadir mean sea level. The<br />
Brown model (Equation 3.12) uses a Gaussian specular point PDF to describe the<br />
statistics of the ‘linear’ ocean wave heights. As stated, the ‘linear’ ocean surface is<br />
defined as an incoherent surface (i.e., not including the SSB) in this study. To a firstorder<br />
approximation, the height of the waves and slope of the waves are assumed<br />
zero-mean, Gaussian, uncorrelated random variables. However, higher order effects,<br />
including correlation between heights and slopes, as well as skewness, indicate that<br />
the surface height and slope are not strictly Gaussian (Barrick and Lipa, 1985). For<br />
example, the results obtained by Srokosz (1986) and Rodriguez (1988) regarding the<br />
effects of the SKB on the range to the sea surface show that the SKB causes an error<br />
of 0.8 cm for a skewness of 0.1 and the SWH of 2 m. This error increases to ~4 cm<br />
for the SWH of 10 m.<br />
Because of the effects from both the EMB and SKB on the shape of the leading edge<br />
of the radar returned waveform, as well as the prospect of extracting information<br />
about the sea state of a non-linear scattering surface, numerous studies have<br />
attempted to estimate the SKB (e.g., Lipa and Barrick, 1981; Parsons, 1979;<br />
Rodriguez, 1988; Barrick and Lipa, 1985; Hayne, 1980), and the EMB (e.g.,<br />
Rodriguez, 1988) using a modified expression for the specular point PDF.<br />
Longuet-Higgins (1966, cited in Challenor, 1989) obtained the PDF of surface height<br />
using an approximation of the Gram-Charlier series (cf. Jeffrey, 1995) for the nonlinear<br />
surface:<br />
⎫ ⎛ ⎞<br />
⎨<br />
⎧ 2<br />
1 λ<br />
= + s<br />
η<br />
qs ( ζ ) 1 H<br />
3( η) ⎬exp⎜<br />
− ⎟<br />
2π<br />
σ ⎩ 6 ⎭ ⎝ 2<br />
(3.19)<br />
s<br />
⎠<br />
where<br />
ξ<br />
η = (3.20)<br />
σ s<br />
and<br />
H<br />
n<br />
are the Hermite polynomials (e.g., Jeffrey, 1995) of degree n with
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 48<br />
H<br />
n<br />
n<br />
n 2 d<br />
2<br />
( x)<br />
= ( −1)<br />
exp( x ) exp( −x<br />
)<br />
(3.21)<br />
n<br />
dx<br />
Equation (3.19) introduces the height skewness parameter ( λ<br />
s<br />
) of the sea surface,<br />
which is a measure of the non-linearity of the wavefield, and is a dimensionless<br />
quantity (Hayne and Hancock III, 1990). Barrick and Lipa (1985) and Srokosz (1986)<br />
extend the result in Equation (3.19) to obtain the joint PDF of the elevation and slope,<br />
which contains another non-linear parameter of the cross-skewness (δ ) as<br />
2<br />
1 ⎧ λ<br />
⎛ ⎞<br />
s<br />
δ ⎫ η<br />
qs ( ζ ) = ⎨1<br />
+ H<br />
3( η) − H1( η) ⎬exp⎜<br />
− ⎟.<br />
(3.22)<br />
2π<br />
σ ⎩ 6 2 ⎭ ⎝ 2<br />
s<br />
⎠<br />
In addition, Hayne (1980) uses a four-term series as the expression of radar-observed<br />
surface height PDF from the laboratory measurements of the surface elevation<br />
density function for a wind-generated wavefield, which contains Hermite<br />
polynomials of H<br />
3<br />
, H<br />
4<br />
, and H<br />
6<br />
(see Hayne, 1980, Equation 12). Two non-linear<br />
parameters of skewness and kurtosis are introduced (discussed below).<br />
The skewness parameter contributes significantly to the SSB (cf. AVISO/Altimeter,<br />
1996, p.45; Chelton, 2001), while the cross-skewness term is thought to be related to<br />
the EMB and thus cause a shift of the electromagnetic surface with respect to the true<br />
mean sea surface (Rodriguez, 1988). A rewritten form of the PDF by Rodriguez<br />
(1988) makes the effect of the EMB more apparent:<br />
⎫ ⎛ ⎞<br />
⎨<br />
⎧ 2<br />
1 = + s<br />
( ) ⎬ ⎜ −<br />
R<br />
q ( ζ λ<br />
η<br />
s<br />
) 1 H<br />
3<br />
ηR<br />
exp ⎟<br />
2π<br />
σ ⎩ 6 ⎭ ⎝ 2<br />
s<br />
⎠<br />
(3.23)<br />
where<br />
η<br />
R<br />
( ζ − ζ γσ / 2)<br />
+ 0 s<br />
= (3.24)<br />
σ<br />
s<br />
where γ is the parameter related to the surface shift partly caused by the EMB and<br />
ζ<br />
0<br />
is the altimeter tracker error. The expression in Equation (3.24) indicates that the<br />
effect of γ is to shift the whole surface by the amount of − γσ / 2)<br />
, which is the<br />
(<br />
s
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 49<br />
magnitude of the EMB. This means that the total altimeter height estimation error,<br />
i.e., the shift of the altimeter-observed sea surface with respect to the true mean sea<br />
surface, consists of the altimeter tracker error ( ζ<br />
0<br />
) and the shift caused by the EMB.<br />
Therefore, Equation (3.23) differs from Equations (3.19) and (3.22).<br />
The SKB can be estimated from the above expressions of the specular point PDF<br />
using the altimeter waveform data (e.g., Hayne, 1980; Rodriguez, 1988; Lipa and<br />
Barrick, 1981; Barrick and Lipa, 1985). Using retracked TOPEX data after removing<br />
the EMB, Rodriguez and Martin (1994) estimate that the dimensionless skewness is<br />
very small and ranges from 0.01-0.12.<br />
Theoretically, determining the EMB requires knowledge of the specular point PDF.<br />
Unfortunately, the current theoretical understanding of the EMB is not sufficiently<br />
well developed to estimate the EMB from a purely theoretical basis or from altimeter<br />
data themselves without additional geophysical information or assumptions (Chelton,<br />
2001; Rodriguez, 1988). There are two possible reasons for this. The first is because<br />
two shifts of the EMB and SKB are introduced in the specular point PDF, and they<br />
are not independent to all orders of approximation. Thus, they cannot be<br />
distinguished. The second is that the EMB depends on the characteristics of the<br />
wavefields, while only the SWH can be unambiguously extracted directly from the<br />
altimeter waveform. Therefore, in practice, empirical estimation of the EMB is a<br />
more appropriate method (Chelton, 2001).<br />
3.3.4 Approximate Expression of the System PTR<br />
The altimeter-transmitted pulses have a shape that is described by the radar system<br />
PTR. However, the PTR also includes effects due to the bandwidth of the receiver<br />
(Hayne, 1980). It is generally expressed by Equations (3.9) and (3.10). In the Brown<br />
model (Equation 3.12), the PTR is expressed by a Gaussian function (Equation 3.10).<br />
In reality, the PTRs are not exactly Gaussian (Hayne, 1980; Rodriguez, 1988).<br />
Rodriguez (1988) compares measured PTRs of SEASAT and GEOSAT with an<br />
idealised PTR ( sinc<br />
2 x)<br />
. The results (Rodriguez, 1988) show that the radar system<br />
PTR presents a lack of symmetry and large side lobes, suggesting thus the PTR is not
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 50<br />
2<br />
sinc x or a Gaussian function. The effect of the Gaussian-approximated PTR will<br />
change the shape of the return waveform, especially in the early parts of the leading<br />
edge. To obtain realistic non-Gaussian ocean parameters, Rodriguez (1988)<br />
concludes that the measured PTR must be used in the estimation process to obtain<br />
realistic sea state parameters.<br />
The modified PTR has been developed by Hayne (1980), and Rodriguez and Martin<br />
(1994). Hayne (ibid.) expresses the PTR using a nearly Gaussian function (a fourterm<br />
series). Rodriguez and Martin (1994) expand the PTR as a series of Gaussian<br />
functions. They both appear to improve the estimates of the non-Gaussian ocean<br />
surfaces (Chelton et al., 2001).<br />
3.3.5 A Modified Solution of the Radar Returns Developed in This Study<br />
Following Brown (1977), modified convolutional representations of the waveform<br />
for non-Gaussian ocean surfaces have been given by Hayne (1980), Rodriguez<br />
(1988), Rodriguez and Chapman (1989), Challenor and Srokosz (1989), and Barrick<br />
and Lipa (1985). The main difference among the convolutional solutions is that<br />
researchers take different modified expressions of the surface-height PDF, the radar<br />
system PTR, and the Bessel function. Following this idea, and according to the<br />
specific application of this study, a modified solution of the radar returns developed<br />
by the author will be introduced in this Subsection.<br />
The model introduced in this study is slightly different from the Brown model and<br />
the modified models (e.g., the Hayne (1980) model). The main differences are as<br />
follows:<br />
(1) As stated, the current theory is not sufficiently well developed to accurately<br />
estimate the EMB using only altimeter data. This term will be neglected from<br />
the specular point PDF. Instead, Equation (3.19) is employed as the expression<br />
of the specular point PDF.<br />
(2) Equation (3.18) is used as the function describing the smooth surface response,<br />
because it takes account of the effects of the Earth’s curvature. It is also<br />
computationally simple, but without significant loss of accuracy (Rodriguez,
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 51<br />
1988), provided that the Bessel function has been expanded appropriately<br />
(ibid.).<br />
(3) Since this study does not focus on estimating non-linear parameters of the non-<br />
Gaussian ocean surface, the Gaussian function will still be assumed to be an<br />
adequate expression of the radar system PTR for the purpose of this study.<br />
Using a variable<br />
t = −2ζ / c , Equation (3.19) can be rewritten in the time domain as:<br />
q s<br />
( t)<br />
=<br />
1 ⎧ λs<br />
⎨1<br />
+<br />
2π<br />
σ ⎩ 6<br />
s<br />
2<br />
3 ⎫ ⎛ η ⎞<br />
( − 3 ) exp⎜<br />
−<br />
t<br />
η η<br />
⎟<br />
t<br />
t<br />
⎬<br />
⎭<br />
⎝<br />
2<br />
⎠<br />
(3.25)<br />
in which<br />
c<br />
ηt<br />
= − t<br />
(3.26)<br />
2σ<br />
s<br />
Equation (3.25) is the surface height PDF in the time domain. It can be reduced to<br />
the linear case when the skewness parameter λ<br />
s<br />
is omitted, i.e., by setting λ<br />
s<br />
=0.<br />
The radar system PTR used is assumed to be a Gaussian function, given by<br />
P<br />
PTR<br />
( t)<br />
≈<br />
1<br />
⎡<br />
1 ⎛ ⎞<br />
⎢ ⎜<br />
t<br />
exp − ⎟<br />
2π<br />
σ ⎢ 2<br />
p<br />
⎣ ⎝σ<br />
p ⎠<br />
2<br />
⎤<br />
⎥<br />
⎥<br />
⎦<br />
(3.27)<br />
As stated, for typical short pulse radar altimeters, the width of the PTR is of the order<br />
20 ns or less (Brown, 1977). This approximation of the PTR is found to be adequate<br />
for these widths by Brown (1977).<br />
From Equations (3.18), (3.25), and (3.27), describing the three terms in the<br />
convolutional representation, a modification of Equation (3.1) will be given. The<br />
convolution of the specular point PDF and the system point PTR can be written as<br />
(Bracewell, 2000)
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 52<br />
B(<br />
t)<br />
= qs(<br />
t)<br />
∗ PPTR(<br />
t)<br />
1 ⎡ 1 ⎛ t ⎞<br />
= exp⎢−<br />
⎜ ⎟<br />
2π<br />
σ ⎢⎣<br />
2 ⎝σ<br />
⎠<br />
2<br />
⎤⎡<br />
1<br />
⎥⎢1<br />
+ λcsH<br />
⎥⎦<br />
⎣ 6<br />
3<br />
⎛ t ⎞⎤<br />
⎜ ⎟⎥<br />
⎝σ<br />
⎠⎦<br />
(3.28)<br />
in which λ cs<br />
and σ are the composite skewness and risetime (i.e., the information<br />
relative to the sea surface roughness through the SWH) given by<br />
2 2 ⎛ 2<br />
= σ p<br />
+ ⎜ s<br />
2<br />
σ ⎞<br />
⎟<br />
⎝ c σ<br />
⎠<br />
(3.29)<br />
3<br />
⎛σ<br />
s ⎞<br />
λcs = λs⎜<br />
⎟<br />
(3.30)<br />
⎝ σ ⎠<br />
Then, the waveform can be obtained as:<br />
P(<br />
t)<br />
= P ( t)<br />
∗ B(<br />
t)<br />
=<br />
s<br />
∫ ∞ −∞<br />
= A<br />
P ( x)<br />
B(<br />
t − x)<br />
dx<br />
A<br />
s<br />
exp<br />
0 ξ<br />
τ<br />
[ − d(<br />
+ d / 2) ](<br />
C1<br />
+ C2<br />
) + PN<br />
(3.31)<br />
where<br />
1 ⎡ ⎛ τ ⎞ ⎤<br />
C<br />
1<br />
= ⎢erf<br />
⎜ ⎟ + 1⎥<br />
(3.32)<br />
2 ⎣ ⎝ 2 ⎠ ⎦<br />
3<br />
2<br />
λ ⎧<br />
⎫ ⎛ ⎞<br />
cs<br />
⎡ ⎛ τ ⎞ ⎤ d 1<br />
2<br />
2 −τ<br />
C2<br />
= ⎨⎢erf<br />
⎜ ⎟+ 1⎥<br />
+ ( 1−<br />
3d<br />
− 3d<br />
⋅τ<br />
−τ<br />
) ⎬exp⎜<br />
⎟<br />
(3.33)<br />
6 ⎩⎣<br />
⎝ 2 ⎠ ⎦ 2 2π<br />
⎭ ⎝ 2 ⎠<br />
with<br />
t − t<br />
τ = σ<br />
0<br />
− d<br />
(3.34)<br />
2<br />
⎛ β ⎞<br />
d = ⎜α<br />
⎟<br />
−<br />
σ<br />
(3.35)<br />
⎝ 4 ⎠<br />
where P<br />
N<br />
is an additive noise term that represents the altimeter’s thermal noise, the<br />
time shift t 0<br />
represents the mean level of the specular points on the ocean surface at
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 53<br />
nadir, and σ represents the standard deviation (in delay time units) of the specular<br />
point about this mean level. Equation (3.31) is an extension of the expression<br />
obtained by Brown (1977) to include the effects of the skewness and Earth curvature.<br />
It is also different from the Hayne (1980) model, which includes another non-linear<br />
kurtosis term to estimate parameters for the non-Gaussian surface.<br />
The effects of Equations (3.32) to (3.35) on the waveform can be discussed<br />
qualitatively. The function C 1<br />
is the predominant expression of the leading edge of<br />
the return waveform, while the function C 2<br />
represents the effects of the skewness<br />
and the off-nadir angle on the leading edge. Because of small magnitude of the<br />
skewness 0.01< λ
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 54<br />
significantly affects the waveform in the leading edge when τ ≤ 2 . It has no effect<br />
on the regions of the thermal noise and trailing edge of the waveform (Figure 3.3).<br />
0.12<br />
0.10<br />
ERS<br />
POSEIDON<br />
TOPEX<br />
Values of d<br />
0.08<br />
0.05<br />
0.03<br />
0.00<br />
0.0 5.0 10.0 15.0 20.0<br />
SWH (metres)<br />
Figure 3.1 Values of d ( ξ = 0 ), showing that d
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 55<br />
0.04<br />
0.02<br />
Values of C2<br />
0.00<br />
−0.02<br />
−0.04<br />
−0.06<br />
ERS<br />
POSEIDON<br />
TOPEX<br />
−0.08<br />
0.0 20.0 40.0 60.0<br />
Bin or Gate<br />
Figure 3.3 Values of C<br />
2<br />
( d = 0 , λ<br />
cs<br />
= -1, and SWH = 4 m) showing the small<br />
magnitude of effects on the leading edge of the waveform, and nearly overlapped<br />
curves for ERS-1/2 and TOPEX.<br />
As can be seen from Figure 3.3, for the times near the tracking gate when τ ≤ 1, the<br />
effect of the skewness is to lower the power, and to raise it when the leading edge<br />
starts to rise or near the plateau of the waveform.<br />
Equation (3.31) can be considered as a fundamental model of the ocean return<br />
waveforms. From the above qualitative and quantitative analysis, it is found that the<br />
effects of non-Gaussian ocean surface through skewness λ cs<br />
on the waveforms are<br />
very small (~0.01-0.12). In this case, when retracking waveform to estimate the<br />
range correction, the skewness term can be ignored without significantly affecting<br />
the results in the complete convolutional form. In Chapter 5, Equation (3.31) will be<br />
used to obtain a linear analytical function for retracking waveforms in coastal regions.<br />
3. 4 Ocean Waveform Retracking<br />
In general, a satellite radar altimeter’s on-board range estimation is limited by the onboard<br />
computational capability and by real-time data processing constraints. As the
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 56<br />
orbit errors are reduced by using geometric, dynamic and combined methods (e.g.,<br />
T/P orbit determination), it becomes increasingly desirable to also reduce the errors<br />
due to the on-board range estimation algorithm. In addition to the on-satellite range<br />
estimates, the satellite also transmits the waveform measured by the radar altimeter<br />
to the ground. This waveform contains information about the scattering surface, but<br />
is retracked later to get improved ranges.<br />
The convolutional representation of the radar return waveform, its shape, and the<br />
precise position of the waveform in the range window allow for the retrieval of more<br />
accurate estimates of the range and the SWH, as well as additional parameters such<br />
as the antenna off-nadir angle and the skewness of the surface elevation distribution<br />
using more sophisticated algorithms. Hence, it has long been recognised that<br />
reprocessing altimeter waveform data will lead to improved geophysical estimates<br />
using waveform retracking over ocean surfaces (Brenner et al., 1993; Barrick and<br />
Lipa, 1985; Rodriguez and Martin, 1994a; Rodriguez, 1988).<br />
Ocean waveform retracking is based upon the convolutional representation of the<br />
waveform, such as the Brown model (Equation 3.12), or the modified model (e.g.,<br />
Equation 3.31) for non-linear ocean surfaces. The retracking algorithms used to<br />
estimate the geophysical parameters over ocean surfaces can mainly be categorised<br />
into two classes of least squares estimation and deconvolution, as follows.<br />
3.4.1 Least Squares Fitting Procedure<br />
As stated, an analytic representation of the waveform can be obtained by formulating<br />
a convolution of three terms. The most commonly used algorithm is to fit a<br />
convolution model of the waveform (e.g., Equation 3.31) to the measured waveforms.<br />
An iterative least squares fitting procedure (Hayne and Hancock III, 1990; Hayne<br />
and Hancock III, 1982) or a maximum likelihood estimation (MLE) algorithm<br />
(Challenor and Srokosz, 1989; Parsons, 1979; Rodriguez and Martin, 1994a;<br />
Tokmakian et al., 1994) is performed to extract the parameter estimates.<br />
Hayne and Hancock III (1990) use an iterative least squares fitting procedure to<br />
retrack GEOSAT waveform data. In their algorithm, Equations (3.16), (3.19) and<br />
(3.27) express three terms in the convolution Equation (3.1). The model of the
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 57<br />
convolutional solution is fitted to the waveforms to produce six independent<br />
estimates. These are:<br />
(1) Waveform amplitude, A ;<br />
(2) The time shift of the range-tracker-positioned waveform tracking gate relative<br />
to the true tracking gate of the waveform, t 0<br />
;<br />
(3) SWH;<br />
(4) Noise baseline, P<br />
n<br />
, (i.e., thermal noise);<br />
(5) Skewness of the radar-observed surface elevation PDF, λ<br />
cs<br />
; and<br />
(6) Off-nadir pointing angle, ξ .<br />
The waveform has noise. It can be seen in the trailing edge, but is more difficult to<br />
identify in the leading edge. The parameters affected by the noise in the tailing edge<br />
will degrade the geodetic parameter estimates in the leading edge if the least squares<br />
fitting procedure is applied to an individual waveform. The fitting method in this<br />
case may appear to give statistically small error but does not provide a consistent<br />
accuracy to the individual parameters. Therefore, to reduce the noise on the return<br />
waveforms, effective fitting is usually carried out using waveform averages in<br />
several seconds rather than individual waveforms (e.g., Challenor and Srokosz,<br />
1989). In the procedure of Hayne and Hancock III (1990), ten-second average<br />
waveforms are used. Since the closed form solutions of the theoretical waveform<br />
cannot be deduced from realistic forms of the radar PTR, the fitting algorithm suffers<br />
from the drawback that the fitting function is analytic only if the radar system PTR is<br />
Gaussian (Brown, 1977; Challenor and Strkosz, 1989), near-Gaussian (Hayne, 1980),<br />
or approximated by a set of Gaussians (Rodriguez and Martin, 1994a).<br />
A typical example of using the MLE algorithm is from the UK Earth Observation<br />
Data Centre (EODC). EODC retracks ERS-1 data to re-estimate all parameters<br />
calculated by the on-board algorithm on the ground using the MLE algorithm (e.g.,<br />
Tokmakian, 1994). The convolution function includes unknown parameters A , t 0<br />
,<br />
σ , ξ , and non-linear wave parameters λ cs<br />
and δ . The MLE algorithm has been<br />
tested by EODC with three forms of the convolution functions as follows:<br />
(1) Non-linear model containing both non-linear wave parameters λ<br />
cs<br />
and δ .
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 58<br />
(2) Non-linear model containing only one of non-linear wave parameters λ<br />
cs<br />
or δ .<br />
(3) Linear model without λ<br />
cs<br />
and δ .<br />
Their results show that the MLE algorithm is successful with the linear model for the<br />
estimation of the geophysical parameters, but cannot accurately estimate all<br />
parameters if the non-linear model is used. From EODC’s tests, it has been shown<br />
that the 20 Hz individual waveform data does not produce as accurate results as<br />
when averaging the waveforms to 1 Hz (ibid.). This means an averaging procedure is<br />
essential as well.<br />
Theoretically, the MLE algorithm is more accurate than least squares (Chelton et al.,<br />
2001) and gives asymptotically minimum variance unbiased estimates of the<br />
parameters (Challenor and Srokosz, 1989). However, in practice, because of<br />
unreliability of the statistical weights due to unmodelled errors in the waveforms<br />
(Chelton, 2001) and further complications introduced by numerics of the model<br />
(Tokmakian et al., 1994), MLE estimates are not significantly better than those<br />
estimated by least squares. Therefore, nonlinear least-squares fitting procedure<br />
appears to still be an adequate method of waveform retracking.<br />
3.4.2 Deconvolution Method<br />
The second method estimates parameters from the specular point PDF obtained by<br />
deconvolution, which was introduced by Priester and Miller (1979), Lipa and Barrick<br />
(1981), and refined by Rodriguea and Chapman (1989). The deconvolution method<br />
uses a fact that the average altimeter return waveform is the convolution of the ocean<br />
surface specular point PDF, which is an unknown function, and two functions of the<br />
radar PTR and the smooth surface response, which are known and depend only on<br />
the radar design parameters. By taking the fast Fourier transform (FFT) of the slope<br />
function, the specular point PDF can be obtained from dividing the Fourier transform<br />
of the data by the Fourier transform of the convolution of the smooth surface<br />
response and the radar PTR. The specular PDF will then allow estimation of three<br />
independent sea-surface-state parameters (e.g., Lipa and Barrick, 1981; Barrick and<br />
Lipa, 1985; Rodriguez, 1988; Rodriguez and Chapman, 1989). These are:<br />
(1) the ocean surface standard deviation;
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 59<br />
(2) ocean surface skewness; and<br />
(3) shift of the altimeter tracking gate with respect to the mean electromagnetic<br />
surface measured by an altimeter.<br />
For the convenience of representation, altimeter return power Equation (3.1) can be<br />
thought of as a matrix equation<br />
y = Mx<br />
(3.36)<br />
where y is the vector of return power and M is the convolution given by<br />
M = P ( t)<br />
∗ P ( t)<br />
(3.37)<br />
fs<br />
PTR<br />
and x is the specular point PDF vector<br />
x = q(t)<br />
(3.38)<br />
Thus, the specular point PDF can be obtained by inverting Equation (3.36).<br />
In its most direct implementation, Lipa and Barrick (1981) suggest that<br />
deconvolution can be performed on the slope of the leading edge of the waveform,<br />
rather than the full waveform. This suggestion means that the radar antenna pattern<br />
and the off-nadir angle can be ignored, and the slope of the mean waveform is given<br />
by the convolution of the specular point PDF and the radar PTR. Thus, data can be<br />
neglected in the trailing edge of the waveform, where the noise level is higher<br />
(Section 3.3.5), and the returns depend weakly on the non-Gaussian ocean surface.<br />
This algorithm has the advantages that it is computationally simple and fast, and it is<br />
not necessary to assume any functional form for the radar specular point PDF.<br />
However, it is impossible in general to ignore the radar antenna pattern and the offnadir<br />
angle for altimeters, such as GEOSAT and TOPEX, because it introduces<br />
errors which are of the same order as the quantities to be estimated (cf. Rodriguez,<br />
1988). Another disadvantage of the slope deconvolution is that the algorithm is<br />
highly sensitive to high-frequency noise in the waveform. This leads the<br />
deconvolution of Lipa and Barrick (1981) being carried out only for waveforms that
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 60<br />
had been averaged over a long time period (e.g., 10 seconds) for which an acceptably<br />
low noise level was obtained.<br />
In order to overcome these problems, Rodriguez (1988) and Rodriguez and Chapman<br />
(1989) introduce a "corrected slope convolution model" as<br />
2<br />
1 ⎡ ⎛ β ⎞ ⎤<br />
M = ⎢P′<br />
( t)<br />
+<br />
⎜α<br />
−<br />
⎟P(<br />
t)<br />
⎥<br />
(3.39)<br />
A ⎣ ⎝ 4 ⎠ ⎦<br />
where P′ (t)<br />
is the derivative of the return power with respect to the return time. This<br />
model differs from the expression derived by Lipa and Barrick (1981, Equation 3),<br />
because the second term on the right-hand side represents the radar antenna pattern<br />
and off-nadir angles. Since the return waveform is also a function of the thermal<br />
noise and the off-nadir pointing angle, they must be estimated separately when using<br />
Equation (3.39).<br />
Both deconvolution algorithms presented by Lipa and Barrick (1981) and Rodriguez<br />
and Chapman (1989) have advantages and disadvantages. Rodriguez and Chapman<br />
(ibid.) investigate the use of both methods. They recommend that a spectral filtering<br />
procedure be used to reduce the high-frequency noise in the estimated parameters.<br />
3. 5 Ice-Sheet Waveform Retracking<br />
Ice-sheet surfaces are characterised by the existence of topography, often with a<br />
significant surface slope with respect to the ocean. Over such topographic surfaces,<br />
the altimeter is generally unable to follow abrupt altitude changes and small-scale<br />
slopes, thus losing signal lock. To maintain the satellite track over large surface<br />
slopes, ERS-1/2 altimeters operate in ice mode over ice sheets at a reduced range<br />
resolution or increased sampling time (12.12 ns) compared with the ocean mode<br />
(3.03 ns). The form of the return is determined by the range to the different reflectors,<br />
their scattering properties, and their position within the antenna beamwidth. This<br />
results in a return waveform of generally unpredictable shape, which changes rapidly<br />
and is offset from the tracking gate designed for the altimeter over oceans. The<br />
present theory of altimetry, developed to describe scattering from the ocean surface<br />
such as the Brown (1977) model, does not deal properly with the geometry of ice-
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 61<br />
sheet surfaces. Therefore, onboard range measurements in altimeter records do not<br />
always correspond to the reflecting surface at nadir.<br />
retracking gate<br />
tracking gate<br />
Returned Power (counts)<br />
offset<br />
32.5<br />
Time or Bins<br />
Figure 3.4 The concept of the observed waveform retracking for ERS-1/2, in which<br />
( retracking gate − tracking ) × 0. 45<br />
offset = gate (m).<br />
Estimation of the true ranges to the reflecting surfaces can be obtained using<br />
waveform retracking techniques. Waveform retracking determines the offset<br />
(Figure 3.4) of the actual tracking gate, which is related to the half-power point on<br />
the leading edge of the waveform, from the pre-designed tracking gate, and corrects<br />
the range calculated by the on-board algorithm accordingly. Retracking algorithms<br />
developed over ice sheets can be categorised into four classes:<br />
(1) Fitting algorithm: β -parameter retracking (Martin et al., 1983);<br />
(2) The off centre of gravity (OCOG) technique (Wingham et al., 1986);<br />
(3) Threshold retracking (e.g., Bamber, 1994; Davis, 1997; Ridley and Partington,<br />
1988; Partington et al., 1989; Partington et al., 1991; Femenias et al., 1993;<br />
Zwally, 1996; Remy and Minster, 1993); and<br />
(4) A surface/volume scattering retracking (Davis, 1993a).
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 62<br />
In addition, Brooks et al. (1978) assume that the actual tracking point on the leading<br />
edge of the waveform is related to the half value of the maximum return power. This<br />
retracking method is, in fact, the same as the 50% threshold retracking. Wingham<br />
(1995b) develops a method for determining the average height of a large topographic<br />
ice sheet from altimeter observations. This method is extended to cover the ice-sheet<br />
geometry, as well as including the effects of the penetration of the surface by the<br />
radar waves, and thus differs from all above methods. A Volterra type of integral<br />
equation is derived (Wingham, 1995a), which allows the uniqueness of the solution<br />
for the average ice-sheet height to be estimated simply. However, a discrete<br />
representation is not given, so that practical implementation is not yet available.<br />
3.5.1 Fitting Algorithm - β-Parameter Retracking<br />
Martin et al. (1983) developed the first retracking algorithm for processing altimeter<br />
return waveforms over continental ice sheets. The algorithm was used to retrack all<br />
SEASAT radar altimeter waveforms to obtain corrected surface elevation estimates.<br />
This algorithm fits a 5- or 9-parameter function to the altimeter waveform, which is<br />
based partly on Brown’s surface scattering model (discussed in Section 5.4.3). The 5-<br />
parameter function is used to fit single-ramp returns (Figure 3.5), while the 9-<br />
parameter function is used to fit double-ramp returns (Figure 3.6) (Martin et al.,<br />
1983). This retracking algorithm is also known as β-parameter retracking or the<br />
NASA algorithm (e.g., Davis, 1995). It is a surface-scattering model that can deal<br />
with complex waveforms reflected from one or two scattering surfaces over ice<br />
sheets. The Ice Altimetry Group at NASA’s Goddard Space Flight Centre (GSFC)<br />
has developed a series of retracking algorithms for ice-sheet waveforms based upon<br />
Martin’s functions (Zwally, 1996). The general parameter function fitting the radar<br />
returns is given as (Martin et al., 1983; Zwally, 1996):<br />
n<br />
⎛ t − β<br />
3i<br />
⎞<br />
y(<br />
t)<br />
= β + ∑ +<br />
⎜<br />
⎟<br />
1<br />
β<br />
2i<br />
(1 β<br />
5iQi<br />
) P<br />
(3.40)<br />
i=<br />
1 ⎝ β<br />
4i<br />
⎠<br />
where
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 63<br />
Q<br />
i<br />
⎧ 0<br />
= ⎨<br />
⎩t<br />
− ( β<br />
3i<br />
+ 0.5β<br />
)<br />
4i<br />
for t < β + 0.5β<br />
3i<br />
3i<br />
4i<br />
for t ≥ β + 0.5β<br />
4i<br />
(3.41)<br />
⎛ −<br />
∫ −∞<br />
⎟ ⎞<br />
= x 2<br />
1 q<br />
P(<br />
x)<br />
exp<br />
⎜ dq<br />
2π ⎝ 2 ⎠<br />
(3.42)<br />
where n = 1 or 2 is the number of the ramp in the waveform range window which<br />
corresponds to single or double reflecting surfaces, respectively (Figures 3.5 and 3.6).<br />
Double ramps indicate that two distinct, nearly equidistant surfaces are observed by<br />
the altimeter. The unknown parameters are as follows.<br />
(1) β<br />
1: the thermal noise level of the return waveform.<br />
(2) β<br />
2i<br />
: return signal amplitude.<br />
(3) β<br />
3i<br />
: the mid-point on the leading edge of the waveform, which is chosen as the<br />
correct range point in the retracking procedure.<br />
(4) β<br />
4i<br />
: the return waveform risetime.<br />
(5) β<br />
5i<br />
: the slope of the trailing edge.<br />
When the linear trailing edge is replaced by an exponential decay term (cf., Zwally,<br />
1996), Equation (3.40) is adapted to give<br />
n<br />
⎛ t − β<br />
3i<br />
⎞<br />
y(<br />
t)<br />
= β + ∑ −<br />
⎜<br />
⎟<br />
1<br />
β<br />
2i<br />
exp( β<br />
5iQi<br />
) P<br />
(3.43)<br />
i=<br />
1 ⎝ β<br />
4i<br />
⎠<br />
where<br />
Q<br />
i<br />
⎧0<br />
= ⎨<br />
⎩t<br />
− ( β<br />
3i<br />
+ 0.5β<br />
)<br />
4i<br />
for<br />
for<br />
t < β<br />
3i<br />
t ≥ β<br />
3i<br />
− 2β<br />
− 2β<br />
4i<br />
4i<br />
(3.44)<br />
This function is used to fit the waveform with a fast-decaying trailing edge over seaice<br />
or ice sheets, which are caused by the beam attenuation (cf. Zwally, 1996). It can<br />
simulate the antennae attenuation as the pulse expands on the surface beyond the<br />
pulse-limited footprint. This function is designed to fit the fast-decaying ice mode<br />
return waveforms (ibid.). Another fitting function developed at GSFC is to fit a<br />
simple linear trailing edge to the first ramp and the exponential decay-trailing edge to
Returned Power (counts)<br />
Returned Power (counts)<br />
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 64<br />
retracking gate<br />
tracking gate<br />
β<br />
51<br />
(trailing edge slope)<br />
β<br />
11<br />
β<br />
21<br />
β<br />
41<br />
(risetime)<br />
β<br />
31<br />
32.5<br />
Time or Bins<br />
Figure 3.5 Five- β -parameter model, fitting a single ramp function to a signal<br />
leading edge of the waveform samples.<br />
tracking<br />
gate<br />
1st retracking<br />
gate<br />
2nd retracking<br />
gate<br />
β<br />
51<br />
(1st trailing<br />
edge slope)<br />
β42<br />
(2nd risetime)<br />
β52<br />
(2nd trailing<br />
edge slope)<br />
β<br />
1<br />
β<br />
41<br />
(1st risetime)<br />
β<br />
21<br />
β<br />
22<br />
β<br />
31<br />
β<br />
32<br />
32.5<br />
Time or Bins<br />
Figure 3.6 Nine- β -parameter model, fitting a double-ramp function to two leading<br />
edges of the waveform samples.
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 65<br />
the second ramp, for which detailed information can be also found in Zwally (1996).<br />
The β parameter retrackers will be employed in the coastal zone for this project<br />
(Chapter 5).<br />
3.5.2 Off-Centre of Gravity (OCOG) Retracking Algorithm<br />
The OCOG retracking method was developed by Wingham et al. (1986), which uses<br />
full waveform samples (excluding the samples in the bins at the beginning and end<br />
due to the wrap-around error that occurs within the digital single processing). It is<br />
based on the definition of a rectangle about the effective centre of gravity (COG) of<br />
the waveform and the amplitude (A) and width (W); see Figure 3.7. The area of the<br />
rectangle equals that of the waveform. The height of the COG is defined as half the<br />
amplitude of the centre of gravity of the waveform. To reduce the effect of lowamplitude<br />
samples in front of the leading edge, the squares of the sample values are<br />
used in the computation (Wingham et al., 1986). The equation used to compute A is<br />
given by<br />
A =<br />
64−n<br />
a<br />
∑<br />
i=<br />
1+<br />
n<br />
a<br />
4<br />
i<br />
64−n<br />
a<br />
∑<br />
P ( t)<br />
P ( t)<br />
(3.45)<br />
i=<br />
1+<br />
n<br />
a<br />
2<br />
i<br />
where P i<br />
(t)<br />
is the value of waveform sample at the i th bin, n a<br />
is the number of<br />
aliased bins at the beginning and end of the waveform (Figure 3.7) depending on the<br />
altimeter (typically n<br />
a<br />
= 4 for ERS-2 and n<br />
a<br />
= 0 for TOPEX/POSEIDON). The range<br />
window is i = 1, 2, ···, 64 sample bins. The waveform width, W, is then obtained by<br />
the relation<br />
∑<br />
AW = P<br />
i<br />
(t)<br />
with the result<br />
W<br />
⎛<br />
= ⎜<br />
⎝<br />
64−na<br />
2<br />
∑ Pi<br />
i=<br />
1+<br />
na<br />
⎞<br />
( t)<br />
⎟<br />
⎠<br />
2<br />
64−na<br />
4<br />
∑ Pi<br />
i=<br />
1+<br />
na<br />
( t)<br />
(3.46)<br />
the location of the COG is formed by<br />
COG =<br />
64−n<br />
a<br />
∑<br />
i=<br />
1+<br />
n<br />
iP ( t)<br />
a<br />
2<br />
i<br />
64−n<br />
a<br />
∑<br />
i=<br />
1+<br />
n<br />
P ( t)<br />
a<br />
2<br />
i<br />
(3.47)
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 66<br />
Finally, the location of the half-power point or mid-point on the leading edge of the<br />
waveform, LEG , is given by<br />
W<br />
LEG = COG −<br />
2<br />
(3.48)<br />
W<br />
Returned Power<br />
50% retracking point estimate<br />
COG<br />
A<br />
Time or Bins<br />
Figure 3.7 Schematic description of the OCOG method applied to an observed<br />
waveform. Using the OCOG method to calculate the amplitude (A) and width (W) of<br />
the rectangle, and the position of the 50% of amplitude interpolated at the leading<br />
edge as the estimate of expected tracking gate.<br />
3.5.3 Threshold Retracking Algorithm<br />
The OCOG algorithm is simple to implement, though it is purely statistical and is not<br />
based upon any physical model of the reflecting surfaces. Moreover, it is sensitive to<br />
the waveform shape affected by surface undulations and off-nadir pointing, because<br />
it uses the full samples in waveform bins. However, when dealing with the returns at<br />
which the slope of the leading edge is smaller (i.e., longer pulse rise times), an<br />
erroneous estimate is produced because the LEG cannot be precisely positioned at<br />
the mid-point of the leading edge of the waveform (e.g., Partington et al., 1989).<br />
To improve the estimation, an empirical method of threshold retracking was<br />
developed and is used by ESA to process altimeter data from the ERS satellite<br />
missions (Davis, 1995). It has also been adopted by NASA/GSFC as an alternative<br />
retracking algorithm for the production of ice-sheet altimeter data sets (Davis, 1997).
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 67<br />
This algorithm is based upon the dimensions of the rectangle (i.e., amplitude, width,<br />
and hence the ‘centre of gravity’) computed using the OCOG method. The threshold<br />
value is then referenced to the amplitude (Bamber, 1994) or the maximum waveform<br />
sample (e.g., Davis, 1997; Zwally, 1996) estimate of the rectangle at 25%, 50%, and<br />
75% of the waveform amplitude. The retracking gate estimate is determined by<br />
linearly interpolating between adjacent samples of a threshold crossing at a steep part<br />
of the leading edge slope of the waveform. This algorithm keeps the same<br />
advantages as the OCOG, but it can determine a more accurate tracking gate position<br />
than the OCOG does (Partington et al., 1989). A disadvantage is that, like the OCOG<br />
method, it is not based on a physical model.<br />
The selection of an optimum threshold level is important when applying this method<br />
to waveforms, because the range to the surface is determined from it. Davis (1995;<br />
1997) tests different threshold levels (10%, 20%, and 50% of the waveform<br />
amplitude) for the purposes of measuring ice-sheet elevation change. The<br />
comparison is performed on SEASAT, GEOSAT-GM, and GEOSAT-ERM<br />
crossover data. It was found that a 10% threshold level can produce more repeatable<br />
elevation estimates, a 20% threshold level is suitable for obtaining the true ice-sheet<br />
elevation in only an average sense, and a 50% threshold level is appropriate only<br />
when surface scattering dominates the return waveform shape.<br />
Thus far, no account has been taken of the thermal noise effect (described in Section<br />
3.4.1) in Equations (3.45), (3.46) and (3.47). However, since the magnitude of the<br />
thermal noise varies with location and time in a given waveform dataset, these<br />
variations cause the leading-edge retracking point to vary if the threshold level is<br />
referenced only to the amplitude in which the thermal noise is included. To<br />
overcome this problem, the mean thermal noise estimated by averaging over earlier<br />
gates 6-10 should first be subtracted from the amplitude (Zwally, 1996).<br />
3.5.4 Surface/Volume Scattering Retracking Algorithm<br />
In general, the leading edge of altimeter waveforms is characterised by a rapid rise in<br />
return power and continuing to increase to the plateau (Figures 2.3 or 3.7). Most<br />
leading edges over ice sheets, however, show first a rapid rise in the return power.<br />
Then, they continue to rise but at a reduced rate, indicating the influence from
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 68<br />
subsurface volume scattering (e.g., Ridley and Partington, 1988; Davis, 1993a). The<br />
reason is that the radar pulse can significantly penetrate the snow over most of the ice<br />
sheet, resulting in elevation estimation errors of up to a few metres (e.g., Partington<br />
et al., 1991) if seeking to determine the height of these surfaces. Therefore, when<br />
retracking is carried out using the β-parameter retracking algorithm, there is an<br />
implicit assumption that the ice sheet surface scatters the radiation in a similar<br />
manner to the ocean surface.<br />
To take into account radar penetration over ice sheets, Partington et al. (1991) use a<br />
threshold retracking method but the position of the retracking gate is selected from<br />
an average waveform derived from all SEASAT waveforms from the region of<br />
interest. To a first order, this method overcomes the penetration problem (ibid.).<br />
Using Gaussian approximations for the altimeter’s antenna pattern and transmitted<br />
pulse shape (i.e., the radar system PTR), Davis and Moore (1993b) derive a closedform<br />
analytical solution for the received power due to volume scattering. Because of<br />
this closed-form, they combine the volume-scattering model with the Brown surfacescattering<br />
model (Equation 3.12) to develop an algorithm capable of retracking<br />
altimeter waveforms over ice sheets. The combined surface and volume-scattering<br />
model is (Davis and Moore, 1993b)<br />
Am<br />
⎡ K ⎤<br />
SV ( n)<br />
= DC + ⎢S(<br />
n)<br />
+ V ( n)<br />
⎥<br />
S<br />
2 ⎣ S1<br />
⎦<br />
(3.49)<br />
in which DC is the thermal noise baseline, A m<br />
is the maximum amplitude of the<br />
model waveform, and S (n)<br />
is an adjusted form of the Brown model (cf. Equation<br />
3.12) given by<br />
1 ⎡ ⎛ t − t ⎡−<br />
⎤<br />
0 ⎞⎤<br />
4c<br />
S( n)<br />
= ⎢1<br />
+ erf ⎜ ⎟⎥exp⎢<br />
( t − t0)<br />
2<br />
⎥<br />
⎣ ⎝ 2σ ⎠⎦<br />
⎣ γ h ⎦<br />
(3.50)<br />
The derivation of Equation (3.50) from Equation (3.12) assumes that the off-nadir<br />
angle ξ = 0 , and the modified Bessel function is expanded to order zero ( n = 0 in<br />
Equation 3.17). The volume-scattering model, V (n)<br />
, is given by
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 69<br />
2 2 2<br />
[ β k c − 2k<br />
c ( t − t )]<br />
2 2<br />
⎛ c β c(<br />
t t0<br />
) ⎞<br />
τ<br />
−<br />
V ( n)<br />
= exp⎜<br />
⎟<br />
− − exp<br />
2 4<br />
2<br />
e s e s 0<br />
4h<br />
c<br />
h<br />
τ<br />
⎝ β β<br />
c ⎠<br />
(3.51)<br />
where constants S 1<br />
and S<br />
2<br />
are normalising factors that ensure the quantities V (n)<br />
and S ( n)<br />
+ V ( n)<br />
range from zero to one, K represents the correct proportion of<br />
volume scattering, c s<br />
is the speed of light in snow, β c<br />
is a constant related to the<br />
antenna beamwidth, β τ<br />
is a constant that determines the 3 dB width of the<br />
transmitted pulse, and k e<br />
is the extinction coefficient of snow. This is a fitting<br />
function, and there are six unknown parameters in the model ( DC , σ , t 0<br />
,<br />
A<br />
m<br />
, K<br />
and k<br />
e<br />
). This model will not be used for the coastal zone retracking in this study,<br />
because there is no known penetration problem present at the coast.<br />
3. 6 Land Waveform Retracking<br />
In spite of the fact that the satellite radar altimeter was developed to measure the<br />
ocean surface, and the on-board algorithms used to control the receiver gain and<br />
height tracker were not designed to cope with terrain signals, substantial quantities of<br />
seemingly useless data were obtained over land. However, SEASAT was the first<br />
altimeter to show its mapping capability over land (and ice-sheet) surfaces. Over<br />
about three months of its lifetime, it collected data over 34% of the Earth’s land<br />
surface (Tapley et al., 1987, cited in Guzkowska et al., 1988). GEOSAT data were<br />
also collected over land both in the 17-day ERM orbit and during period of the GM<br />
(Brenner et al., 1990). Moreover, with the advent of ERS-1/2, the ice mode with a<br />
longer sampling time on the altimeter and different orbit repeat phases provided far<br />
better spatial coverage of data (~80% of the Earth’s surface) for topographic<br />
purposes than previous satellite missions (Berry, 2000).<br />
Waveform data from different satellite altimeters collected over land surfaces have<br />
been analysed by Guzkowska et al. (1988), Brenner et al. (1990) and Berry (2000a).<br />
Guzkowska et al. (1988) categorise the SEASAT waveform data into seven classes,<br />
in which each type of data shows unique characteristics corresponding to the<br />
reflecting surface. Berry (2000) also presents a selection of 10 classes of waveforms<br />
with different characteristics for ERS-1 over land. These results indicate that specific
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 70<br />
problems have been associated with the attempt to retrieve an accurate range to<br />
surface over land. One of them, which is the same as that encountered over ice-sheet<br />
surfaces, is that the on-board satellite processor cannot calculate an accurate range to<br />
surface, because the altimeter is in general unable to centre the return reflected from<br />
the rapidly varying topographic surface in the range window. Thus, in order to work<br />
with these data, waveform retracking must be employed.<br />
Land waveform retracking is mainly conducted via two approaches. The first is to<br />
retrack waveforms using an expert system depending on the specific waveform<br />
categories, which was developed by Berry et al. (1998), Berry (2000a) and Berry et<br />
al. (2000b). ERS-1/2 geodetic missions, plus their operation in the ice mode, resulted<br />
in the acquisition of large and effective volumes of radar returns over land surfaces.<br />
Berry et al. (1998) generate orthometric heights from ERS-1 geodetic mission dataset<br />
using retracked waveform data from an expert systems approach. Data from the<br />
entire ERS-1 geodetic mission has been reprocessed to generate a Regional Altimeter<br />
Result database, and the data are then used to generate orthometric heights.<br />
Land topography, in particular topography with a non-uniform slope, and the nonuniform<br />
reflectivity of the land are main courses that lead to the complicated return<br />
waveform shapes over land. The expert system, on the other hand, contains several<br />
retrackers (Berry, 2000a, 2001). Each retracker is specifically designed and<br />
developed for a particular type of return waveform, for which ten retrackers have<br />
been created for a land waveform-retracking system (Berry, 2001). Some of them are<br />
(Berry et al., 1998):<br />
(1) Spline retracker, which is more effective with pre-peaked waveforms by fitting<br />
bicubic spline functions to the leading edge of the waveform;<br />
(2) Narrow waveforms, for high power specular returns from inland;<br />
(3) Bor retracker, for waveforms showing slope information;<br />
(4) Bog retracker, for Bor-type waveforms; and<br />
(5) Default retracker, i.e., threshold retracker, for waveforms that cannot be<br />
retracked by other retrackers.<br />
Two steps have been taken when reprocessing waveform data from land. The first<br />
step is to sort the waveforms based on their shape and retain all waveforms which
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 71<br />
show a leading edge regardless of complexity of the whole waveform shape. Next,<br />
choosing the optimised retracker to estimate the range to the land surface at the nadir<br />
based upon the individual waveform shapes. More detailed information about the<br />
expert system of land waveform retracking can be found in Berry et al. (1998) and<br />
Berry (2000).<br />
The second approach to land waveform retracking is achieved by retracking all<br />
waveforms using a single retracker, typically the OCOG technique (e.g., Bamber et<br />
al, 1997, cited in Berry, 2000) or one optimised for double-peaked waveforms (eg.,<br />
Brenner et al., 1997, cited in Berry, 2000). Before the retracker is applied to<br />
waveform data, they are edited and filtered so that the single retracker can succeed<br />
for a larger percentage of data. Shape corrections are then applied to individual<br />
retracked ranges to surface and combined to derive mean or median height values.<br />
Results from this approach represent the land topography with a spatial resolution of<br />
15' or 5' (Berry, 2000).<br />
The mathematical methods for these have not been presented because 1) they will not<br />
be used in this dissertation, and 2) moreover, they are not presented by authors.<br />
3. 7 Brief Introduction to Coastal Waveform Retracking<br />
The potential of the altimeter waveform data to be used in coastal regions has been<br />
the subject of comparatively little study (Anzenhofer et al., 2000; Brooks and<br />
Lockwood, 1990; Brooks et al., 1997). So far, considering the existing retracking<br />
algorithms, two approaches have been applied to retracking coastal waveforms,<br />
which are the fitting procedure and threshold technique. The first is to fit an<br />
analytical waveform model to the observed waveform data for determining the more<br />
accurate range estimate. The fitting function used from open literature<br />
(e.g., Anzenhofer et al., 2000) is the five-β-parameter function Equations (3.40) and<br />
(3.43), choosing either the linear trailing edge or the exponential trailing edge. When<br />
using the function fitting algorithm, usually n = 1 in Equations (3.40) and (3.43), an<br />
iterative least squares procedure is used.<br />
An alternative method used by Brooks et al. (1997) in coastal regions is a slightly<br />
modified threshold retracking technique, which tracks the ocean return from the
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 72<br />
hybrid waveforms of water and land close to the coastline. This retracking involves a<br />
three-stage process:<br />
(1) Returns are observed using a visual examination method to identify the ocean<br />
and land returns.<br />
(2) A threshold level is selected for each individual pass such that the threshold is<br />
higher than both the thermal noise level and the minor waveform leakage in the<br />
early gates, and is sufficiently low to intercept the ocean return part in the<br />
waveform. Then, the actual tracking gate position of the ocean return is<br />
obtained by converting the intercepted amplitude on the leading edge of the<br />
waveform to the gate number.<br />
(3) The position of the expected tracking gate is ascertained by averaging midpoint<br />
positions on the leading edge of the waveforms from several seconds of openocean<br />
waveforms in the same area, pass and date. The difference between this<br />
expected tracking gate and the actual tracking gate is computed and then<br />
converted to the range correction in metres. This range correction is applied to<br />
the range measurement from the on-board satellite-tracking algorithm.<br />
This method is appropriate for retracking the mixed waveforms in coastal regions.<br />
However, it cannot be conducted automatically (Brooks, et al., 1997). The method<br />
developed to select the threshold level by the author in Chapter 5 will solve this<br />
problem.<br />
3. 8 Summary<br />
An outline of waveform retracking algorithms, both for ocean and non-ocean<br />
surfaces, has been presented in this Chapter. Starting from the basic convolutional<br />
representation of the waveform, the Brown model is introduced. Then, it is extended<br />
to a form that includes additional corrections (i.e., skewness and Earth’s curvature)<br />
so that it can be used to estimate more parameters over non-linear ocean surfaces.<br />
Finally, a model which will be used in this study is deduced. A detailed qualitative,<br />
and some quantitative, analysis is carried out on this model.<br />
Over non-ocean surfaces, waveform-retracking algorithms have been developed for<br />
obtaining reasonably accurate elevation estimates from altimeter data over ice-sheet
Chapter 3. RADAR ALTIMETER WAVEFORM RETRACKING METHODS 73<br />
and land surfaces. Over ice-sheets, retracking algorithms are mainly based upon<br />
either a linear surface scattering model (i.e., the Brown model) or the threshold<br />
method. In addition, a volume and surface scattering model extends the Brown<br />
model to deal with the returns from snow-covered ice sheets. Over land surfaces,<br />
existing retracking algorithms are either those used over ice sheets or a retracking<br />
system. Because a physically-based model is impossible over land due to the more<br />
rugged and rapidly varying surface, an expert system that depends on the waveform<br />
shapes is currently the only apparently effective retracking method.<br />
It is important to note that improvement of the altimeter data in coastal regions has<br />
been the subject of comparatively little study, although many previous studies have<br />
reported the data are in error in these regions (e.g., Mantripp, 1996; Brooks and<br />
Lockwood, 1990; Hwang, 1998; Featherstone, 2003). The existing retracking<br />
algorithms are also either the β-parameter fitting procedure or the threshold<br />
algorithm for ice-sheet surfaces. This suggests that it is necessary to investigate<br />
further the coastal retracking algorithms, which is the objective of this dissertation.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 74<br />
4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS<br />
4. 1 Introduction<br />
In order to avoid contaminated data in coastal regions, Nerem (1995) and Shum et al.<br />
(1997) in their studies of analysis of global mean sea level and ocean tidal models<br />
over open oceans, simply eliminate altimeter SSH data in areas with water depths<br />
less than 200 m. Shum et al. (1998) show that it is necessary to improve correction<br />
algorithms for data near coastal regions and semi-enclosed seas. Brooks et al. (1990;<br />
1997) investigate land effects on GEOSAT and TOPEX radar altimeter<br />
measurements in the Pacific Rim coastal zone, and provide a slightly modified<br />
threshold retracking algorithm for the data user to rectify and recover the sea surface<br />
topography in close proximity to the land. Although some of these efforts have been<br />
reported previously in the open literature, empirical evidence is not yet published to<br />
show how far away from the shoreline altimeter measurements are poor.<br />
This Chapter uses two approaches of the threshold retracking (Section 3.5.3) and<br />
mean waveform analysis of one cycle of ERS-2 and five cycles of POSEIDON 20<br />
Hz waveform data to quantify the distance from the Australian shoreline that<br />
altimeter measurements are contaminated. This distance is obtained by using the 50<br />
percent threshold retracking point estimates and analysing mean waveform and its<br />
standard deviation across a zone extending out to 350 km offshore from Australia. In<br />
this Chapter, the modelling (and retracking) of the altimeter waveform will not be<br />
addressed. Instead, only the distance from the coastline in which sea surface heights<br />
are poorly estimated by the on-satellite tracking algorithm will be examined.<br />
Quantified distance results will be presented in Section 4.5. In addition, several<br />
examples from waveforms along satellite ground tracks near the coastline are taken<br />
to analyse in detail how waveforms are contaminated and what the characteristics of<br />
contaminated waveforms are in coastal regions (Section 4.6). A preliminary test that<br />
estimates the land effects on waveforms in the vicinity of land will be given in<br />
Section 4.7.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 75<br />
4. 2 Background<br />
When the altimeter pulse interacts with the Earth’s surface, the shape of a returned<br />
waveform depends on the topography of the reflecting surface. Over open oceans,<br />
the altimeter tracker operates as designed and thus uses the Brown (1977) ocean<br />
statistics model (Section 3.2). The on-satellite tracking algorithm can well centre the<br />
ocean return leading edge at a ‘tracking gate’, which is a pre-designed position at the<br />
range window corresponding to the range to the nadir mean sea surface. Over land<br />
and ice sheets, however, the reflection of a satellite radar altimeter signal differs from<br />
the Brown model owing to the non-Gaussian-reflecting surface, where waveforms<br />
are different to the expected/ideal waveform shape over oceans are generally<br />
received (Sections 3.5 and 3.6). This leads to an error in the on-satellite range<br />
estimates that must be corrected through retracking techniques. As stated, retracking<br />
altimeter data aims to estimate the departure of the midpoint on the waveform’s<br />
leading edge from the pre-designed tracking gate, and thus correct the on-satellite<br />
range measurements accordingly (Martin et al., 1983; Bamber, 1994; Berry et al.,<br />
1998).<br />
In coastal regions, the situation differs from open oceans mainly because of the close<br />
proximity to land and the complex spatial-temporal variation of coastal water<br />
surfaces. As a satellite ground track approaches, recedes or runs parallel to the<br />
coastline, even though the altimeter’s nadir point is over the sea, the altimeter may<br />
also track the off-nadir returns reflected from the higher land and any brighter still<br />
inland water. This causes a problem in that the altimeter cannot follow the apparent<br />
accelerations in the surface shape. As a result, the waveforms recorded on board the<br />
satellite will become misaligned with respect to the expected tracking gate (i.e.,<br />
contaminated). Therefore, a postprocessing or retracking is necessary as well on the<br />
on-satellite range calculation (Brooks and Lockwood, 1990; Mantripp, 1996; Quartly<br />
et al., 2001), which will be covered in Chapters 5, 6 and 7. In this Chapter, the spatial<br />
extent of this contamination is quantified so that efficiency can be achieved by<br />
learning over what range to retrack coastal data.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 76<br />
4.2.1 ERS-2 and POSEIDON Altimeter Footprints<br />
The altimeter-illuminated ground region (i.e., along-track footprint) from which most<br />
power returns for each 20 Hz measurement is typically oval in shape and its size<br />
depends on the satellite altitude, surface slope and roughness, and the width of<br />
waveform gates used to compute its diameter (e.g., Chelton et al., 2001, p.36). Two<br />
types of altimeter along-track footprint are important, which are the middle gate and<br />
the automatic gain control (AGC) along-track footprints. The middle gate is the<br />
average of 2, 4, 8 or 16 waveform samples centred on the leading edge of the ocean<br />
return, and the middle-gate along-track footprint is defined by the half-width of the<br />
middle-gate after the tracking gate. For the sea surface with SWH=0 m (i.e., a flat sea<br />
surface) and SWH=15 m, the size of the middle-gate along-track footprint varies<br />
from 8.5 km to 15 km for ERS-2 and 8.5 km to 16 km for POSEIDON. The AGC<br />
gate is the average of waveform samples 17 through 48 centred on the leading edge<br />
of the ocean return. Similarly, the AGC along-track footprint is computed from 15.5<br />
AGC gates after the tracking gate. Its diameter for POSEIDON is about 13.9 km for<br />
a flat sea surface, increasing to about 16.9 km for SWH=15 m (Chelton et al., 2001,<br />
p. 36 and Figure 22).<br />
The on-satellite tracking algorithm compares the returned signal from the middle<br />
gate to that from the AGC gate to estimate the range to nadir sea surface. Since the<br />
AGC footprint is larger than the middle gate footprint, a sharp change in scattering<br />
surface will be first sensed by the AGC gate before the middle gate is affected. In<br />
addition, the footprint diameter estimated by all waveform gates after the tracking<br />
gate is much larger than the above-mentioned sizes, such as 22 km for TOPEX<br />
mission (Brooks et al., 1997). Therefore, it is necessary to consider a larger footprint<br />
when investigating waveform contamination in coasts. However, the exact distance<br />
from the coast at which this occurs is unknown. Hence, this is a secondary<br />
justification for this Chapter.<br />
4.2.2 ERS-2 and POSEIDON Altimeter Waveforms<br />
Figures 1.2 and 4.2 show typical examples of observed ERS-2 and POSEIDON<br />
waveforms over oceans, respectively. As can be seen, they are somewhat different<br />
from the ideal waveform (Figure 1.1), due mainly to reasons of the ‘wrap-around’
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 77<br />
200<br />
150<br />
Power (counts)<br />
100<br />
50<br />
0<br />
0 8 16 24 32 40 48 56 64<br />
Bins or Gates<br />
Figure 4.1 An example of the observed POSEIDON waveform over deep oceans<br />
(tracking gate = 31.5).<br />
error and ‘fading’ (Rayleigh) noise (cf. Partington et al., 1991; Quartly et al., 2001).<br />
The former causes a reduction in power in the last few bins and a commensurate rise<br />
at the beginning of the waveform, and the later causes an undulating trailing edge in<br />
the waveform. The noise affects the leading edge as well, but it is more difficult to<br />
identify in the leading edge and is clearly seen in the trailing edge. In addition,<br />
Figure 1.2 does not show the thermal noise because the ERS-2 altimeter artificially<br />
suppresses this noise (cf. Quartly et al., 2001).<br />
Since altimeter-observed waveforms are different from the Brown model (Figure 1.1),<br />
it is hard to correctly define the tracker point to be tracked by altimeter. Therefore, it<br />
is displaced in the observed waveform by the tracking gate. After subtracting the<br />
noise baseline (i.e., the thermal noise level) from each waveform gate, the tracking<br />
gate is defined by a gate (or bin) number that corresponds to the midpoint on the<br />
leading edge of the returned waveform. On-satellite algorithms try to centre the ramp<br />
of the waveform at the tracking gate, but this generally succeeds only over open<br />
oceans. The tracking gate is specified prior to launch as part of the overall design of<br />
the radar system and corresponds to the mean surface range over the ocean (Brown,
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 78<br />
1977). The tracking gate is chosen to align between waveform sample bins 31-32 for<br />
POSEIDON and bins 32-33 for ERS-2 (e.g., Quartly et al., 2001). The parameters<br />
linked to the reflecting surface can be estimated from the detailed shape of the<br />
leading edge, trailing edge and precise position of the waveform in the sampled<br />
window relative to the altimeter tracking gate location.<br />
4. 3 Processing Methodology for Determination of Contamination Area<br />
Whenever land is within the altimeter footprint, there is a likelihood that the land<br />
return contaminates the waveforms, depending on the height and type of terrain<br />
along the shoreline. In order to quantify the contaminated distance from the shoreline,<br />
it is necessary to calculate waveform anomalies with respect to the ideal waveform<br />
shape. Fitting an ocean-like function to each individual waveform then calculating<br />
the standard deviation of the differences is a reasonable measure of determining if<br />
the waveform contains any departure from the ocean-type response. However, it is<br />
not a good way of calculating a better sea surface height in coastal regions because<br />
the model does not fit the actual return. Therefore, two processing techniques are<br />
used in this Chapter: threshold retracking and arithmetic averaging. The<br />
contaminated area is determined by computing the tracking gate estimates and<br />
analysing the shape of the mean waveform, instead of only focussing upon the<br />
individual waveform shapes. Waveform retracking at the Australian coast will be<br />
covered in Chapter 6.<br />
4.3.1 Threshold Retracking<br />
The threshold retracking technique (see Section 3.5.3) is employed in this Chapter<br />
due to its perceived advantage of sensitivity to the surface topography (Partington et<br />
al., 1989). To estimate a contaminated boundary for coastal waveforms, it is essential<br />
to distinguish whether the returned waveforms are affected by coastal topography<br />
(land and sea-states). Unlike the returned waveform over land or ice, where the<br />
waveform shape is significantly different from that over the ocean, most waveforms<br />
affected by land in coastal regions usually show a combination of water and land<br />
(Brooks et al., 1997). Therefore, if the actual position of the mid-point on the leading
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 79<br />
edge of the ocean waveform can be determined accurately, the purpose of estimating<br />
a contaminated boundary for coastal waveforms will be achieved.<br />
As stated, assuming that the reflecting surface approximates a uniform, spherically<br />
smooth, and diffuse surface with a symmetrical distribution of small scale slopes, the<br />
range from the satellite to the mean surface at nadir corresponds to the location of the<br />
half-power point (i.e., the midpoint) on the leading edge of the waveform (Brown,<br />
1977). These assumptions of statistical homogeneity are usually satisfied over open<br />
ocean surfaces. Thus, the 50% threshold value can be used as the accurate estimate<br />
of the mid-point on the leading edge of the ocean waveform. Since the 50% retracked<br />
position should agree with the tracking gate for ocean waveforms, it may be different<br />
from the tracking gate for the waveform affected by non-ocean surfaces. This can, in<br />
turn, be used to assess the contamination of the waveform in coastal regions. This is<br />
the principle on which the identification of contaminated altimeter data is based in<br />
this study.<br />
When applying the threshold-retracking algorithm, the formulae used to compute A,<br />
W and COG are given by Equations (3.45), (3.46) and (3.47). The 50% threshold<br />
value of the amplitude of the rectangle is used as the estimate of the mid-point on<br />
leading edge of the waveform in this Chapter. It is linearly interpolated between the<br />
bins adjacent to the threshold value at the leading edge of the waveform. Then, its<br />
position at the range window (i.e., the bin or gate number) can be computed from this<br />
estimate. This position is then compared with the expected tracking gate (bins 32-33<br />
for ERS-2) to determine the contaminated distance.<br />
For the ERS-2 altimeter, there is no need to account for the thermal noise effect.<br />
However, attention must be paid to it for the POSEIDON altimeter waveform data.<br />
To overcome this problem, the mean thermal noise estimated by averaging over gates<br />
5-10 is subtracted from the amplitude; the threshold-retracking scheme adopted here<br />
is referenced to 50% of the amplitude above the mean thermal noise level. This<br />
approach has been applied to the POSEIDON data used in this study.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 80<br />
4.3.2 Mean Power of the Waveform<br />
In order to provide a statistically meaningful assessment of the contaminated area, an<br />
averaging procedure was carried out around the Australian coast. This aims to<br />
calculate a broad contaminated area by analysing shape variations between the mean<br />
coastal waveform and the theoretical waveform due to the Brown model. It is likely<br />
that the variation in scattering surfaces within the coastal footprint of the altimeter<br />
creates the error in the individual returned waveforms. Averaging can reduce this<br />
‘noise’ and leave a mean waveform, which should represent the typical shape of the<br />
waveform over the reflecting surfaces (cf. Partington et al., 1989; Quartly et al.,<br />
2001). The mean waveform generated by this averaging technique is then compared<br />
with the theoretical or ideal waveform to estimate the general differences and hence<br />
the contaminated distance from the coastline.<br />
There are a number of possibilities for the variation of the mean waveform shape.<br />
Firstly, the altimeter measurement is both a waveform and a gain measured by the<br />
AGC. The signal processor maintains the value of the AGC gate at a specified value<br />
by adjusting the attenuator in the receiver. Although the AGC applies this variable<br />
attenuation to the return signal, the mean waveform shape will still vary with sharp<br />
changes in the scattering surface. Secondly, the statistical homogeneity property of<br />
the scattering surfaces within coastal areas is changed and probably invalid. Land<br />
returns may contribute to the mean waveform, thus causing a variation. If this is the<br />
case, the variation in different areas can provide the indication of the contaminated<br />
distance from the coast. Finally, the SWH and the satellite off-nadir pointing angle<br />
may cause the variations in the slope of the leading edge and in the trailing edge of<br />
the mean waveform. However, mean waveforms with different slopes and trailing<br />
edges should cross their leading edges at the same position as the tracking gate over<br />
oceans (Brenner et al., 1993; Hayne, 1980). In this case, the contaminated distance<br />
can be obtained from combining the position of the retracking point with the<br />
variation of the mean waveform. Therefore, using the variation of the mean<br />
waveform shape and its standard deviation in different coastal regions should give a<br />
useful indication of the areas in which the waveform data may be contaminated.<br />
For each waveform-sampled bin, the mean power of waveform M (i)<br />
and its<br />
standard deviation STD (i)<br />
are calculated by
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 81<br />
N<br />
1<br />
M<br />
i<br />
( t)<br />
= ∑ Pij<br />
( t)<br />
(4.1)<br />
N<br />
j=<br />
1<br />
STD ( t)<br />
=<br />
i<br />
n<br />
∑<br />
j=<br />
1<br />
( P ( t)<br />
− M<br />
ij<br />
n −1<br />
i<br />
( t))<br />
2<br />
(4.2)<br />
where n is the number of waveforms used in the averaging procedure. Averaging<br />
was carried out over different distance bands with increasing distance from the<br />
Australian coast. These distance bands were used to investigate broad variations in<br />
waveform shape over the spatial-scale of the coastal region and over the time-scale<br />
of a cycle.<br />
4. 4 Data and Editing<br />
4.4.1 ERS-2<br />
One cycle (March to April 1999) of ERS-2 35-day repeat orbit individual waveforms<br />
was used in this study (Section 2.3.2). The geographical distribution of the<br />
observations out to 350 km around the Australian coast is displayed in Figure 4.2.<br />
Each ERS-2 waveform record consists of 1 Hz and 20 Hz data. Geophysical and<br />
environmental corrections are supplied at 1 Hz, and observations, such as the range,<br />
0<br />
location, σ , and waveforms, are supplied at 20 Hz (Section 2.4.2).<br />
Since the purpose of this Chapter is to analyse waveform contamination over coastal<br />
regions, all land returns must be exactly edited out when near the shoreline. For<br />
ERS-2, a flag is set to record the working mode changes based on the location of the<br />
sub-satellite point. When the satellite is over the land, it is flagged as in ice mode.<br />
Therefore, it might be possible to estimate the location of shoreline by observing the<br />
changes of the flag. However, the locations near the Australian coastline are usually<br />
flagged incorrectly (Figure 4.3). This means the flag supplied with the data is not<br />
appropriate for determining the location of the coastline.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 82<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
0˚<br />
0˚<br />
-10˚<br />
-10˚<br />
-20˚<br />
-20˚<br />
-30˚<br />
-30˚<br />
-40˚<br />
-40˚<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
Figure 4.2 ERS-2 ground tracks (35-day repeat orbit, from March to April 1999) to a<br />
distance of 350 km from Australian shoreline (Lambert projection).<br />
0˚<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
0˚<br />
-10˚<br />
-10˚<br />
-20˚<br />
-20˚<br />
-30˚<br />
-30˚<br />
-40˚<br />
-40˚<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
Figure 4.3 The ocean-ice-mode flag in ERS-2 data supplied close to the Australian<br />
shoreline compared to the Wessel (2000) coastline, showing that most flags around<br />
Australia are on the land (Lambert projection).
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 83<br />
Instead, to identify if the sub-satellite point is over ocean or land near the coastal<br />
regions, the DS759.2 (5′×5′ resolution) ocean depth model (Dunbar, 2000) and a<br />
water/land grid (0.5′×0.5′ resolution) derived from GMT high-resolution shoreline<br />
data (Wessel, 2000) have been used together, instead of using the ERS-2 flags (see<br />
Section 2.4).<br />
4.4.2 POSEIDON<br />
The geographical distribution of the POSEIDON coastal observations around<br />
Australia’s coasts are shown in Figure 4.4. A POSEIDON waveform record consists<br />
of 1 Hz and 20 Hz data. However, all data are supplied at 1 Hz except for 20 Hz<br />
waveform measurements (AVISO/Altimeter, 1998). The land/water flag for<br />
POSEIDON is provided by the AVISO in the waveform products. It is directly used<br />
in this study due to its fairly good agreement with the actual shoreline (Section 2.4.3).<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
0˚<br />
0˚<br />
-10˚<br />
-10˚<br />
-20˚<br />
-20˚<br />
-30˚<br />
-30˚<br />
-40˚<br />
-40˚<br />
110˚<br />
120˚<br />
130˚<br />
140˚<br />
150˚<br />
160˚<br />
Figure 4.4 POSEIDON ground tracks (10-day repeat orbit, from January 1998 to<br />
January 1999) to a distance of 350 km from Australian shoreline (Lambert<br />
projection).
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 84<br />
4. 5 Determination of Contamination Area<br />
Before calculating the retracking point and averaging the waveforms, geophysical<br />
and environmental corrections were applied to all altimeter range measurements. The<br />
GSHHS shoreline model (Wessel and Smith, 1996) has been employed for<br />
determining the position of the shoreline with the exception of the POSEIDON data.<br />
The distance referred to below is the shortest distance (on the sphere) between the<br />
point observed on the satellite ground track and the shoreline point from the GSHHS.<br />
Since 1 Hz range measurement is usually taken by averaging 20 waveforms, the<br />
whole 20 Hz data record is kept in the calculation, though some waveforms may be<br />
returned from only the land in coastal regions. The purpose is to analyse how the 1<br />
Hz range data averaged from 20 Hz data are contaminated in coastal regions, since<br />
these 1 Hz data are used by most users from GDRs.<br />
4.5.1 Using 50% of Threshold Retracking Points of ERS-2<br />
The Australian coastal area is subdivided into 10 areas of 15°×15°, which will be<br />
used in Sections 5.6 and 5.8. Within these areas, if the value of the 50% threshold<br />
retracking point is between bins 31-33, the mid-point is considered to be at the<br />
expected location, otherwise the data are considered to be contaminated by land or<br />
coastal sea surface states. The criterion of bins 31-33 is referenced to the statistics of<br />
50% retracking points over ocean areas (i.e., 50-350 km offshore Australia), where<br />
96% of the values of 50% retracking points lie within bins 31-33. This proportion<br />
agrees with that in open oceans (cf. Quartly, 2001). This criterion is used to analyse<br />
the impact from land or coastal scattering surfaces on the individual waveform when<br />
a satellite ground track is within a few kilometres of the shoreline. The calculations<br />
were carried out using ERS-2 data and the results summarised in Table 4.1. Figure<br />
4.5 shows contaminated waveforms along satellite ground tracks in a small, but<br />
representative, coastal area (135° ≤ λ < 150°, -45° ≤ ϕ < -30°) based on these<br />
calculations.<br />
It can be seen from Table 4.1 that although the contaminated waveforms comprise a<br />
small percentage (from 0.2% to 2.9%, depending on different sub-areas), they are all<br />
in close proximity to the coastline. The more contaminated data and the longest<br />
contaminated distances occur off the north Australian coast and off the south
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 85<br />
Table 4.1 Contaminated 1 Hz ocean depths and 20 Hz distances from the nearest shoreline in different<br />
Australian coastal regions for the ERS-2 waveform data (20Hz, March to April 1999).<br />
a<br />
Areas<br />
Max.<br />
Distance a<br />
(km)<br />
Min.<br />
Distance<br />
(km)<br />
Mean<br />
Distance<br />
(km)<br />
Max.<br />
Ocean<br />
Depth b<br />
(m)<br />
% of<br />
Contaminated<br />
Waveforms c<br />
Coastal Types d<br />
120°≤λ
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 86<br />
Figure 4.5 One cycle of ERS-2 contaminated waveforms (highlighted by circles)<br />
along ground tracks in the southeast coasts of Australia.<br />
Australian coast, near Tasmania. There are two reasons for this. The first is the<br />
complex shoreline in these coastal regions. The second is that there are more satellite<br />
ground tracks in north and south of Australia because of the inclination of the<br />
satellite orbit with respect to the Australia landmass. The eastern and western<br />
Australian coasts show less contaminated data and a shorter contaminated distance.<br />
Overall however, the shortest and the longest contaminated distances from the<br />
coastline are 0.02 km and 21.76 km, respectively; and the mean contaminated<br />
distance is 6.85 km.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 87<br />
It can be seen from Figure 4.2 that some ERS-2 ground tracks are not available for<br />
this cycle near the Australian coast. This is another (lesser) reason why a small<br />
percentage of waveforms appears to be affected to the east and west. Thus, it is<br />
necessary to estimate a broad boundary around Australia, which contains the most of<br />
contaminated data, by using waveform data from both ERS-2 and POSEIDON<br />
altimeters.<br />
The ocean depths interpolated from an Australian 30" bathymetric model (Buchanan,<br />
1991) for these contaminated data range from 12.2 m to 361.5 m (Table 4.1). From<br />
Table 4.1, it is not evident that the contaminated distance correlates with the ocean<br />
depth, while 50% retracking points show some correlation with ocean depths (Figure<br />
4.6). Also, correlation with depth varies from one area to another. Waveforms over<br />
areas where the ocean depth is less than 50 m (Figure 4.6a) and<br />
100 m (Figure 4.6c) appear to be highly contaminated, while some contaminated data<br />
are found in the deeper (up to ~300 m) ocean (Figure 4.6b). Therefore, if shallow<br />
water data close to the shoreline with water depth less than 200 m are simply<br />
removed (e.g., Nerem, 1995; Shum et al., 1998), some valuable data are probably<br />
omitted. Moreover, some contaminated data may still be kept in the data set in<br />
coastal regions while they should be removed.<br />
The rightmost column in Table 4.1 describes the general nature of the coastal terrain<br />
(cf. Thom, 1984; Kelleher et al., 2002). According to Brooks et al. (1997) and<br />
Brooks (2002), the contaminated distance should be correlated with the coastal<br />
topography, in which the correlation is not only related to the height of the coastal<br />
terrain, but also the slope of the terrain.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 88<br />
50% retracking points<br />
64<br />
7a<br />
48<br />
32<br />
16<br />
0<br />
-400 -350 -300 -250 -200 -150 -100 -50 0<br />
Ocean Depth (m)<br />
50% Retracking Points<br />
64<br />
48<br />
7b<br />
32<br />
16<br />
0<br />
-400 -350 -300 -250 -200 -150 -100 -50 0<br />
Ocean Depth (m)<br />
50% Retracking Points<br />
64<br />
7c<br />
48<br />
32<br />
16<br />
0<br />
-400 -350 -300 -250 -200 -150 -100 -50 0<br />
Ocean Depth (m)<br />
Figure 4.6 Distributions of 50% retracking points plotted versus DS759.2 ocean<br />
depths for contaminated waveforms in areas of 105°≤λ
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 89<br />
4.5.2 Using the Mean Waveform and its Standard Deviation<br />
The mean waveforms, and the standard deviation of these means, were computed<br />
over three distance bands in coastal regions, which range in distance of 0-350 km, 0-<br />
50 km, and 0-10 km off the Australian coast. Each has been divided into several<br />
annuli with intervals of 50 km, 10 km and 2 km, respectively. The time span of the<br />
data used to average is a cycle (35 days for ERS-2 and 10 days for POSEIDON). For<br />
POSEIDON, similar results were obtained from five cycles. Therefore, only the<br />
results from cycle 197, which are representative of all the data examined, are shown<br />
here.<br />
(a) 0-350 km off the Shoreline<br />
Figures 4.7 and 4.8 show averaged waveforms and standard deviations, respectively,<br />
in the distance-bands of 0-350 km with the interval of 50 km around the Australian<br />
coast for ERS-2 and POSEIDON. It is important to note that the coastal effects on<br />
the waveforms come from both Australia and Indonesia when the data are taken to<br />
the north of Australia. Therefore, the data thought to be affected by land near<br />
Indonesia have not been used to calculate or plot the mean waveform. Despite the<br />
different number of measurements in each distance band, the mean waveforms are<br />
generally in agreement with the theoretical model (Figure 1.1). The slope of the<br />
leading edge of the ERS-2 and POSEIDON mean waveforms is similar to that of the<br />
theoretical model. However, it can be seen that the trailing edges show some slight<br />
departure feature from the theoretical model (Figure 1.1) caused by fading noise and<br />
the wrap-around feature for both altimeters (cf. Figure 1.2 and Figure 4.1).<br />
The standard deviations of individual waveforms with respect to the mean waveform<br />
for ERS-2 show a peak in the waveform bins 38 and 39 and two ramps at the leading<br />
edge. Since this occurs in all distance bands, it seems to be a systematic anomaly.<br />
The reason for this should be investigated from more ERS-2 altimeter waveform data.<br />
Compared with ERS-2, the standard deviation of the POSEIDON data shows a<br />
smoother shape (Figure 4.8). Figures 4.7 and 4.8 show a large standard deviation in<br />
the distance band of 0-50 km compared with other standard deviations, implying that<br />
there is more contamination in this area. This will be studied next.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 90<br />
Mean<br />
Wavwform<br />
(counts)<br />
1200<br />
900<br />
600<br />
300<br />
0<br />
64<br />
55<br />
46<br />
37<br />
Bins<br />
28<br />
19<br />
10<br />
1<br />
0- 50 km<br />
50-100 km<br />
100-150 km<br />
150-200 km<br />
200-250 km<br />
250-300 km<br />
300-350 km<br />
STD (counts)<br />
1200<br />
900<br />
600<br />
300<br />
0<br />
64<br />
55<br />
46<br />
37<br />
Bins<br />
28<br />
19<br />
10<br />
1<br />
0- 50 km<br />
50-100 km<br />
100-150 km<br />
150-200 km<br />
200-250 km<br />
250-300 km<br />
300-350 km<br />
Figure 4.7 The mean waveform (top) and the standard deviation of the mean<br />
waveform (bottom) for 1 cycle of ERS-2 data in seven 50 km-wide<br />
bands around the Australia’s coastline.<br />
Mean<br />
Waveform<br />
(counts)<br />
200<br />
150<br />
100<br />
50<br />
0<br />
55<br />
46<br />
Bins<br />
37<br />
28<br />
19<br />
10<br />
1<br />
0- 50 km<br />
50-100 km<br />
100-150 km<br />
150-200 km<br />
200-250 km<br />
250-300 km<br />
300-350 km<br />
STD (counts)<br />
200<br />
150<br />
100<br />
50<br />
0<br />
55<br />
46<br />
Bins<br />
37<br />
28<br />
19<br />
10<br />
1<br />
0- 50 km<br />
50-100 km<br />
100-150 km<br />
150-200 km<br />
200-250 km<br />
250-300 km<br />
300-350 km<br />
Figure 4.8 The mean waveform (top) and the standard deviation of the mean<br />
waveform (bottom) for POSEIDON cycle 197 in seven 50 km-wide<br />
bands around the Australia’s coastline.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 91<br />
(b) 0-50 km off the Shoreline<br />
Mean and standard deviations in the distance bands of 0-50 km with an interval of 10<br />
km are shown in Figures 4.10 and 4.11 for ERS-2 and POSEIDON, respectively. For<br />
both altimeters, the leading edges of the mean waveforms in the bands of 10-20 km,<br />
20-30 km, 30-40 km and 40-50 km have a similar slope of leading edge, but this<br />
becomes less in band 0-10 km compared with the others.<br />
From Figure 4.9, the ERS-2 standard deviations for 30-40 km and 40-50 km agree<br />
well with these beyond 50 km (cf. Figures 4.8 and 4.9), while other standard<br />
deviations also show similar shapes, even if there are some larger values around the<br />
leading edge in other distance bands. The standard deviations for the 0-10 km band<br />
are the largest, with a peak value of 1288 counts. From Figure 4.10 for POSEIDON,<br />
the large standard deviation is evident in the 0-10 km band, especially close to the<br />
tracking gate near the peak of the plot.<br />
Mean Power of Waveform (counts)<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
STD (counts)<br />
1600<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
0−10 km<br />
N=9370<br />
10−20 km<br />
N=14202<br />
20−30 km<br />
N=14998<br />
30−40 km<br />
N=13995<br />
40−50 km<br />
N=12514<br />
Figure 4.9 Mean power of the waveform (left) and the standard deviation of the mean<br />
waveform (right) for ERS-2 in five 10 km-wide bands from the Australian coastline.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 92<br />
Mean Power of Waveform (counts)<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
STD (counts)<br />
60<br />
40<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
60<br />
40<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
60<br />
40<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
60<br />
40<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
60<br />
40<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
0−10 km<br />
N=4400<br />
10−20 km<br />
N=5160<br />
20−30 km<br />
N=3660<br />
30−40 km<br />
N=4620<br />
40−50 km<br />
N=4060<br />
Figure 4.10 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for POSEIDON cycle 197 in five 10 km-wide bands from the<br />
Australian coastline.<br />
(c) 0-10 km off the Shoreline<br />
Figures 4.11 and 4.12 show mean waveforms and their standard deviations 0-10 km<br />
off the shoreline with an interval of 2 km for ERS-2 and POSEIDON, respectively. It<br />
can be seen from Figure 4.11 that mean ERS-2 waveforms depart from the<br />
theoretical waveform in distance bands of 0-2 and 2-4 km. In the band of 8-10 km,<br />
the mean waveform agrees well with the theoretical or ocean waveform and with<br />
others at greater distances (cf. Figures 4.7, 4.9 and 4.11). It is evident that the<br />
standard deviation is large in gates for all distance bands. The variation of the<br />
standard deviation in the 0-2 km and 2-4 km distance bands is very large, up to about<br />
3000 counts in the peak.<br />
From Figure 4.12, it is obvious that nearly the same conclusion for the mean<br />
POSEIDON waveforms as for ERS-2 can be obtained. However, POSEIDON’s<br />
standard deviation does not show the obvious large values that the ERS-2 does in the<br />
4-6 km and 6-8 km bands. The largest standard deviation is also in the distance band
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 93<br />
0-2 km and the next largest is in the 2-4 km band. The standard deviation of the<br />
mean waveform increases with decreasing distance.<br />
From Figures 4.11 and 4.12, it can be seen that the larger standard deviation from the<br />
bands of 0-2 km and 2-4 km is dominant, which implies that this is the area of main<br />
contamination of coastal waveforms, but some lesser contamination also occurs in<br />
the other bands as well.<br />
The above results suggest that it is possible to estimate a typical contaminated area<br />
around coasts by simply analysing the variations of the mean waveform from the<br />
ideal and the standard deviation of the mean. Using this approach around Australia<br />
shows that the waveform data will mainly be contaminated when the ground track is<br />
within a distance of about ~4 km from the coast, while there is some smaller level of<br />
contamination to about 4-8 km for POSEIDON and 4-10 km for ERS-2.<br />
Mean Power of Waveform (counts)<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
STD (counts)<br />
3000<br />
2000<br />
1000<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
1200<br />
800<br />
400<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
0−2 km<br />
N=1098<br />
2−4 km<br />
N=1495<br />
4−6 km<br />
N=1826<br />
6−8 km<br />
N=2323<br />
8−10 km<br />
N=2505<br />
Figure 4.11 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for ERS-2 in five 2 km-wide bands from the Australian<br />
coastline.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 94<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
0−2 km<br />
N=560<br />
Mean Power of Waveform (counts)<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
STD (counts)<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
2−4 km<br />
N=740<br />
4−6 km<br />
N=840<br />
6−8 km<br />
N=1000<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
8−10 km<br />
N=1260<br />
Figure 4.12 Mean power of the waveform (left) and the standard deviation of the<br />
mean waveform (right) for POSEIDON cycle 197 in five 2 km-wide bands from the<br />
Australian coastline.<br />
4.5.3 Using Statistics of the 50% Retracking Point<br />
Statistics of the 50% threshold retracking points have been computed using both<br />
POSEIDON and ERS-2 waveform data in a coastal area of 10 km from the<br />
Australian shoreline. This distance is subdivided into 5 bands with an interval of<br />
2 km. As stated, if there is no effect from land and coastal sea states on the returned<br />
waveform, the threshold value of 50% retracking point should be very close to the<br />
pre-designed tracking gate when it is over oceans. Five cycles of POSEIDON<br />
waveform data and 1 cycle of ERS-2 waveform data are used to compute the 50%<br />
threshold retracking point in the above bands.<br />
Figure 4.13 shows the 50% retracking point distribution for these waveform data. For<br />
POSEIDON, the 50% retracking points comprise 15%, 50% and 59% in bins 30-32<br />
for bands of 0-2 km, 6-8 km and 8-10 km, respectively. For ERS-2, they are 20%,<br />
45% and 45% in bins 31-33 for bands of 0-2 km, 6-8 km and 8-10 km, respectively.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 95<br />
POSEIDON has a higher percentage than that of ERS-2 near the tracking gate with<br />
increasing distance. It was found from the calculation that 94% of POSEIDON<br />
waveforms represent their mid-point of leading edge very close to the tracking gate<br />
in bins 30-32 when beyond 10-20 km of the shoreline, while 84% of ERS-2 data<br />
show in bins 31-33 for the same distance.<br />
% of observations<br />
Poseidon<br />
60<br />
40<br />
0−2 km N=2814<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
2−4 km N=3349<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
4−6 km N=4207<br />
40<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
6−8 km N=4895<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
8−10 km N=6187<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
% of observations<br />
ERS−2<br />
60<br />
40<br />
0−2 km N=1075<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
2−4 km N=1457<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
4−6 km N=1830<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
6−8 km N=2316<br />
40<br />
20<br />
0<br />
0<br />
60<br />
10 20 30 40 50 60<br />
40<br />
8−10 km N=2504<br />
20<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
Figure 4.13 Fifty percent retracking point frequency distributions for 5 cycles of<br />
Poseidon waveform data (left) and 1 cycle of ERS-2 waveform data (right). They are<br />
shown as the percentage of observations in each bin for five 2 km-wide bands around<br />
the Australian coastline.<br />
Tables 4.2 and 4.3 show the statistics of the differences between 50% retracking<br />
point and the tracking gate for POSEIDON and ERS-2, respectively. The 50%<br />
retracking points are obviously offset from the tracking gate by 2.5±5.9 bins for<br />
POSEIDON and by 2.2±6.3 bins for ERS-2 in the band of 0-2 km. These offsets lead<br />
to a mean range difference of about 1m for both altimeters. POSEIDON’s mean
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 96<br />
range difference drops below 10 cm in the 6-8 km band, whereas ERS-2 does not<br />
decrease to below 10 cm until 10-20 km from the shoreline.<br />
Table 4.2 Differences between 50% retracking points and the tracking gate (bin 31.5)<br />
for 5 cycles (January 1998 to January 1999) of Poseidon waveforms (each bin has<br />
the range distance of ~0.469 m).<br />
Band<br />
(km)<br />
Mean<br />
(bins)<br />
Range difference<br />
(cm)<br />
RMS<br />
(bins)<br />
STD<br />
(bins)<br />
Number of<br />
20 Hz Points<br />
0 – 2 2.46 115 ±5.94 ±5.40 2814<br />
2 – 4 1.14 53 ±3.72 ±3.54 3349<br />
4 – 6 -0.38 -18 ±2.49 ±2.46 4207<br />
6 – 8 -0.15 7 ±1.67 ±1.66 4895<br />
8 - 10 0.10 5 ±0.96 ±0.96 6187<br />
10 - 20 0.14 6 ±0.72 ±0.74 23820<br />
Table 4.3 Differences between 50% retracking points and the tracker point (bin 32.5)<br />
for 1 cycle (March to April 1999) of ERS-2 waveforms (each bin with the range<br />
width of ~0.454m).<br />
Band<br />
(km)<br />
Mean<br />
(bins)<br />
Range difference<br />
(cm)<br />
RMS<br />
(bins)<br />
STD<br />
(bins)<br />
Number of<br />
20 Hz Points<br />
0 – 2 2.20 100 ±6.32 ±5.93 1075<br />
2 – 4 0.27 12 ±5.82 ±5.81 1457<br />
4 – 6 -0.47 -21 ±4.74 ±4.72 1830<br />
6 – 8 -0.29 -13 ±3.27 ±3.25 2316<br />
8 - 10 -0.23 -10 ±2.56 ±2.55 2504<br />
10 - 20 -0.14 -6 ±1.94 ±1.95 14224<br />
From Figure 4.13 and the values in Tables 4.2 and 4.3, this 50% retracking analysis<br />
indicates that a distance band of ~8 km and ~10 km are evidently the contaminated<br />
area for POSEIDON and ERS-2, respectively. The effect even extends to ~10-20 km<br />
for some ERS-2 data. Thus, there is a clear difference between POSEIDON and<br />
ERS-2 results from this analysis. It seems that contaminated distance for POSEIDON<br />
is shorter than that for ERS-2.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 97<br />
Finally, from reviewing the waveform products, it was found that ERS-2 provides<br />
20 Hz ground locations for 20 Hz ground waveforms, while POSEIDON provides<br />
1 Hz locations for 20 Hz waveforms. Thus, if a location for POSEIDON is given in a<br />
band, the waveforms 3 km either side of the band also contribute to the calculation in<br />
the band. This means that the results in Table 4.2 are actually computed for a larger<br />
band than 2 km, which is a plausible explanation for the different contaminated<br />
distances, as well as the different sized footprint and altimetric characteristics (also<br />
see Section 4.2.1).<br />
4. 6 Waveform Contamination Analysis<br />
The waveform contamination from near the coastline differs among the altimeter<br />
ground tracks, depending on whether the groundtrack is crossing the shoreline from<br />
water to land or from land to water (e.g., Brooks et al., 1997). In this Section,<br />
examples of ERS-2 altimeter ground tracks at the Australian coast (two water-to-land<br />
and two land-to-water) are analysed to understand the effects of the coastal<br />
topography on waveforms. Both SSH profiles (without retracking) and relevant<br />
waveforms are used for this analysis. Unlike the previous Section, distances in this<br />
Section are defined as the along-track distance to the coastline. The analysis in this<br />
Section focuses on the waveform itself. A comparison of the altimeter-derived SSH<br />
with the geoid heights will be given in Sections 1.2.1 and 5.8.2 to support the results<br />
obtained in this Section.<br />
4.6.1 Waveform Contamination for Water-to-Land Ground Tracks<br />
(a) Tasman Sea (Eastern Australia)<br />
As an ERS-2 altimeter groundtrack from cycle 43 (ascending pass 21792)<br />
approaches land northwestward (Figure 4.14) in the Tasman Sea, the 20 Hz SSH<br />
profile initially shows small but smooth undulation, and then rises greatly in the five<br />
late SSH samples near the shoreline. These relevant 20 successive waveforms are<br />
plotted via the bins (or gates) in Figure 4.15, which correspond to a ~7 km alongtrack<br />
distance. The locations of waveforms in Figure 4.15 correspond to the last 20<br />
SSH measurements numbered in Figure 4.14.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 98<br />
Land - 20<br />
SSH (m)<br />
30<br />
Tasman Sea<br />
- 15<br />
- 10<br />
- 5<br />
-30˚ 18'<br />
153˚ 06' 153˚ 09' 153˚ 12'<br />
- 1<br />
-30˚ 21'<br />
-30˚ 24'<br />
Figure 4.14 The 20 Hz SSH profile of pass 21792 (ascending track, cycle 43 of<br />
ERS-2), approaching the eastern Australian shoreline from the Tasman Sea. The<br />
number beside the SSH samples corresponds to the waveform in Figure 4.15.<br />
1000<br />
1<br />
1000<br />
2<br />
1000<br />
3<br />
1000<br />
4<br />
Counts<br />
500<br />
500<br />
500<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
5<br />
1000<br />
500<br />
6<br />
1500<br />
1000<br />
500<br />
7<br />
1000<br />
500<br />
8<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
9<br />
1000<br />
500<br />
10<br />
1000<br />
500<br />
11<br />
1000<br />
500<br />
12<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
13<br />
2000<br />
1000<br />
14<br />
1500<br />
1000<br />
500<br />
15<br />
2000<br />
1000<br />
16<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
2000<br />
1000<br />
17<br />
2000<br />
1000<br />
18<br />
4000<br />
2000<br />
19<br />
8000<br />
6000<br />
4000<br />
2000<br />
20<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.15 20 Hz waveforms of pass 21792 (ascending track, cycle 43 of ERS-2),<br />
approaching Australian eastern shoreline in the Tasman Sea. The number in the top<br />
left-hand corner is related to that in Figure 4.14.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 99<br />
Waveform one in Figure 4.15 occurs when the groundtrack is ~5.4 km from land,<br />
and the land reflection does not yet appear. The land reflections appear clearly first in<br />
the gate (or bin) near gate 45 in the waveform 14 (Figure 4.15), then moving<br />
gradually closer to the tracking gate 32.5 in the successive waveforms 15-20. In the<br />
16th waveform (Figure 4.15), the ocean return is discernible near the tracking gate,<br />
while the land returns begin near the tracking gate. As the groundtrack approaches<br />
the land, the ocean return remains invisible on the leading edge and appears in the<br />
late gates, while the land returns are around the tracking gate in the range window.<br />
Waveforms 1 to 13 (Figure 4.15) basically show the ocean waveform characteristics<br />
with a clear single ramp. From waveforms 14 to 18, a single peak can be observed,<br />
but the shape is different from the open-ocean waveform. The last two (19 and 20)<br />
have a narrow width and significant peaks of the power, showing sharply rising and<br />
descending shape. Beginning with the waveform 16, the altimeter tracker appears to<br />
be responding more to the off-nadir higher and brighter land return than the at-nadir<br />
water return. As a result, the measured range is too short, and calculated at-nadir sea<br />
surface heights corresponding to waveforms 16-20 will be biased too high (Figure<br />
4.14).<br />
(b) Indian Ocean (Western Australia)<br />
Another example of an ERS-2 altimeter groundtrack (cycle 43 and ascending pass<br />
21450) approaching the land is shown in Figure 4.16, where the coast is dominated<br />
by sand dunes (Thom, 1984). In the area where the groundtrack intersects the<br />
shoreline, the land elevation in this area is between ~0-65 m above mean sea level<br />
from the Australian 9" DEM (version 2, Hutchinson, 2001, described in Section<br />
2.5.4). The SSH profile (related to a ~80 km along-track distance) shows no apparent<br />
change in Figure 4.16 when the groundtrack is offshore further than ~3.2 km (before<br />
the SSH sample 10). As the groundtrack gets closer to the shoreline after the tenth<br />
SSH samples, the SSH begins to rise.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 100<br />
SSH (m)<br />
-25<br />
-26˚ 24'<br />
20 -<br />
Land<br />
10 -<br />
-26˚ 27'<br />
1 -<br />
-26˚ 30'<br />
-26˚ 33'<br />
-26˚ 36'<br />
Indian<br />
Ocean<br />
-26˚ 39'<br />
113˚ 18' 113˚ 21'<br />
Figure 4.16 The 20 Hz SSH profile of pass 21450 (ascending track, cycle 43 of<br />
ERS-2), approaching the western Australian shoreline from the Indian Ocean. The<br />
number beside the SSH samples corresponds to the waveform in Figure 4.17.<br />
Counts<br />
10000<br />
5000<br />
1<br />
1000<br />
500<br />
2<br />
1000<br />
500<br />
3<br />
1000<br />
500<br />
4<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1500<br />
1000<br />
500<br />
5<br />
1000<br />
500<br />
6<br />
1000<br />
500<br />
7<br />
1000<br />
500<br />
8<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1500<br />
1000<br />
500<br />
9<br />
1500 10<br />
1000<br />
500<br />
1000<br />
500<br />
11<br />
1500<br />
1000<br />
500<br />
12<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
13<br />
1500<br />
1000<br />
500<br />
14<br />
1000 15<br />
500<br />
1000<br />
500<br />
16<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
17<br />
2000<br />
1000<br />
18<br />
4000<br />
2000<br />
19<br />
4000<br />
2000<br />
20<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.17 The 20 Hz waveforms of pass 21450 (ascending track, cycle 43 of ERS-<br />
2), approaching Australian western shoreline in the Indian Ocean. The number is<br />
related to that in Figure 4.16.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 101<br />
Figure 4.17 shows 1-second (20 Hz) waveforms related to the last 20 SSHs in<br />
Figure 4.16. The distance to the coastline from the location of the first waveform is<br />
~3.2 km. The first waveform shows a specular type with a high peak of ~10,000<br />
counts, implying that it is reflected from calm or standing water. As a result, the<br />
relative SSH (the first one in Figure 4.17) is measured too short, because the range is<br />
too long. The following waveforms (two to nine) show ocean-like shapes, but do not<br />
contain obvious land returns until the tenth waveform. The higher coastal land does<br />
not show significant effects on the waveforms. This is because the range window of<br />
64 gates is only related to a difference between surface elevations no larger than<br />
29 m for ERS-1/2. Therefore, the land topography more than ~30 m with respect to<br />
mean sea level around the groundtrack at this area does not appear as obvious land<br />
returns in the waveform range window.<br />
However, from waveforms 10 to 17 (Figure 4.18), the land returns appear to have<br />
affected the altimeter tracker response, and the waveforms have moved slightly to the<br />
left with respect to the tracking gate in the range window. In the last three waveforms<br />
(18-20) in Figure 4.17, the ocean returns have moved to the later gates and the land<br />
returns moved close to the tracking gate. This means that the shorter range<br />
measurement is obtained from the land returns reflected from higher surfaces. As a<br />
result, the calculated SSH will be too high.<br />
4.6.2 Waveform Contamination for Land-to-Water Ground Tracks<br />
(c) Indian Ocean (Northwestern Australia)<br />
Figure 4.18 shows an ERS-2 20 Hz SSH profile from an ascending pass 21636 (cycle<br />
43), which is going from land to water northwestward. This is a tidal plain coast area<br />
(Thom, 1984). The land-to-water crossover occurs at the 16th SSH in Figure 4.18. At<br />
this time, the terrain is higher than the ocean with respect to the radar pulse front, and<br />
the land reflections appear dominantly in the waveforms so that the SSH results are,<br />
probably, calculated from the land returns. As the satellite goes further from the<br />
shoreline, because of the off-nadir angle, the altimeter still measures the closest slant<br />
range to the higher land. As a result, in this case the effects of land on the calculated<br />
SSHs last nearly three seconds (~21 km along-track distance).
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 102<br />
The three-second successive waveforms, which correspond to the SSHs from 1-60 in<br />
Figure 4.18, have been plotted via the bins or gates in Figures 4.20, 4.21 and 4.22.<br />
Most waveforms seem to be specular due to still water in Figure 4.20. From the first<br />
to fifth waveforms, still water produces waveforms with higher peaks and narrow<br />
width. Lower power signals, which begin around the 40th gate from the sixth to 19th<br />
waveforms, are ocean returns as the groundtrack crosses the shoreline from the land<br />
to water.<br />
Beginning with waveform 20 in Figure 4.19, the initial radar returns are from the sea<br />
surface and the land returns move towards the late gates. Between waveforms 28 to<br />
33 in Figure 4.20, the off-nadir terrain and at-nadir sea surfaces are approximately<br />
equidistant from the altimeter, producing waveforms with double ramps. By<br />
waveform 34 when the groundtrack is 3.6 km to the shoreline (Figure 4.20), the land<br />
returns have disappeared.<br />
Although the sea surface returns are dominant in waveforms 34 through 40 in Figure<br />
4.20 and from 41 to 54 in Figure 4.21, these waveforms are misaligned with respect<br />
to the pre-designed tracking gate (32.5) such that the measured ranges are too long,<br />
and the calculated SSHs are too low (see Figure 4.18). In the rest of waveforms from<br />
55 to 60 in Figure 4.21, the half power point shifts slightly to the left with respect to<br />
the tracking gate. After three seconds (about ~21 km along the track or ~6 km from<br />
shore), waveforms become centred in the range window (i.e., at the tracking gate).<br />
(d) Southern Ocean (Southern Australia)<br />
The 20 Hz SSH profile from cycle 43 descending pass 21886 traverses<br />
southwestward across the shoreline from land to the Southern Ocean is shown in<br />
Figure 4.22. There is low land elevation along the coastline (Hutchinson et al., 2001).<br />
The SSH profile drops fast after the groundtrack enters the water, and then rises<br />
gradually. It changes to the normal ocean waveform and thus normal SSH after<br />
~2.5 seconds (~17 km from the shoreline) in this example.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 103<br />
SSH (m)<br />
-20<br />
Indian Ocean<br />
-22˚ 03'<br />
-22˚ 06'<br />
-22˚ 09'<br />
-22˚ 12'<br />
60 -<br />
-22˚ 15'<br />
50 -<br />
-22˚ 18'<br />
40 -<br />
30 -<br />
20 -<br />
10 -<br />
1 -<br />
Land<br />
114˚ 24' 114˚ 27' 114˚ 30'<br />
Figure 4.18 The 20 Hz SSH profile of pass 21636 (ascending track, cycle 43 of<br />
ERS-2), receding from the northwestern Australian shoreline to the Indian Ocean.<br />
x 10 4<br />
Counts<br />
15000<br />
10000<br />
5000<br />
1<br />
2<br />
1<br />
2<br />
2000<br />
1000<br />
3<br />
15000<br />
10000<br />
5000<br />
4<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
3000<br />
2000<br />
1000<br />
5<br />
1000<br />
500<br />
6<br />
1500 7<br />
1000<br />
500<br />
1500<br />
1000<br />
500<br />
8<br />
Counts<br />
2000<br />
1000<br />
0<br />
0 16 32 48 64<br />
9<br />
2000<br />
1000<br />
0<br />
0 16 32 48 64<br />
10<br />
8000<br />
6000<br />
4000<br />
2000<br />
0<br />
0 16 32 48 64<br />
11<br />
6000<br />
4000<br />
2000<br />
0<br />
0 16 32 48 64<br />
12<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
6000<br />
4000<br />
2000<br />
13<br />
3000 14<br />
2000<br />
1000<br />
2000<br />
1000<br />
15<br />
1000<br />
500<br />
16<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
17<br />
1000<br />
500<br />
18<br />
1500<br />
1000<br />
500<br />
19<br />
1000<br />
500<br />
20<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.19 20 Hz waveforms of pass 21636 at the 1st second (~0-7 km along-track<br />
distance) from the shoreline (ascending track, cycle 43 of ERS-2), receding from the<br />
northwestern Australian shoreline to the Indian Ocean.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 104<br />
Counts<br />
1000<br />
500<br />
21<br />
1500<br />
1000<br />
500<br />
22<br />
1500 23<br />
1000<br />
500<br />
1500 24<br />
1000<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
2000<br />
1000<br />
25<br />
2000 26<br />
1000<br />
1500 27<br />
1000<br />
500<br />
1500 28<br />
1000<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1500<br />
1000<br />
500<br />
29<br />
1500<br />
1000<br />
500<br />
30<br />
1000 31<br />
500<br />
1500<br />
1000<br />
500<br />
32<br />
Counts<br />
1500<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
33<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
34<br />
0<br />
0 16 32 48 64<br />
1000 35<br />
500<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
36<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
37<br />
0<br />
0 16 32 48 64<br />
Bins<br />
1000<br />
500<br />
38<br />
0<br />
0 16 32 48 64<br />
Bins<br />
800<br />
600<br />
400<br />
200<br />
39<br />
0<br />
0 16 32 48 64<br />
Bins<br />
1000<br />
500<br />
40<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.20 20 Hz waveforms of pass 21636 at the 2nd second (~7-14 km alongtrack<br />
distance) from the shoreline (ascending track, cycle 43 of ERS-2), receding<br />
from northwestern Australian shoreline to the Indian Ocean.<br />
Counts<br />
800 41<br />
600<br />
400<br />
200<br />
0<br />
0 16 32 48 64<br />
1000 42<br />
500<br />
0<br />
0 16 32 48 64<br />
1000 43<br />
500<br />
0<br />
0 16 32 48 64<br />
1500<br />
1000<br />
500<br />
44<br />
0<br />
0 16 32 48 64<br />
1000<br />
45<br />
1000 46<br />
1000<br />
47<br />
1000 48<br />
Counts<br />
500<br />
500<br />
500<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
49<br />
1000 50<br />
500<br />
1000 51<br />
500<br />
1000<br />
500<br />
52<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
53<br />
1000<br />
500<br />
54<br />
1000<br />
500<br />
55<br />
1000<br />
500<br />
56<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
57<br />
1000<br />
500<br />
58<br />
1000<br />
500<br />
59<br />
1000<br />
500<br />
60<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.21 20 Hz waveforms of pass 21636 at the 3rd second (~14-21 km alongtrack<br />
distance) from the shoreline (ascending track, cycle 43 of ERS-2), receding<br />
from the northwestern Australian shoreline to the Indian Ocean.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 105<br />
When the groundtrack exits the land, the first three seconds of 60 waveforms related<br />
to the SSH profile have been plotted in Figures 4.24, 4.25 and 4.26. The first five<br />
waveforms in Figure 4.23 show mostly the response to the land reflections. After<br />
waveform six, which is at a distance of ~2.1 km from the shoreline, the leading edges<br />
of the waveforms shift to the earlier gates. This leftward shift continues to the<br />
waveform 25 in Figure 4.24. After that, the waveforms move back gradually towards<br />
the alignment of the ocean waveform through the rest of waveforms in Figure 4.24<br />
and all waveforms in Figure 4.25. Waveforms after 54 in Figure 4.25, when the<br />
ground track is more than ~17.9 km from shore, become stably aligned with respect<br />
to the tracking gate, appearing to be normal open-ocean returns.<br />
Waveforms in this example do not show obvious high power and the land returns<br />
except for the first six waveforms in Figure 4.23, because there is a beach area here<br />
without the steep slope and sharp changes of the topography (Thom, 1984). However,<br />
apparent waveform contamination still lasts about three seconds (~21 km along-track<br />
distance). This will be discussed in Section 4.6.4<br />
Land<br />
-15<br />
- 20<br />
- 10<br />
- 1<br />
-32˚ 18'<br />
SSH (m)<br />
- 60<br />
- 50<br />
- 40<br />
- 30<br />
-32˚ 21'<br />
-32˚ 24'<br />
Southern Ocean<br />
126˚ 45' 126˚ 48'<br />
-32˚ 27'<br />
-32˚ 30'<br />
-32˚ 33'<br />
Figure 4.22 The 20 Hz SSH profile of pass 21886 (descending track, cycle 43 of<br />
ERS-2), receding from the southern Australian shoreline to the Southern Ocean.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 106<br />
4000<br />
1<br />
4000 2<br />
4000<br />
3<br />
3000 4<br />
Counts<br />
2000<br />
2000<br />
2000<br />
2000<br />
1000<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
4000<br />
5<br />
3000 6<br />
2000 7<br />
2000 8<br />
Counts<br />
2000<br />
2000<br />
1000<br />
1000<br />
1000<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1500<br />
1000<br />
500<br />
9<br />
1000<br />
500<br />
10<br />
1000 11<br />
500<br />
1000<br />
500<br />
12<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
13<br />
0<br />
0 16 32 48 64<br />
1000<br />
500<br />
14<br />
0<br />
0 16 32 48 64<br />
600 15<br />
400<br />
200<br />
0<br />
0 16 32 48 64<br />
800<br />
600<br />
400<br />
200<br />
16<br />
0<br />
0 16 32 48 64<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
17<br />
600<br />
400<br />
200<br />
18<br />
600<br />
400<br />
200<br />
19<br />
600<br />
400<br />
200<br />
20<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.23 20 Hz waveforms of pass 21886 at the 1st second (0-7 km along-track<br />
distance) from the shoreline (descending track, cycle 43 of ERS-2), receding from<br />
the southern Australian shoreline to the Southern Ocean.<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
21<br />
800<br />
600<br />
400<br />
200<br />
22<br />
600<br />
400<br />
200<br />
23<br />
800<br />
600<br />
400<br />
200<br />
24<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
25<br />
800 26<br />
600<br />
400<br />
200<br />
800<br />
600<br />
400<br />
200<br />
27<br />
800<br />
600<br />
400<br />
200<br />
28<br />
Counts<br />
0<br />
0 16 32 48 64<br />
800<br />
600<br />
400<br />
200<br />
29<br />
0<br />
0 16 32 48 64<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
30<br />
0<br />
0 16 32 48 64<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
31<br />
0<br />
0 16 32 48 64<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
32<br />
0<br />
0 16 32 48 64<br />
Counts<br />
800<br />
600<br />
400<br />
200<br />
33<br />
0<br />
0 16 32 48 64<br />
1000<br />
500<br />
34<br />
0<br />
0 16 32 48 64<br />
1000 35<br />
500<br />
0<br />
0 16 32 48 64<br />
1000 36<br />
500<br />
0<br />
0 16 32 48 64<br />
1000<br />
37<br />
1000<br />
38<br />
1000<br />
39<br />
1000<br />
40<br />
Counts<br />
500<br />
500<br />
500<br />
500<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.24 20 Hz waveforms of pass 21886 at the 2nd second from the shoreline<br />
(ascending track, cycle 43 of ERS-2), receding Australian southern shoreline to the<br />
Southern Ocean.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 107<br />
1000<br />
41<br />
1000 42<br />
1000 43<br />
1000 44<br />
Counts<br />
500<br />
500<br />
500<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
45<br />
1000 46<br />
500<br />
1000<br />
500<br />
47<br />
1000<br />
500<br />
48<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
1000<br />
49<br />
1000 50<br />
1000<br />
51<br />
1000<br />
52<br />
Counts<br />
500<br />
500<br />
500<br />
500<br />
Counts<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
53<br />
1500<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
54<br />
1000<br />
0<br />
0 16 32 48 64<br />
500<br />
55<br />
0<br />
0 16 32 48 64<br />
1000 56<br />
500<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
0<br />
0 16 32 48 64<br />
Counts<br />
1000<br />
500<br />
57<br />
1000<br />
500<br />
58<br />
1000<br />
500<br />
59<br />
1000<br />
500<br />
60<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
0<br />
0 16 32 48 64<br />
Bins<br />
Figure 4.25 20 Hz waveforms of pass 21886 at the 3rd second from the shoreline<br />
(ascending track, cycle 43 of ERS-2), receding Australian southern shoreline to the<br />
Southern Ocean.<br />
4.6.3 Contamination Caused by the Altimeter’s Operation<br />
As can be seen from Figures 4.19 to 4.26, after obvious returns from the land or<br />
inland coastal still water disappear from the range window, waveforms continue to<br />
take a second or more of time to align to the tracking gate. The corresponding alongtrack<br />
distance can be longer to ~21 km from the shoreline, which cannot be<br />
explained by the land effects within the altimeter footprint. The reasons for this are<br />
not clear, but it may be linked to the operation of the altimeter. The altimeter works<br />
often in the normal track mode over sea surfaces, while this is not the case over nonocean<br />
surfaces. After a water-to-land or land-to-water transition, the altimeter usually<br />
takes a second or more of time to reacquire the signal over the changed surface<br />
(Brooks et al., 1997; Strawbridge and Laxon, 1994, cited in Scott et al., 1994).<br />
In addition, the ocean waveform should show a linear trailing edge (see Section 5.4.3)<br />
based upon the theoretical waveform model. This means that the ocean waveform<br />
samples can be shifted rightwards in the range window without a change of the<br />
relevant position of the retracking gate. To demonstrate this, 20 waveforms at a
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 108<br />
second time span along descending groundtrack 21886 related to waveforms 21-40 in<br />
Figure 4.24 are chosen. This group of waveforms is characterised to have retracking<br />
gates from 23.3-27.3, which all depart from the pre-designed tracking gate (32.5).<br />
These waveforms are shifted rightwards with respect to the centre of the range<br />
window by 0, 5, 10, 15, 20, or 25 gates (corresponding to the range of 0 m, 2.27 m,<br />
4.54 m, 6.81 m, 9.08 m, or 11.36 m, respectively). Using the least squares fitting<br />
algorithm (Chapter 5), an ocean model (see Section 5.4.1) is fitted to each waveform<br />
to estimate the retracking gate. Differences between the retracking estimates before<br />
and after shift are computed. Descriptive statistics of the differences are then<br />
calculated and listed in Table 4.4.<br />
Table 4.4 Descriptive statistics of the differences of retracking gate estimates before<br />
and after shifting rightward waveform samples in the range window (20 Hz<br />
waveforms from pass 21886 at the 2nd second from the coastline, 1 gate≈0.4542 m)<br />
Shift<br />
(gates)<br />
Number of<br />
waveforms<br />
Min<br />
(gates)<br />
Max<br />
(gates)<br />
Mean<br />
(gates)<br />
STD<br />
(gates)<br />
5 20 -0.10 0.03 -0.02 0.03<br />
10 20 -0.13 0.07 -0.02 0.05<br />
15 19 -0.22 0.09 -0.01 0.08<br />
20 19 -0.20 0.24 0.01 0.09<br />
25 19 -0.26 0.23 -0.01 0.13<br />
As can be seen from Table 4.4, when shifting waveforms 5 and 10 gates rightward in<br />
the range window, the retracking gate estimates have a good agreement with those<br />
before shifting. When shifting 15, 20, or 25 gates, one waveform fails to obtain the<br />
retracking gate estimate, while others still agree well with the estimate before<br />
shifting. The STD increases with increasing the shifting gates. Because of the fading<br />
noise, the maximum range error caused by this shifting is ~6 cm. The fitting<br />
algorithm keeps working until shifting the position of the leading edge rightward in<br />
the range window to gate 57. This testing result indicates that the retracking gate can<br />
be successfully estimated by the fitting procedure.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 109<br />
Therefore, these test results support that the original tracking mode still continues for<br />
some time, when the returns reflected from the surface changed present in the range<br />
window. The results also demonstrate that the altimeter’s operating feature is one of<br />
waveform contamination sources in coasts, and this operating feature influences<br />
mainly waveforms along the ground tracks that leave land to water (cf. Figures 4.19<br />
and 4.23). It has been noted from previous discussions (Sections 4.6.1 and 4.6.2) that<br />
this problem does not appear to happen to the data from water to land ground tracks<br />
where the altimeter probably remains in the normal ocean mode till getting very<br />
close to coast. These data are, thus, less contaminated.<br />
This result also implies a tremendous influence on the use of untracked SSH data in<br />
coastal regions. Only the data from water to land ground tracks could be used in the<br />
area vicinity (e.g., ~5 km) of the coastline without waveform retracking<br />
4.6.4 Summary of Waveform Contamination in Coasts<br />
Considering the groundtrack approaching land, as it gets closer to the coasts, the land<br />
returns within the altimeter footprint begin to appear in the waveform range window.<br />
As long as the range to the nadir ocean surface is shorter than the range to the offnadir<br />
land surfaces, the ocean returns will appear earlier in the range window.<br />
However, as the land returns move into the AGC gate, the altimeter tracker tends to<br />
track the off-nadir terrain, leading the calculated at-nadir SSH too high. The affected<br />
distance is about 7 km along the groundtrack from the coastline for ERS-1/2.<br />
On the other hand, for ground tracks leaving land to water, the altimeter tracker<br />
continues to track the off-nadir land surfaces and responds more to the off-nadir<br />
shorter range at the time when the ground track is just receding from land to water.<br />
When the altimeter ground track goes further into the ocean, land returns move back<br />
to the late gates. Waveforms are, meanwhile, shifted left first and then right with<br />
respect to the centre of the range window (e.g., Figures 4.21 and 4.22). This shift can<br />
last three seconds or more of time, indicating a longer along-track contaminated<br />
distance of ~21 km (cf. Section 4.6.2).<br />
If considering only the shape variations of the waveform in the proximity of the<br />
coasts, the following three typical characteristics can be summarised:
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 110<br />
(1) The land returns appear in the range window together with the ocean surface<br />
scattering. Because of the different surface reflectivity and elevations, the<br />
power superposition of the ocean returns and land returns changes both the<br />
waveform shape and magnitude of the amplitude with increasing the<br />
importance of the land return in the range window. This, in turn, causes the<br />
gate location of the mean-surface retracking point to vary from 50% of the<br />
amplitude (i.e., the half-power point) as the land return signal begins to<br />
increasingly dominate the waveform shape. The ocean returns cannot be<br />
recovered when the land returns predominate in the waveform.<br />
(2) Waveforms shift towards the left or right as a function of along-track distance<br />
with respect to the tracking gate in the range window, but do not significantly<br />
change the magnitude of the waveform’s amplitude. Land returns are not<br />
obviously visible, and the contamination cannot be explained by the land<br />
effects, because the distance to the shoreline may be larger than the radius of<br />
the footprint. In this case, waveform contamination results from the altimeteroperating<br />
feature.<br />
(3) Inland standing water causes very high peaks in the waveform. This type of<br />
specular waveform appears with the large magnitude of the amplitude and the<br />
narrow width of power. These should not be retracked unless the information<br />
on the inland water is required.<br />
4. 7 An Experiment of the Land Effects on Waveforms Using the DEM Data<br />
A basic feature of the waveform is that it can be considered as the cumulative energy<br />
returned from each illuminated surface scatterer, sampled by a series of gates in the<br />
range window (Femenias et al., 1993; Martin et al., 1983). Although most coastal<br />
waveforms have an ocean-like shape with a single ramp, there are a percentage of<br />
waveforms (~20%) that show non-ocean-like shapes, particularly over areas closer to<br />
land (Section 6.2). These waveforms consist of ocean and land returns, hence leading<br />
to a mixed waveform shape. There are two possibilities for these waveforms. The<br />
first is that the ocean return could be recovered from these observed waveforms<br />
using waveform retracking. The alternative is that it might be impossible to recover<br />
the ocean return if it is obscured by the land return when the ground track is very
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 111<br />
close to the coastline, say, ~0-2.5 km, and range to the ocean surface is no longer<br />
recoverable.<br />
In most cases of waveform retracking, the relation between the derived surface and<br />
the actual surface are unknown, as independent measures of the ground truth are not<br />
generally available. A digital elevation model (DEM), however, is a representation of<br />
the terrain using the mean elevation information. Moreover, it includes the coastline<br />
information including the main outline of the land, bays, the outer edge of mangrove,<br />
swamps, closed-off narrow inlets, and watercourses at or near their mouths<br />
(Hutchinson et al., 2001). Therefore, a high-resolution DEM can provide the land<br />
elevation for dealing with mixed waveforms, though the disadvantage is that it is<br />
generated by the land-averaged elevation at each cell. Other alternatives are the<br />
optical satellite remote sensing, topographic maps or air photography, but these are<br />
time consuming. Thus, a high-resolution DEM is still a useful external data source.<br />
Because the well-developed ocean surface scattering model (e.g., Brown, 1977) is<br />
deduced under the assumptions of the surface statistical homogeneity (see Section<br />
3.2), these assumptions are not valid over non-ocean surfaces. Thus, to use the<br />
independent DEM data, some other methods must be developed. The idea is that if<br />
the land returns contained in the waveform can be determined separately using an<br />
independent data source, they might be removed from the shape-mixed waveform to<br />
leave the ocean return alone. Thus, the ocean surface can be recovered. To achieve it,<br />
however, detailed information about the topography along the coastline is necessary.<br />
The topography should include the land and inland water (e.g., lakes and rivers). Part<br />
of the effort, which has been performed and designed in this study, will be presented<br />
in this Section.<br />
4.7.1 Concept of Land Effects Using the DEM Model<br />
As stated in Chapter 2, the altimeter’s beam-limited footprint is the area within<br />
which the beam attenuation is 3 dB or less. It has a larger footprint size than that of<br />
the pulse-limited footprint (see Section 4.2.1). It is in this larger footprint size that<br />
the land returns have higher probability of appearing in the range window. Therefore,<br />
when discussing the land effects on the waveform, it is necessary to account for the<br />
beam-limited footprint rather than the pulse-limited footprint.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 112<br />
Consider two different reflected surfaces of land and water within the beam-limited<br />
footprint from which the power backscatters to the altimeter. The land elevation<br />
above local instantaneous sea level is H . The coordinate of this location can be<br />
defined by the radial distance from the altimeter to the point on the sea surface, the<br />
off-nadir angle ξ , the azimuth angle χ about the axis defined by the line between<br />
the altimeter and the nadir point. Because of the approximate spherical geometry of<br />
the Earth, it is convenient to replace the angle ξ with the colatitude angle φ<br />
subtended by rays from the centre of the Earth to the nadir point and the off-nadir<br />
'<br />
point H ( χ , ξ,<br />
t)<br />
= H ( χ,<br />
φ,<br />
t)<br />
(see Figure 4.26). Define Rφ ( χ,<br />
t)<br />
to be the radial<br />
distance from the altimeter to H ( χ , φ,<br />
t)<br />
. The reflected power is received from this<br />
specular reflector at the two-way travel time t if H ( χ , φ,<br />
t)<br />
falls within the footprint.<br />
To the lowest order for the small off-nadir angles ξ relevant to the altimeter, the<br />
'<br />
distance Rφ ( χ,<br />
t)<br />
is related to the distance R from the altimeter to the mean sea level<br />
at the angle φ by<br />
R ( χ,<br />
φ,<br />
t)<br />
≈ R − H ( χ,<br />
φ,<br />
t)<br />
(4.3)<br />
'<br />
φ<br />
A more precise relation is easily obtained from the geometry in Figure 4.26, but this<br />
'<br />
approximate solution is adequate for the present purpose. In addition, Rφ ( χ,<br />
t)<br />
can<br />
be also related to the satellite altitude alt(t) and the geoid height N( χ , φ , t)<br />
above a<br />
reference ellipsoid as<br />
R ( χ,<br />
φ,<br />
t)<br />
≈ alt(<br />
t)<br />
− H ( χ,<br />
φ,<br />
t)<br />
− N(<br />
χ,<br />
φ,<br />
t)<br />
(4.4)<br />
'<br />
φ<br />
Thus, on the one hand, the range to the land surface (<br />
'<br />
R<br />
φ<br />
) can be estimated by<br />
Equation (4.4), On the other hand, it can be obtained from the altimeter-measured<br />
range ( χ , φ , t)<br />
to the surface and the retracking correction dr 2 (Equation 5.7, see<br />
R s<br />
Section 5.4 for details) estimated by the nine-parameter function (Equation 3.41) as<br />
R ( χ,<br />
φ,<br />
t)<br />
= R<br />
'<br />
φ<br />
s<br />
( χ , φ , t)<br />
+ dr2<br />
= R ( χ , φ , t)<br />
+ G<br />
s<br />
2M<br />
( β − g )<br />
32<br />
0<br />
(4.5)
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 113<br />
alt<br />
R<br />
ξ<br />
R'φ<br />
Rφ<br />
H<br />
Re<br />
N<br />
mean sea<br />
level<br />
reference<br />
ellipsoid<br />
φ<br />
Figure 4.26 A schematic geometrical description of the radar return from a point on<br />
the land surface at off-nadir angle ξ and associated colatitude angle φ where the<br />
land elevation above the mean sea level is H . The distance from the altimeter to the<br />
'<br />
land surface is R<br />
φ<br />
.<br />
where β<br />
32<br />
is the retracking gate estimate in units of waveform gates (related to the<br />
second ramp),<br />
G 2 M<br />
= t p<br />
⋅ c / 2 is the conversion from numbered gates to metres, t p<br />
is<br />
the pulse width, and g<br />
0<br />
is the location in gates of the altimeter tracking gate. From<br />
Equations (4.4) and (4.5), β<br />
32<br />
can be related to the land elevation H ( χ , φ,<br />
t)<br />
by<br />
β = a + bH ( χ , φ , )<br />
(4.6)<br />
32<br />
t<br />
where<br />
a<br />
alt( t)<br />
− N(<br />
χ , φ , t)<br />
− R ( χ , φ , t)<br />
s<br />
= g<br />
0<br />
+<br />
(4.7)<br />
G2M
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 114<br />
and<br />
= 1/<br />
(4.8)<br />
b G2M<br />
Equation (4.6) describes the relationship between the retracking gate estimate and the<br />
land elevation. At a certain time t , β<br />
32<br />
is only linearly related to the land elevation<br />
H ( χ , φ,<br />
t)<br />
at the reflector’s location. Therefore, if the location of β 32<br />
at the range<br />
window can be determined by the independent DEM data source, it can be used as a<br />
constrained condition or initial value of the nine-parameter function. This is because<br />
an appropriate initial value or constraint will make the fitting algorithm converge<br />
rapidly and (probably) accurately. However, it is also clear from Equation (4.6) that<br />
several surface reflectors at the time t within the satellite footprint can have the same<br />
slant range to the satellite. Thus, β<br />
32<br />
is not a unique function of the land elevation.<br />
This can be also presented from the simple test shown below<br />
4.7.2 A Preliminary Test of Land Effects Using the DEM Model<br />
When testing this method, the following steps have been taken for each observing<br />
location ( λ<br />
0<br />
,ϕ<br />
0<br />
) along satellite ground tracks near land.<br />
(1) The algorithm estimates the shortest spherical distance s to the coastline and<br />
extracts the location ( λ , ϕ ) at the point using a shoreline model, at which the<br />
c<br />
c<br />
ground track intercepts the coastline.<br />
(2) A search radius centred at the location ( λ<br />
c<br />
, ϕ<br />
c<br />
) is then determined according to<br />
the value of s and the size of the satellite footprint, which can range from<br />
2-8 km. The land elevations and geoid heights within the search area are<br />
extracted at the point of each elevation cell using the Australian 9" DEM<br />
(Hutchinson, 2001) and AUSGeoid98 (Featherstone et al., 2001) geoid grid<br />
(2´×2´), respectively.<br />
(3) Retracking gate estimates β<br />
32<br />
( χ i<br />
, φ i<br />
, t)<br />
are obtained using Equation (4.6),<br />
where i = 1, 2, ···, n is the number of the 9" cells within the search radius.<br />
Because of the width of the range window, the maximum difference of the
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 115<br />
elevations must be less than 28 m. Thus, elevations higher than 28 m were not<br />
used in the computation.<br />
Figure 4.27 shows two plots of β<br />
32<br />
( χ i<br />
, φ i<br />
, t)<br />
via the land elevation H ( χ<br />
i<br />
, φi<br />
, t)<br />
. The<br />
discontinuous retracking gate estimates in Figure 4.27 are caused by the non-uniform<br />
elevation distribution within the search radius. As can be seen, all land elevations<br />
within the search radius contribute to the return power, thus causing the problem of<br />
non-uniqueness. This is a typical problem for non-ocean surfaces, because several<br />
targets with the same slant range to the altimeter exist within the footprint.<br />
Retracking gate eatimates<br />
(bins)<br />
60<br />
55<br />
50<br />
45<br />
40<br />
35<br />
30<br />
0 1 2 3 4 5 6 7 8<br />
Land elevations (m)<br />
Retracking gate estimates<br />
(bins)<br />
60<br />
55<br />
50<br />
45<br />
40<br />
35<br />
30<br />
0 1 2 3 4 5 6 7 8<br />
Land elevations (m)<br />
Figure 4.27 Retracking gate estimates corresponding to the land elevation. The<br />
search radius is 2.7 km (top) and 3.5 km (bottom).<br />
Thus, although Equation (4.6) is simple, it is difficult to effectively implement.<br />
Theoretically, the altimeter starts to track off-nadir higher and brighter land when the
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 116<br />
distance between the satellite and the land surface is shorter than that to the sea<br />
surface. In practice however, of these land reflectors illuminated by the satellite, it is<br />
hard to determine, based on only the DEM, which one importantly contributes to the<br />
relevant mid-point on the second ramp of the waveform. Also, information about the<br />
surface reflectivity of land and water is important, because the radar returns are<br />
directly related to the nature of the scattering surface illuminated by the transmitted<br />
electromagnetic single. However, the DEM cannot provide the water information on<br />
land.<br />
To overcome these problems, a more detailed function based directly on the radar<br />
equation is necessary. Wingham’s method (1995) may be a way to solve the<br />
problem. However, the DEM contains only land elevation and no water information<br />
over land. This is another difficulty when trying to create the function for the land<br />
return, because the water reflector, even a layer of very shallow water (e.g., after<br />
rain), can make the return power much higher (Berry, 2002). In addition, the DEM<br />
represents the average elevation of a cell (cf. Hutchinson, 2001), while returns reflect<br />
from the facets with different elevations. Another condition that needs to be thought<br />
about is the land slope. Brooks (2002) has found that the land slope affects the<br />
returns reflected from the land near the coasts as well. Some higher land may scatter<br />
forward the radar pulse so that it does not show land return on the waveform. Since<br />
20 Hz waveform data require a spatial resolution of ~350 m, the high-resolution<br />
DEM and detailed topography information are necessary to estimate land returns.<br />
Because of the time limitation of this study, this can only be a topic of future<br />
research.<br />
4. 8 Summary<br />
Five cycles of POSEIDON and one cycle of ERS-2 20 Hz waveform data have been<br />
used to quantify a broad, contaminated boundary around the Australian coast. This<br />
contamination is assumed to have been caused by backscatter from the land and the<br />
more variable coastal sea surface states, coupled with the footprint size of the radars<br />
and the incorrect positioning of the groundtrack.<br />
Since bins 31-33 for the ERS-2 altimeter and bins 30-32 for the POSEIDON<br />
altimeter are the expected bins for the tracking gate over oceans, three major
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 117<br />
conclusions can be drawn from this study. First, the ERS-2 altimetry shows<br />
differences in contamination as a function of the sub-area along the Australian<br />
shoreline. The maximum contaminated distances vary from ~8 km to ~22 km,<br />
depending upon locations and features of the shoreline. Second, both altimeters show<br />
differences in contamination as a function of the distance from the shoreline. Finally,<br />
the contaminated distance for POSEIDON is less than that for ERS-2. A mean<br />
contaminated distance of ~8 km for POSEIDON and ~10 km for ERS-2 can be<br />
observed using the 50% of threshold retracking points in the whole Australian<br />
coastal region.<br />
The mean waveforms and standard deviations of the mean in different distance bands<br />
provide an indication of the contamination influencing the shape and variability of<br />
the returned waveforms in coastal regions. They can provide a reasonable<br />
explanation for the observed variations of the waveform in the proximity of coasts. A<br />
typical contaminated boundary can be ascertained from Australian shoreline to<br />
~8 km and ~10 km for POSEIDON and ERS-2, respectively. Beyond 8 or 10 km (to<br />
350 km), the mean waveform shapes for both altimeters match the mean waveform<br />
shape observed over ocean surfaces, though they do not agree exactly with the<br />
theoretical waveform shape (Section 4.5).<br />
The different reflector characteristics lead to different types of waveforms. In<br />
addition, the contamination distance depends on how the satellite ground track<br />
crosses the shoreline (i.e., approaching or leaving the land). In general, the<br />
waveforms along the altimeter groundtrack leaving land to water suffer longer<br />
contaminated distances than those along the track approaching land from water. In<br />
both cases, the ocean surface sometimes can no longer be recovered from<br />
contaminated waveforms obscured significantly by the land return, when the ground<br />
track is much closer to the coastline, say, ~0-2.5 km, because the land returns<br />
become predominant in the range window.<br />
A preliminary test has been performed to calculate the retracking gate estimate<br />
related to the land returns in the range window using an external independent data<br />
source of the DEM and AUSGeoid98 models in coastal regions. Analysing results<br />
show that non-uniqueness of the estimate over land near coasts is still a main<br />
problem that may need to be solved in the future.
Chapter 4. WAVEFORM CONTAMINATION CLOSE TO AUSTRALIAN COASTS 118<br />
Contamination leads to errors in the on-board calculation of the range measurements.<br />
It is recommended from the results in this Chapter that such contaminated<br />
measurements must be detected prior to their being included in geodetic and<br />
oceanographic solutions in coasts. Until then, altimeter SSH or range data should be<br />
interpreted with some caution for distances less than, say, ~22 km from a coastline,<br />
and discarded altogether for distances less than 4 km.<br />
This Chapter utilises different methods (e.g., statistic and threshold analysis, and<br />
using additional DEM data source) to show that the land topography does influence<br />
the waveform shape and thus the data. This can also be achieved using some other<br />
methods, such as to plot the distance from the coastline against near shore land<br />
height (within 0.5 km of coast) to see if there is a correlation between them.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 119<br />
5. COASTAL WAVEFORM RETRACKING SYSTEM: DESIGN AND<br />
IMPLEMENTATION<br />
5. 1 Introduction<br />
The techniques used to retrack ocean and non-ocean waveforms have been discussed<br />
in Chapter 3. They can also be applied to the coastal waveform (e.g., Anzenhofer et<br />
al., 2000; Deng et al., 2001). During waveform data processing and analysis, it has<br />
been found in this study that fitting the waveform to a well-developed analytical<br />
function (i.e., the Brown (1977) ocean model) is capable of retracking most of the<br />
individual altimeter waveforms (~99% of data, see Table 6.1 in Section 6.2.2) over<br />
open oceans. Similarly, using a 50% threshold level can give reasonable estimates of<br />
the absolute range to the at-nadir sea surface. Using only a single retracker in coasts,<br />
however, limits the precision of the recovered SSH due mainly to the disagreement<br />
of the contaminated waveform with the theoretical/ideal waveform.<br />
Therefore, it has become clear that no single retracking algorithm can always deal<br />
with the diverse waveforms in coastal regions. A coastal waveform retracking system<br />
which consists of different retrackers (e.g., the parametric fitting and the threshold<br />
algorithms), thus, is necessary when attempting to extract the precise SSH from<br />
retracked range measurements.<br />
This Chapter presents the implementation of both fitting and threshold retracking<br />
algorithms, and develops a coastal retracking system that is based upon a detailed<br />
pre-analysis of the coastal waveforms. Issues involved in the algorithm development<br />
are fitting functions, the iterative least squares procedure, linearisation of the<br />
parameters, determination of the initial estimates of parameters, and the weight<br />
scheme. A new method will be introduced that uses iterative weights to detect the<br />
outliers in the waveforms ensuring the effective convergence of the fitting procedure.<br />
For the threshold retracking algorithm, a method will be described that selects an<br />
appropriate threshold level for coastal waveforms. Finally, a coastal retracking<br />
system will be developed, which makes it possible to improve the SSH precision by<br />
reprocessing the altimeter waveforms, correcting the range measurements, and<br />
extracting precise SSH data from these corrected ranges.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 120<br />
5. 2 Coastal Waveform Retracking System Design<br />
A comprehensive analysis of waveform shapes at the Australian coast (Chapter 4)<br />
led to the need to develop a coastal waveform retracking system. The primary idea of<br />
the retracking system is that retrackers must cover most of the coastal waveforms,<br />
and they can complement each other. Since the SSH data can only be recovered from<br />
ocean returns, it is impossible for the system to reprocess the waveforms dominated<br />
by the land return near land to extract SSHs. This means that it is not essential to<br />
have as many retrackers as used over land, where much more diverse waveforms are<br />
found (cf. Berry, 2000). However, it is significant that the system can correctly<br />
categorise the waveforms, so that they can be retracked using an adequate retracker.<br />
This concern generated an additional waveform classification.<br />
Ideally, a physically realistic model should be used to retrack coastal waveforms,<br />
because the waveform measured by a radar altimeter links not only to the geometric<br />
surface of the ocean, but also to oceanic geophysical features. An appropriate model<br />
will provide a correct understanding of the relationship between the geophysical<br />
parameters of interest and the radar return from coastal regions.<br />
With this consideration, the ocean model without non-linear wave parameters<br />
(presented in Section 5.5.1) is first used to retrack ocean-like waveforms. To deal<br />
with narrow, high-power waveforms reflected from calm coastal or inland water<br />
surfaces, the threshold retracking technique with 50% threshold level is used as well.<br />
It was found from an analysis of retracked results that both retrackers could not deal<br />
with some waveforms in close proximity to the land. Fitting an ocean model to nonocean-like<br />
waveforms causes failure of the fitting algorithm. This led to the use of<br />
other retrackers of the five- and nine-β-parameter fitting functions because of their<br />
ability to retrack non-ocean waveforms (presented in Section 5.5.3). The 50%<br />
threshold level cannot always give improved results, particularly in the area closer to<br />
the coastline. To validate the threshold level of the retracking algorithm, a statistical<br />
analysis has been performed to suggest the quality of the retracked SSH data (Section<br />
5.7.2). The evaluation turned out a detailed selection of the threshold levels.<br />
When retracking individual waveforms in coasts using all possible retrackers, results<br />
show improved SSHs, but biases exist among different algorithms. Thus, an
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 121<br />
Figure 5.1 Block diagram of the coastal retracking system
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 122<br />
algorithm assessment has been implemented to analyse and estimate these biases. All<br />
these investigations motivate the development of the coastal retracking system.<br />
Finally, a coastal retracking system has been developed to retrack altimeter<br />
waveforms in Australian coasts, which involves several external input data,<br />
waveform data editing, waveform classification, fitting algorithms and threshold<br />
techniques. The flow diagram in Figure 5.1 illustrates how altimeter waveforms are<br />
reprocessed to give the corrected SSH data. The procedure follows the main steps as:<br />
(1) Data editing - edit out the waveforms that do not show the leading edge in the<br />
range window, retain waveforms that show a leading edge regardless of<br />
complexity of the whole waveform;<br />
(2) Waveform classification - sort waveforms based on different shapes, and<br />
indicate which retracker will be used for each;<br />
(3) Fitting algorithms - make initial estimates of the parameters, determine<br />
observational weights, and run iteration procedure;<br />
(4) Threshold retracking - retrack waveforms with the narrow width and high peak,<br />
as well as the waveforms that failed with the fitting algorithms; and<br />
(5) Corrections - include the biases between each method, range correction, and<br />
the geophysical and environmental corrections.<br />
The description of each step will be presented in this Chapter and the ancillary input<br />
data sets needed by the system are described in Section 2.5. An important factor in<br />
the system is to accurately classify each return waveform so that it can be performed<br />
by an appropriate retracker. Retracking results and analysis at the Australian coast<br />
will be shown in Chapter 6.<br />
5. 3 Extraction of the Geodetic Parameter: the Sea Surface Height<br />
After retracking, the retracked range is then<br />
Retracked Range = Observed Range + Retracking Range Correction<br />
Since the retracking system consists of fitting algorithms and threshold retracker, the<br />
biases between them (discussed later in Section 5. 8) must be applied to the range<br />
measurements. Biases can be calculated by retracking the same cycle of ocean
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 123<br />
waveform data. The corrected range in this study is based upon the retracked range<br />
from the fitting algorithm so that the bias is applied to the range which is retracked<br />
using the threshold retracker by<br />
Corrected Range = Retracked Range + Bias<br />
In addition, calculating an accurate SSH from the altimetric measurement involves<br />
accounting for not only the waveform range corrections, which are discussed in this<br />
study, but also correcting the range for atmospheric propagation delay and applying<br />
appropriate ocean-tide correction to remove the time-varying tidal effect. After these<br />
corrections are applied to the range, the SSH can then be calculated from the satellite<br />
altitude and corrected range as<br />
Corrected SSH = Orbit - Corrected Altimeter Range - Corrections.<br />
Up to now, the corrected SSH data are used at 20 Hz. They can produce 1 Hz or 2 Hz<br />
data products for geodetic applications.<br />
5. 4 Analysis of Waveform Shapes<br />
5.4.1 Determination of the Peaks in the Waveform<br />
The peaks in the waveform (cf. Figure 5.2) can be estimated from an analysis of the<br />
second derivative of the waveform. The first derivative, W ′(t)<br />
, is approximated by<br />
the first forward differences with respect to the gate number<br />
W ( tk<br />
+ 1)<br />
−W<br />
( tk<br />
)<br />
W '( t)<br />
=<br />
∆t<br />
(5.1)<br />
where W t ) is the waveform sample at the time t , ∆ t is the time interval between<br />
( k<br />
waveforms t<br />
k + 1<br />
and t<br />
k<br />
, and k = 1L64<br />
(depending on the altimeter) is the gate<br />
number of the waveform.<br />
Using results of contiguous first forward differences, the second derivative of the<br />
waveform, W ′′(t<br />
) , can be approximated by second forward differences as
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 124<br />
W ′(<br />
t W ′<br />
k + 1)<br />
− ( tk<br />
)<br />
W ′ ( t)<br />
=<br />
∆t<br />
(5.2)<br />
If there are n p peaks (n p = 6 in this study) in the waveform, the gate locations of all<br />
possible peaks L max (i) (i = 1, 2, ···, n), then, can be determined, where the second<br />
derivatives cross zero from negative to positive, and vice versa for all minimum<br />
locations L min (i).<br />
Finally, the peak amplitudes, W amp (i), can be found from the waveform<br />
corresponding to the gates of L max (i). All peaks are checked to determine whether<br />
they are larger than an input criterion, which is obtained prior to the retracking<br />
procedure by analysing ocean waveform amplitude features. Peaks less than this<br />
criterion are not significant. In addition, if the interval between two peaks is too<br />
small, only the larger one is selected as the peak. After editing, the remaining peak(s)<br />
will be used to classify the waveforms. The maximum number of waveform peaks<br />
(ERS-1/2) in Australian coastal regions is found to be three. Results of a detailed<br />
analysis will be given later in Chapter 6.<br />
5.4.2 Waveform Classification<br />
The shape of the radar altimeter waveform depends strongly on the reflecting surface<br />
properties. Small amplitude roughness and slopes of the surface will cause a<br />
broadening of the leading edge of the return. Multiple reflecting surfaces within the<br />
altimeter footprint will cause the waveform to be the superposition of the different<br />
returns from each surface elevation, thus forming the multi-peaks. Over these<br />
multiple reflecting surfaces near the land, the waveform shape cannot follow the<br />
Brown (1977) model.<br />
Waveform classification is based upon a wide ranging analysis of the waveform<br />
shapes at the Australian coast. After observing a number of ERS-2 coastal<br />
waveforms, they can be sorted into nine categories (Figure 5.2) depending on the<br />
waveform shape features, such as amplitude, width, and measure of peaks. Of these<br />
waveform features, the peaks can be ascertained using the method described in<br />
Section 5.4.1, and waveform amplitude and width can be determined using the
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 125<br />
OCOG method (Wingham et al., 1986). A detailed waveform analysis by<br />
geographical area will be given in Chapter 6.<br />
The shape characteristics of the waveform revealed in Figure 5.2 are generally<br />
associated with a particular type of surface. Of these waveforms in coastal regions,<br />
most of them (~80%) present a shape of ocean-like single ramp without the<br />
significant peak in the range window (Figure 5.2a), though the waveform profiles<br />
may shift left or right with respect to the predesigned tracking gate. This type of<br />
return has been found in this study over ocean surfaces at any distance up to 300 km<br />
from the coastline. Waveforms shown in Figure 5.2b to Figure 5.2g are usually found<br />
when the altimeter simultaneously illuminates both water and land surfaces, where<br />
the elevation difference between both surfaces within the pulse-limited footprint is<br />
less than ~28 m (due to the scale of the range window). Because of two or more<br />
elevation surfaces in the footprint, such waveforms occur ~0-10 km from the<br />
coastline (Chapter 3). The narrow-peaked waveform (Figure 5.2h) occurs over the<br />
standing water near coastline, such as bays, gulfs, estuaries, and harbours, as well as<br />
inland lakes. A distinct feature of this type of waveform is its extremely high<br />
amplitude (typically ~8400 counts in average from Section 6.2.2).<br />
Therefore, waveform categories in coastal regions indicate the specific reasons why a<br />
retracking system is appropriate in coasts. The primary factor is that many waveform<br />
shapes observed in coastal regions are not encountered in data sets over open oceans.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 126<br />
(a) No Significant Peak<br />
(b) Single Pre-peak<br />
Power (counts)<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
B i n s<br />
Power (counts)<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
B i n s<br />
Power (counts)<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
(c) Single Post-peak<br />
Power (counts)<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
(d) Single Middle-peak<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
B i n s<br />
B i n s<br />
Power (counts)<br />
4000<br />
3000<br />
2000<br />
1000<br />
0<br />
(e) Double Pre-peaks<br />
Power (counts)<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
(f) Double Post-peaks<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
B i n s<br />
B i n s<br />
(g) Multi Peaks (h) Sharp Peak<br />
Power (counts)<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
Power (counts)<br />
20000<br />
15000<br />
10000<br />
5000<br />
0<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
1<br />
6<br />
11<br />
16<br />
21<br />
26<br />
31<br />
36<br />
41<br />
46<br />
51<br />
56<br />
61<br />
B i n s<br />
B i n s<br />
Figure 5.2 Typical ERS-2 waveform categories in Australian coasts
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 127<br />
5. 5 Fitting Functions<br />
To use fitting algorithms, an analytic expression of coastal waveforms is essential,<br />
which describes the interaction of the radar pulse with the scattering surface. As<br />
stated, estimating non-linear wave parameters is not the aim of this study. In fact, it<br />
is impossible to accurately estimate non-linear wave parameters using contaminated<br />
waveforms in coasts. Therefore, the fitting functions used in this study aim at the<br />
determination of the range correction through the revised estimate of the tracking<br />
gate. In the proposed system, an ocean model and the β-parameter models will be<br />
used as fitting functions, described next.<br />
5.5.1 The Ocean Model<br />
In Chapter 3, a modified expression (Equation 3.31) for the waveform over non-<br />
Gaussian ocean surface was presented. The sea surface skewness λ cs<br />
in Equation<br />
(3.33) usually takes a small magnitude. Rodriguez and Martin (1994) estimate the<br />
skewness using TOPEX Ku-band waveforms (cycles 3-27) as between 0 - 0.11 for<br />
both ascending and descending passes, which is obtained by averaging estimates<br />
over all cycles. Therefore, it can be safely neglected in the waveform fitting function<br />
without the loss of the retracking accuracy. For λ = 0 (i.e., neglecting skewness),<br />
Equation (3.31) is the linear surface case and the result can be reduced to a form<br />
similar to the Brown (1977) model as<br />
cs<br />
1 ⎡ ⎛ τ ⎞ ⎤ ⎡ ⎛ d ⎞⎤<br />
P( t)<br />
= PN + A0<br />
Aξ ⎢erf<br />
⎜ ⎟ + 1⎥<br />
exp⎢−<br />
d⎜τ<br />
+ ⎟<br />
2<br />
⎥<br />
(5.3)<br />
⎣ ⎝ 2 ⎠ ⎦ ⎣ ⎝ 2 ⎠⎦<br />
in which five fitting parameters are:<br />
(1) The thermal noise level P<br />
N<br />
,<br />
(2) The amplitude scaling term A 0<br />
,<br />
(3) The tracking-gate-related time t 0<br />
(i.e., the time corresponding to the mean sea<br />
surface),<br />
(4) The risetime σ , and<br />
(5) The off-nadir angle ξ (represented by Aξ<br />
Equation 3.7).
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 128<br />
Equation (5.3) will be the model used in this study for the purpose of waveform<br />
retracking. For simplicity of presentation, it will be called the revised ocean model<br />
hereafter.<br />
After waveform retracking, the retracking range correction, dr , is given as<br />
dr = G ⋅ t / t − )<br />
(5.4)<br />
2M<br />
(<br />
0 p<br />
g<br />
0<br />
where<br />
G 2 M<br />
= t p<br />
⋅ c / 2 is the conversion from numbered gates to metres, t p<br />
is the<br />
pulse width, and<br />
g<br />
o<br />
is the location in gates of the altimeter tracking gate.<br />
The revised ocean model (Equation 5.3) describes the waveform reflected from the<br />
linear Gaussian ocean surface. Compared to the Brown model (Equation 3.12), the<br />
differences between them are that the ocean model includes the effects of the Earth’s<br />
curvature and the modified Bessel function in Equation (3.2) has been replaced by an<br />
approximated expression.<br />
It is recalled briefly here about the series expansion of the Bessel function (Equation<br />
3.17). In practice, it has been expanded to different orders (Parsons, 1979; cf. Davis,<br />
1993; Hayne, 1980; Rodriguez, 1988). When retracking waveforms to determine the<br />
range correction using the Brown (1977) model, it is generally expanded to zero<br />
order (Parsons, 1979; e.g., Davis, 1993). This means that I ( β t ) 1 , and the<br />
0<br />
=<br />
Bessel function does not change its value with varying pulse return times and<br />
corresponds to a zero off-nadir angle.<br />
The approximation used in Equation (5.3) comes from Rodriguez (1988), who<br />
2<br />
expands the Bessel function to first order and keeps terms of order β t and lower,<br />
then replaces series by exp( β<br />
2 t / 4)<br />
. The error introduced from this approximation is<br />
less than 1% (Rodriguez, 1988). This approximation is more reasonable for actual<br />
waveforms, because it takes account of the non-zero off-nadir angle and different<br />
pulse return times.<br />
In Equation (5.3), the sea surface roughness (through the composite risetime) and the<br />
off-nadir angle are two main factors that affect the shape of the waveform. Mean<br />
ocean-return waveforms in Figure 5.3 show the family of the curves, which result
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 129<br />
from varying the RMS wave height (SWH from 0 - 20 m with a 2 m interval). As can<br />
be seen from Figure 5.3, varying SWH changes the rise time (or the slope) of the<br />
leading edge of the waveform. The rougher the sea state, the longer the rise time (i.e.,<br />
lower gradient) of the waveform’s leading edge.<br />
550<br />
500<br />
450<br />
SWH = 0 m<br />
400<br />
Power (counts)<br />
350<br />
300<br />
250<br />
200<br />
SWH = 20 m<br />
150<br />
100<br />
50<br />
0<br />
0 10 20 30 40 50 60<br />
Bins<br />
Figure 5.3 Variation of the leading edge’s slope with varying SWH (0 - 20 m),<br />
modelling waveforms using the ocean model ( ξ = 0 °).<br />
In addition to the surface roughness, another factor that can affect the waveform rise<br />
time is the surface slope or altimeter antenna off-nadir pointing angle. Figure 5.4<br />
shows the effects of the off-nadir angle on the waveform, for which SWH = 4 m is a<br />
typical situation of the sea surface and the numbers give the values of the off-nadir<br />
angles. As can be seen from Figure 5.4, there are three effects:<br />
(1) The off-nadir angle ξ changes not only the slope of the trailing edge, but also<br />
the slope of the leading edge. Both slopes of the leading and trailing edges<br />
decrease with increasing ξ . However, the location (or range to the mean sea<br />
surface) of the mid-point on the leading edge of the waveform does not change<br />
with varying ξ .
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 130<br />
500.0<br />
Power (counts)<br />
400.0<br />
300.0<br />
200.0<br />
ξ=0.0<br />
ξ=0.2<br />
ξ=0.4<br />
ξ=0.6<br />
100.0<br />
ξ=0.8<br />
ξ=1.0<br />
0.0<br />
0 8 16 24 32 40 48 56 64 72 80<br />
Bins<br />
Figure 5.4 Effect of the off-nadir angle (in degrees) on the amplitude, leading edge,<br />
and especially the slope of the trailing edge, modelling waveforms using the ocean<br />
model (SWH = 4 m).<br />
(2) The magnitude of the amplitude of the waveform is affected as well. Since an<br />
off-nadir angle changes the decay of the trailing edge, it will affect the value of<br />
the AGC gate and hence the AGC scaling of the waveform.<br />
(3) The slope of the trailing edge changes approximately linearly with off-nadir<br />
angle (Barrick and Lipa, 1985). The trailing edge’s slope changes its direction<br />
at ξ ≈ 0.6° (Figure 5.4). The reason is that as the surface slope increases from<br />
zero, the pulse-limited footprint moves from the centre of the beam and the<br />
leading edge part of the waveform (both leading edge and amplitude) is<br />
attenuated by the antenna gain function. At steeper angles ( ξ >1°), the<br />
attenuation becomes so great that the AGC loop can no longer compensate the<br />
loss in signal strength (Martin et al, 1983; Brrick and Lipa, 1985). As a result,<br />
loss of tracking will occur.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 131<br />
5.5.2 Five- and Nine-β-Parameter Models<br />
The β-parameter models (Equations 3.40 and 3.43) are also employed in the<br />
retracking system because of their capability of dealing with complex waveforms<br />
reflected from single or multiple scattering surfaces. They are developed based upon<br />
a modified version of the Brown (1977) ocean return function (Parsons, 1979), but<br />
add an additional slope parameter after the ramp (discussed later in Section 5.5.3). A<br />
description of the five- and nine-β-parameter models can be found in Section 3.5.1.<br />
The retracking corrections dr can be computed by the following formulas.<br />
(a) The Five-β-Parameter Model<br />
dr = G 2 M<br />
⋅ ( β − g<br />
0<br />
)<br />
(5.5)<br />
where β<br />
30<br />
is the retracking gate estimate in units of waveform gates.<br />
30<br />
(b) The Nine-β-Parameter Model<br />
dr = G 2 M<br />
⋅ ( β − g )<br />
(5.6)<br />
1<br />
30 0<br />
dr = G 2 M<br />
⋅ ( β − g )<br />
(5.7)<br />
2<br />
60 0<br />
where β<br />
60<br />
is the retracking gate estimate in units of waveform gates (related to the<br />
second ramp), dr 1 is the correction related to the range to at-nadir mean sea level,<br />
and dr 2 is the correction related to the range to the off-nadir (assumed) land surface.<br />
5.5.3 Comparison between the Ocean and Five-β-Parameter Models<br />
Both ocean and five-β-parameter models are fundamentally based upon the Brown<br />
(1977) model. Comparison between them will give qualitative guidance to selecting<br />
an adequate fitting function for the retracking procedure. Therefore, an assessment of<br />
these functions will be discussed in this Section. In order to compare them directly, it<br />
is necessary to simplify the ocean model (Equation 5.3). By assuming a zero offnadir<br />
angle (i.e., A = 0 ), Equation (5.3) can be rewritten as<br />
ξ
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 132<br />
2<br />
⎡<br />
⎤<br />
2<br />
1 ⎛ t − t ⎞ ⎡ ⎛<br />
⎞⎤<br />
0<br />
− ασ<br />
ασ<br />
P(<br />
t)<br />
= P + ⎢ ⎜<br />
⎟<br />
+ 1⎥<br />
exp⎢−<br />
⎜ − −<br />
⎟<br />
N<br />
PA<br />
erf<br />
α t t0<br />
⎥<br />
(5.8)<br />
2 ⎢⎣<br />
⎝ 2σ<br />
⎠ ⎥⎦<br />
⎣ ⎝ 2 ⎠⎦<br />
2<br />
in which P A<br />
= A0<br />
is the amplitude scaling term. The term ασ in Equation (5.8) can<br />
be computed by substituting the altimeter parameters from Table 2.1. The computed<br />
2<br />
values of ασ as a function of SWH are listed in Table 5.1. It can be seen, assuming<br />
2<br />
that a typical situation of the ocean surface (i.e., SWH = 4 m), the values of ασ are<br />
approximately 0.16 ns for both ERS-1/2 and TOPEX. Thus, this term may be safely<br />
neglected for the purpose of the waveform retracking, and Equation (5.8) can be<br />
reduced further to<br />
Table 5.1 ασ² Values for ERS-1/2 (ocean mode) and TOPEX<br />
SWH (m) ERS-1/2 (ns) TOPEX (ns)<br />
0.5 0.01 0.01<br />
2.0 0.05 0.05<br />
4.0 0.16 0.16<br />
6.0 0.35 0.35<br />
8.0 0.62 0.61<br />
10.0 0.96 0.95<br />
12.0 1.38 1.36<br />
14.0 1.88 1.85<br />
16.0 2.45 2.41<br />
18.0 3.10 3.05<br />
20.0 3.82 3.76<br />
P<br />
1<br />
2<br />
⎡<br />
⎢<br />
⎣<br />
⎛ t − t<br />
[ − ( t − )]<br />
0<br />
( t)<br />
= PN<br />
+ PA<br />
erf ⎜ ⎟ + 1 exp α t0<br />
⎝<br />
⎛ t − t<br />
⎝ σ<br />
⎞<br />
2σ<br />
⎠<br />
⎤<br />
⎥<br />
⎦<br />
0<br />
= PN<br />
+ PA<br />
P⎜<br />
⎟exp<br />
α<br />
⎞<br />
⎠<br />
[ − ( t − t )]<br />
0<br />
(5.9)<br />
where P (x)<br />
is the normal probability distribution (Equation 3.42).
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 133<br />
Comparing Equation (5.9) with Equations (3.40) and (3.43) [set i = 1 ], the only<br />
difference between the two functions is in the trailing edge of the waveform. In the<br />
region of the trailing edge, the normal probability distribution P (x)<br />
is equal to unity<br />
for SWHs less than 10 m. Thus, in this region, Equations (5.9), (3.40), and (3.43) can<br />
be written as P t<br />
(t)<br />
, y t<br />
(t)<br />
and y te<br />
(t)<br />
as<br />
[ − ( t − )]<br />
P ( t)<br />
= P + P exp α t<br />
(5.10)<br />
t<br />
y t<br />
N<br />
A<br />
1<br />
+ β<br />
21<br />
1<br />
( β )<br />
51<br />
1<br />
0<br />
( t)<br />
= β + Q<br />
(5.11)<br />
and<br />
y t<br />
1<br />
+ β<br />
21<br />
exp<br />
( − β )<br />
( t)<br />
= β<br />
Q<br />
(5.12)<br />
51<br />
1<br />
From the linear ocean surface (Equation 5.11), β<br />
51<br />
is the slope of the trailing edge.<br />
In addition, Equations (5.10) and (5.12) can be rewritten more simply as<br />
( P ( t)<br />
P ) = [ ln( P ) +α t ] −α<br />
t<br />
ln (5.13)<br />
t<br />
−<br />
N<br />
A 0<br />
and<br />
ln<br />
( y t<br />
( t)<br />
β1 ) = ln( β<br />
21<br />
) − β<br />
51Q1<br />
− (5.14)<br />
where α and β<br />
51<br />
are also the slope of the trailing edge of the ocean model and the<br />
five-β-parameter model with a exponential decay trailing edge, respectively.<br />
The slope of the trailing edge in the ocean model is related to the physical parameters<br />
of the radar antenna pattern, the Earth’s radius, and the off-nadir angle, according to<br />
4c<br />
1<br />
α =<br />
cos(2ξ<br />
)<br />
γ h (1 + h / R)<br />
(5.15)<br />
However, it can be seen from Equations (5.11) and (5.14) that the slope of the<br />
trailing edge depends on only a non-physical (or empirical) parameter, β 51<br />
. In this<br />
sense, the five-β-parameter model is derived partly from the ocean model (Equation<br />
5.3) or the Brown (1977) model (Equation 3.12) in the case of a linear ocean surface.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 134<br />
The slope of the waveform trailing is presented using an empirical parameter rather<br />
than the antenna off-nadir angle in the waveform. Therefore, the five-β-parameter<br />
model may be defined as a semi-empirical surface-scattering model.<br />
Variations of the slope of the waveform’s trailing edge modelled using the five-βparameter<br />
function (with linear trailing edge) for SWH = 4 m, β 21<br />
= 100 (counts),<br />
and − 0.03<br />
≤ β<br />
51<br />
≤ 0. 03 (0.01 interval) are shown in Figure 5.5. From this, two<br />
effects are evident:<br />
1000.0<br />
β5=0.03<br />
800.0<br />
β5=0.02<br />
Power (counts)<br />
600.0<br />
400.0<br />
β5=0.01<br />
β5=0.0<br />
β5=−0.01<br />
200.0<br />
β5=−0.02<br />
0.0<br />
β5=−0.03<br />
0 8 16 24 32 40 48 56 64 72 80<br />
Bins<br />
Figure 5.5 Variations of the trailing edge’s slope with varying parameter β 5 ,<br />
modelling waveforms using the five-β-parameter parameter function (linear trailing<br />
edge, SWH = 4 m, β<br />
21<br />
= 500 counts, β 5 = β<br />
51).<br />
(1) As the value of β 51<br />
increases, the power of the waveform increases, i.e., the<br />
values of β 51<br />
affect the amplitude. Similarly as for the ocean model, the<br />
position of the mid-point on the leading edge of the waveform is not influenced<br />
by variations of the slope in the trailing edge.<br />
(2) The slope of the trailing edge increases with increasing values of β 51<br />
. The<br />
descending trailing edge of the waveform changes to a rising trailing edge after<br />
β<br />
51= 0.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 135<br />
The results in Figure 5.5 show a common feature with Figure 5.4, in which the<br />
trailing edge region is nearly linear. However, β<br />
51<br />
can vary to a greater extent than<br />
the slope of the ocean model. This makes it possible that the five-β-parameter can fit<br />
more complex waveform shapes over non-ocean surfaces.<br />
Figure 5.6 shows modelling ERS-2 waveforms using the five-β-parameter function<br />
(Equation 3.43, with an exponential decayed trailing edge) for SWH = 4 m, β 21<br />
=<br />
100 counts, and − 0.03<br />
≤ β<br />
51<br />
≤ 0. 03 (0.01 interval). From Figure 5.6:<br />
(1) Although β<br />
51<br />
mainly affects the slope of the trailing edge and the amplitude of<br />
the waveform, it slightly changes the slope of the leading edge. However, this<br />
change does not vary the mid-point location on the leading edge.<br />
(2) Unlike the effects in the case of the linear trailing edge, the exponential decay<br />
trailing edge decreases both the slope and the amplitude with increasing β 51<br />
.<br />
Again, the slope of the trailing edge changes its direction after β<br />
51= 0.<br />
1500.0<br />
β5=−0.03<br />
Power (counts)<br />
1000.0<br />
500.0<br />
β5=−0.02<br />
β5=−0.01<br />
β5=0.0<br />
β5=0.01<br />
β5=0.02<br />
β5=0.03<br />
0.0<br />
0 8 16 24 32 40 48 56 64 72 80<br />
Bins<br />
Figure 5.6 Variations of the trailing edge’s slope with varying parameter β 5 ,<br />
modelling waveforms using the five-β-parameter parameter function (exponential<br />
decayed trailing edge, SWH = 4 m, β<br />
21<br />
= 500 counts, β 5 = β<br />
51).
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 136<br />
In addition, if the above analysis is performed on Equation (5.3), one can obtain the<br />
trailing edge part as<br />
2<br />
⎡ ⎛ ασ ⎞⎤<br />
P(<br />
t)<br />
= P + exp⎢−<br />
⎜ − −<br />
⎟<br />
N<br />
A α t t0<br />
⎥<br />
(5.16)<br />
⎣ ⎝ 2 ⎠⎦<br />
and<br />
ln<br />
[ P(<br />
t)<br />
− P ]<br />
N<br />
⎡ ⎛<br />
= ln( A)<br />
+ ⎢−α<br />
⎜t<br />
− t<br />
⎣ ⎝<br />
⎡ ⎛<br />
= ⎢ln(<br />
A)<br />
+ α<br />
⎜t<br />
⎣ ⎝<br />
0<br />
0<br />
2<br />
ασ ⎞⎤<br />
−<br />
⎟⎥<br />
2 ⎠⎦<br />
2<br />
ασ ⎞⎤<br />
+<br />
⎟⎥<br />
−αt<br />
2 ⎠⎦<br />
(5.17)<br />
in which<br />
A<br />
0<br />
= A A , and α is the slope of the trailing edge of the ocean waveform.<br />
ξ<br />
Since only ξ is a variable in the function α (ξ ) related to an altimeter<br />
(Equation 5.15), the trailing edge’s slope varies with variation of ξ . This is the same<br />
result as that discussed in Section 5.5.1, in which ξ has been found to be a dominant<br />
factor of the trailing edge’s slope.<br />
By comparison, it is evident that different models have different slopes of the trailing<br />
edge and variations in the slope of the trailing edge do not change the position of the<br />
mid-point on the leading edge of the waveform. In coastal regions, however,<br />
waveforms show diverse shapes with different slopes (see Figure 5.2) of the trailing<br />
edge. An inappropriate slope description may degrade the parameter estimates of the<br />
mid-point (i.e., retracking gate) on the leading edge or cause a failure of the fitting<br />
procedure. The above analysis is significant for understanding the differences among<br />
the fitting functions. More importantly, it is the different slope characters of the<br />
trailing edges that make these functions complement each other for retracking<br />
different waveform shapes in coastal regions.<br />
5. 6 Iterative Nonlinear Fitting Procedure<br />
No matter whether or not the convolution functions contain non-linear wave<br />
parameters, the solution of the fitting functions is characterised by a non-linear<br />
model. Thus, there is no straightforward method to optimally fit the model to the
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 137<br />
return waveforms and extract the model parameters. This is further complicated due<br />
to the noisy and shape-mixed features of measured waveforms in coastal regions. All<br />
fitting models describe the mean (or idealised) power, whereas real waveforms<br />
measured at the altimeter fluctuate about the mean because of the fading noise,<br />
variations in the surface geometry, and inhomogeneities near the land. Therefore, the<br />
determination of model parameters is based on an iterative least squares approach<br />
and an appropriate weighting scheme.<br />
5.6.1 Solutions of the Waveform Retracking Fitting Function<br />
In general, non-linear models in the standard statistical techniques are linearised by<br />
using approximate values via a first-order Taylor series expansion of the unknown<br />
parameters. The corrections to the approximate values become the unknown<br />
parameters and solutions for the corrections are iterated in order to improve the<br />
approximate values. Taking y i<br />
as a function of the estimated parameter<br />
k = 1,<br />
2, K,m<br />
, it can be written as:<br />
x<br />
k<br />
,<br />
y<br />
=<br />
( x )<br />
i<br />
F i<br />
m×1<br />
(5.18)<br />
where<br />
i = 1,<br />
2, L,<br />
n is the bin-number of the waveform samples in range window<br />
(e.g., n = 64 for ERS-1/2). Assuming x = x 0<br />
+ δ x , y<br />
i<br />
is approximated at x 0<br />
in the<br />
parameter-space by the Taylor series expansion to the first order as<br />
∧<br />
y ≈ y = F x ) + F′<br />
( x ) δ x<br />
i<br />
i<br />
i<br />
(<br />
0 i 0<br />
(5.19)<br />
where F x ) is an approximation computed by the fitting function using the<br />
i<br />
( 0<br />
approximated values of the unknowns, and<br />
F′<br />
( x<br />
i<br />
0<br />
)<br />
⎛<br />
⎜ ∂y<br />
=<br />
⎜ ∂x1<br />
⎝<br />
∂y<br />
∂x<br />
L<br />
∂y<br />
∂x<br />
x 2<br />
0 x<br />
m<br />
0<br />
x0<br />
⎞<br />
⎟<br />
⎟<br />
⎠<br />
(5.20)<br />
is a matrix populated by the partial derivatives of the functional model with respect<br />
to the unknowns. From Equations (5.19) and (5.20), the function is formed as
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 138<br />
V = Aδ x − L<br />
(5.21)<br />
where A = F′<br />
x ) is the design matrix, L = l − F x ) is the vector of the n<br />
( 0<br />
observations determined by the measured waveforms and the first-order<br />
approximations at x<br />
0<br />
. Equation (5.21) can be solved by the least squares approach:<br />
( 0<br />
T −1<br />
T<br />
δ x = ( A PA)<br />
( A PL)<br />
(5.22)<br />
where P is the weight matrix of the observations. The standard deviations of the<br />
estimated parameters are given as<br />
2 T −1<br />
Q x<br />
= σ<br />
0<br />
( A PA)<br />
,<br />
(5.23)<br />
σ<br />
2<br />
0<br />
=<br />
T<br />
V PV<br />
n − m<br />
.<br />
(5.24)<br />
The initial approximation of x<br />
0<br />
is corrected to produce revised estimates<br />
x = x 0<br />
+ δ x . Then, the x replaces the original x 0<br />
as new a priori values and the<br />
iterative procedure is repeated N times until a defined relative error criterion is<br />
satisfied, or a maximum number of iterations is reached.<br />
5.6.2 Linearisation of the Waveform Retracking Fitting Function<br />
Having expressed the return power as analytic functions, the linearisation of these<br />
functions can be explicitly implemented by computing the partial derivatives of each<br />
parameter. Taking the revised ocean model (Equation 5.3) as an example, the five<br />
partial derivatives of the modelled waveform with respect to the parameters are given<br />
by<br />
∂P<br />
∂<br />
P N<br />
= 1<br />
(5.25)<br />
∂P =<br />
∂A<br />
P<br />
0<br />
A 0<br />
(5.26)
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 139<br />
∂P<br />
∂t<br />
0<br />
=<br />
A0<br />
⎧ 2<br />
2 ∂u<br />
∂v<br />
Aξ<br />
exp( −v)<br />
⎨ exp( −u<br />
) − [ erf ( u)<br />
+ 1]<br />
2 ⎩ π ∂t0<br />
∂t0<br />
⎫<br />
⎬<br />
⎭<br />
(5.27)<br />
∂P<br />
=<br />
∂σ<br />
A<br />
2<br />
0<br />
⎧<br />
Aξ<br />
exp( −v)<br />
⎨<br />
⎩<br />
2<br />
exp( −u<br />
π<br />
2<br />
∂u<br />
) −<br />
∂σ<br />
[ erf ( u)<br />
+ 1]<br />
∂v<br />
⎫<br />
⎬<br />
∂σ<br />
⎭<br />
(5.28)<br />
∂P<br />
2<br />
∂ξ<br />
=<br />
A0<br />
2<br />
+<br />
⎪⎧<br />
1<br />
Aξ<br />
exp( −v)<br />
⎨<br />
⎪⎩ A<br />
2<br />
exp( −u<br />
π<br />
2<br />
ξ<br />
∂u<br />
)<br />
2<br />
∂ξ<br />
[ erf ( u)<br />
+ 1]<br />
−<br />
∂Aξ<br />
2<br />
∂ξ<br />
[ erf ( u)<br />
+ 1]<br />
∂v<br />
2<br />
∂ξ<br />
⎫<br />
⎬<br />
⎭<br />
(5.29)<br />
where<br />
u =<br />
τ<br />
2<br />
(5.30)<br />
⎛ d ⎞<br />
v = d⎜τ<br />
+ ⎟<br />
⎝ 2 ⎠<br />
(5.31)<br />
It should be noted that the derivative with respect to ξ 2 has been computed rather<br />
than the derivative with respect to ξ in Equation (5.29). The reason for this is that<br />
the derivative with respect to ξ could be zero if ξ → 0<br />
, and hence the off-nadir<br />
angle cannot be estimated. Thus, assuming that the two sinusoidal terms in Equations<br />
(3.14) and (3.15) can be approximated to the first order for small angles, ξ can be<br />
2<br />
replaced by ξ to avoid the problem. An additional problem with the introduction of<br />
2<br />
ξ is that the estimate of ξ 2 could contain negative values due to the presence of<br />
noise. However, this effect can be neglected in the study, as the aim is to extract the<br />
retracking gate estimate rather than the off-nadir angle. For research that is interested<br />
only in the off-nadir estimate, other approaches can be used (cf. Barrick and Lipa,<br />
1985).<br />
5.6.3 Initial Estimates of the Unknown Model Parameters<br />
High-quality initial estimates of the parameters are important to reduce the number<br />
of iterations and guarantee a convergence in the least squares estimation (and hence
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 140<br />
computational time) that the non-linear least squares procedure takes to converge. If<br />
the initial parameters cannot be estimated precisely, the parameter corrections after<br />
one iteration might be very large, causing the algorithm to diverge rapidly or to<br />
converge on the incorrect result. However, there is no predefined method that<br />
determines the optimum a priori values for the altimeter waveforms. Therefore, to<br />
generate initial parameter estimates, different schemes were tested and evaluated<br />
based on the speed of convergence, the mean-squared error, and ability to fit a wide<br />
variety of waveforms. In accordance with the testing results of these three factors, the<br />
OCOG method can also provide additional information when making initial<br />
estimates of the parameters (see below).<br />
(1) The Ocean Model<br />
Before ascertaining the initial parameter estimates, the initial estimate of the thermal<br />
noise (or noise baseline),<br />
0<br />
P<br />
N<br />
, is determined by calculating the average amplitude of<br />
the waveform samples over five of earlier gates (e.g., gates 6-10 for POSEIDON)<br />
P<br />
1<br />
10<br />
0<br />
N<br />
= ∑W i<br />
5 i=<br />
6<br />
(5.32)<br />
where<br />
W<br />
i<br />
is the i -th measured waveform sample. This value is then subtracted from<br />
each waveform sample and the OCOG method is applied to the waveforms to<br />
calculate relative parameters of the rectangle (described in Section 3.5.2). The results<br />
from the OCOG method are used to determine the initial approximations of the<br />
amplitude term (<br />
0<br />
A<br />
0<br />
) and the time corresponding to the location of the mid-point at<br />
the range bins ( t 0 0<br />
). The waveform amplitude ( A ) is set to be the initial estimate of<br />
0<br />
A<br />
0<br />
, and t 0 0<br />
can be obtained from the gate number of the leading edge ( LEG ) from<br />
the OCOG by<br />
t<br />
0<br />
0<br />
= LEG ⋅T<br />
(5.33)<br />
where T is the duration of the radar pulse ( T = 3. 03ns for ERS-1/2 and T = 3. 125ns<br />
for TOPEX/POSEIDON).
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 141<br />
The initial approximation of the RMS surface roughness (i.e., the composite<br />
0<br />
risetime), σ<br />
cr<br />
, can be estimated by substituting SWH = 2cσ<br />
s<br />
into Equation 3.13 to<br />
give<br />
σ<br />
0<br />
cr<br />
⎡<br />
= ⎢<br />
⎢⎣<br />
( f t<br />
T )<br />
2<br />
2<br />
1/ 2<br />
⎛ SWH ⎞<br />
2 ⎥ ⎥ ⎤<br />
+ ⎜ ⎟<br />
⎝ c ⎠ ⎦<br />
(5.34)<br />
where factor f t<br />
is an approximation (0.425 for T/P and 0.526 for ERS-1/2) of the<br />
compressed pulse shape by a Gaussian function (Marth et al., 1993; Batoula et al.,<br />
1999), SWH = 4<br />
m, which is the common case over ocean surfaces. The initial<br />
2<br />
approximation of the (small) off-nadir angle squared, ξ , can be set to zero.<br />
(2) The Five-β-parameter Model<br />
The initial approximations for the five- β-parameter model are determined using the<br />
method from Anzenhofer et al. (2001), but modified slightly to account for the<br />
thermal noise. The initial estimate of the parameter β<br />
1<br />
is set using Equation (5.32).<br />
The maximum amplitude of the waveform after removing the thermal noise (item 2)<br />
is defined as the initial estimate of the parameter β<br />
2<br />
. The gate number of the leading<br />
edge from the OCOG is taken for the initial approximation of the parameter β<br />
3<br />
. The<br />
initial estimate of the waveform risetime β<br />
4<br />
is set empirically to 1.3 gates, and the<br />
slope of the trailing edge β<br />
5<br />
can be set to zero.<br />
(3) The Nine-β-parameter Model<br />
Measured waveform shapes related to this function usually show double ramps<br />
(Figure 3.6). Of these parameters, the initial estimate of the thermal noise can be set<br />
using Equation (5.32). The initial estimate for the location of the first ramp (usually<br />
related to the mean sea surface) is determined by locating the first sample gate,<br />
whose slope exceeds a threshold value (e.g., 50 counts/gate for ERS-1/2). Once this<br />
threshold level is exceeded, the next five gates are checked to find the maximum<br />
slope. The gate number corresponding to this maximum slope is used as the initial<br />
approximation for the location of the first ramp. The same method can be used to<br />
determine the initial estimate for the second ramp location. Other initial estimates
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 142<br />
can be defined using the similar method described above for the five-β-parameter<br />
model.<br />
5.6.4 Data Weight and Outlier Detection Scheme<br />
Before fitting, a priori weights are determined using the method from Anzenhofer et<br />
al. (2000). It is acknowledged that the heavier weight emphasis should be put on<br />
fitting to the ramp locations which define the retracking correction (e.g., Zwally,<br />
1996; Brenner et al., 1993; Anzenhofer et al., 2000). Using the parameters estimated<br />
from the OCOG retracker, the location of the leading edge, tp , can be estimated.<br />
Then a priori weights of the ERS-1/2 waveform can be determined empirically for<br />
the ocean model and the five-parameter model as<br />
0<br />
p i<br />
= 200 for t ≤ tp -1<br />
(5.35)<br />
= 100 for tp-1<br />
< t ≤ tp + 2<br />
0<br />
p i<br />
(5.36)<br />
0<br />
p i<br />
= 50 for tp + 2 < t ≤ 64<br />
(5.37)<br />
This is a reasonable weight scheme because it accounts for the characteristics and<br />
distribution of the useful information in the waveform. The heaviest weights are put<br />
on the waveform samples in the first plateau (i.e., thermal noise region) before the<br />
leading edge so that they will be thought to be relatively accurate measurements in<br />
the adjustment. The second heaviest weights are applied to the samples of the leading<br />
edge itself. The lightest weights are given to the waveform samples in the trailing<br />
edge, making them contribute less to the solutions (cf. Figure 3.5).<br />
In general, the data in the trailing edge show obvious undulations caused by the<br />
‘fading’ noise, which results in the incoherent superposition of signals from different<br />
reflecting facets (Partington, 1991; Marth et al., 1993; Quartly and Srokosz, 2001).<br />
The noise in the trailing edge greatly influences the iterative fitting procedure and<br />
even accurate results cannot be estimated (Tokmakian et al., 1994). To solve the<br />
problem, two approaches have been used. The first is to average waveforms in the<br />
time span of several seconds, and then to fit a model to this averaged waveform (e.g.,
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 143<br />
Tokmakian et al., 1994; Hayne and Hancock III, 1990). The second is to lightly<br />
weight the waveform samples in the trailing edge (Anzenhofer et al., 2000; Zwally,<br />
1996).<br />
However, the averaging procedure is not always appropriate in coastal regions,<br />
because ocean waveforms will be distorted by contaminated waveforms within the<br />
averaging period. Therefore, the second method is used in the study. According to<br />
Anzenhofer’s (2000) weight scheme, the data after the ramp have been<br />
downweighted in the solution (see Equation 5.37). However, it is found in this study<br />
that some unexpected undulations in the trailing edge significantly affect the solution.<br />
To overcome this problem, an outlier detection approach from an iterative weight<br />
scheme is used in this study.<br />
In this iterative procedure, the initial least squares adjustment is conducted using the<br />
0<br />
a priori values ( β<br />
k 0<br />
) and a priori weights ( p<br />
i<br />
). After the first adjustment, a weight<br />
function<br />
w<br />
i<br />
is given as (Li, 1988)<br />
⎧ 1<br />
⎪ for vi<br />
> 0.7σ<br />
0<br />
vi<br />
w<br />
i<br />
= ⎨<br />
(5.38)<br />
⎪<br />
1<br />
for vi<br />
≤ 0.7σ<br />
0<br />
⎪⎩<br />
0.7σ<br />
0<br />
2<br />
where v<br />
i<br />
and σ<br />
0<br />
, which is the estimation of the unit weight variance, are computed<br />
by Equations (5.21) and (5.24), respectively. The next step then is to determine the<br />
weight (<br />
j+1<br />
p<br />
i<br />
) for the next adjustment using wi<br />
as follows:<br />
p =<br />
j+1<br />
i<br />
p<br />
j<br />
i<br />
w<br />
i<br />
(5.39)<br />
where j means the j-th iteration in the adjustment. In the meantime, estimates of the<br />
β parameters take the place of the originals and repeat the entire cycle to produce<br />
improved estimates of the β parameters. In this way, the observations considered to<br />
contain the outlier will be downweighted but still be kept in the dataset, while nonoutlier<br />
observations will maintain full weight (though downweighted with respect to<br />
the other data in the range window) after the iterative cycle.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 144<br />
5.6.5 Iterative Procedure<br />
The procedure taken here is similar to that used by Anzenhofer et al. (2000), Zwally<br />
(1996) and Davis (1993). The parameter estimates are obtained by an iterative<br />
procedure as follows:<br />
(1) Estimate initial values of the model parameters and the weights of the<br />
measured waveforms.<br />
(2) Compute the new values for the least squares approach using the first-order<br />
Taylor expansion and measured waveforms (Equation 5.21).<br />
(3) Compute a correction to the current estimates of the unknown parameters<br />
(Equation 5.22), and the standard deviations of the parameters (Equation 5.23).<br />
(4) Update the parameter estimates for the new initial estimates.<br />
(5) Repeat steps 2-4 until some exit criterion in Table 5.2 is satisfied.<br />
From the flowchart of the fitting algorithm (Figure 5.1), after corrected parameter<br />
estimates are obtained from Equation (5.22), several exit criteria are applied to<br />
determine whether the fitting iteration continues.<br />
(1) The first exit criterion is based upon an absolute difference specified for each<br />
unknown parameter between ( i +1)<br />
and i iterations. Table 5.2 lists the<br />
absolute difference values for the ocean model parameters. If the absolute<br />
difference of all parameter estimates from two successive iterations is less than<br />
these criteria, the fitting iteration terminates.<br />
(2) The second exit criterion is based upon the relative change in the standard<br />
2<br />
deviation of the unit weight, σ<br />
0<br />
. If the relative change of the σ<br />
0<br />
is less than<br />
2% from one iteration to the next, the fitting loop is ended.<br />
(3) The final exit criterion is the maximum number of the iterations. The<br />
maximum number in this study is set to 20 to avoid infinite looping situation in<br />
the case of divergence. In practice, a flag which records the status of the fitting<br />
loop will be returned with the parameter estimates. The parameter estimates are<br />
retained if the algorithm exits before 20 iterations and the flag does not show<br />
iteration failure, while they are discarded if the other criteria fail in 20<br />
iterations or the loop ends at the maximum iterative number.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 145<br />
In addition, for each iteration step, some constraints are applied, such as amplitude<br />
parameter estimate must be larger than zero.<br />
Table 5.2 Criteria of the absolute differences between two iterations (the ocean<br />
model for ERS-1/2)<br />
Parameters Absolute Differences Units<br />
P<br />
N<br />
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 146<br />
Equation (5.32) using five unaliased waveform samples before doing retracking.<br />
Once the estimate of P<br />
N<br />
is obtained, the threshold level, T<br />
l<br />
, is then set to<br />
T = P + T<br />
(5.40)<br />
l<br />
N<br />
LO<br />
where T<br />
LO<br />
is the threshold level determined using the selection scheme (d) in Section<br />
5.7.2. The retracking location on the leading edge of the waveform (or at the range<br />
window),<br />
g<br />
r<br />
, is linearly interpolated between the bins adjacent to T<br />
l<br />
using<br />
=<br />
T<br />
−1+<br />
−W<br />
l k −1<br />
g<br />
r<br />
g<br />
k<br />
(5.41)<br />
Wk<br />
−Wk<br />
−1<br />
where<br />
g<br />
k<br />
is the location of the first gate exceeding T l<br />
. The g r<br />
is set to ( g<br />
k<br />
−1)<br />
in<br />
the case of W k<br />
W<br />
1<br />
in Equation (5.41). Then, the range correction can be obtained<br />
= k −<br />
from a bin-to-metre conversion factor timing the gate difference between this<br />
estimate and the designed tracking gate as<br />
dr = G ⋅ ( g −<br />
0<br />
)<br />
(5.42)<br />
2M<br />
r<br />
g<br />
where g<br />
0<br />
is the expected tracking gate in units of waveform gates.<br />
5.7.2 Selection of the Threshold Level<br />
As stated, Brooks et al. (1997) select the threshold level for mixed waveform shapes<br />
in coastal regions based upon a visual examination method. These are two concerns<br />
related to this selection. The first is that threshold values are higher than both the<br />
noise and the minor waveform leakage in the early gates. The second is that the<br />
values are sufficiently lower than the sought-after ocean returns. Brooks et al. (ibid.)<br />
determine different threshold values for 18 TOPEX ground tracks near land using<br />
both concerns. Then, they are used to retrack the coastal waveform along the same<br />
pass. This method is good for selecting the threshold level because it takes account<br />
for the strength of the power backscattered from different reflecting surfaces.<br />
However, because an automatic selection of the threshold level is not available, this<br />
restrains the technique from retracking a wide range of the waveforms in coastal<br />
regions.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 147<br />
Considering the shape variations of the waveform in the proximity of the coasts<br />
analysed in Section 4.6, the selection of the threshold level used in the coastal<br />
retracking system differs from these previously published methods used over ice<br />
sheets (Zwally, 1996; Davis, 1995; Davis, 1997) and coastal regions (Brooks et al.,<br />
1997). Four threshold schemes are investigated before arriving at the final threshold<br />
level.<br />
(a) Fixed Counts<br />
The threshold level of the fixed count is defined here as the 50% of the meanwaveform<br />
amplitude in counts. For each satellite ground pass, the mean waveform is<br />
formed from one or two seconds of purely ocean waveforms prior to the coastal<br />
waveforms along it. The power in counts corresponding to the 50% of the meanwaveform<br />
amplitude is the fixed count for the pass. This fixed count is, then, used as<br />
the threshold level to retrack the waveforms along the same pass into coastal regions.<br />
The gate location corresponding to this value is linearly interpolated between the<br />
bins adjacent to the crossing of the fixed counts on the leading edge of the coastal<br />
waveform, being the estimate of the retracking gate. This scheme, in fact, determines<br />
the retracking gate estimates using the 50% of the ocean-waveform amplitude, but it<br />
assumes that the amplitude of the ocean returns does not change with respect to the<br />
reflecting surfaces.<br />
(b) 50% of the Waveform Amplitude<br />
Over open oceans where the waveform is completely dominated by surface scattering,<br />
the half-power point (i.e., the 50% threshold retracking point) best represents the<br />
expected tracking-gate estimate which corresponds to the range to the at-nadir<br />
surface within the altimeter pulse-limited footprint. Therefore, the second selecting<br />
scheme is to choose the 50% threshold point as the threshold level.<br />
(c) 30% and 50% of the Waveform Amplitude<br />
Factors that affect the waveform’s shape (or amplitude) are primarily the off-nadir<br />
angle or the surface slope and the surface geometry. Compared with brighter and<br />
higher land, where returns usually show higher power, ocean returns that appear in<br />
the range window tend to be lower power. Therefore, in the case of the mixed<br />
waveform of land and water, an appropriate low threshold level is suitable such that<br />
retracking can locate the half-power point of the leading edge of ocean returns.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 148<br />
Mean Power and STD (counts)<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
Mean waveform<br />
STD<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64<br />
Bins<br />
Mean Power and STD (counts)<br />
1200<br />
1000<br />
800<br />
600<br />
400<br />
200<br />
0<br />
Mean waveform<br />
STD<br />
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64<br />
Bins<br />
Figure 5.7 Mean waveforms and the standard deviation in two 5 km-wide bands of<br />
0-5 km (top) and 5-10 km (bottom) from the Australian coastline.<br />
Therefore, in this third scheme, two threshold levels are selected depending on the<br />
distance between the observing point along the satellite groundtrack and the coast.<br />
The 30% threshold level is only selected for the waveforms within 5 km distance<br />
from the coast, while a threshold level of the 50% of the waveform amplitude is<br />
chosen after 5 km from the coastline. The distance of 5 km is determined based upon<br />
an analysis of the mean waveforms (Section 4.5.2). To determine the proper<br />
threshold level within different distance bands from the coast, the average waveform<br />
is formed using one cycle of ERS-2 20 Hz waveforms from the coastal regions of
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 149<br />
interest within six 5 km-wide bands from the Australian coastline. Figure 5.7 shows<br />
the mean waveforms and STD in two 5 km-wide bands of 0-5 km and 5-10 km from<br />
the coastline. Beyond 10 km from the coastline, the mean waveforms show no<br />
significant change to the 5-10 km band.<br />
The position of the threshold crossing corresponds to the mid-point of the leading<br />
edge. The 30% of the waveform amplitude (or 20% of the amplitude in some areas,<br />
discussed below) is selected to be the threshold level for retracking waveforms<br />
within 0-5 km from the coasts. When the distance from the coast is larger than 5 km,<br />
the ocean scattering becomes predominant in the waveform returns so that the 50%<br />
of the waveform amplitude can be used as an estimate of the threshold level.<br />
(d) Varying Threshold Level<br />
Although the 30% threshold level should effectively be a low-power strength<br />
retracker in coastal regions for the mixed waveforms of land and water, it might<br />
cause two problems. The first one is that waveforms without contamination will be<br />
retracked inaccurately by this lower-level retracker. The second is that changing the<br />
threshold level at 5 km distance from the coast might cause a step jump to the<br />
retracked range measurements. To avoid both problems, a scheme of varying<br />
threshold level is developed.<br />
Different threshold levels are set depending on whether or not the waveform is<br />
contaminated. If the waveform classification indicates a distorted waveform, the<br />
threshold level is set to 30% (or 20%) of the waveform amplitude, otherwise the 50%<br />
of waveform amplitude is set. Compared with the waveform shape, the distance from<br />
the coast is not significant in this criterion. It links the threshold level to the<br />
contaminated waveform, avoiding the step change that might occur in corrected<br />
range measurements from the threshold level (c).<br />
5.7.3 Results of Selecting the Threshold Levels<br />
In order to ascertain the appropriate threshold level, one cycle of 20 Hz ERS-2<br />
waveform data (March to April 1999) and ten (15º×15º) areas around the Australian<br />
coast are used (Figure 5.8). The AUSGeoid98 (2'×2') geoid grid is used as an<br />
external ‘ground truth’ for the result comparisons (see also Section 2.5.1). Because
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 150<br />
the waveform contamination strongly depends on both the distance from the coast<br />
and land surface topography along the coastline, computations are conducted in six 5<br />
km-wide bands around the coasts. Waveforms have been threshold retracked using<br />
the above schemes, respectively. The range correction calculated from waveform<br />
retracking and geophysical corrections from the data products are then applied to the<br />
measured range. The retracked SSH data are then obtained from the satellite altitudes<br />
and corrected range measurements.<br />
105˚<br />
0˚<br />
120˚<br />
135˚<br />
150˚<br />
165˚<br />
0˚<br />
1 2<br />
-15˚<br />
3<br />
4 5<br />
6<br />
-15˚<br />
-30˚<br />
-30˚<br />
7<br />
8 9<br />
10<br />
-45˚<br />
105˚<br />
120˚<br />
135˚<br />
150˚<br />
-45˚<br />
165˚<br />
Figure 5.8 Ten 15º×15º coastal areas around Australia, showing the area number<br />
(Lambert projection).<br />
The comparison between the instantaneous SSH and the geoid height is an<br />
appropriate tool to investigate the overall quality of the threshold-level selecting<br />
schemes. For each along-track altimeter SSH measurement after retracking, the<br />
gridded AUSGeoid98 geoid height is bispline-interpolated to the longitude and<br />
latitude of the point. The difference between the mean SSH and geoid height can be<br />
obtained by subtracting the interpolated geoid height from the corresponding SSH.<br />
Then, the descriptive statistics are computed from this difference. The instantaneous<br />
SSH, of course, is not equivalent to the geoid (see Section 2.5.2). The difference
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 151<br />
between them includes the time variation (e.g., tides) and dynamic SST. However,<br />
averaging can remove the time variation and so the remaining difference is between<br />
the mean SSH and the geoid height. Since the precisely regional geoid model<br />
AUSGeoid98 is used, the mean difference can be a good approximation of the<br />
dynamic SST, which is of the order of ~1-2 m (e.g., Chelton et al., 2001). Therefore,<br />
the verification of any improvement in the SSH will only be seen in the standard<br />
deviations of the mean difference, with a smaller value indicating an improvement in<br />
the determination of the threshold level.<br />
Figures 5.9 and 5.10 show mean differences of the SSH and geoid heights and<br />
standard deviations in a distance out to 5 km around Australia’s coasts, respectively.<br />
Before retracking, the magnitudes of mean differences in ten areas are less than 2 m,<br />
but the standard deviations (STD) are large in most of the areas. In area one, where<br />
the land topography involves rocks and small barrier reefs in the north Australian<br />
coasts, the STD is even larger (±19.93 m). The STDs are small (±0.47 m and ±0.49<br />
m) in south-western and south-eastern areas seven and ten, where the coastal relief<br />
comprises rocks, cliffs and large barrier reefs (cf. Thom, 1984).<br />
After threshold retracking using scheme (b), the STD decreases in six areas.<br />
However, it increases in the other four areas compared with the STD before<br />
retracking. This suggests that the 50% threshold level is not an appropriate selection<br />
for the waveforms within 5 km from the coast. The reason for this is that the mostly<br />
varying land topography and standing water within this distance can heavily<br />
contaminate the waveforms, so that many waveform shapes do not follow the ocean<br />
model. After retracking using schemes (c) and (d), the STD shows improvements in<br />
seven areas. Of them, the results from the scheme (d) show the most significant<br />
improvement than any others. In contrast, scheme (a) gives the least improvement,<br />
and even worse results in most areas after retracking compared with the results<br />
before retracking. For areas 5, 8 and 10, an appropriate threshold level of 20% of the<br />
waveform amplitude has been found from the tests.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 152<br />
2.5<br />
2<br />
unretracked<br />
50% threshold level<br />
30% threshold level<br />
varying threshold level<br />
fixed counts<br />
1.5<br />
Mean Differences (m)<br />
1<br />
0.5<br />
0<br />
−0.5<br />
−1<br />
1 2 3 4 5 6 7 8 9 10<br />
Areas<br />
Figure 5.9 Mean differences between the SSH data after retracking and AUSGeoid98<br />
geoid heights in ten areas 5 km from the Australian coastline.<br />
25<br />
20<br />
unretracked<br />
50% threshold level<br />
30% threshold level<br />
varying threshold level<br />
fixed counts<br />
15<br />
STD (m)<br />
10<br />
5<br />
0<br />
1 2 3 4 5 6 7 8 9 10<br />
Areas<br />
Figure 5.10 Standard deviations of the mean differences in ten areas 5 km from the<br />
Australian coastline.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 153<br />
Figures 5.11 and 5.12 show mean differences of the SSH and geoid heights and<br />
standard deviations in a distance band of 5-10 km from Australia’s coasts,<br />
respectively. As can be seen from the ordinate scale in Figures 5.11 and 5.12, both<br />
mean differences and the STDs decrease with increasing the distance from the coasts.<br />
It is evident that the SSH is improved after retracking no matter which threshold<br />
level is used. The schemes (b) and (d) show similar results in most areas. This<br />
implies that less waveform contamination occurs in these areas, which gives the<br />
same result as that shown in Chapter 4. The results from scheme (a) present<br />
improvements in this distance band as well. However, the results from Figures 5.11<br />
and 5.12 suggest that scheme (d) is the best threshold level.<br />
The mean differences and the STD in four other 5 km-wide distance bands (10-30<br />
km from the coast) were computed and analysed. When the distance between the<br />
ground track and coastline is far more than 10 km, schemes (b) and (d) generate a<br />
very good agreement for both the mean and STD from the results after retracking.<br />
Results suggest that less waveform contamination occurs in these areas and the<br />
threshold level can be set to 50% of the waveform amplitude.<br />
During testing, it became evident that the varying threshold level can be used to<br />
effectively threshold retrack most waveforms in coastal regions, since it takes into<br />
account the coastal waveform characteristics. The count-fixed level (scheme a)<br />
cannot follow the variations of the scattering surfaces. It assumes that the retracking<br />
amplitude of the ocean waveform does not change with varying the surface<br />
roughness or topography. However, this is not the actual case of the waveform even<br />
over ocean surfaces. The 50% threshold retracking point is set too high to catch the<br />
ocean returns for the contaminated waveforms in coasts. Setting the 30% threshold<br />
level within 5 km from the coast and the 50% threshold after 5 km (scheme c)<br />
presents an improved result. However, changing the threshold level suddenly at a<br />
distance of 5 km causes a step change in the range measurements after retracking. In<br />
contrast, the varying threshold level cannot only generate improved SSH data, but<br />
also avoids this step change.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 154<br />
1.4<br />
1.2<br />
unretracked<br />
50% threshold level<br />
varying threshold level<br />
fixed counts<br />
1<br />
Mean Differences (m)<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
−0.2<br />
1 2 3 4 5 6 7 8 9 10<br />
Areas<br />
Figure 5.11 Mean differences between the SSH data after retracking and<br />
AUSGeoid98 geoid heights in ten areas 5-10 km from the Australian coastline.<br />
12<br />
10<br />
unretracked<br />
50% threshold level<br />
varying threshold level<br />
fixed counts<br />
8<br />
STD (m)<br />
6<br />
4<br />
2<br />
0<br />
1 2 3 4 5 6 7 8 9 10<br />
Areas<br />
Figure 5.12 Standard deviations of the mean differences in ten areas 5-10 km from<br />
the Australian coastline.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 155<br />
It should be pointed out that the majority of coastal altimeter waveforms exhibit an<br />
ocean-like shape with a single ramp (see Chapter 6). Only a small percentage of them<br />
show a significant peak or a combination of ocean and land returns. Consequently,<br />
the use of the 30% and 50% threshold levels over large areas of the coastal regions is<br />
more reasonable from these results. In the test, it has also been found that only 30%<br />
of threshold level sometimes might not be sufficiently low to intercept the soughtafter<br />
ocean returns from the waveforms with double ramps. The 20% retracking point<br />
has been found in this investigation to be the best threshold level for areas five, eight,<br />
and ten (cf. Figure 5.8).<br />
5. 8 An Assessment of Biases among Waveform Retracking Algorithms<br />
In this study, waveform retracking is carried out in coastal regions. Thus, the first<br />
concern in this Section is whether or not the retracked data in coasts matches the<br />
unretracked data in open oceans, and if not, to find the difference (or bias) between<br />
them. In fact, biases have been found in this investigation and in a previous study<br />
(Anzenhofer et al., 2000). In the results of Anzenhofer et al. (2000), a bias<br />
(~66.4±10.7 cm) has shown between the ERS-1 GDR unretracked range and the<br />
retracked range using the five-β-parameter function.<br />
In addition, there is no ‘universal’ retracking algorithm that can deal with all types of<br />
the waveform data (Berry et al., 1998; Martin et al., 1983; e.g., Berry, 2000).<br />
Therefore, a bias question is: if different retrackers are used to retrack individual<br />
waveform data, is there any bias exiting between among retrackers? The second<br />
concern is to estimate if there is a bias among different retracking algorithms, and to<br />
quantify and correct for it.<br />
5.8.1 Analysis Methods and Data<br />
Both the iterative least squares method (Section 5.6.5) and threshold retracking<br />
methods (Section 5.7.1) are used to assess the performance of different retracking<br />
algorithms in coastal regions through waveform retracking. These retrackers include<br />
the 50% and 30% threshold levels, the ocean model and five-β-parameter model. The<br />
five- β -parameter function with an exponential trailing edge is not used, because the
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 156<br />
results of Anzenhofer et al. (2000) indicate that the bias between the five- β -<br />
parameter functions with the linear and exponential trailing edges is insignificant<br />
(~1.3±10.8 cm). For convenience of presentation, hereafter the five- β -parameter<br />
model is called the five-parameter model.<br />
The data used are cycles 42 and 43 of ERS-2 20 Hz waveforms over an ocean area<br />
around Australia out to 350 km from the coastline. Contaminated waveforms in<br />
coasts (obtained in Chapter 4) are retracked first to investigate how retracking<br />
improves the range and thus the SSH, and to find biases. Then, waveforms in an<br />
ocean area of 50-350 km off the coastline are retracked to estimate biases between<br />
both algorithms and SSHs before and after retracking. Waveforms are retracked<br />
using each retracker to estimate the range correction and then the retracked SSH.<br />
Retracked SSHs are used to compute the SSH differences between different<br />
retrackers at each geographical location. The bias is finally estimated by fitting a<br />
cubic polynomial function to the SSH differences.<br />
Both before and after waveform retracking, it is likely that there is still a small<br />
percentage (~4% in this study) of SSH data that may present as outlier measurements.<br />
Therefore, the Z-score is employed as an outlier detection method when computing<br />
the bias<br />
where<br />
∆SSH i<br />
− ∆SSH<br />
Z<br />
i<br />
=<br />
(5.1)<br />
STD<br />
∆ SSH is the difference of the SSH data at the same geographical position<br />
(assumed being normal distribution), STD is the standard deviation of the mean<br />
∆ SSH , Z<br />
i<br />
is the Z-score and<br />
the variable<br />
Z<br />
i<br />
< 3 is selected. The value of<br />
Z<br />
i<br />
> 3 indicates that<br />
SSH<br />
i<br />
is not within 3 STD of the sample mean, and may be considered<br />
as an outlier and deleted.<br />
The above analysis method is similar to the collinear analysis technique, which is<br />
usually applied to repeated altimeter data to assess the mesoscale variability of the<br />
sea surface (e.g., Cheney and Marsh, 1983; Sandwell and McAdoo, 1990), average<br />
repeated SSHs to estimate a mean sea level and its variations (Nerem, 1995; Leben et<br />
al., 1990; e.g., Deng et al., 1997), or compute gravity anomalies (e.g., McAdoo,
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 157<br />
1990). Since the inter-comparison here is conducted between different retracking<br />
algorithms using the same data set, the analysis is also different from the collinear<br />
technique, which involves the time variation and the small separation of the same<br />
ground track (< 1 km) occurring between different repeat orbits. This makes it more<br />
effective to estimate the biases among the different retracking methods.<br />
5.8.2 Retracking Results and Analysis<br />
All contaminated ERS-2 waveform (20 Hz, cycles 42 and 43, from March to May<br />
1999) estimated in Chapter 4 at the Australian coast have been retracked using the<br />
fitting and threshold algorithms. Figure 5.13 shows two typical examples of the<br />
retracking results, plotting the SSH (20 Hz) profile versus along-track distances. In<br />
order to evaluate how the waveform is contaminated near the coast and improved by<br />
the retracking techniques, the geoid height has been bispline interpolated at the each<br />
location along the track using the AUSGeoid98 grid. The geoid height profiles are<br />
also plotted versus along track distances in Figure 5.13. The profiles at the top of<br />
Figure 5.13 represent a ground track 21085 approaching land from water, while those<br />
at the bottom of the Figure is another ground track 21364 leaving land to water.<br />
As stated in Sections 2.5.2 and 5.7.3, it is acknowledged that the geoid height does<br />
not equal the SSH measured by the altimeter. There is a dynamic SST between them,<br />
which can be of the order of ~1-2 m. This difference is evidenced in Figure 5.13.<br />
However, the absolute SST cannot be determined at this time due to the uncertainty<br />
in the geoid undulation (cf. Chelton et al., 2001). In this case, applying a SST<br />
correction may cause an additional error to the SSH, which might be larger than the<br />
errors of other corrections (Hwang, 2003). Therefore, the SST is not applied to the<br />
SSH data in Australian coastal regions (Chapters 5 and 6) when an absolute height<br />
comparison is conducted. The difference between the SSH and geoid height is not<br />
taken as an indicator for the purpose of comparison or analysis. Instead, the profile’s<br />
trends of the SSH and geoid or the along-track slopes will be considered as a more<br />
appropriate indicator to examine the waveform contamination and the improvement<br />
of the retracking procedure.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 158<br />
−18.00<br />
−20.00<br />
−22.00<br />
−24.00<br />
SSH (m)<br />
−26.00<br />
−28.00<br />
−30.00<br />
−32.00<br />
Geoid height<br />
Unretracked<br />
Ocean model<br />
Five−parameter model<br />
50% threshold level<br />
−34.00<br />
0.0 5.0 10.0 15.0 20.0 25.0<br />
Along−track distance from the coastline (km)<br />
−8.00<br />
−10.00<br />
SSH (m)<br />
−12.00<br />
−14.00<br />
−16.00<br />
Geoid height<br />
Unretracked<br />
Ocean model<br />
Five−parameter model<br />
50% threshold level<br />
−18.00<br />
0.0 5.0 10.0 15.0 20.0 25.0<br />
Along−track distance from the coastline (km)<br />
Figure 5.13 Geoid height and SSH profiles along ground tracks 21085 (top) and<br />
21364 (bottom) (ERS-2, cycle 42) before and after retracking using the ocean model,<br />
the five-parameter model and the 50% threshold level in coastal regions.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 159<br />
As can be seen from Figure 5.13, the typical near-land effects on unretracked SSH<br />
profiles begins at the along-track distance of ~20 km and ~17.5 km from the<br />
coastline for the track 21085 (top picture) and track 21364 (bottom picture),<br />
respectively. When approaching the coastline, unretracked SSHs decrease and then<br />
increase gradually (see also Section 4.6). The difference of the minimum SSH height<br />
and the normal SSH height can be large, up to ~ -5 m within ~10 km for both tracks.<br />
In contrast to the change of unretracked SSH profile, the geoid heights show a linear<br />
trend within the whole distance.<br />
By comparison of the geoid heights with unretracked SSHs, it is evident that<br />
unretracked SSHs are affected by waveform contamination in coastal regions.<br />
In addition, according to the shape of the unretracked SSH profiles, the along-track<br />
SSH and geoid gradients are computed and listed in Tables 5.3 and 5.4. The results<br />
of the track 21085 (Table 5.3) show that along-track geoid gradients computed from<br />
the AUSGeoid98 grid are small, from 9.2 ppm to 10.9 ppm when the track is<br />
0-7.5 km and 7.6-19.5 km from the coastline. Corresponding to the same distances,<br />
SSH gradients computed from unretracked SSHs are 975.6 ppm and -509.2 ppm,<br />
respectively. They are much larger than the maximum geoid gradient value of ~150<br />
ppm in Australia (Johnston and Featherstone, 1998; Friedlieb et al., 1997; cf.<br />
Featherstone et al., 2001). After 19.5 km from the coastline, geoid gradients<br />
computed using AUSGeoid98 geoid grid and SSH gradients computed using<br />
unretracked SSHs show small values of 12.5 ppm and -6.4 ppm, respectively. These<br />
results, thus, suggest as well that unretracked SSHs are in error within an along-track<br />
distance of 0-19.5 km from the coastline.<br />
Table 5.3 Along-track geoid and SSH gradients (track 21085)<br />
Gradients<br />
Names<br />
(ppm)<br />
ds = 0.0 - 7.5 km ds = 7.6 - 19.5 km ds = 19.8 - 26.0 km<br />
Geoid height 9.2 10.9 12.5<br />
Unretracked SSH 975.6 -509.2 -6.4<br />
Analysing the along-track SSH and geoid gradients listed in Table 5.4 for the track<br />
21364 can also find that unretracked SSHs are affected within a distance of 0-20 km
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 160<br />
from the coastline. After 20 km, geoid gradients computed from AUSGeoid98 geoid<br />
grid and SSH gradients computed from unretracked SSHs take small values of<br />
8.5 ppm and -21.4 ppm, respectively.<br />
Table 5.4 Along-track geoid and SSH gradients (track 21364)<br />
Names<br />
Gradients (ppm)<br />
ds = 0.0 - 4.7 km ds = 4.8 - 17.5 km ds = 17.5 - 23.5 km<br />
Geoid height 14.2 16.4 8.5<br />
Unretracked SSH 803.3 -363.8 -21.4<br />
It is also apparent from Figure 5.13 that the retracking techniques, irrespective of the<br />
method used, have extended the SSH profile several kilometres shoreward. For the<br />
track 21085 (top picture in Figure 5.13), retracking improves the SSHs over a<br />
distance of ~5-20 km from the coastline. A similar improvement of the SSH can be<br />
found from the track 21364 (bottom figure in Figure 5.13), where the SSH profile<br />
has been extended shoreward ~14 km along track. Although the analysis is only for<br />
two tracks, they are good representative of the ground tracks closer to the coastline.<br />
5.8.3 Biases among Retracking Algorithms<br />
Apart from the improvement in SSHs after retracking, it is evident that there is the<br />
bias among different retracking algorithms, in particular between the fitting<br />
algorithm and the threshold retracking (50% threshold level) algorithm (Figure 5.13).<br />
In order to investigate the biases among different algorithms and even between the<br />
SSH data before and after retracking, waveform data over a open ocean area<br />
(50-350 km from the coastline, -30°S
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 161<br />
A typical example of the SSH profiles before and after retracking is plotted as a<br />
function of the latitude in Figure 5.14. It is evident that the differences of the SSH<br />
are caused by the bias. The histograms of the differences ( ∆ SSH ) between the SSH<br />
data retracked by different methods are shown in Figures 5.15 and 5.16. From<br />
Figure 5.15, the differences of the ocean model and the five-parameter model are<br />
small, between -5 cm and +5 cm (see also Table 5.5 and 5.6) and nearly unidentified,<br />
though the differences do not exactly follow a normal distribution. The results from<br />
the differences between 50% threshold retracking and two fitting functions<br />
(Figure 5.16) show, however, an obvious bias.<br />
Figure 5.14 Retracked and unretracked SSH profiles over open oceans, showing the<br />
biases among them.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 162<br />
Figure 5.15 Histogram of the differences between the ocean model and fiveparameter<br />
model.<br />
Figure 5.16 Histograms of the differences between the 50% threshold retracking<br />
technique and the ocean model as well as the five-parameter model.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 163<br />
Using cycles 42 and 43 of ERS-2 waveform data, the biases have been estimated and<br />
the results are listed in Tables 5.5 and 5.6, respectively. The bias between two fitting<br />
models is small and almost statistically insignificant (Table 5.5). This is because they<br />
are both derived from the Brown model. However, it is significant to find the bias<br />
values of ~57 cm and ~56 cm between the 50% threshold algorithm and the fiveparameter<br />
model and the ocean model, respectively (Table 5.6). These values<br />
indicate that the range retracked by the fitting model is ~57 cm or ~56 cm longer<br />
than that retracked by 50% threshold retracking algorithm.<br />
Table 5.5 Biases between different retracking techniques from fitting a quadratic<br />
function to the ERS-2 SSH (cycle 42) differences over an ocean area (-30°S
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 164<br />
account of noise in the waveform (especially the trailing edge) and how the noise is<br />
affecting the leading edge from one waveform to the next, while the fitting algorithm<br />
is still affected by the noise in the trailing edge. The limited 3-4 sample bins of the<br />
leading edge are also restricted the precision of the estimates for both algorithms.<br />
Thus, the bias problem results from the noise effect in the fitting algorithm and too<br />
simplistic assumptions in the threshold retracking. The biases will be studied further<br />
in the next section.<br />
5.8.4 Biases between the SSH Data before and after Retracking<br />
From Figure 5.13, it is also apparent that there is a difference between the SSH data<br />
before and after retracking. Figures 5.17 and 5.18 show histograms of differences of<br />
the SSH data before and after retracking using different algorithms. It is evident that<br />
the differences of the SSHs before and after retracking appear to follow a normal<br />
distribution, but their means obviously are not equal to zero and indicate the biases.<br />
Figure 5.17 Histograms of the differences between unretracked SSH data and<br />
retracked SSH data using the ocean model and the five-parameter model.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 165<br />
Figure 5.18 Histogram of the differences between untracked SSH data and retracked<br />
SSH dada using the 50% threshold retracking technique.<br />
The biases have also been estimated and the results from cycles 42 and 43 are listed<br />
in Table 5.7 and Table 5.8, respectively. The differences computed from all<br />
retracking algorithms indicate that there are biases between retracked and<br />
unretracked the SSH data.<br />
From Table 5.7 and Table 5.8, a bias of ~30 cm exists between retracked SSH data<br />
(by 50% threshold retracking) and the unretracked SSH data, indicating the measured<br />
range is too long. The bias of ~27 cm between the SSH data retracked by fitting<br />
approaches and the unretracked SSH data suggests the measured range is too short.<br />
Theoretically, it is not necessary to retrack waveforms over open oceans as the onboard<br />
tracking algorithm is based on the Brown model and the assumptions of the<br />
homogeneous statistics of the model are generally satisfied over ocean surfaces.<br />
However, a bias still found between the SSH data retracked by the ocean model and<br />
the unretracked SSH data in this study. Therefore, the subsequent analysis here aims<br />
at understanding the reasons that cause the biases.
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 166<br />
Table 5.7 Biases between the ERS-2 (cycle 42) SSHs before and after retracking<br />
over an ocean area (-30°S
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 167<br />
samples in the range window. Since a sample or gate corresponds to a distance of<br />
~0.4542 m (ERS-1/2), it should be possible that the error is caused by the retracking<br />
procedure itself.<br />
From the above analysis, if one is interested in using data closer to the coastline and<br />
retracking the coastal waveform data, the bias between the SSH data before and after<br />
retracking should be removed when combining the retracked and unretracked data.<br />
Three approaches can be used:<br />
• Accept the already-applied instrument corrections and use the range provided<br />
by the GDR products over open oceans. The corrected range in coastal regions<br />
is obtained by applying the range correction estimated from retracking<br />
procedure to the measured range from the waveform products. Then, these two<br />
types of range measurements can be used together without the bias between<br />
them.<br />
• Use the range measurements from waveform products, retrack all waveforms<br />
(both over open ocean and coastal regions), and then apply the range<br />
corrections obtained from the retracking procedure to all measurements.<br />
• Retrack only the waveform data in coastal regions, but estimate an absolute<br />
bias between the altimeter-derived SSH (without retracking) and independent<br />
sea surface measured at the tide gauge. The bias, then, is used to correct the<br />
SSH over open oceans.<br />
5. 9 Summary<br />
A coastal retracking system has been developed. Central to the system is the use of<br />
two retracking techniques to derive an improved geodetic parameter, namely SSH,<br />
which include both the iterative least squares fitting algorithm and the threshold<br />
retracking algorithm. Retrackers adopted in the system are the ocean fitting model,<br />
five-parameter fitting models, nine-parameter fitting models, and the varying<br />
threshold levels, in which five- and nine-parameter models are used for the<br />
waveforms with both the linear and exponential trailing edges.<br />
By quantitative comparison of the ocean model and the five-parameter model, it has<br />
been found that the slope of the trailing edge modelled by the five-parameter model<br />
has a larger span of variation than the slope modelled by the ocean model. This
Chapter 5. COASTAL WAVEFORM RETRACKING SYSTEM 168<br />
advantage, in turn, gives the five-parameter model greater capability of fitting nonocean-like<br />
waveforms in coasts than the ocean model.<br />
A waveform classification procedure is given in this Chapter, which enables the<br />
waveform to be sorted and retracked by an appropriate retracker. To overcome the<br />
‘noise’ in waveforms caused by fading noise and sources, a weight iterative scheme<br />
has been developed.<br />
Threshold retracking can be an improved method in coastal regions, if an appropriate<br />
threshold level is selected for the retracking. Comparison of retracked SSH data with<br />
available ground truth of the AUSGeoid98 geoid heights confirms that the 50%<br />
threshold level is the best level for open-ocean waveform, but not an adequate level<br />
for the contaminated ocean waveforms in coastal regions. Instead, a varying<br />
threshold level, which is the 30% threshold level for contaminated waveforms and<br />
50% threshold level for uncontaminated waveforms, can make a good agreement<br />
between the retracked SSH data and the geoid heights.<br />
Biases exist both between retracked SSH data from different retracking algorithms<br />
and SSH data before and after retracking. Biases have been analysed and estimated<br />
by retracking ocean waveforms in this Chapter using two cycles of ERS-2 waveform<br />
data readily available at the time of this study. This finding suggests that biases must<br />
be estimated before retracking and applied to the data to obtain consistent results<br />
after retracking.
Chapter 6. WAVEFORM RETRACKING APPLICATION 169<br />
6. WAVEFORM RETRACKING APPLICATION TO THE AUSTRALIAN<br />
COAST<br />
6. 1 Introduction<br />
The principal objective of this Chapter is to determine to what extent the SSH data<br />
can be improved by the retracking system designed and developed in Chapter 5 in<br />
coastal regions. The quantitative distance estimation and detailed analysis of<br />
waveform contamination for the Australian coastal regions has been addressed in<br />
Chapter 4. As part of the development and implementation of the retracking system,<br />
it is important to apply the system to the coastal regions and to assess the quality of<br />
the retracked altimeter-derived SSH data. Therefore, the coastal retracking system<br />
will be tested over a small part of the Australian coast (0-30 km from the coastline,<br />
Figure 4.3) using two cycles (42 and 43, March to May 1999) of ERS-2 20 Hz<br />
waveform data. To obtain a general idea of the impact of retracking on the altimeterderived<br />
SSH, the quality checks of the retracking results consist of the comparisons<br />
with external data source and validation of internal results. The external independent<br />
reference used is the AUSGeoid98 geoid model. The results of retracking and<br />
comparisons with unretracked and external control data will be discussed in this<br />
Chapter.<br />
Following the steps of the retracking block diagram in Figure 5.1, results of<br />
waveform classification and waveform retracking will be presented and discussed in<br />
Sections 6.2 and 6.3, respectively. The internal validation of the quality of the<br />
retracked altimeter data will be conducted by validation of the single track SSH data<br />
(Section 6.4), comparisons between SSH data before and after retracking (Section<br />
6.5) and a collinear analysis (Section 6.6). SSH data before and after retracking will<br />
be compared with the geoid heights interpolated from the AUSGeoid98 model along<br />
ground tracks for the external comparison (Sections 6.7).<br />
In addition, validating the retracked data will involve using ~2.5 years of ERS-1<br />
20 Hz waveform data from different phases in the Taiwan Strait. Retracked SSH data<br />
will be used to estimate the gravity anomalies. The results will then be compared
Chapter 6. WAVEFORM RETRACKING APPLICATION 170<br />
with the external ground control of the ship-track gravity anomalies. The results will<br />
be given in Chapter 7.<br />
6. 2 Waveform Classification in Australian Coastal Regions<br />
As shown in Chapter 4, most of contaminated waveforms in coastal regions present<br />
an ocean-like single ramp. However, the returns also present some other non-oceanlike<br />
shapes, such as a superposition of the energy reflected from each scattering<br />
surface within the satellite footprint, which represent each distinct elevation when<br />
approaching the shoreline. Therefore, sorting data based on waveform shapes is<br />
essential before retracking them. In this Section, the contaminated waveform data<br />
estimated in Chapter 4 will be classified by using the methods presented in Section<br />
5.3.<br />
6.2.1 Typical Waveform Shapes<br />
Contaminated waveforms shown in Chapter 4 around the Australian coast have been<br />
categorised. A detailed example of waveform categorisation given here is from two<br />
ascending (21636 and 21679) and one descending (21815) ground tracks in a<br />
15º×15º area (105°≤λ
Chapter 6. WAVEFORM RETRACKING APPLICATION 171<br />
108˚<br />
110˚<br />
112˚<br />
114˚<br />
116˚<br />
118˚<br />
120˚<br />
-16˚ -16˚<br />
-18˚ 21815<br />
-18˚<br />
21679<br />
-20˚ -20˚<br />
-22˚ -22˚<br />
Indian<br />
Ocean<br />
21636<br />
-24˚ -24˚<br />
-26˚ -26˚<br />
-28˚ -28˚<br />
-30˚ -30˚<br />
108˚<br />
110˚<br />
112˚<br />
114˚<br />
116˚<br />
118˚<br />
120˚<br />
Figure 6.1 Part of ERS-2 ground tracks off the northwest Australian coast, showing<br />
the orbit number of each ground track. The highlighted diamonds along tracks will<br />
be shown in detail waveform shapes later in this Section.<br />
Figure 6.3 also shows waveforms from the pass 21815 but crossing an island<br />
(cf. Figure 6.1). The blank area (~ -20.9°
Chapter 6. WAVEFORM RETRACKING APPLICATION 172<br />
reflected from both land and standing water, and finally change to the typical ocean<br />
shapes at around -22.1° latitude. The waveform shifting in the range window is<br />
obviously observed in Figure 6.4, which is probably due to the operation manner of<br />
the altimeter discussed in Section 4.6.3, and thus causing a longer contaminated<br />
distance for this track.<br />
Figure 6.5 shows waveforms from an ascending ground track 21679 (cf. Figure 6.1)<br />
passing a coastal area of Shark Bay (113°≤λ
Chapter 6. WAVEFORM RETRACKING APPLICATION 173<br />
1000 2000<br />
Counts<br />
-18.60 -18.55 -18.50 -18.45 -18.40 -18.35 -18.30 -18.25<br />
Latitude (degrees)<br />
16 32 48 64<br />
Bins<br />
Figure 6.2 Typical ocean waveforms from pass 21815 (cf. Figure 6.1).<br />
Counts<br />
2500 5000<br />
-21.0 -20.9 -20.8 -20.7 -20.6 -20.5 -20.4 -20.3 -20.2<br />
Latitude (degrees)<br />
16 32 48 64<br />
Bins<br />
Figure 6.3 Waveforms recorded when the ground track 21815 is approaching an<br />
island at latitude ~ -20.5º and leaving the island to water at ~ -20.9º. The data gap is<br />
due to the island (cf. Figure 6.1).
Chapter 6. WAVEFORM RETRACKING APPLICATION 174<br />
Counts<br />
2000 4000 6000 8000 10000<br />
-22.35 -22.30 -22.25 -22.20 -22.15 -22.10<br />
Latitude (degrees)<br />
16 32 48 64<br />
Bins<br />
Figure 6.4 Waveforms leaving land to water (pass 21636 in Figure 6.1). The highpeaked<br />
specular responses are probably due to off-nadir brighter calm water. The<br />
waveform shifting (~ -22.25°
Chapter 6. WAVEFORM RETRACKING APPLICATION 175<br />
Counts<br />
5000 10000 15000<br />
64<br />
48<br />
32<br />
Bins<br />
16<br />
-26.0 -25.9 -25.8 -25.7 -25.6 -25.5 -25.4 -25.3 -25.2 -25.1<br />
Latitude (degrees)<br />
Figure 6.5 Typical ERS-2 radar altimeter contaminated waveforms (pass 21679)<br />
across water, land, and calm/still sea surfaces, showing high-peaked waveform<br />
shapes reflected from the calm water surfaces, mixed shapes of land and water, and<br />
ocean-like shapes.
Chapter 6. WAVEFORM RETRACKING APPLICATION 176<br />
6.2.2 Results of Waveform Categories<br />
The detailed categories of contaminated waveforms in ten Australian coastal regions<br />
(see Figure 5.8) have been conducted and the general statistics of the waveform<br />
categories for these ten coastal areas are listed in Table 6.1. The following<br />
conclusions can be drawn from Table 6.1.<br />
Table 6.1 Statistics of ERS-2 (cycle 42) contaminated waveform shapes in ten<br />
Australian coastal areas (cf. Figure 5.8)<br />
Waveform<br />
categories<br />
% of<br />
waveforms<br />
Mean distance<br />
from coastline<br />
(km)<br />
Mean width of<br />
waveforms<br />
(bins)<br />
Mean peak of<br />
waveforms<br />
(counts)<br />
Ocean * 99.97 168.04 25.8 1304.1<br />
Ocean-like 80.19 10.50 24.9 1380.3<br />
Single pre-peaked 4.64 2.31 10.7 2785.1<br />
Single mid-peaked 0.54 2.54 9.8 2615.0<br />
Single post-peaked 1.35 3.12 11.8 3027.0<br />
Double pre-peaked 0.55 3.08 14.8 2280.9<br />
Double post-peaked 0.73 2.15 16.1 2271.9<br />
Multi-peaked 0.18 2.30 13.0 2117.5<br />
Sharp/High-peaked 4.35 1.73 3.3 8439.5<br />
Unusable 7.51 6.03 34.7 1234.6<br />
* Ocean waveforms over open oceans 50-350 km from the coastline, with 50%<br />
threshold retracking point estimates between bins/gates 31-33.<br />
1. Most contaminated waveforms (80.19%) have ocean-like shapes. The mean<br />
width is similar to that of ocean waveforms, whereas the mean peak value is<br />
slightly larger than ocean waveforms.<br />
2. Of other shapes, the highest peak of the power is the high-peaked waveform<br />
and the next is the single post-peaked waveform, but mean peak values of all<br />
other shapes are larger than the mean peak of the ocean-like waveform.<br />
3. The high-peaked waveform has the narrowest mean waveform width (3.3 bins).<br />
The second narrowest width (9.8 bins) belongs to the single mid-peaked<br />
waveform. Other non-ocean-like waveform widths are still much smaller than<br />
the ocean-like waveform.
Chapter 6. WAVEFORM RETRACKING APPLICATION 177<br />
4. Although ~80% of contaminated waveforms in coastal regions take an oceanlike<br />
shape, other shapes of waveforms are found much closer to the coastline<br />
from this data set (1.73 - 3.12 km in average), in particular the high-peaked<br />
waveform (1.73 km from the coastline in average). This fact indicates further<br />
that using only the ocean model or five-parameter model cannot re-process all<br />
waveforms in coastal regions. Other retrackers are more significant in such<br />
areas to remove errors caused by waveform contamination.<br />
In addition, from a waveform categorisation in this study, high-peaked waveforms<br />
have been found to take a larger percentage of contaminated waveforms (from 5.4%<br />
to 6.9%) in north-west Australian coastal areas 1 and 2 because of the rugged<br />
coastline, complex sea states and inland/standing water.<br />
6. 3 Waveform Retracking Results and Discussion<br />
Given the above observations, 20 Hz waveforms from cycles 42 and 43 offshore<br />
Australia are re-processed using the coastal retracking system developed in Chapter 5<br />
out to a distance of 30 km. Choosing 0-30 km is based on the observation in Chapter<br />
4, where the maximum contaminated distance can occur out to ~22 km from the<br />
coastline.<br />
6.3.1 Data Editing<br />
The first step of retracking is data editing to check whether or not the return presents<br />
a leading edge (i.e., ramp) in the range window. This is conducted through the simple<br />
heuristic assessment method used by Scott et al. (1994). The maximum 8.2% of noramp<br />
(i.e., unusable) waveforms are filtered out from this step (see Table 6.2 and<br />
Table 6.3). The second step is to classify each return waveform so that it can be<br />
retracked by an appropriate retracker.<br />
6.3.2 Waveform Retracking Using Different Algorithms<br />
Next, a specific retracker based upon the classification is applied to each waveform,<br />
which is either a fitting or threshold retracking algorithm. When using the iterative<br />
procedure, an adequate fitting function must be selected and the procedure is
Chapter 6. WAVEFORM RETRACKING APPLICATION 178<br />
repeated until a pre-defined existing criterion is satisfied (see Section 5.5.5). Since a<br />
linear solution to a non-linear problem is used, sometimes the fitting algorithm does<br />
not work (Zwally, 1996; Davis, 1995). Therefore, the threshold-retracking algorithm<br />
is used to replace the iterative least squares fit to compute the retracking correction if<br />
and when the fitting procedure fails. The threshold retracking technique thus works<br />
for both the high-peaked waveforms and those fail to the fitting algorithm. The<br />
system will select automatically the threshold level, which is 30% or 50% of the<br />
waveform amplitude, for each waveform depending on its shape. As stated in Section<br />
5.6.3, the 50% threshold level is used to retrack the ocean-like waveforms, while<br />
other shapes of waveforms will be retracked by the 30% threshold level.<br />
The percentages of the waveforms retracked by fitting and threshold algorithms are<br />
listed in Tables 6.2 and 6.3. It can be seen that 77.7% - 94.9% (cycle 42) and 83.5% -<br />
96.2% (cycle 43) of waveforms in this area can be retracked by the iterative least<br />
squares fitting procedure. The rest of the waveforms (the last column) still require<br />
threshold retracking. They are 4.8% - 18.1% of waveforms for cycle 42 and 3.8% -<br />
15.6% of waveforms for cycle 43, respectively. The percentages of using different<br />
retrackers vary from area to area depending on the reflecting surface topography.<br />
Table 6.2 Waveform data status and percentage of the waveforms retracked using<br />
fitting and threshold retracking algorithms in Australian coastal regions (cycle 42)<br />
Area<br />
No. of<br />
Unusual<br />
Fit<br />
Threshold<br />
number<br />
Areas<br />
data<br />
data<br />
%<br />
%<br />
%<br />
1 120°≤λ
Chapter 6. WAVEFORM RETRACKING APPLICATION 179<br />
Table 6.3 Waveform data status and percentage of the waveforms retracked using<br />
fitting and threshold retracking algorithms in Australian coastal regions (cycle 43)<br />
Area<br />
No. of<br />
Unusual<br />
Fit<br />
Threshold<br />
number<br />
Areas<br />
data<br />
data<br />
%<br />
%<br />
%<br />
1 120°≤λ
Chapter 6. WAVEFORM RETRACKING APPLICATION 180<br />
Table 6.4 Percentages of ERS-2 waveforms fitted using different functions in<br />
Australian coastal regions (cycle 42).<br />
Area<br />
No. of fit<br />
Ocean<br />
E-five-<br />
Nine-<br />
number<br />
Areas<br />
data<br />
model<br />
parameter a<br />
parameter b<br />
%<br />
%<br />
%<br />
1 120°≤λ
Chapter 6. WAVEFORM RETRACKING APPLICATION 181<br />
Figure 6.6 Typical coastal waveforms fitted using the following models:<br />
(a): the five-parameter model with an exponential trailing edge,<br />
(b) and (c): the ocean model,<br />
(d) and (f): the nine-parameter model with linear trailing edge,<br />
(e): the nine-parameter model with an exponential trailing edge.<br />
waveforms are retracked by the fitting algorithm, while only 2% of waveforms need<br />
to be threshold retracked. The percentage of fitting algorithm increases with<br />
increasing distance from the coastline, confirming that the waveform shapes agree<br />
well with the typical ocean waveform after 5 km offshore. This was also seen in<br />
Chapter 4.<br />
Figure 6.6 shows six typical coastal waveforms and the functional fits to them. It also<br />
shows that different fitting functions work well to deal with different shapes of<br />
waveforms in coastal regions. As stated, most of waveforms are single peaked (or<br />
ramp) pulses. The ocean model fits these waveforms very well as shown in Figure<br />
6.6(b) and (c). Some waveforms near land show higher power and double ramps,<br />
which need to be fitted by the five-parameter function with an exponential trailing<br />
edge as shown in Figure 6.6(a), or the nine-parameter function with either a linear
Chapter 6. WAVEFORM RETRACKING APPLICATION 182<br />
trailing edge (cf. Section 3.5) in Figure 6.6(d) and (e) or an exponential trailing edge<br />
(Figure 6.6(e), cf. Section 3.5).<br />
6. 4 Analysis of the Single-Track Retracked Results<br />
After retracking, two types of corrections from atmospheric propagation delays and<br />
geophysical models supplied with the waveform data by NRSC (1995) are applied to<br />
both retracked and unretracked range measurements to obtain the SSH. Since the EM<br />
bias correction caused by ocean surface waves is not supplied with the waveform<br />
data products used in this study, it is not applied to unretracked range measurements.<br />
However, the retracking procedure removes the need for this correction (Brenner et<br />
al., 1993; Hayne, 2002). The geophysical and environmental corrections applied are<br />
dry and wet tropospheric range corrections, ionospheric correction, and tide<br />
correction. The tide correction is the sum of the ocean tide from the Schwiderski<br />
model and the ocean loading tide from the Ray-Sanchez model (NRSC, 1995). The<br />
Doppler range correction is not applied to the range measurement, because most<br />
values were found to exceed the error criteria of ±55 cm given by NRSC (1995) as<br />
stated in Section 2.4.2.<br />
As mentioned in Section 5.8.3, the relative biases among different retracking<br />
algorithms are determined by retracking the waveforms over open oceans. They are<br />
then applied to the range measurements retracked by the threshold retracking<br />
algorithm. Finally, the altimeter-derived SSH is based on a reference of the ocean<br />
model. This aims to remove the biases caused by reprocessing the waveform data<br />
using different algorithms. As the determined biases depend strongly on the<br />
algorithm, they can also be determined by other methods. For example, the method<br />
used by Dong et al. (2002) may be employed for this purpose, which validates the<br />
absolute bias between the altimeter-derived sea level and that derived by the tide<br />
gauge data using the local geoid model, SST model, and data measured at the tide<br />
gauge equipped with Global Positioning System (GPS) receivers. By using a similar<br />
method, it is possible to estimate the absolute biases between different retracking<br />
algorithms, but this remains an open issue for future work because of data<br />
availability limitations at the time of this study.
Chapter 6. WAVEFORM RETRACKING APPLICATION 183<br />
To compare the results from the retracking system with those retracked by a single<br />
retracker, two ground tracks (21085 and 21364) are chosen, which are the same<br />
tracks as these presented in Figure 5.13 (Section 5.8.2). The SSH (20 Hz) and geoid<br />
height profiles (AUSGeoid98) are also plotted via the along-track distance from the<br />
coastline in Figure 6.7. As stated in Sections 2.5.2 and 5.8.2, this aims to compare<br />
the profile trends of the SSH and geoid height for investigating the contamination<br />
caused by coastal sea states.<br />
To compare results with those in Figure 5.13 (Section 5.8.2), the pictures have the<br />
same X- and Y-scales as those in Figure 5.13. It is evident when comparing Figure<br />
6.7 with Figure 5.13 that erroneous SSHs caused by the fitting algorithm when<br />
approaching the coastline (0-5 km) in Figure 5.13 have been removed in Figure 6.7.<br />
The bias between fitting and threshold retracking algorithms is also removed after<br />
applying the bias correction determined in Section 5.8 (cf. Figures 6.7 and 5.13).<br />
Because of use of the retracking system, more SSH data can be derived from<br />
corrected range measurements than the SSH data from a single fitting retracker, thus<br />
keeping the useful high-frequency singles in coastal regions. The numbers of SSH<br />
data obtained from the system and a single retracker (the ocean model) are N system<br />
retracking = 78 and N single retracker = 59 for track 21085 and N system retracking = 73 and N single<br />
retracker = 63 for track 21364.<br />
As can be seen from Figures 5.13 and 6.7, both single retracker and the retracking<br />
system can improve the SSH profile shoreward several kilometres. More importantly,<br />
the retracking system comprising several retrackers can improve further the SSHs<br />
shoreward ~2.5 km from ~5 km to ~2.5 km for both tracks. However, it is important<br />
to note that the SSH profiles cannot be closed completely to the coastline because of<br />
land-dominated returns discussed in Chapter 4. In the case of tracks 21085 and<br />
21363, there are still an along-track distance of ~2.5 km that cannot be recorded by<br />
waveform retracking.
Chapter 6. WAVEFORM RETRACKING APPLICATION 184<br />
−18.00<br />
−20.00<br />
−22.00<br />
−24.00<br />
SSH (m)<br />
−26.00<br />
−28.00<br />
−30.00<br />
−32.00<br />
Geoid height<br />
Unretracked<br />
Retracked<br />
−34.00<br />
0.0 5.0 10.0 15.0 20.0 25.0<br />
Along−track distance from the coastline (km)<br />
−8.00<br />
−10.00<br />
SSH (m)<br />
−12.00<br />
−14.00<br />
−16.00<br />
Geoid height<br />
Unretracked<br />
Retracked<br />
−18.00<br />
0.0 5.0 10.0 15.0 20.0 25.0<br />
Along−track distance from the coastline (km)<br />
Figure 6.7 Geoid height and SSH profiles (ERS-2, cycle 42) along ground track<br />
21085 (top) and 21364 (bottom) before and after retracking using the coastal<br />
retracking system, showing the same profiles as those in Figure 5.13.
Chapter 6. WAVEFORM RETRACKING APPLICATION 185<br />
Comparing the SSH profiles before and after retracking in Figure 6.7, the SSH<br />
profiles after retracking show a good agreement with the geoid-height profile from<br />
~2.5 km to ~25 km to the coastline. However, it is also apparent that the retracking<br />
has added some measurement noise. Compared with unretracked SSHs, the retracked<br />
SSH profile becomes nosier. The noise in the retracked SSH profile might be caused<br />
by, firstly, the shorter wavelength signal, which suggests that on-board tracking<br />
algorithm cannot follow well temporal variations of the scattering surface; secondly,<br />
unmodelled non-linear parameters in this study (e.g., the skewness). Thus, a low-pass<br />
filtering procedure is still essential when using 20 Hz SSH data after retracking for<br />
the geophysical application in coastal regions. The search radius of the filter depends<br />
on the applications, say 5-7 km for the recovery of gravity anomalies.<br />
6. 5 Analysis of the SSH Data Before and After Retracking<br />
The comparison is performed by calculating the difference dh between the retracked<br />
SSH h r and unretracked SSH h g at the same geographic location along each ground<br />
track. This difference (dh=h r − h g ) represents the sum of the bias of SSH data before<br />
and after retracking, and tracking error that can be removed by waveform retracking.<br />
The mean difference and standard deviation (STD) are calculated at six<br />
5 km wide distance bands. Choosing different distance bands allows one to see how<br />
close to the coastline the altimeter-observed ranges are improved by the retracking<br />
procedure. Results are listed in Tables 6.6 and 6.7 for cycles 42 and 43, respectively.<br />
It should be noted that the bias between SSH before and after retracking (described<br />
in Section 5.8.4) is not applied to the individual SSH data. Instead, the biases can be<br />
applied to the mean difference between the SSHs before and after retracking in<br />
Tables 6.6 and 6.7 to analyse the error removed by the retracking procedure.<br />
Therefore, results in Tables 6.6 and 6.7 contain the biases, which are -27.8±12.4 cm<br />
(cycle 42) and -27.5±10.8 cm (cycle 43) (ibid.). It is recalled here that the bias could<br />
be estimated and removed by using the methods mentioned in Section 5.8.<br />
Tables 6.6 and 6.7 show that STD of the SSH differences decreases with increasing<br />
the offshore distance. The STD decreases to less than 1 m for distances greater than<br />
15 km from the coastline. Because the comparison is conducted at the same along-
Chapter 6. WAVEFORM RETRACKING APPLICATION 186<br />
track point and same time, the satellite orbit error and any other random errors can be<br />
removed by taking the difference of the retracked and unretracked SSH. Therefore,<br />
the large values of the STD indicate, on one hand, that the significant error occurs<br />
due to waveform contamination near the coastline. On the other hand, the retracking<br />
algorithm itself introduces the additional error in the SSH data near land, because the<br />
ocean returns are difficult to recover from land-dominated waveforms.<br />
Table 6.6 Descriptive statistics of ERS-2 20 Hz SSH differences before and after<br />
retracking (dh = h r – h g , cycle 42).<br />
Distance<br />
to<br />
coastline<br />
(km)<br />
Number<br />
of<br />
SSH<br />
Mean<br />
difference<br />
(m)<br />
Max<br />
difference<br />
of SSH<br />
(m)<br />
Min.<br />
difference<br />
of SSH<br />
(m)<br />
STD<br />
±(m)<br />
0-5 5074 -0.79 9.25 -9.99 2.35<br />
5-10 6517 -0.20 9.43 -9.82 1.31<br />
10-15 7416 -0.24 6.62 -8.15 0.99<br />
15-20 7313 -0.25 7.43 -6.89 0.71<br />
20-25 7728 -0.30 7.71 -8.38 0.57<br />
25-30 7496 -0.28 5.69 -9.13 0.57<br />
Table 6.7 Descriptive statistics of ERS-2 20 Hz SSH differences before and after<br />
retracking (dh = h r – h g , cycle 43).<br />
Distance<br />
to<br />
coastline<br />
(km)<br />
Number<br />
of<br />
SSH<br />
Mean<br />
difference<br />
(m)<br />
Max<br />
difference<br />
of SSH<br />
(m)<br />
Min.<br />
difference<br />
of SSH<br />
(m)<br />
STD<br />
±(m)<br />
0-5 6218 -0.79 8.41 -9.99 2.43<br />
5-10 8924 -0.14 9.89 -9.82 1.39<br />
10-15 9820 -0.31 6.63 -9.47 1.02<br />
15-20 9850 -0.30 8.11 -9.91 0.84<br />
20-25 10140 -0.31 4.71 -5.70 0.65<br />
25-30 9660 -0.32 7.05 -6.31 0.56<br />
If considering only the amount of the bias, the important amount of error removed by<br />
waveform retracking is -51.2 cm and 7.8 cm for cycle 42 in distance bands of 0-5 km<br />
and 5-10 km, respectively, while the results do not show a significant amount of
Chapter 6. WAVEFORM RETRACKING APPLICATION 187<br />
error in the rest of the four distance bands. Similar results can be found for cycle 43,<br />
for which the amount of error in distance bands of 0-5 km and 5-10 km is -51.5 cm<br />
and 13.5 cm, respectively. Analysis of the results suggests that waveform retracking<br />
can effectively remove the altimeter tracking error in the vicinity of coastal regions.<br />
When the satellite ground track goes further to the open ocean, the tracking error<br />
becomes less significant. In the case of this study, the amount of tracking error is<br />
~-3 cm. This agrees well with the result (-2.37 cm) of Brenner et al. (1993).<br />
6. 6 Collinear Analysis<br />
From the analysis in Sections 6.4 and 6.5, it is evident that retracking can remove<br />
errors caused by altimeter on-board tracking algorithm due to coastal sea states and<br />
land topography. In this Section, the collinear differences, which are computed using<br />
altimeter-derived SSH profiles obtained along identical ground tracks, are employed<br />
to estimate how the retracking procedure improves the altimeter data in coastal<br />
regions. The differences are also used to qualify altimeter data before and after<br />
retracking.<br />
6.6.1 Methodology<br />
Over open oceans, ERS-2 SSH profiles from different cycles contain primarily geoid<br />
and SST information, as well as the random measurement error, which includes<br />
errors introduced by satellite orbit error, altimeter range measurement error, and<br />
temporal variability of the ocean surface, such as tides (Cheney et al., 1983). In<br />
coastal regions, however, waveform contamination causes an additional error. The<br />
relative collinear offset between the profiles is<br />
dH<br />
= H 2<br />
− H1<br />
+ ε , where the SSH<br />
measurements at the same point, H<br />
1<br />
and H<br />
2<br />
, occur at different times of cycles t 1<br />
and t 2<br />
, and ε is a random measurement error.<br />
The collinear differences remove the long-wavelength geoid, the quasi-timeindependent<br />
SST and variations of the ocean surface, representing thus the amount of<br />
orbit error, altimeter range measurement error, and the temporal variations of the<br />
surface over open oceans. As stated, the random error includes the tracking error<br />
caused by waveform contamination for unretracked data in coastal regions. After
Chapter 6. WAVEFORM RETRACKING APPLICATION 188<br />
removing the satellite orbit error, SSH data from repeat cycles can also be used to<br />
determine the mean sea surface (e.g., Deng et al., 1997; Nerem, 1995; Wang and<br />
Rapp, 1991). Thus, the STD of the difference dH here is used as a measure of the<br />
quality of a retracking algorithm’s range estimate. If errors caused by waveform<br />
contamination can be removed by retracking algorithms, this will result in smaller<br />
STDs. By comparing the STD of the collinear difference dH before and after<br />
retracking, it can be determined whether or not the retracked SSHs are improved by<br />
the retracking system.<br />
The requirement of the collinear analysis is that data from different cycles must be<br />
reduced to a reference orbit, so that the comparison can be conducted on the same<br />
basis (Cheney et al., 1983; Wang and Rapp, 1991). In reality, the satellite ground<br />
tracks do not repeat exactly due to orbital variations caused by gravitational<br />
influences and drag effects. The cross-track deviation is ~1 km for most satellites (cf.<br />
Brenner et al., 1990; Wang et al., 1991). Therefore, 1-second normal points are<br />
usually generated (ibid.), and the geoid gradient correction is applied to data (Wang<br />
and Rapp, 1991) for the reduction of the observed data to the reference orbit to<br />
obtain exactly ‘collinear’ data. The cross-track geoid gradient correction is applied to<br />
correct the effects of the geoid slope caused by non-repeated tracks (ibid.).<br />
6.6.2 Discussion of Results<br />
In this study, 20 Hz SSH data are retained to see possible high-frequency signals in<br />
coastal regions. Since cycle 43 contains more orbits and thus more complete data<br />
coverage, it is chosen as a reference. SSH data from cycle 42 are then reduced to<br />
relevant points along ground tracks of cycle 43 using the method developed to<br />
compute the collinear difference between SSHs from different cycles by Deng et al.<br />
(1996). The cross-track geoid gradient correction is not applied to the data here,<br />
because data came from two adjacent cycles and the effects of the geoid slope can be<br />
neglected.<br />
The descriptive statistics of the collinear SSH differences (20 Hz) computed from<br />
cycles 42 and 43 have been calculated both before and after retracking. Results have<br />
been plotted via six 5-km-wide distance bands in Figures 6.8 and 6.9. There are high<br />
values of STD for dH using both unretracked and retracked data. The reasons may
Chapter 6. WAVEFORM RETRACKING APPLICATION 189<br />
be the satellite orbit error and temporal variability of the ocean surface. As can be<br />
seen from Figure 6.9, the STDs of dH after retracking are smaller than STDs before<br />
retracking in all distance bands, in particular 0-15 km from the coastline. It is clear<br />
from this comparison that retracking improves the precision of altimeter range<br />
estimates in all distance bands to a different extent, in particular in two distance<br />
bands of 0-5 km and 10-15 km. After 15 km from the coastline and in 5-10 km<br />
distance, the STDs after retracking are still slightly smaller than STD values before<br />
retracking, indicating improvement from waveform retracking as well.<br />
Before Retracking<br />
After retracking<br />
Mean differences (m)<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.8 Mean differences of the collinear SSH data of cycles 42 and 43 before and<br />
after retracking in six 5 km wide distance bands.<br />
Before retracking<br />
After retracking<br />
STD (m)<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.9 STD (positive values) of the collinear 20 Hz SSH differences of cycles 42<br />
and 43 before and after retracking in six 5 km wide distance bands, showing<br />
improvement from SSH data after retracking.
Chapter 6. WAVEFORM RETRACKING APPLICATION 190<br />
6. 7 Evaluation of SSH Data Using Geoid Heights<br />
6.7.1 Comparison with the AUSGeoid98 Geoid Model<br />
The collinear analysis of SSH data along repeating tracks can determine the internal<br />
consistency (i.e., precision) of altimeter measurements. In order to assess the<br />
accuracy of the retracked data and to determine how the altimeter retracked SSH<br />
profiles represent the true sea surface, an independently surveyed reference surface,<br />
referenced to the same ellipsoid, is required. The surface used here is the<br />
AUSGeoid98 2′×2′ geoid grid (Featherstone et al., 2001). This geoid height model,<br />
actually, is not an absolutely independent ground truth, because it uses altimeterderived<br />
free-air gravity anomalies offshore (ibid.). However, as stated in Section<br />
2.5.1, altimeter-derived gravity and ship track gravity data were used for<br />
computation of the geoid in a ‘draping’ method such that the dense ship-track gravity<br />
contribute significantly to the grid signal near the Australian coast, especially<br />
0-30 km from the coastline, while the altimeter-derived gravity data contribute<br />
mainly to the area which is far away from the Australian coast and without the shiptrack<br />
data (cf. Kirby and Forsberg, 1998; Featherstone et al., 2003, Figure 1).<br />
Therefore, the AUSGeoid98 geoid model can be thought of as a quasi-independent<br />
reference.<br />
To first order, the altimeter-observed mean sea surface represents the geoid<br />
(Mantripp, 1996). This means that comparison of the geoid height and SSH can<br />
provide assessment of the quality of retracked SSH data. It is also recalled that the<br />
SST is not removed from the SSH data around the Australian coast due to the reasons<br />
stated in Section 2.5.2. As stated in Section 5.7.3, since the quasi-independent<br />
reference AUSGeoid98 is used, again, the verification of any improvement in the<br />
SSH will only be seen in the standard deviations of the mean difference, with a<br />
smaller value indicating an improvement in the determination of the altimeterderived<br />
SSH.<br />
For each along-track altimeter-derived SSH both before and after retracking, the<br />
gridded geoid height is bispline interpolated with longitude and latitude of the<br />
location. Then it is subtracted from the corresponding SSH to give the difference<br />
between the SSH and geoid height. The statistics of differences is used to quantify
Chapter 6. WAVEFORM RETRACKING APPLICATION 191<br />
the quality of the retracking system. Again, the STD is used to measure how the<br />
waveform retracking system works rather than the mean difference between the<br />
collinear differences.<br />
6.7.2 Results of the Comparison between SSH Data and Geoid Heights<br />
The descriptive statistics of differences between SSH data (20 Hz) before and after<br />
retracking and gridded geoid height has been computed. For convenience of<br />
comparison, mean differences and the STD of the differences have been plotted via<br />
six 5-km-wide distance bands in Figures 6.10 and 6.11 for cycle 42, and Figures 6.12<br />
and 6.13 for cycle 43, in which the STD is presented only using the positive value.<br />
Again, the STDs show large values for both SSH data before and after retracking.<br />
The reasons may be the temporal variations of the ocean surface and other incorrect<br />
corrections, such as ocean tides and wet tropospheric range corrections, but retracked<br />
SSH data show a higher precision than unretracked data.<br />
Before retracking<br />
After retracking<br />
Mean difference (m)<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.10 Mean differences of geoid heights (AUSGeoid98) and SSH data before<br />
and after retracking in six 5 km wide distance bands (cycle 42).<br />
For cycle 42, Figures 6.10 and 6.11 indicate that both mean difference and STD<br />
decrease when using retracked SSH data. The values of STD drop down to ~1 m and<br />
the significant improvement occur in distances 0-15 km to the coastline. After 15 km<br />
from the coastline, there is no significant decrease in the value of the STD of<br />
differences. From Figures 6.12 and 6.13 for cycle 43, similar results can be obtained.<br />
These results suggest that more precise SSH profile can be obtained from retracked<br />
SSH data, especially in distance of 0-15 km from the coastline.
Chapter 6. WAVEFORM RETRACKING APPLICATION 192<br />
Before retracking<br />
After retracking<br />
STD (m)<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.11 STD (positive values) of the mean difference before and after retracking<br />
in six 5 km wide distance bands, showing improvement on SSH data after retracking<br />
(cycle 42).<br />
Before retracking<br />
After retracking<br />
Mean differences (m)<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.12 Mean differences between geoid heights (AUSGeoid98) and SSH data<br />
before and after retracking in six 5 km wide distance bands (cycle 43).<br />
Before retracking<br />
After retracking<br />
STD (m)<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
0-5 5-10 10-15 15-20 20-25 25-30<br />
Distances from the coastline (km)<br />
Figure 6.13 STD (positive values) of the mean difference before and after retracking<br />
in six 5 km wide distance bands, showing improvement on SSH data after retracking<br />
(cycle 43).
Chapter 6. WAVEFORM RETRACKING APPLICATION 193<br />
Figures 6.11 and 6.13 show that the STD of differences after retracking drops down<br />
to ±1 m when the distance to the coastline is larger than 10 km for cycle 42 and 5 km<br />
for cycle 43, respectively. However, the STD before retracking decreases to less than<br />
±1 m when the distance to the coastline is more than 20 km for both cycles. This<br />
indicates clearly that the retracked SSH can be determined accurately to the same<br />
level as SSHs over open oceans after offshore 10 km or even 5 km (e.g., cycle 43),<br />
while unretracked SSH data can have the same accuracy after only 15 km from the<br />
coastline. This means that these improved SSHs after retracking in coastal regions<br />
can be used with sufficient accuracy to compute coastal geoidal undulations, coastal<br />
ocean circulations, gravity anomalies and ocean tides. The results from Figures 6.11<br />
and 6.13 also show a better accuracy at retracked SSH data in Australian coastal<br />
regions than unretracked SSH data. This demonstrates that the coastal waveform<br />
retracking system developed in this study has improved and expended altimeterobserved<br />
measurements ~10 km shoreward in generally in Australian coastal regions<br />
comparing with unretracked SSH data when using AUSGeoid98 as a quasiindependent<br />
reference.<br />
6. 8 Summary<br />
Regional investigations around the Australian coast have demonstrated that the<br />
waveform retracking system developed in Chapter 5 can effectively reprocess 20 Hz<br />
waveform data and remove most errors caused by waveform contamination in these<br />
areas. Two cycles (42 and 43, March to May 1999) of ERS-2 20 Hz waveform data<br />
sets offshore 30 km around the Australia’s coast have been used to test the coastal<br />
retracking system developed in this study. The whole coastal region is divided into<br />
10 15°×15° sub areas. SSH data both before and after retracking are derived from<br />
uncorrected and corrected range measurements for the purpose of comparison and<br />
validation. Altimeter-derived SSH data used are 20 Hz without applying any filtering<br />
procedure to avoid any loss of high frequency signals. Using these SSH data and the<br />
AUSGeoid98 model, several analyses have been conducted in this Chapter.<br />
The results of waveform classification show that most coastal waveforms (80.19%)<br />
have ocean-like shapes with a single ramp but without significant peaks. Of the<br />
remaining ~12.34% of waveforms, predominant waveforms have single pre-peaked
Chapter 6. WAVEFORM RETRACKING APPLICATION 194<br />
(4.65%) and high-peaked (4.35%) waveform shapes. Others, such as single midpeaked<br />
and post-peaked, double pre-peaked and post-peaked, and multi-peaked<br />
waveforms, are from 0.73% to 1.35%.<br />
The categorised waveform is then retracked by an appropriate retracker depending on<br />
its shape. Retracking results at the Australian coast show that 77.7% - 96.2% of<br />
waveforms can be retracked using least squares iterative fitting algorithm, while<br />
3.8% - 18.1% of waveforms are needed to be retracked by the threshold retracking<br />
algorithm. The percentages of using different retrackers vary from area to area. Of<br />
these waveforms retracked by fitting algorithm, over 98.3% of waveforms are fitted<br />
using the ocean model and the rest of waveforms are fitted by the five-parameter<br />
model with an exponential trailing edge and the nine-parameter model with either a<br />
linear trailing edge or an exponential trailing edge. It has been found that waveforms<br />
retracked by the threshold retracking algorithm decrease from 35% in a distance of<br />
0-5 km from the coastline to 4% in a distance of 5-10 km from the coastline. After<br />
10 km offshore, only 2% of waveforms need to be threshold retracked.<br />
By comparison of along-track altimeter-derived SSH data before and after retracking,<br />
the largest errors caused by waveform contamination are found in a distance band of<br />
0-5 km, which is ~-51.2 cm (cycle 42) and ~-51.5 cm (cycle 43). Errors of 7.8 cm<br />
and 13.5 cm are found in the distance band of 5-10 km from the coastline. Beyond 10<br />
km offshore, the error has been found to be less significant (~-3 cm).<br />
Results from single track analysis indicate that the coastal retracking system can deal<br />
with more waveforms to derive thus more SSH data than a single retracker (e.g., the<br />
ocean model). More importantly, the system can make further the SSH profile closer<br />
to the coastline ~2.5-5 km. This is an improvement of the single retracker, at which<br />
the SSH profile presents without showing effects of waveform contamination after<br />
5 km from the coastline.<br />
In general, by comparing altimeter-derived along-track SSH data before and after<br />
retracking with AUSGeoid98 geoid heights, results show that the altimeter-derived<br />
SSH profiles can be extended ~5-10 km shoreward by the use of a coastal retracking<br />
system and range corrections. These improved SSHs show a same accuracy level as<br />
those over open oceans, thus indicating a great potential to apply them with sufficient
Chapter 6. WAVEFORM RETRACKING APPLICATION 195<br />
accuracy to altimetric geodetic applications in coastal regions. However, SSH<br />
profiles at a distance of 0-5 km to the coastline cannot be recovered by the waveform<br />
retracking procedure at the same precision level as SSH data after 5 km from the<br />
coastline.<br />
The problem found from this research is that, in general, the retracked SSHs become<br />
nosier compared to the unretracked data. Therefore, a low-pass filtering process is<br />
necessary before applying these retracked data.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 196<br />
7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES NEAR<br />
TAIWAN FROM WAVEFORM RETRACKING<br />
7. 1 Introduction<br />
Satellite altimetry has provided the most valuable of data sets for the recovery of the<br />
marine gravity field. Numerous local and global marine gravity anomalies have been<br />
created using a variety of successful techniques. Papers reporting current results<br />
include those of Hwang and Parsons (1995), Sandwell and Smith (1997), Andersen<br />
and Knudsen (1998), Tziavos et al. (1998), Hwang (1998), Andersen et al. (2001),<br />
and Hwang et al. (2002b). The precision and spatial resolution of marine gravity<br />
anomalies derived from satellite altimetry have also been significantly improved in<br />
open oceans over the past 25 years (Tapley and Kim, 2001; Rapp and Yi, 1997).<br />
Using the high-density data collected from multi-satellite missions, the precision of<br />
the derived global gravity fields is reported to have a range of ~±3 mgal to<br />
~±14 mgal from the comparisons of ship-track gravity and altimeter-derived gravity<br />
measurements (e.g., Tapley and Kim, 2001; Hwang et al., 2001).<br />
However, the precision of altimeter-derived gravity anomalies in coastal regions is<br />
not at the same level as those in open oceans. The differences between predicted<br />
anomalies from altimeter data and ship-track gravity data tend to become larger in<br />
coastal zones (e.g., Hwang et al., 1998b; Andersen and Knudersen, 2000;<br />
Featherstone, 2002; Hwang et al., 2002b). Hwang et al. (2002b) estimate a global<br />
gravity anomaly grid using Seasat, GEOSAT, ERS1-2, TOPEX/POSEIDON<br />
altimeter data and report a precision of ~±13 mgal in the Fiji Plateau and<br />
Mediterranean Sea, where the SSH data are likely affected by land and islands.<br />
Featherstone (2002) investigates the differences around the Australian coastal<br />
regions between ship-track gravity data and the most recent altimeter-derived gravity<br />
grids of GMGA97 (Hwang et al., 1998b), KMS01 (Andersen et al., 2001), and<br />
Sandwell’s version 9.2 (Sandwell and Smith, 1997). The result shows that larger<br />
differences (>100 mgal in magnitude) occur in areas closer to the coastal regions,<br />
which are consistent for all altimeter grids (see Figures 11-14 of Featherstone, 2002).
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 197<br />
The reasons for these larger differences observed near the coastal regions are<br />
probably as follows.<br />
• Land topography and rugged sea states contaminate coastal altimeter waveform<br />
measurements, causing SSHs to be in error. For example, TOPEX/POSEIDON<br />
altimeter waveform data are affected by land to ~0-34.8 km from the coastline<br />
in some coastal regions (Brooks et al., 1997); ERS-2 altimeter waveform data<br />
can be contaminated to a distance range of ~0-22 km off the Australian<br />
coastline (cf. Chapter 4).<br />
• Altimeter measurements cannot be properly corrected due to inaccurate<br />
geophysical corrections such as shallow-sea tides and the atmospheric delay.<br />
The recent analyses in accuracy of these corrections are given, e.g., by Shum et<br />
al. (1998), Andersen and Knudsen (2000), and Hwang et al. (1998b). In<br />
addition, waveform contamination introduces errors to the SWH for the SSB<br />
correction, which will be discussed later.<br />
• There are still large biases in ship-track gravity data due to improper or<br />
impossible data cleaning (i.e., crossover analysis, poor navigation in position,<br />
and Eötvös corrections) so that they cannot always offer definitive control on<br />
the accuracy of the altimeter-derived gravity anomalies.<br />
• Lack of altimeter data over the land near the coastline presents a problem that<br />
the gravity computation over coastal waters will be theoretically inaccurate.<br />
This is because the gravity recovery procedure requires uniform data around<br />
the computation point (see Eq. 15 of Hwang et al., 1998b), but this may not be<br />
the case when the point is close to the coastline.<br />
• The FFT technique requires the data to be on a regular grid. Incomplete data<br />
with gaps in coastal regions will cause the edge effect (i.e., Gibbs phenomenon)<br />
in recovery of gravity anomalies.<br />
In considering the first reason above, the error caused by altimeter waveform<br />
contamination in coastal regions can be mostly removed using waveform-retracking<br />
(e.g., Brooks et al., 1997; Deng et al., 2001). This Chapter focuses on improving the<br />
accuracy of altimeter-derived gravity anomalies through retracking ERS-1 altimeter<br />
waveform data in the Taiwan Strait over an area bound by 119°E ≤ λ ≤ 122°E and<br />
23°N ≤ ϕ ≤ 26°N. For the third cause above, there are some well-adjusted ship-track
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 198<br />
gravity data in the study area (see Section 2.5.3) which can be used to conduct an<br />
analysis between them and altimeter-derived gravity anomalies (not withstanding the<br />
errors in these data; mentioned above).<br />
The Taiwan Strait is characterised by shallow water (ocean depth less than 200 m),<br />
small islands, and rugged coastlines. Thus, it is assumed reasonably representative of<br />
the altimeter waveform features that may be encountered in most coastal regions.<br />
Nearly 2.5 years of ERS-1 altimeter waveform data over the area include the<br />
measurements from the ice, multi-disciplinary and geodetic phases, producing denser<br />
data coverage for the recovery of the gravity anomalies. These are the reasons that<br />
this area is chosen.<br />
The data and area signatures, waveform retracking methods and SSH extraction will<br />
be described in Sections 7.2, 7.3 and 7.4, respectively. In order to assess how the<br />
retracking procedure improves the accuracy of the altimeter-derived gravity<br />
anomalies, two SSH sets (i.e., before and after retracking) are extracted from<br />
waveform data using the same method except for the retracking procedure.<br />
Retracking is carried out over a larger area including the China Seas (i.e., the East,<br />
South, Yellow Seas and Bohai). The description in Section 7.5 will relate to the<br />
recovery of gravity anomalies from altimeter-derived data both before and after<br />
waveform retracking from the inverse Vening Meinesz formula (Hwang, 1998a)<br />
using the 1-D FFT technique. Comparisons of recovered anomalies from estimates<br />
by these two data sets to ship-track gravity data will be described. It should be<br />
acknowledged here that although these ship-track gravity data may contain errors,<br />
they are the only external data that can be used to indirectly verify the tracking<br />
algorithm’s differences.<br />
7. 2 Data and Area<br />
As stated in Section 2.4.1, ~2.5 years of 20 Hz ERS-1 waveform data from AWF<br />
products are used in a coastal region close to China and Taiwan bounded by<br />
110°E ≤ λ ≤ 129°E and 20°N ≤ ϕ ≤ 42°N. Figure 7.1 shows the geographical<br />
distribution of the observations, where the study area for the gravity anomalies in this<br />
Chapter is represented by a small rectangle (118°E ≤ λ ≤ 123°E and
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 199<br />
22°N ≤ ϕ ≤ 27°N). Among the ERS-1 repeat cycles, the geodetic mission (GM)<br />
enables the acquisition of a high density of altimeter measurements and thus<br />
improves the spatial resolution of the altimeter-derived SSH and hence the<br />
determination of marine gravity anomalies.<br />
Figure 7.1 Distribution of ERS-1 ground tracks<br />
As the original purpose of the data used in this area was to investigate the shallow<br />
water tides, observations with the ocean depth deeper than 200 m were omitted,<br />
making no data in the eastern area of Taiwan (Figure 7.1). This presents a problem<br />
for the recovery of the gravity anomaly in this part of the study area, where the final<br />
marine gravity field will be dominated by the reference gravity field and cannot show<br />
the high frequency components (i.e., ~20 km wavelength) because of the lack of data.<br />
The study area is a narrow strait (roughly ~110 km wide) between Taiwan and China<br />
linking the East China Sea to the South China Sea (named the Taiwan Strait). The<br />
computation area is 1° larger than the final gridded gravity field (119°E ≤ λ ≤ 122°E
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 200<br />
and 23°N ≤ ϕ ≤ 26°N) at the edges to avoid the Gibbs effect of the FFT technique.<br />
Table 7.1 shows a summary of altimeter observations in the area.<br />
Table 7.1 Summary of ERS-1 observation cycles bounded by 118°E ≤ λ ≤ 123°E and<br />
22°N ≤ ϕ ≤ 27°N<br />
Repeat cycle<br />
Time period<br />
Number<br />
Ascending<br />
Descending<br />
(days)<br />
of cycles<br />
pass<br />
pass<br />
3 23/12/1993 - 10/04/1994 36 0 1<br />
35 13/07/1993 - 23/12/1993<br />
21/03/1995 - 02/06/1996<br />
5<br />
13<br />
8 6<br />
168 04/10/1994 - 24/03/1994 1 35 24<br />
Although only ERS-1 altimeter data are used, an adequate density (4×4 minutes) of<br />
measurements is the main characteristic of the data set readily available at the time of<br />
this research. From Table 7.1, the 3-day repeat cycle shows only one descending<br />
ground track, but the 168-day repeat cycle presents 35 ascending and 24 descending<br />
tracks. Together with eight ascending and six descending tracks from the 35-day<br />
repeat cycle, satellite ground tracks from three types of repeat cycles create a denser<br />
coverage of data than that using only single repeat cycle of data over this area,<br />
increasing greatly the spatial resolution (Figure 7.1). This will contribute most to the<br />
high-frequency parts of the coastal marine gravity field.<br />
To assess the accuracy of the gravity anomaly from satellite altimeter, the ship-track<br />
gravity data at 397 points described in Section 2.5.3 around Taiwan are used in this<br />
study (Figure 2.5).<br />
7. 3 Waveform Retracking<br />
To retrack the return waveforms in this area, the iterative least squares procedure<br />
(Section 5.5) is used to fit an analytic function representing the expected return to the
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 201<br />
waveform. The forms of the analytic function used in this study are the fiveparameter<br />
model (Equation 3.40 and n =1) and the ocean model (Equation 5.3).<br />
7.3.1 Retracking Scheme<br />
Waveform contamination occurs mostly in the vicinity of land and islands (see<br />
Chapter 4). Over ocean surfaces, the contamination occurs only over regions where<br />
the rugged sea state or the off-nadir satellite attitude excursions affect the shape of<br />
the return (Brenner et al., 1993). However, research conducted by Fairhead et al.<br />
(2001) and Sandwell (2003) indicates that waveform retracking is required over all<br />
oceans to improve the resolution and quality of the marine gravity field. Therefore,<br />
all 2.5 years of ERS-1 20 Hz waveforms shown in Figure 7.1 are retracked. Before<br />
retracking, an edit criterion (Section 5.2) was applied to remove waveforms in which<br />
there is no leading edge in the waveform sample window, which removed ~2% of the<br />
data.<br />
When fitting, a priori values and weights are determined using the method presented<br />
in Section 5.5.4. Retracking results are obtained by retracking the waveform over<br />
ocean surfaces and in coastal regions using the fitting functions of either the ocean<br />
model or five-parameter model. Each waveform is first classified into ocean-like and<br />
non-ocean shapes so that it can be retracked by the appropriate functional fit. The<br />
iterative procedure is repeated until the unknown parameter estimate ( t 0<br />
or β<br />
3<br />
)<br />
converges to within 0.1% or<br />
−4<br />
σ<br />
0<br />
−σ<br />
0<br />
< 10 . It converges in typically 2-4<br />
i<br />
i −1<br />
iterations. In addition, the five-parameter function replaces the ocean model to<br />
compute the retracking correction when the ocean model fails.<br />
The retracking procedure is conducted by using either the five-parameter model or<br />
the ocean model. Retracking was first performed using both models. A collinear<br />
analysis was applied to one cycle of 35-day repeat of 1 Hz SSH data (from July to<br />
August 1993) to investigate whether there is a serious bias between both methods.<br />
The difference ( dh ) between the retracked SSH data calculated from the ocean<br />
model and five-parameter model was computed and analysed. Figure 7.2 shows a<br />
histogram of dh in which the mean SSH difference is -0.7±13.5 cm (3070<br />
observations). Of the residuals, 95% are less than 10 cm and 83% are less than 5 cm,
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 202<br />
indicating a good agreement between both methods and a similar result as ERS-2<br />
(see Section 5.8). This result also means that the combination of both ocean and fiveparameter<br />
models can maintain a consistent measurement noise level across the<br />
entirety of the sea surface profile. Since the ocean model physically represents the<br />
returns over oceans, this is the reason that both models are finally chosen to be the<br />
main retracking functions in this study.<br />
Frequency of data<br />
800<br />
600<br />
400<br />
200<br />
0<br />
-1.45<br />
-1.27<br />
-1.08<br />
-0.90<br />
-0.71<br />
-0.52<br />
-0.33<br />
-0.14<br />
0.05<br />
dh (m)<br />
0.23<br />
0.42<br />
0.61<br />
0.80<br />
0.99<br />
1.18<br />
1.37<br />
Figure 7.2 The histogram of dh (the SSH difference between the five-parameter<br />
model and the ocean model (mean -0.7 cm; STD 13.5 cm; 3070 observations).<br />
7.3.2 Examples of Waveform Retracking Results<br />
Figure 7.3 shows one ascending (north-westwards) and one descending ground track<br />
(south-westwards) in the study area. The ascending track (A1) goes from water to<br />
land. The descending track (D1) recedes from land to water. It is parallel to the<br />
coastline at ~37°N latitude, and then intercepts the land at ~32.4°N latitude. The<br />
track then re-commences as D2 and goes to the Taiwan Strait. The 20 Hz waveforms<br />
have been retracked using the methods mentioned above and compressed to 1 Hz<br />
SSH data; the before/after retracking 1 Hz SSH results, plotted versus latitude, are<br />
shown in Figure 7.4 for these three tracks. The profiles from top to bottom of<br />
Figure 7.4 show the SSH from tracks A1, D1 and D2. The SSH derived from the<br />
original (i.e., not retracked) and retracked (using the five-parameter and ocean<br />
models) data are shown. It is apparent that the retracking has removed some<br />
measurement noise near the coastal regions caused by land topography and<br />
rugged/rough sea states.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 203<br />
Figure 7.3 Three-ground tracks in the study area<br />
As for the SSH derived from waveforms before retracking, Figure 7.4a does not<br />
show the obvious land effect on the unretracked SSH data when the ascending track<br />
(A1) approaches land at ~23.5°N latitude, while the contamination which may be<br />
caused by rough sea states occurs from ~21.8°N to ~22.4°N latitudes. Descending<br />
track (D1, Figure 7.4b) shows obvious sharp changes in unretracked SSH results<br />
when the track gets close to the land. Similar sharp changes can also be found in<br />
unretracked SSH data along the descending track (D2, Figure 7.4c). It is evident in<br />
Figure 7.4 that the retracking, no matter which method, has removed the ‘error’<br />
(unexpected sharp changes) in the SSH and the highly variable measured sea surface<br />
is replaced by the smoothed profile. The SSH data from retracked waveforms using<br />
the five-parameter model or ocean model show similar results (Figure 7.4).
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 204<br />
Sea Surface Height (metre)<br />
14.0<br />
12.0<br />
10.0<br />
8.0<br />
Five−parameter<br />
Ocean + five−parameter<br />
without retracking<br />
6.0<br />
21.7 22.2 22.7 23.2<br />
Latitude (degree)<br />
18.0<br />
Sea Surface Height (metre)<br />
16.0<br />
14.0<br />
12.0<br />
10.0<br />
8.0<br />
Five−parameter<br />
Ocean + five−parameter<br />
without retracking<br />
6.0<br />
32.0 34.0 36.0 38.0 40.0<br />
Latitude (degree)<br />
sea Surface Height (metre)<br />
21.0<br />
20.0<br />
19.0<br />
18.0<br />
17.0<br />
16.0<br />
15.0<br />
Five−parameter<br />
Ocean + five−parameter<br />
without retracking<br />
14.0<br />
22.0 23.0 24.0 25.0 26.0<br />
Latitude (degree)<br />
Figure 7.4 SSHs from ground track A1 (a), D1 (b), and D2 (c) in Figure 7.3 before<br />
and after retracking using the retrackers of the five-parameter and ocean models.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 205<br />
The results in Figure 7.4 show that it is necessary to retrack all waveforms in this<br />
area due to its characteristics of shallow water, rugged coastlines and rough sea states.<br />
It also can be seen from Figure 7.4 that the retracking methods used can extend the<br />
reliability of the measurements shoreward. From this analysis, there is a small bias<br />
between the SSH before and after retracking (cf. Figure 7.4). A similar problem has<br />
also been found when using ERS-2 waveform data around Australian coastal regions<br />
(Section 5.8.4), where a bias of ~27 cm is estimated from two cycles of waveform<br />
data. Since the along-track geoid slope will be used in this study to calculate the<br />
gravity anomaly, the differentiation operation can effectively remove this bias. Thus,<br />
its effect on the SSH can be reduced. However, it cannot be neglected when<br />
analysing the absolute temporal variation in SSHs or mixing the original data (before<br />
retracking) and retracked data.<br />
7. 4 The SSH Extracted from Waveforms<br />
To assess the results from the above retracking, the SSH is extracted using two<br />
approaches. The first is to create the 1 Hz SSH data set from the 20 Hz data source<br />
without waveform retracking, while the other is to retrack the 20 Hz waveform and<br />
then to ‘compress’ the retracked SSH to the 1 Hz SSH set. No matter which method<br />
is used, a quadratic function is fitted to 20 Hz SSHs and Pope’s (1976) tau-test is<br />
applied to detect the erroneous SSH data. Final 1 Hz SSHs are calculated at the midpoint<br />
of the fitting function after removing the outlying observations.<br />
7.4.1 Corrections<br />
To obtain the data set of the SSH, geophysical and environmental corrections are<br />
applied to the range measurement. The corrections for atmospheric propagation<br />
delays are obtained directly from the AWF waveform product. The AWF product did<br />
not supply the inverse barometer correction, which is related to the hydrostatic<br />
response of the sea surface with increasing or decreasing atmospheric pressure.<br />
Instead, it was computed using the surface atmospheric pressure which is available<br />
indirectly via the dry tropospheric correction obtained from the AWF. The tidal<br />
correction applied is the superposition of the elastic ocean tides (i.e., the ocean tide<br />
and loading tide) derived from CRS4.0 global ocean tide model (Eanes, 1999), and
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 206<br />
the solid Earth tide supplied by the AWF. The vertical (radial) velocity of the<br />
satellite causes a frequency Doppler shift which affects the time delay measurement<br />
and thus the range. For each elementary measurement, the Doppler correction is<br />
added to the altimeter range. Table 7.2 lists the variability of the corrections applied<br />
in the area and their maximum and minimum magnitude. From Table 7.2, the change<br />
of the SSB was found to cause an error in the SSH data in the study area.<br />
Table 7.2 The variability of the range corrections to deliver SSH in the study area<br />
(cf. Figure 7.1)<br />
Correction Min (m) Max (m)<br />
Dry tropospheric delay 2.27 2.30<br />
Wet tropospheric delay 0.12 0.40<br />
Ionospheric delay 0.05 0.09<br />
Inverted barometer 0.05 0.17<br />
SSB -1.23 -0.02<br />
Elastic Ocean tides -2.15 0.22<br />
Solid Earth tides -0.10 0.03<br />
Doppler shift -0.01 0.01<br />
As stated in Section 3.3.3, the SSB is the difference between the apparent sea level<br />
measured by an altimeter and the true mean sea level. Theoretical understanding of<br />
the SSB, particularly the EMB, remains limited, so the SSB correction is calculated<br />
from the SWH and the wind speed using an empirical model derived from analysis of<br />
altimeter data itself (AVISO/Altimeter, 1998; Chelton et al., 2001). The SWH, as<br />
computed by the on-satellite processor (MacArthur, 1980, cited in Martin et al.,<br />
1983), is a measure of the ocean wave amplitude. Over the coastal regions near the<br />
land, the SWH computed on-satellite is not useful due to the effects of land<br />
topography and generally more rough sea states. Therefore, the SSB may no longer<br />
be correct in these areas.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 207<br />
The SSB’s effect on the SSH is analysed through SSH profiles along the three<br />
ground tracks in Figure 7.3. SSH profiles along tracks are plotted based upon<br />
whether they are retracked or unretracked, applied by all geophysical corrections<br />
including SSB or not. Besides the SSB, parts of corrections of dry tropospheric delay<br />
and elastic ocean tides are chosen to plot along each track (from Figures 7.5 to 7.7),<br />
because they also show the larger variety in magnitude (Table 7.2). As can be seen,<br />
the SSB shows an obvious step change in this coastal region, but it is not expected to<br />
change significantly and suddenly over a few kilometres unless a rougher sea state is<br />
present.<br />
Compared to the SSB, other corrections of dry tropospheric delay and ocean tides<br />
show a relatively smooth change. As a result, the SSH profiles are shifted upward a<br />
value of the sum of corrections after they are applied to the range measurements.<br />
However, since the SSB should be subtracted from the SSH, it doubles its effect on<br />
the SSH when the step change occurs. Therefore, the SSB is not applied to the range<br />
measurements in this study. This may leave an error for both unretracked and<br />
retracked SSH data. However, as the retracking procedure partly removes the ocean<br />
wave influence on the waveform, the need of the range measurement for the SSB has<br />
been reduced. This is the reason that the retracked SSH shows the smoothing profiles<br />
in Figures 7.5, 7.6 and 7.7. In addition to the SSB, there are some sharp changes to<br />
be found in SSH profiles without retracking even in ocean areas far away from the<br />
coastline (see, Figures 7.5 and 7.6), while waveform retracking, again, obviously<br />
removes these unexpected changes. This implies that the SSH cannot be precisely<br />
computed from the on-satellite tracking algorithm in this coastal and semi-enclosed<br />
sea.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 208<br />
14.0<br />
Sea Surface Height (metre)<br />
10.0<br />
6.0<br />
2.0<br />
−2.0<br />
21.7 22.2 22.7 23.2<br />
Latitude (degree)<br />
Figure 7.5 SSH profiles and parts of corrections along the track A1.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 209<br />
18.0<br />
Sea Surface Height (metre)<br />
14.0<br />
10.0<br />
6.0<br />
2.0<br />
−2.0<br />
32.0 34.0 36.0 38.0 40.0<br />
Latitude (degree)<br />
Figure 7.6 SSH profiles and parts of corrections along the track D1.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 210<br />
22.0<br />
Sea Surface Height (metre)<br />
18.0<br />
14.0<br />
10.0<br />
6.0<br />
2.0<br />
−2.0<br />
22.0 23.0 24.0 25.0<br />
Latitude (degree)<br />
Figure 7.7 SSH profiles and parts of corrections along the track D2.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 211<br />
7.4.2 The Weight Function<br />
After waveform retracking, both retracked and unretracked 20 Hz SSH data derived<br />
from waveforms along ground tracks are created to 1 Hz SSH sets. Before the<br />
calculation of the gravity anomaly using these SSH data, a procedure of data filtering<br />
is necessary to delete possible outliers. The 1 Hz data are then separately smoothed<br />
and edited by a low-pass filter to remove the noise in the measurements and ensure<br />
the accuracy of altimeter-derived gravity anomalies. Because of the advantage of the<br />
data filter, a Gaussian gridding kernel is chosen as the weight function for removing<br />
outliers in SSH data and computing the average gradients along ascending and<br />
descending tracks (stated below). For a point ( ϕ , λ)<br />
from the observation ds<br />
kilometres, the formula is<br />
2<br />
⎡ 1 ⎛ ds ⎞ ⎤<br />
w(<br />
ϕ , λ)<br />
= exp⎢−<br />
⎜ ⎟ ⎥<br />
ds ≤ R s<br />
. (7.1)<br />
⎢⎣<br />
2 ⎝ σ ⎠ ⎥⎦<br />
where R s<br />
= D / 2 is the search radius of the weight function, D is the search<br />
window size, and σ = D / 6 . The value of D is important for this function due to the<br />
degree of smoothness increasing with it. From Equation (7.1), an influenced point<br />
from the observation will gain a small weight near the edge of the search window no<br />
matter how large the size is. Thus, a large value of D will over smooth the filtered<br />
results, while a small value of D will still leave unreasonable results. In this study,<br />
D =70 km is chosen for the along track SSH smoothing. This value ensures that there<br />
are about 10 seconds [time] of SSH data to be included in each filtering procedure<br />
for obtaining a robust result. For calculating along-track average gradients on a 2'×2'<br />
grid, the value of D is chosen to be the diagonal of the cell (~5 km) so that all SSH<br />
data can be searched (see Section 7.5.2).<br />
7.4.3 Data Filtering and Editing<br />
Data filtering will smooth out any high-frequency signal that is larger than the<br />
proposed resolution to avoid data aliasing. SSH data filtering is a moving average<br />
procedure, which uses a weight function (Equation 7.1, D = 70 km) and slides along<br />
the track by 1 second of data (i.e., 20 SSHs) for each calculation. This procedure is
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 212<br />
carried out at all observation points, resulting in a filtered SSH set. The final SSH at<br />
( ϕ , λ)<br />
is given as:<br />
SSH ( ϕ , λ)<br />
= ∑(<br />
SSH )<br />
i<br />
wi<br />
∑ wi<br />
.<br />
(7.2)<br />
i<br />
i<br />
where<br />
w represents the value of the weight at the point ( ϕ , λ)<br />
for the particular i th<br />
i<br />
observation within the distance D .<br />
Data that disagree by more than three times of STD are flagged as outliers and<br />
replaced by the filtered values. In addition, when the number of SSH data within D<br />
are less than 4, the smoothing procedure is not conducted, and all data are deleted.<br />
Approximately 1% of the unretracked and 2% of retracked initial data sets are<br />
eliminated using this filter, implying that retracked data show more short-wavelength<br />
signal than the unretracked data. Figures 7.8 and 7.9 show filtered (dark curves) and<br />
unfiltered (light curves) SSH results before and after retracking along ground tracks<br />
in Figure 7.3 through the above Gaussan smooth procedure, respectively. As can be<br />
seen, the Gaussian weight function removes the measurement error in observations.<br />
7. 5 Marine Gravity Anomalies from Satellite Altimetry<br />
There have been several approaches developed to compute gravity anomalies from<br />
satellite altimeter data over the past two decades. The four main types of the<br />
philosophies and computational procedures have been summarised by Featherstone<br />
(2002). Each of them has advantages and disadvantages. Featherstone (ibid.) shows<br />
that there are some significant differences among marine gravity anomalies<br />
computed using different methods from satellite altimetry around Australia, and<br />
these tend to become larger in coastal areas. Therefore, it is still an open question to<br />
say which one is better than others.<br />
One of them is the approach described by Hwang (1998a), which is used in this study<br />
due to its numerically efficient computation and no requirement of crossover<br />
adjustment (e.g., Sandwell and Smith, 1997; Hwang, 1998a; Andersen and Knudsen,<br />
1998). In this approach, the gravity anomaly is calculated from the along-track<br />
deflections of the vertical (or along-track SSH slopes) derived from satellite altimeter
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 213<br />
Figure 7.8 Untracked SSH profiles, showing the filtered SSH data (dark curves) and<br />
unfiltered SSH data (light curves) along the tracks A1 (a), D1 (b) and D2 (c) in<br />
Figure 7.3.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 214<br />
Figure 7.9 Retracked SSH profiles, showing filtered SSH data (dark curves) and<br />
unfiltered SSH data (light curves) along the tracks A1 (a), D1 (b) and D2 (c) in<br />
Figure 7.3.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 215<br />
data using the inverse Vening Meinesz formula (Hwang, 1998a). The use of alongtrack<br />
geoid slope data has been proved to be one of effective techniques used in the<br />
calculation of gravity anomalies from satellite altimeter data, as the differentiation<br />
operation effectively reduces the effect of the long wavelength radial orbit error (e.g.,<br />
Sandwell, 1984; Hwang and Parsons, 1995; Sandwell, 1997).<br />
Before calculating along-track deflections of the vertical, the quasi time-independent<br />
estimates of dynamic ocean topography must be removed from the altimeter-derived<br />
SSHs. Such removal is conceptually required because the ocean surface is not an<br />
equipotential surface, but the geoid is. The remove-compute-restore technique is<br />
used, in which the deflections of the vertical (i.e., ξ north-south and η east-west)<br />
implied by EGM96 (Lemoine et al., 1998) global geopotential model to spherical<br />
harmonic degree 360 is used as a reference field, and it is removed from the raw<br />
deflections of the vertical to yield a residual deflections of the vertical set. These<br />
north and east component of deflections of the vertical are then interpolated onto a<br />
regular grid using least squares collocation (Moritz, 1980) after a procedure of<br />
determining outliers in the data (cf. Hwang et al., 2001). The inverse Vening<br />
Meinesz formula (Hwang, 1998a) is employed to convert these gridded north-south<br />
and west-east components to gravity anomalies via the 1-D FFT technique. The final<br />
gravity anomalies are obtained by adding back the eliminated 360 degree EGM96<br />
reference field.<br />
The approach has been successfully applied to recover marine gravity anomalies<br />
from multi-satellite altimeter data (e.g., SEASAT, GEOSAT, ERS-1 and<br />
TOPEX/POSEIDON) both at global and regional scales (Hwang, 1998a; Hwang et<br />
al., 2002b). The advantage of this computational procedure is that data are carefully<br />
prepared using the gridding and filtering techniques so that the data can be extracted<br />
efficiently. It is noted that ERS-1 altimeter data close to the coastline (~20 km from<br />
the coastline line) had not been used by Hwang (2001) to avoid data contamination<br />
in coastal regions. In this study, however, data used are only from the ERS-1<br />
altimeter in a coastal region. This presents a necessity for this study that some slight<br />
modifications (see Section 7.5.1) should be conducted in the computing procedure to<br />
use all potential retracked SSHs in coastal regions when recovering the gravity<br />
anomalies through the inverse Vening Meinesz formula.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 216<br />
7.5.1 Along-track Deflections of the Vertical<br />
Before computing the along-track deflections of the vertical, in order to obtain<br />
geoidal height from SSHs, the value of the quasi-time independent SST interpolated<br />
from the model of Levitus et al. (1997) is subtracted from the SSH. In addition, the<br />
reference gravity field of the EGM96 to harmonic degree 360 is also removed from<br />
the geoidal height, meaning the along-track deflections of the vertical (generated<br />
from below Equation 7.3) are residual deflections of the vertical or residual geoidgradients.<br />
Taking the first horizontal derivatives of the altimeter-derived SSHs along track<br />
generates the along-track deflections of the vertical (DOV), which is the negative<br />
gradient of the geoid (Sandwell, 1992; Olgiati et al., 1995; and Hwang et al., 1998b).<br />
The along-track gradient, ε , is calculated at the midpoint between two successive<br />
observations by the formula<br />
ε<br />
k<br />
=<br />
N<br />
k +1<br />
− N<br />
∆s<br />
k<br />
(7.3)<br />
where<br />
N<br />
k<br />
is the geoidal height at the observation k , and ∆ s is the distance between<br />
observations k + 1 and k . In addition, the following Equation is less noisy for the<br />
along-track DOV:<br />
ε<br />
N<br />
1<br />
− N<br />
2∆s<br />
k + k −1<br />
k<br />
=<br />
(7.4)<br />
7.5.2 Averaging Gradients<br />
Usually, before computing the along-track geoid slope, the SSHs from repeat<br />
missions are reduced to a reference orbit (see Section 6.6) and then averaged them to<br />
remove the time-variant ocean signals and measurement noise (e.g., Hwang and<br />
Parsons, 1995; Hwang et al., 2002a; Rapp and Yi, 1997; Sandwell and Smith, 1997).<br />
However, lack of a reference orbit makes it difficult to directly average geoid heights<br />
for ERS-1/35 repeat data. Therefore, a new method has been developed to average<br />
along-track geoid gradients from repeat cycles in this coastal region. This procedure<br />
averages the geoid slope rather than the geoid height, thus avoiding the introduction
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 217<br />
of offsets at data gaps (cf. McAdoo, 1990; Kirby, 1996) and reducing the timevariant<br />
signals.<br />
The data used contain 3-day and 35-day repeat missions, and geodetic mission. Of<br />
these missions, the repeat mission repeats ground tracks to within 1 km, while the<br />
ground track of ERS-1/GM is unpredictable (with respect to the 35-day repeat track).<br />
For the repeat missions, the same pass (ascending or descending ground track) from<br />
different cycles does not repeat exactly due to some causes, such as orbital variations<br />
caused by gravitational influence and drag effects. Orbital manoeuvres have been<br />
used in the satellite to maintain a ground track within some specified linear tolerance<br />
of a nominal or reference track (Wang and Rapp, 1991). For most satellite altimeters<br />
such as GEOSAT and T/P, the aim is to achieve cross-track deviations on the order<br />
of 1 km (Wang and Rapp, 1991). This is the same case as the ERS satellite. Taking<br />
this data characteristic in mind, ideally the averaging method should retain the alongtrack<br />
resolution, but also average the data into the spaces between tracks from<br />
different repeat cycles so that a complete along-track gradient set can be generated.<br />
In this new averaging procedure, residual geoid heights at any location are first<br />
sorted and found in a 2′×2′ cell. A weighted mean geoid gradient, ε , is then<br />
computed at this cell using a Gaussian weight function (7.1). The search radius<br />
(~2.5 km) is chosen to be half the diagonal of the cell so that all SSHs in a cell can be<br />
searched. In addition, the standard deviation of the mean geoid gradient is calculated<br />
in the cell by the formula<br />
∑<br />
( ε − ε )<br />
i<br />
2<br />
i<br />
σ<br />
k<br />
=<br />
(7.5)<br />
n −1<br />
where n is the number of the sea surface slopes in the cell. When n =<br />
1, it is<br />
impossible to calculate σ<br />
k<br />
from (7.5). In this case, σ<br />
k<br />
is given a value of 0.21 µrad<br />
which is based upon an empirical noise level of ERS-1 (cf. Hwang et al., 2001). This<br />
usually happens to the mean slopes formed from the geodetic phase.<br />
After averaging gradients and before forming the regular grids for the use of 1-D<br />
FFT algorithm, a similar outlier-removing procedure to Hwang et al. (2002b) is<br />
applied to the mean gradients to remove residual erroneous data. Unlike Hwang’s
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 218<br />
scheme, where a 4′×4′ cell was used due to a high density of the ground coverage<br />
from multi-satellite altimeter data, the size of the cell is increased to 8′×8′. The<br />
reasons for this are as follows. First, the data used comes only from the ERS-1<br />
altimetry, thus a small cell cannot provide enough data points (some 4′×4′ cells may<br />
contain only one data point from the GM phase) and produce larger degrees of<br />
freedom to make the result more reliable, leading to improper detection of outliers.<br />
Second, since the study area is a coastal region, a small cell will cause all data to be<br />
eliminated when close to the coastline. However, it should be acknowledged that a<br />
larger cell might oversmooth the data.<br />
Averaging is carried out separately on the ascending and descending groundtracks.<br />
Table 7.3 lists the descriptive statistics results from averaging gradients in this way<br />
and outlier detection. Before detecting outliers, the mean value of the averaged<br />
slopes (before retracking, both ascending and descending) is 0.13 µrad and their<br />
standard deviation is ±3.01 µrad, while the mean of the slopes (after retracking, both<br />
ascending and descending) is 0.25 µrad and their standard deviation is ±2.61 µrad.<br />
After detecting outliers, the standard deviations reduce to ±2.34 µrad and ±1.82 µrad<br />
before and after retracking, respectively. There are 15% and 13% of outliers which<br />
are detected from geoid gradient data sets before and after waveform retracking,<br />
respectively. From Table 7.3, the number of final residual geoid gradients before<br />
retracking is more than that after retracking. Again, this shows that the retracking<br />
procedure is effective.<br />
Table 7.3 Descriptive statistics of the averaged residual geoid gradients before and<br />
after the outlier detection on a 8′×8′ cell.<br />
Before detecting outliers After detecting outliers<br />
Methods Mean<br />
(µrad)<br />
STD<br />
(µrad)<br />
Number<br />
of slopes<br />
Mean<br />
(µrad)<br />
STD<br />
(µrad)<br />
Number<br />
of slopes<br />
Before retracking 0.13 ±3.01 5949 -0.01 ±2.34 5177<br />
After retracking 0.25 ±2.61 5806 0.14 ±1.82 4953
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 219<br />
7.5.3 Gravity Anomaly Computation<br />
After removing outliers, the along-track geoid slopes are converted to regular grids<br />
of east (η) and north (ξ) vertical deflection using least-squares collocation. The<br />
covariance functions used here are taken from Hwang and Parsons (1995). Then, the<br />
calculations for the determination of residual gravity anomalies are carried out using<br />
the inverse Vening Meinesz formula Equation (23) of Hwang (1998a), and the<br />
innermost zone effects is estimated by Equation (47) of Hwang (ibid.) using 1-D FFT<br />
and the remove-restore technique. As stated, a 1° border is used to avoid Gibbs edge<br />
effects. The flow diagram in Figure 7.10 illustrates how altimeter waveform<br />
observations are processed and converted to gravity anomalies. The procedure<br />
follows the steps described in Section 7.4 for extracting the SSHs and estimating the<br />
gravity anomalies in this Section. Two gravity-anomaly grids (2′×2′) are generated<br />
by this procedure: a grid without waveform retracking; and a grid with waveform<br />
retracking. Figures 7.11 and 7.12 show the equivalent free air anomalies relative to<br />
GRS80 normal gravity estimated by unretracked and retracked ERS-1 SSH data.<br />
7.5.4 Comparisons between Marine and Ship Gravity Anomalies<br />
The accuracy of the gridded free air anomaly in Figures 7.11 and 7.12 can be<br />
assessed by comparing the ship-track gravity data with anomaly fields generated<br />
using altimeter data. The comparison is made along ship tracks in the Taiwan Strait<br />
in the vicinity of Taiwan (Section 2.4.3), where the ship data available is shown in<br />
Figure 2.5 and the altimeter ground tracks are shown in Figure 7.1. Each of the<br />
gridded anomaly data sets derived from altimeter SSH data before and after<br />
retracking is interpolated using a bispline function to the position of the ship-derived<br />
gravity anomaly. The number of the ship-track gravity data used in the comparisons<br />
in this area is 397 points. The descriptive statistics of the differences are shown in<br />
Table 7.4. Considering the retracked results in the comparison, it can be seen that the<br />
solution is better than that which uses unretracked SSHs in the anomaly estimation.<br />
The standard deviation of the difference (altimeter-derived anomalies minus ship<br />
anomalies) of ±13.93 mgal before retracking is larger than that of ±9.92 mgal after<br />
retracking. Therefore, the precision of the gridded altimeter-derived gravity anomaly<br />
has been improved by ~4 mgal.
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 220<br />
Figure 7.10 Flow diagram for the SSH extraction from waveforms and recovery of<br />
gravity anomalies
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 221<br />
119˚<br />
26˚<br />
120˚<br />
121˚<br />
122˚<br />
26˚<br />
183.79<br />
89.96<br />
25˚<br />
25˚<br />
66.44<br />
49.49<br />
34.99<br />
m<br />
gal<br />
21.45<br />
7.90<br />
24˚<br />
24˚<br />
-6.59<br />
-23.55<br />
-47.07<br />
-147.56<br />
23˚<br />
119˚<br />
120˚<br />
121˚<br />
23˚<br />
122˚<br />
Figure 7.11 Gravity anomalies from altimeter data before retracking (units in mgal)<br />
119˚<br />
26˚<br />
120˚<br />
121˚<br />
122˚<br />
26˚<br />
187.70<br />
91.47<br />
25˚<br />
25˚<br />
67.90<br />
50.91<br />
36.39<br />
m<br />
gal<br />
22.82<br />
9.25<br />
24˚<br />
24˚<br />
-5.27<br />
-22.27<br />
-45.83<br />
-143.78<br />
23˚<br />
119˚<br />
120˚<br />
121˚<br />
23˚<br />
122˚<br />
Figure 7.12 Gravity anomalies from altimeter data after retracking (units in mgal)
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 222<br />
It is acknowledged that there is a low accuracy from Table 7.4 for both altimeterderived<br />
gravity anomalies before and after retracking with respect to other studies,<br />
e.g., Rapp and Yi (1997). As stated in Section 2.5.3, however, the accuracy of the<br />
ship-track gravity data after the crossover adjustment is ±11.2 mgal, although these<br />
data has been subjected to a data cleaning processes. This means that there is still the<br />
error existing in the ship-track gravity data. Therefore, the error of the ship-track<br />
gravity data should be a possible reason that causes larger STDs in Table 7.4.<br />
Figures 7.13 and 7.14 show the different results along ship tracks using unretracked<br />
and retracked altimeter data, respectively. As can be seen from Figure 7.13,<br />
numerous larger differences with magnitudes greater than +15 mgal occur in any<br />
ship track not being limited to near the coast. As stated, the ERS-1 waveform data<br />
are contaminated in this area not only in coastal regions but also in the water area far<br />
from the shoreline (cf. Figures 7.5 and 7.6). The larger differences here are more<br />
likely to have been subjected to some form of waveform contamination. This<br />
becomes more evident when comparing Figure 7.13 with Figure 7.14 (after<br />
retracking), where larger differences (>15 mgal in magnitude) appear along the ship<br />
tracks that are very much closer to Taiwan’s coast. These larger differences near the<br />
coastline may be due either to the limited distance that waveform retracking can<br />
improve or the Gibbs edge effect at the coastline.<br />
Table 7.4 Descriptive statistics of the differences between altimeter-derived grids and<br />
the 397 ship-track free-air gravity anomalies (units in mgal)<br />
Grid Max Min Mean STD RMS<br />
Before retracking 32.71 -41.62 -4.86 ±13.93 ±14.74<br />
After retracking 19.82 -41.08 -6.09 ±9.92 ±11.63
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 223<br />
119˚<br />
26˚<br />
120˚<br />
121˚<br />
122˚<br />
26˚<br />
25˚<br />
25˚<br />
35<br />
25<br />
15<br />
5<br />
-5<br />
m<br />
gal<br />
24˚<br />
24˚<br />
-15<br />
-25<br />
-35<br />
-45<br />
23˚<br />
119˚<br />
120˚<br />
121˚<br />
23˚<br />
122˚<br />
Figure 7.13 Differences between altimeter-derived (before retracking) and ship-track<br />
gravity anomalies<br />
119˚<br />
26˚<br />
120˚<br />
121˚<br />
122˚<br />
26˚<br />
25˚<br />
25˚<br />
35<br />
25<br />
15<br />
5<br />
-5<br />
m<br />
gal<br />
24˚<br />
24˚<br />
-15<br />
-25<br />
-35<br />
-45<br />
23˚<br />
119˚<br />
120˚<br />
121˚<br />
23˚<br />
122˚<br />
Figure 7.14 Differences between altimeter-derived (after retracking) and ship-track<br />
gravity anomalies
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 224<br />
In interpreting the anomaly difference here, one should recall that different altimeter<br />
SSH data sets are extracted from the same waveform source except for the retracking<br />
procedure. All other calculations are carried out using the same procedure. Therefore,<br />
the results shown in Table 7.4 reflect the improvements only from waveform<br />
retracking in coastal regions.<br />
7. 6 Summary<br />
Gravity anomaly predictions were made from ~2.5 years of ERS-1 SSH data sets<br />
before and after retracking using the inverse Vening Meinesz formula and the results<br />
were compared with ship-track gravity data in the Taiwan Strait. The comparison of<br />
results demonstrated that waveform retracking has improved the quality of the SSH<br />
in coastal regions near the China Sea and Taiwan Strait, thus improving the accuracy<br />
of the altimeter-derived gravity field. The agreements between the predicted and<br />
ship-track gravity anomalies are ±13.9 mgal before retracking and ±9.9 mgal after<br />
retracking.<br />
The analysis of extracting the SSHs from altimeter waveform data suggests that the<br />
SSB correction is inaccurate when it is computed using the empirical model in the<br />
coastal region, due mainly to the incorrect SWH estimated from the on-satellite<br />
processor. Waveform retracking can remove the need for the ocean wave influence<br />
and thus reduce the SSB error, while without or using inaccurate SSB corrections<br />
causes the error to the unretracked SSH data. In addition, the method of averaging<br />
residual geoid gradients developed for the AWF data product in this study has been<br />
demonstrated as being effective. When applied this averaging method to the<br />
unretracked SSHs, an accuracy of ~ ±14 mgal can be obtained for the altimeterderived<br />
gravity anomalies, which is a similar level to that in Hwang et al. (2001).<br />
Results from this Chapter show a great potential of providing accurate gravity field<br />
information in coastal regions by using satellite altimeter SSH data after waveform<br />
retracking. For further research, to overcome the problem of altimeter data<br />
discontinuity very near to the coastline, combining the uniform coverage of the<br />
altimeter data with ship-borne gravity and land gravity data will provide an accurate<br />
high-resolution gravity field in coastal regions. It is also necessary to improve the un-
Chapter 7. IMPROVED COASTAL MARINE GRAVITY ANOMALIES 225<br />
modelled or poorly modelled geophysical corrections (e.g., SSB, wet tropospheric<br />
delay, sallow-water tides). However, this was not a specific topic of this research.
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 226<br />
8. CONCLUSIONS AND RECOMMENDATIONS<br />
8. 1 Summary of Dissertation<br />
Satellite radar altimetry is able to provide globally homogeneous and repeated sea<br />
surface information with very high accuracy and resolution. However, the accuracy<br />
of the satellite altimeter-based SSH degrades in coastal regions due to altimeter<br />
waveform contamination, inaccurate geophysical corrections, and enhanced ocean<br />
variability. Because of lower precision of the data in coastal areas, improvement of<br />
the quality of altimeter data is an important issue before using them for geodetic and<br />
oceanographic applications. When considering only the improvement of the altimeter<br />
data themselves, the approach used is to retrack waveforms to estimate the range<br />
correction, and then to derive the SSH in coastal regions. This approach has shown<br />
the altimeter’s potential in coastal regions. However, previous research on retracking<br />
procedures reprocess the coastal waveform using either merely an existing retracker,<br />
which is used to retrack data over non-ocean surfaces, or a non-automatic method<br />
that is hard to retrack a wide range of altimeter data near land. Compared with<br />
previous work, this is the first time that this research has carried out the waveform<br />
contamination in such a detailed way in a wide range of coastal regions.<br />
In an effort to precisely recover the altimeter-derived SSH data to the greatest extent<br />
close to the coastline, a dedicated coastal waveform retracking system has been<br />
proposed in this <strong>thesis</strong>. This system is developed based upon a systematic analysis of<br />
altimeter radar waveforms in Australian coastal regions. It has the advantage of<br />
containing several retrackers required to reprocess the waveform data automatically<br />
depending on the shapes of the waveform, while offering the precise SSH data sets<br />
derived from corrected range measurements that are estimated through adequately<br />
modelling the altimeter waveform at each along-track location in coastal regions.<br />
A coastal retracking system has been designed and developed (Chapter 5). Central to<br />
the system is the use of two retracking techniques to derive accurate geophysical<br />
parameters, which include both the iterative least squares fitting algorithm and the<br />
threshold retracking algorithm. Retrackers adopted in the system are the ocean fitting<br />
model, five-parameter fitting models, nine-parameter fitting models, and the varying
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 227<br />
threshold levels, in which five- and nine-parameter models are applied to the<br />
waveforms with both linear and exponential trailing edges. In order to obtain<br />
optimum and consistent estimates of the range correction, detailed investigations<br />
have been performed to test and select appropriate retrackers. A waveform<br />
classification procedure is also developed, which enables the waveform to be sorted<br />
and retracked by an appropriate retracker. To dear with the fading ‘noise’ in the<br />
waveforms, a weight iterative scheme has been developed.<br />
Data used are 20 Hz waveforms from two cycles of ERS-2 altimeter mission, five<br />
cycles of POSEIDON mission, and nearly 2.5 years of ERS-1 multi-missions. The<br />
AUSGeoid98 geoid model and ship-track gravity anomalies in the Taiwan Strait<br />
have been used as (partly) independent ground truth. In conjunction with the<br />
shoreline model, an Australian bathymetric grid, and a water/land grid, these data<br />
sets have been used in this research for the purpose of developing and implementing<br />
this system.<br />
The validation of the retracking algorithms has been tested using these data sources<br />
in both the Australian coast and the Taiwan Strait. Results from a number of<br />
comparisons and tests will be specifically concluded in Section 8.2. Generally<br />
speaking, regional investigations around the Australian coast have demonstrated that<br />
a waveform retracking system can effectively remove errors caused by waveform<br />
contamination in the area due to the presence of land, coastal sea states (e.g. inland<br />
or still water), and even the altimeter-operating feature. This study indicates that the<br />
altimeter-derived SSH data along a ground track, as supplied in GDR products, can<br />
be extended 5-15 km kilometres shoreward by the use of a coastal retracking system<br />
developed in this study.<br />
To investigate how retracked SSH data can improve the accuracy of altimeterderived<br />
coastal marine gravity fields in coastal regions, an application has been<br />
conducted using the ERS-1 20 Hz waveform data (
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 228<br />
averaging geoid gradients has been developed for the data used due to lack of the<br />
reference orbit information. Two altimeter-based gravity anomaly grids are created<br />
using the altimeter data before and after retracking. Results are then compared with<br />
the ship-track gravity data. Compared with the result of ±13.9 mgal before retracking,<br />
an improved accuracy of ±9.9 mgal is obtained after retracking. In addition, it has<br />
been found that the SSB correction is inaccurate in coastal regions because of the<br />
coastal waveform contamination.<br />
8. 2 Specific Conclusions<br />
This research has investigated the use of a coastal retracking procedure to reprocess<br />
satellite radar altimeter waveform data for extracting improved SSH data sets in<br />
coastal regions. The investigations conducted involve a quantitative estimation of the<br />
waveform-contaminated distance, the development and calibration of the coastal<br />
waveform retracking system, a comprehensive analysis of waveform data, the<br />
implementation of the retracking system, and applications of the system in Australian<br />
coastal regions and the Taiwan Strait. The key conclusions of the research are made<br />
below.<br />
8.2.1 The Quantitative Estimation of the Contaminated Distance<br />
Five cycles of POSEIDON and one cycle of ERS-2 20 Hz waveform data offshore<br />
0-350 km have been used to quantify a broad, contaminated boundary around the<br />
Australian coast (Chapter 4). This contamination is assumed to have been caused by<br />
backscatter from the land and the more variable coastal sea surface states, coupled<br />
with the footprint size of the radar altimeters, the direction that the satellite ground<br />
track crosses the coastline, and the operating feature of the radar altimeter. In<br />
addition, a preliminary experiment of land effects on the waveform has been<br />
performed using waveforms, the Australian 9"×9" DEM and the AUSGeoid98 2'×2'<br />
geoid model.<br />
Using the mean waveforms and standard deviations of the mean in different distance<br />
bands (Section 4.5.2), a typical contaminated boundary can be ascertained from the<br />
Australian shoreline to ~8 km and ~10 km for POSEIDON and ERS-2, respectively.<br />
Beyond 8 km or 10 km (to 350 km), the mean waveform shapes for both altimeters
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 229<br />
match the mean waveform shape observed over ocean surfaces, though they do not<br />
agree exactly with the theoretical waveform due to the fading noise and the wraparound<br />
error.<br />
Using the 50% threshold retracking point as the estimate of the pre-design tracking<br />
gate (Sections 4.5.2 and 4.5.3), three main conclusions can be drawn. First, the<br />
ERS-2 altimetry shows differences in contamination as a function of the sub-area<br />
along the Australian shoreline. The maximum contaminated distances vary from<br />
~8 km to ~22 km, depending upon locations and features of the shoreline. Second,<br />
both POSEIDON and ERS-2 altimeters show differences in contamination as a<br />
function of the distance from the shoreline, while ocean depths do not show an<br />
important correlation with the contaminated distance (Section 4.5.1). Finally, the<br />
contaminated distance for POSEIDON is less than that for ERS-2. A mean<br />
contaminated distance of ~8 km for POSEIDON and ~10 km for ERS-2 can also be<br />
observed using the 50% of threshold retracking points in the whole Australian<br />
coastal region.<br />
The ERS-2 altimeter also shows distance differences (Section 4.6.4) in waveform<br />
contamination depending on different manners of the ground track crossing the<br />
shoreline (i.e., approaching or leaving the land). In general, the waveforms along the<br />
altimeter ground track leaving land to water suffer longer contaminated distance than<br />
those along the track approaching land from water. The contaminated distance for a<br />
land-approaching track is ~7 km along the ground track for ERS-2. On the other<br />
hand for a satellite ground track leaving land to water, the along-track contaminated<br />
distance can be as long as ~21 km from the coastline. In both cases, the ocean<br />
surface sometimes can be no longer recoverable from contaminated waveforms<br />
obscured significantly by the land return, when the ground track is very closer to the<br />
coastline, say ~0-2.5 km.<br />
The reasons that cause the distance differences of waveform contamination can be<br />
the specific reflected topography (both land and sea) in coastal regions. From a<br />
ERS-2 waveform shifting test (Section 4.6.3), the result shows that altimeter’s<br />
operating feature should be one of reasons that cause the longer distance of distorted<br />
waveforms along-track leaving land to water. This type of distorted waveform
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 230<br />
usually shows usually the shift forward and backward with respect to the expected<br />
tracking gate in the range window, but do not significantly change the magnitude of<br />
the waveform’s amplitude. Land returns are not obviously visible in these<br />
contaminated waveforms, because the distance between the observing location and<br />
the land is usually beyond the satellite footprint. Using TOPEX waveform data,<br />
Brooks et al. (1997) find also the same problem that longer contaminated distances<br />
of the waveform occur upon the descending tracks. Thus, the result of the shifting<br />
test in this research can apply to not only the ERS-2 altimeter, but also TOPEX.<br />
Contamination leads to errors in the on-board calculation of the range measurements<br />
so that the accuracy of altimeter range measurements in such areas is lower and thus<br />
presents the problem for extending the use of altimeter data to the shoreline. It is<br />
recommended from the quantitative contaminated distance in this study that the SSH<br />
data within the distance provided by the GDR products are not reliable and must be<br />
detected prior to their being included in geodetic and oceanographic solutions in<br />
coastal regions. It is also suggested that waveform data reprocessing is essential to<br />
improve altimeter-derived SSH data in these areas. For most GDRs users without<br />
waveform products, it is recommended that altimeter SSH or range data should be<br />
interpreted with some caution for distances less than say ~22 km from a coastline,<br />
and discarded altogether for distances less than 5 km.<br />
8.2.2 Practical Implementation of the Coastal Waveform Retracking System<br />
Regionally comprehensive investigations of waveform characteristics and validation<br />
of the coastal retracking system around the Australian coast have demonstrated that<br />
diverse waveform shapes exist, and the retracking of 20 Hz waveform data can<br />
effectively remove most errors caused by waveform contamination in these areas.<br />
Two cycles of ERS-2 20 Hz waveform data sets offshore 30 km around Australian<br />
coastal regions have been used for implementation of the coastal retracking system<br />
developed in this study. The AUSGeoid98 model was again used as a ground truth<br />
for the purpose of comparison. In an effort to improve the altimeter-derived SSH<br />
results, the following conclusions can be made:<br />
(1) Results of waveform classification show that most coastal waveforms (80.19%)<br />
have ocean-like shapes with a single ramp but without significant peaks. Of the
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 231<br />
remaining ~20% of waveforms, predominant waveforms have single prepeaked<br />
(4.65%) and high-peaked (4.35%) waveform shapes. Others, such as<br />
single mid-peaked and post-peaked, double pre-peaked and post-peaked, and<br />
multi-peaked waveforms, take the percentages of 0.73% - 1.35%. Although<br />
there is only a small percentage of non-ocean-like waveforms, the statistical<br />
results show that these waveforms are much closer to the coastline (1.73-3.12<br />
km in average), indicating that a retracking system containing several<br />
retrackers is necessary.<br />
(2) Retracking results in Australian coastal regions show that 77.7% - 96.2% of<br />
waveforms can be retracked using a least squares iterative fitting algorithm,<br />
while 3.8% - 18.1% of waveforms have to be retracked by the threshold<br />
retracking algorithm. The waveforms show differences in percentages of using<br />
different retrackers as a function of the sub-area along ground tracks. Of these<br />
waveforms retracked by the fitting algorithm, over 98.3% of waveforms are<br />
fitted using the ocean model and the rest of waveforms are fitted by the fiveparameter<br />
model with an exponential trailing edge and the nine-parameter<br />
model with either a linear trailing edge or an exponential trailing edge. It has<br />
been found that the use of the threshold retracker decreases with increasing<br />
offshore distance from 35% in an offshore distance of 0-5 km to 4% in a<br />
offshore distance of 5-10 km. After 10 km offshore, only 2% of waveforms are<br />
threshold retracked, which agrees well with the estimation of the contaminated<br />
distance in Chapter 4 and indicates that the method developed to categorise<br />
coastal waveforms is effective.<br />
(3) Results from single track analysis indicate that the coastal retracking system<br />
has a capability of dealing with diverse waveforms in coastal regions. This<br />
advantage means that the system derives more accurate SSH data than a single<br />
retracker. By comparing altimeter-derived along-track SSH data before and<br />
after retracking with AUSGeoid98 geoid heights, results show that the<br />
altimeter-derived SSH profiles can be extended ~5-10 km shoreward by the use<br />
of coastal retracking system and range corrections. However, SSH profiles at a<br />
distance of 0-5 km to the coastline cannot be recovered by waveform<br />
retracking procedure at the same precision level as SSH data after 5 km from
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 232<br />
the coastline. This is most probably due to predominant land returns in the<br />
waveform range window and the incorrect geophysical corrections in these<br />
near land areas.<br />
In summary, the effect of a coastal waveform retracking system, which contains a set<br />
of methods, computing programs and retrackers for reprocessing waveform data, has<br />
been demonstrated by the improvement observed in Australian coastal regions. When<br />
using two cycles of ERS-2 20 Hz waveform data and the AUSGeoid98 geoid grid as<br />
an independent ground reference, results after retracking provide a typical<br />
improvement of ~5 km on the offshore distance. Compared with the contaminated<br />
distance of ~10 km before retracking, the use of the retracking system is able to<br />
reduce this contaminated distance to ~5 km.<br />
8.2.3 Development and Validation of a Coastal Retracking System<br />
The development and validation of the coastal retracking system includes design and<br />
investigation of algorithms, selection of the threshold levels, estimation of biases<br />
between different retrackers and the SSH data before and after retracking, as well as<br />
the tests on the effectiveness of the system.<br />
(d) Retracker Investigation<br />
An ocean return model has been deduced and is a modified convolutional solution of<br />
the radar returns based upon the Brown (1977) model. This ocean model without the<br />
non-linear ocean wave parameter has been used as a main retracker in this research<br />
due to its clearly physical description of the ocean surface. By quantitative<br />
comparison to the five-parameter model, it has found that the slope of the trailing<br />
edge modelled by the five-parameter model has a larger span of variation than the<br />
slope modelled by the ocean model. This advantage makes, in turn, the fiveparameter<br />
model hold a greater capability of fitting non-ocean-like or irregular<br />
waveforms in coastal regions than the ocean model. Thus, the five-parameter model<br />
replaces the ocean model for these irregular shapes of waveforms in the<br />
implementation of the system.<br />
Selection of the threshold level has been performed by comparing an available<br />
ground truth of the AUSGeoid98 geoid heights with the SSH data before and after
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 233<br />
threshold retracking, in ten 15°×15° sub-areas around the Australian coastal regions<br />
and six 5 km wide distance bands. Results confirm that the 50% threshold level is the<br />
best level for open-ocean waveform, but not an adequate level for the contaminated<br />
ocean waveforms in coastal regions. Instead, a varying threshold level, which is the<br />
30% threshold level for contaminated waveforms and 50% threshold level for<br />
uncontaminated waveforms, can give a good agreement between the retracked SSH<br />
data and the geoid heights. This result of the threshold level broadly agrees with that<br />
used by Brooks et al. (1997).<br />
(b) Bias Evaluation<br />
Biases exist both between retracked SSH data from different retracking algorithms<br />
and SSH data before and after retracking. Biases have been analysed and estimated<br />
by retracking ocean waveforms in Section 5.8 using two cycles of ERS-2 waveforms<br />
and can be concluded as follows, which suggest that biases must be estimated before<br />
retracking and applied to the data to obtain consistent results after retracking.<br />
(1) Bias between the SSH data retracked by the ocean and five-parameter fitting<br />
algorithms is −0.85±1.35 cm, which is not statistically significant.<br />
(2) Bias between the SSH data retracked by the 50% threshold retracker and the<br />
fitting algorithms (both the ocean and five-parameter models) is<br />
+56.35±5.84 cm.<br />
(3) Biases between altimeter-derived SSH data before and after retracking are<br />
+27.58±8.78 cm for fitting algorithms and −30±3.58 cm for the 50% threshold<br />
retracking algorithm.<br />
It should be pointed out that the differences between the standard deviations are<br />
partly due to the different numbers of the waveform data used for the comparison.<br />
This means that it is not reliable to determine which retracker is better though a<br />
direct comparison of the STDs. Thus, an independently external ocean surface<br />
reference is necessary to estimate absolute biases precisely. However, as any<br />
gravimetric determination of the geoid is deficient in the zero- and first-degree terms<br />
(Heiskanen and Moritz, 1967; Featherstone et al., 1996), the geoid heights<br />
interpolated from AUSGeoid98 are not taken as a reference to ascertain the absolute<br />
bias between the SSH before and after retracking.
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 234<br />
For the biases existing between the SSH data before and after retracking, however,<br />
there is alternative approach that can be adopted to remove them. Because the bias is<br />
caused by the instrument and the on-board tracking algorithm (Hayne and Hancock<br />
III, 1990; Chelton, 2001; Hayne, 2002), it can be corrected from the instrument<br />
correction. The correction can be computed by a ground data processing system or<br />
by waveform retracking. The recent research of Dong et al. (2002) show that the<br />
absolute bias can also be determined using altimeter data and GPS levelling at tide<br />
gauges.<br />
(c) Algorithm Validation<br />
To test how effective the retracker is, retracking using only a single retracker has<br />
been conducted in two distinct coastal areas of Australia and the Taiwan Strait. The<br />
AUSGeoid98 geoid heights and the ship track gravity anomalies are used as external<br />
input data sets of the ground truth.<br />
When comparing the geoid heights with the SSH data in Australian coastal regions<br />
(Section 5.8.2), the emphasis is on the shapes of the height profiles and the changes<br />
of the along-track slope. Results from single track comparison show that the SSH<br />
gradients in the vicinity of the land before retracking are large varying from<br />
~363.8-975.6 mm/km over offshore along-track distances of 0-19.5 km, being much<br />
larger than the maximum geoid gradient value of ~150 mm/km in Australia (cf.,<br />
Johnston and Featherstone, 1998; Friedlieb et al., 1997; Featherstone et al., 2001).<br />
After retracking, the SSH gradients decrease by ~16.0 mm/km over offshore alongtrack<br />
distance of 4.7 km − 19.5 km, agreeing well with the geoid gradient<br />
(~16.4 mm/km) at the same distance. Results indicate strongly that retracking, no<br />
matter which retracker is used, can improve the altimeter-derived SSH in coastal<br />
regions.<br />
Using ~2.5 years of ERS-1 SSH data sets before and after retracking, gravity<br />
anomaly predictions in the Taiwan Strait are made from the inverse Vening Meinesz<br />
formula (Hwang, 1998). Retrackers used are the fitting algorithm including the ocean<br />
and five-parameter models. Results are then compared with limited ship-track gravity<br />
data. The comparison of results demonstrated that waveform retracking has improved<br />
the quality of the SSH in coastal regions near the China Sea and Taiwan Strait, thus
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 235<br />
improving the accuracy of the altimeter-derived gravity field. Improved residual<br />
geoid gradients and later on gridded gravity anomaly estimation are obtained in this<br />
coastal region when the SSHs after retracking are used instead of unretracked SSH<br />
data. The STD of agreements between the predicted and ship-track gravity anomalies<br />
are ±13.9 mgal before retracking and ±9.9 mgal after retracking. This shows a great<br />
potential of providing accurate gravity field information in coastal regions by using<br />
satellite altimeter SSH data after waveform retracking.<br />
8.2.4 Other Conclusions<br />
A preliminary test has been performed to calculate the retracking gate estimate<br />
related to the land returns in the range window using external independent data<br />
source of the DEM model in coastal regions. Analysing results show that nonuniqueness<br />
of the estimate over land near coastal regions is still a main problem that<br />
should be solved in the future.<br />
The analysis on extracting the SSHs from altimeter waveform data suggests that the<br />
SSB correction is inaccurate when it is computed using the empirical model in the<br />
coastal region due mainly to the incorrect SWH estimated from the on-satellite<br />
processor. Waveform retracking can remove the need for the ocean wave influence<br />
and thus reduce the SSB error, while an error is caused to the unretracked SSH data<br />
by the need of this correction no matter whether or not the SSB is applied. In<br />
addition, the method of averaging residual geoid gradients developed for the AWF<br />
data product in this study has been demonstrated being effective, because a similar<br />
accuracy level (~±14 mgal) as Hwang et al. (2001) is obtained using unretracked<br />
altimeter data.<br />
8. 3 Recommendations for Future Work<br />
The improvement of the altimeter SSH data in Australian waters and the Taiwan<br />
Strait has demonstrated that the coastal waveform retracking system developed in<br />
this research has the potential of being applied to other coastal regions, as well as<br />
other satellite altimeter waveform data sets. With higher precision of the altimeter<br />
measurements over open oceans, demand for improving the precision of the data in<br />
coastal regions will become increasingly desirable to realise the global coverage of
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 236<br />
the altimetric missions. Therefore, recommendations and further work will be given<br />
in this Section.<br />
8.3.1 Extraction of Ocean Returns from the Mixed Waveform Shape<br />
Current retracking algorithms including those in the coastal system cannot recover<br />
the ocean returns that appear in the late gates behind the land returns in the range<br />
window when the ground track is very much closer to the coastline (Brooks et al.,<br />
1997). However, if the land return in the waveform can be estimated and removed, it<br />
should be possible to recover the ocean return. As stated, this can be achieved only<br />
when other external and high-resolution reflected surface topography are available.<br />
With new satellite altimetry missions, improved DEM created from satelliteobserved<br />
data, and detailed topography information, developing new waveform<br />
retrackers for this type of waveform will become an area requiring further research<br />
effort.<br />
8.3.2 Improvement of the Correction Algorithms near Land<br />
This research has demonstrated that waveform contamination can be removed in a<br />
certain offshore distance using waveform retracking techniques. However, waveform<br />
retracking can improve only the range measurements. The incorrect geophysical and<br />
atmospheric corrections still affect the precision of the altimeter-derived SSH data<br />
sets, which are the direct measurements in geodetic applications. Improvement of the<br />
corrections in coastal regions has received attention from several researchers, such as<br />
Shum et al. (1998) and Anzenhofer et al. (2000). To fulfil the satellite altimetric<br />
potential in coastal regions, developing geophysical and environmental correction<br />
algorithms will be another future research field.<br />
As concluded in this research, the SSB correction causes a large error in the SSH<br />
data near the coastline. In general, the SSH data before used for geodetic applications<br />
must apply an instrument correction to remove the effects of the SSB. However, how<br />
can the SSH be improved by such incorrect SSB correction in coastal regions?<br />
Fortunately, retracking procedures also removes the need for the SSB correction.<br />
Therefore, retracking in a wide range of coastal regions will also be one for future<br />
research.
Chapter 8. CONCLUSIONS AND RECOMMENDATIONS 237<br />
8.3.3 Improvement of the Altimeter Operation<br />
From this research, the operation of the altimeter affects the waveform, in particular<br />
waveforms along ground tracks leaving land to water. This result suggests that it is<br />
essential to improve the on-board tracking algorithm based on this operating feature<br />
or the design of the altimeter operation.<br />
8.3.4 Geodetic Application<br />
For further research of determining gravity anomalies in coastal regions, and to<br />
overcome the problem of altimeter data discontinuity very near to the coastline,<br />
combining the uniform coverage of the altimeter data with shipborne gravity and<br />
land gravity data will provide a very accurate high-resolution gravity field in coastal<br />
regions.
238<br />
REFERENCES<br />
Andersen, O. B. and Knudsen, P. (1998) Global Marine Gravity from the ERS-1 and<br />
Geosat Geodetic Mission Altimetry, Journal of Geophysical Research, Vol.<br />
103, pp. 8129-8137.<br />
Andersen, O. B. and Knudsen, P. (2000) The Role of Satellite Altimetry in Gravity<br />
Field Modelling in Coastal Areas, Physics and Chemistry of the Earth (A), Vol.<br />
25, No. 1, pp. 17-24.<br />
Anzenhofer, M., Shum, C. K. and Renstch, M. (2000) Coastal Altimetry and<br />
Applications, The Ohio State <strong>University</strong>, Columbus, USA, 36 pp.<br />
Andersen, O. B., Knudsen, P. and Trimmer, R. (2001) The KMS2001 Global Mean<br />
Sea Surface and Gravity Field, Proceedings of the IAG2001 Scientific<br />
Assembly, Budapest, Hungary, September [CD-ROM].<br />
AVISO/Altimeter (1996) AVISO User Handbook: Merged TOPEX/POSEIDON<br />
Products, AVI-NT-020101-CN, Edition 3.0, Toulouse, France, 196 pp.<br />
AVISO/Altimeter (1998) Preliminary AVISO User Handbook: POSEIDON<br />
Waveform Product, AVI-NT-02-555-CN, Edition 1.0, Toulouse, France, 101 pp.<br />
Bamber, J. L. (1994) A Digital Elevation Model of the Antarctic Ice Sheet Derived<br />
from ERS-1 Altimeter Data and Comparison with Terrestrial Measurements,<br />
Annals of Glaciology, Vol. 20, pp. 48-54.<br />
Bamber, J. L. (1994) Ice Sheet Altimeter Processing Scheme, International Journal<br />
of Remote Sensing, Vol. 15, No. 4, pp. 925-938.<br />
Bamber, J. L., Ekholm, S. and Krabill, W. (2001) A new, high-resolution digital<br />
elevation model of Greenland fully validated with airborne laser altimeter<br />
data, Journal of Geophysical Research - Solid Earth, Vol. 106, No. B4, pp.<br />
6733-6746.<br />
Barrick, D. E. and Lipa, B.J. (1985) Analysis and Interpretation of Altimeter Sea<br />
Echo, Advances in Geophysics, Vol. 27, pp. 61-100.<br />
Berry, P. A. M., Sanders, R. F., Leenmans, C. and Bron, E. (1998) Generating<br />
Orthometric Heights from the ERS-1 Altimeter Geodetic Mission Dataset:<br />
Results From an Expert Systems Approach, in: Geodesy on the Move, Schward,<br />
K. (ed.), Springer Verlag, Berlin, Germany, pp. 113-118.<br />
Berry, P. A. M. (2000a) Topography from Land Radar Altimeter Data: Possibilities<br />
and Restrictions, Physics and Chemistry of the Earth (A), Vol. 25, No. 1, pp.<br />
81-88.
239<br />
Berry, P. A. M., Hoogerboord, J.E. and Pinnock, R.A. (2000b) Identification of<br />
Common Error Signatures in Global Digital Elevation Models Based on<br />
Satellite Altimeter Reference Data, Physics and Chemistry of the Earth (A),<br />
Vol. 25, No. 1, pp. 95-99.<br />
Berry, P. A. M. (2001) Personal Communication, Faculty of Computing and<br />
Engineering, De Montfort <strong>University</strong>, U.K.<br />
Berry, P. A. M. (2002) Personal Communication, Faculty of Computing and<br />
Engineering, De Montfort <strong>University</strong>, U.K.<br />
Bracewell, R. N. (2000) The Fourier Transform and Its Applications (Third Edition),<br />
McGraw Hill, Boston, USA, 597 pp.<br />
Brenner, A. C., Bindschadler, R. A., Thomas, R. H. and Zwally, H. J. (1983) Slope-<br />
Induced Errors in Radar Altimetry over Continental Ice Sheets, Journal of<br />
Geophysical Research, Vol. 88, No. C3, pp. 1617-1623.<br />
Brenner, A. C., Frey, H. V. and Zwally, H. J. (1990) Comparisons between GEOSAT<br />
and SEASAT Tracking over Non-Ocean Surfaces, Geophysical Research<br />
Letters, Vol. 17, No. 10, pp. 1537-1540.<br />
Brenner, A. C., Koblinsky, C. J. and Zwally, H. J. (1993) Postprocessing of Satellite<br />
Return Signals for Improved Sea Surface Topography Accuracy, Journal of<br />
Geophysical Research, Vol. 98, No. C1, pp. 933-944.<br />
Brenner, A., Frey, H., Dimarzio, J., and Tsaoussi, L. (1997) Topography over South<br />
America from ERS, http://earth.esa.int/symposia/papers/frey/ [last accessed: 8<br />
March 2004].<br />
Brooks, R. L., Campbell, W. J., Ramseier, R. O., Stanley, H. R. and Zwally, H. J.<br />
(1978) Ice Sheet Topography by Satellite Altimetry, Nature, Vol. 274, No.<br />
5671, pp. 539-543.<br />
Brooks, R. L. and Lockwood, D. W. (1990) Effects of Islands in the Geosat Footprint,<br />
Journal of Geophysical Research, Vol. 95, No. C3, pp. 2849-2855.<br />
Brooks, R. L., Lockwood, D. W. and Lee, J. E. (1997) Land Effects on TOPEX<br />
Radar Altimeter Measurements on Pacific Rim Coastal Zones,<br />
http://topex.wff.nasa.gov/docs.html [last accessed: 8 March 2004].<br />
Brooks, R. L. (2002) Personal Communication, Computer Sciences Corporation,<br />
Wallops Island, VA 23337, USA.<br />
Brown, G. S. (1977) The Average Impulse Response of a Rough Surface and Its<br />
Applications, IEEE Transactions on Antennas and Propagation, Vol. AP-25,<br />
No. 1, pp. 67-74.<br />
Buchanan, C. (1991) Bathymetric 30 Arc Second Grid of the Australian Region,<br />
Geoscience Australia, Canberra, Australia [CD-ROM].
240<br />
Challenor, P. G. and Srokosz, M. A. (1989) The Extraction of Geophysical<br />
Parameters from Radar Altimeter Return from a Nonlinear Sea Surface, in:<br />
Mathematics in Remote Sensing, Brooks, S. R. (ed.), Oxford <strong>University</strong> Press,<br />
Oxford, UK, pp. 254-268.<br />
Chelton, D. B., Ries, J.C., Haines, B. J., Fu, L.-L. and Callahan, P.S. (2001) Satellite<br />
Altimetry, in: Satellite Altimetry and Earth Sciences: A Handbook of<br />
Techniques and Applications, Fu, L.-L. and Cazenave, A. (eds.), Academic<br />
Press, San Diego, USA, pp. 1-132.<br />
Cheney, R. E. and Marsh, J. G. (1983) Global Mesoscale Variability from Collinear<br />
Tracks of SEASAT Altimeter Data, Journal of Geophysical Research, Vol. 88,<br />
pp. 4343-4354.<br />
Cooper, A. P. R. (1989) Slope Correction by Relocation for Satellite Radar Altimetry,<br />
International Geoscience and Remote Sensing Symposium, Vancouver, Canada,<br />
July, pp. 2730-2733.<br />
Cudlip, W. and Milnes, M. (1994) Overview of Altimeter Data Processing at the<br />
EODC, International Journal of Remote Sensing, Vol. 15, No. 4, pp. 871-887.<br />
Davis, C. H. (1993a) A Surface and Volume Scattering Retracking Algorithm for Ice<br />
Sheet Satellite Altimetry, IEEE Transactions on Geoscience and Remote<br />
Sensing, Vol. 31, No. 4, pp. 811-818.<br />
Davis, C. H. and Moore, R. K. (1993b) A Combined Surface- and Volume-Scattering<br />
Model for Ice-Sheet Radar Altimetry, Journal of Glaciology, Vol. 39, No. 133,<br />
pp. 675-686.<br />
Davis, C. H. (1995) Growth of the Greenland Ice Sheet: A Performance Assessment<br />
of Altimeter Retracking Algorithms, IEEE Transactions on Geoscience and<br />
Remote Sensing, Vol. 33, No. 5, pp. 1108-1116.<br />
Davis, C. H. (1997) A Robust Threshold Retracking Algorithm for Measuring Ice-<br />
Sheet Surface Elevation Change from Satellite Radar Altimeter, IEEE<br />
Transactions on Geoscience and Remote Sensing, Vol. 35, No. 4, pp. 974-979.<br />
Deng, X., Chao, D. B., Li, J. C. and Chen, J. Y. (1996) A Preliminary Process of<br />
TOPEX/POSEIDON Data in the China Sea and Neighbour Sea, ACTA<br />
Geodaetica et Cartographica Sinica, Vol. 3, pp. 226-321.<br />
Deng, X., Chao, D.B. and Li, J. C. (1997) A Preliminary Estimate of the Mean Sea<br />
Level Variations in the China Sea from TOPEX/POSEIDON Data,<br />
Proceedings of the IAG International Conference of Crustal Movement and<br />
Hazard Reduction in East Asia and South-East Asia, Wuhan, China, November,<br />
pp. 509-516.<br />
Deng, X., Featherstone, W. E., Hwang, C. and Berry, P. A. M. (2001) Retracking<br />
ERS-2 Radar Altimeter Waveform Data in Australia's Coastal Zone,
241<br />
Proceedings of the IAG 2001 Scientific Assembly, Budapest, Hungary,<br />
September [CD-ROM].<br />
Deng, X., Featherstone, W. E., Hwang, C. and Berry, P .A. M. (2002) Estimation of<br />
Contamination of ERS-2 and POSEIDON Satellite Radar Altimetry Close to<br />
the Coasts of Australia, Marine Geodesy, Vol. 25, No. 4, pp. 249-271.<br />
Dong, X., Moore, P. and Bingley, R. (2002) Absolute Calibration of the<br />
TOPEX/POSEIDON Altimeter Using UK Tide Gauges, GPS, and Precise,<br />
Local Geoid-Differences, Marine Geodesy, Vol. 25, pp. 189-204.<br />
Dunbar, P. (2000) DS759.2 – Terrain Base, Global 5-minute Ocean Depth and Land<br />
Elevation, from NGDC, http://dss.ucar.edu/datasets/ds759.2/ [last accessed: 8<br />
March 2004].<br />
Eanes, R. (1999) Improved Ocean Tide Model from Satellite Altimetry, Fall Meeting<br />
1999, American Geophysical Union, San Francisco, USA.<br />
Fairhead, J. D., Green, C.M. and Odegard, M. E. (2001) Satellite-Derived Gravity<br />
Having an Impact on Marine Exploration, The Leading Edge. Vol. 20, pp. 873-<br />
876.<br />
Featherstone, W. E., Alexander, K. and Sideris, M. G. (1996) Gravimetric Geoid<br />
Refinement Using High Resolution Gravity and Terrain Data, Geomatics<br />
Research Australasia, No. 64, pp. 75-99.<br />
Featherstone, W. E., Kirby, J. F., Kearsley, A. H. W., Gilliland, J. R., Johnston, G.<br />
M., Steed, J., Forsberg, R. and Sideris, M. G. (2001) The AUSGeoid98 Geoid<br />
Model of Australia: Data Treatment, Computations and Comparisons with<br />
GPS-Levelling Data, Journal of Geodesy, Vol. 75, pp. 313-330.<br />
Featherstone, W. E. (2003) Comparison of Different Satellite Altimeter-Derived<br />
Gravity Anomaly Grid with Ship-Borne Gravity Data around Australia, in:<br />
Gravity and Geoid 2002, Tziavos, I.N. (ed.), Department of Surveying and<br />
Geodesy, Aristotle <strong>University</strong> of Thessaloniki, pp. 326-331.<br />
Fedor, L. S., Godbey, T. W., Gower, J .F. R., Guptill, R., Hayne, G. S., Rufenace, C.<br />
L. and Walsh, E. J. (1979) Satellite Altimeter Measurements of Sea States - An<br />
Algorithm Comparison, Journal of Geophysical Research, Vol. 84, No. B8, pp.<br />
3991-4001.<br />
Femenias, P., Remy, F., Raizonville, R. and Minster, J. F. (1993) Analysis of<br />
Satellite-Altimeter Height Measurements above Continental Ice Sheets,<br />
Journal of Glaciology, Vol. 39, No. 133, pp. 591-600.<br />
Fernandes, M. J., Bastos, L. and Antunes, M. (2003) Coastal Satellite Altimetry -<br />
Methods for Data Recovery and Validation, in: Gravity and Geoid 2002,<br />
Tziavos, I.N. (ed.), Department of Surveying and Geodesy, Aristotle<br />
<strong>University</strong> of Thessaloniki, pp. 302-307.
Frey, H. and Brenner, A. C. (1990) Australian Topography From Seasat Overland<br />
Altimetry, Geophysical Research Letters, Vol. 17, No. 10, pp. 1533-1536.<br />
242<br />
Friedlieb, O. J., Featherstone, W. E. and Dentith, M. C. (1997) A WGS84-AHD<br />
Profile over the Darling Fault, Western Australia, Geomatics Research<br />
Australasia, Vol. 67, pp. 17-32.<br />
Guzkowska, M. A. J., Rapley, C. G., Cudlip, W. and Mason, I. M. (1988) Satellite<br />
Radar Altimetry over Arid Regions, Proceedings of IGARSS'88 Symposium,<br />
Edinburgh, Scotland, September, pp. 659- 662.<br />
Hayne, G. S. (1980) Radar Altimeter Mean Return Waveform from Near-Normal-<br />
Incidence Ocean Surface Scattering, IEEE Transactions on Antennas and<br />
Propagation, Vol. AP-28, No. 5, pp. 687-692.<br />
Hayne, G. S. and Hancock III, D. W. (1982) Sea-State Related Altitude Errors in the<br />
Seasat Altimeter, Journal of Geophysical Research, Vol. 87, No. C5, pp. 3227-<br />
3231.<br />
Hayne, G. S. and Hancock III, D. W. (1990) Corrections for the Effects of<br />
Significant Wave Height and Attitude on Geosat Radar Altimeter<br />
Measurements, Journal of Geophysical Research, Vol. 95, No. C3, pp. 2837-<br />
2842.<br />
Hayne, G. S., Hancock III, D. W. and Purdy, C. L. (1994) The Corrections for<br />
Significant Wave Height and Attitude Effects in the TOPEX Radar Altimeter,<br />
Journal of Geophysical Research, Vol. 99, No. C12, pp. 24941-24955.<br />
Hayne, G. S. (2002) Personal Communication, NASA Goddard Space Flight Centre,<br />
Wallops Island, Virginia, USA.<br />
Heiskanen, W. A. and Moritz, H. (1980) Physical Geodesy, W.H. Freeman, San<br />
Francisco, USA, 364 pp.<br />
Hilton, R. D., Featherstone, W. E., Berry, P. A. M., Johnston, C.P.D. and Kirby, J.F.<br />
(2003) Comparison of Digital Elevation Models (DEMs) over Australia and<br />
External Validation Using ERS-1 Satellite Radar Altimetry, Australian<br />
Journal of Earth Sciences, Vol. 50, No. 2, pp. 157-168.<br />
Hsu, S., Liu, C., Shyu, C. Liu, S., Sibue, J., Lallemand, S., Wang, C. and Reed D.<br />
(1998) New Gravity and Magnetic Anomaly Maps in the Taiwan-Luzon<br />
Region and Their Preliminary Interpretation, TAO, Vol. 9, No. 3, pp. 509-532.<br />
Hutchinson, M., Stein, J. and Stein, J. (2001) Upgrade of the 9 Second Australian<br />
Digital Elevation Model, Centre for Resource and Environmental Studies,<br />
Australian National <strong>University</strong>, Canberra, http://cres.anu.edu.au/dem/ [last<br />
accessed: 8 March 2004].<br />
Hwang, C. and Parsons, B. (1995) Gravity Anomalies Derived from Seasat, Geosat,<br />
ERS-1 and TOPEX/POSEIDON Altimetry and Ship Gravity: A Case Study
243<br />
over the Reykjanes Ridge, Geophysical Journal International, Vol. 122, pp.<br />
551-568.<br />
Hwang, C. (1998a) Inverse Vening Meinesz Formula and Deflection-Geoid Formula:<br />
Applications to the Predictions of Gravity and Geoid over the South China Sea,<br />
Journal of Geodesy, Vol. 72, pp. 304-312.<br />
Hwang, C., Kao, E. and Parsons, B. (1998b) Global Derivation of Marine Gravity<br />
Anomalies from Seasat, Geosat, ERS-1 and TOPEX/POSEIDON Altimeter<br />
Data, Geophysical Journal International, Vol. 134, pp. 449-459.<br />
Hwang, C. and Wang, C. (2002a) New Free-Air Gravity Anomaly Grid of Taiwan,<br />
Journal of Surveying Engineering, Vol. 44, No. 2, pp. 1-22 (in Chinese).<br />
Hwang, C., Hsu, H. and Jang, R. (2002b) Global Mean Sea Surface and Marine<br />
Gravity Anomaly from Multi-Satellite Altimetry: Applications of Deflection-<br />
Geoid and Inverse Vening Meinesz Formulae, Journal of Geodesy, Vol. 76, pp.<br />
407-418.<br />
Jeffrey, A. (1995) Handbook of Mathematical Formulas and Integrals, Academic<br />
Press, New York, USA, 410 pp.<br />
Jensen, J. R. (1999) Radar Altimeter Gate Tracking: Theory and Extension, IEEE<br />
Transactions on Geoscience and Remote Sensing, Vol. 37, No. 2, pp. 651-658.<br />
Johnston, G. M., and Featherstone, W. E. (1998) AUSGeoid98: A New Gravimetric<br />
Geoid for Australia, Australia's National Mapping Agency, Canberra, Australia,<br />
http://www.ga.gov.au/nmd/geodesy/ausgeoid/docs/iemsgary.pdf [last accessed:<br />
8 March 2004].<br />
JPL (1993a) TOPEX/POSEIDON Project, TOPEX Ground System Software<br />
Interface Specification, Volume 2: Design (SIS-2), Altimeter Sensor Data<br />
Record (SDR)-Alt SDR Data, 633-751-23-001, Rev.C, Jet Propulsion<br />
Laboratory, California Institute of Technology, Pasedena, USA, 77 pp.<br />
JPL (1993b) TOPEX/POSEIDON Project, TOPEX Ground System, Software<br />
Interface Specifications Volume 2: Design (SIS-2), Altimeter Sensor Data<br />
Record (SDR)-Alt SDR, 633-752-23-021, Rev.B, Jpl Propulsion Laboratory,<br />
California Institute of Technology, Pasedena, USA, 38 pp.<br />
Kelleher, G., Bleakley, G., Walls, K. and Dingwall, P. (2002) Marine Region 18:<br />
Australia/New Zealand,<br />
http://www.ea.gov.au/coasts/mpa/nrsmpa/global/volume4/pubs/v4chap18.pdf<br />
[last accessed: 8 March 2004].<br />
Kirby, J. F. (1996) The Development and Application of a New Algorithm for Ocean<br />
Geoid Recovery, PhD Thesis, <strong>University</strong> of Edinburgh, UK, 156 pp.<br />
Kirby, J. F. and Forsberg, R. (1997) A Comparison of Techniques for the Integration<br />
of Satellite Altimeter and Surface Gravity Data for Geoid Determination, in:
244<br />
Geodesy on the Move, Forsberg, R., Feissel, M. and Dietrich, R. (eds.),<br />
Springer-Verlag, Berlin, Germany, pp. 207-212.<br />
Leben, R. R., Born, G. H., Thompson, J. D. and Fox, C. A. (1990) Mean Sea Surface<br />
Variability of the Gulf of Mexico Using Geosat Altimeter Data, Journal of<br />
Geophysical Research, Vol. 95, pp. 3025-3032.<br />
Legresy, B. and Remy, F. (1998) Using the Temporal Variability of Satellite Radar<br />
Altimetric Observations to Map Surface Properties of the Antarctic Ice Sheet,<br />
Journal of Glaciology, Vol. 44, No. 147, pp. 197-206.<br />
Lemoine, F. G., Kenyon, S.C., Factor, J.K., Trimmer, R.G., Pavlis, N.K., Chinn,<br />
D.S., Cox, C.M., Klosko, S.M., Luthcke, S.B., Torrence, M.H., Wang, Y.M.,<br />
Williamson, R.G., Pavlis, E.C., Rapp, R.H., Olson, T.R. (1998) The<br />
Development of the Joint NASA GSFC and the National Imagery and Mapping<br />
Agency (NIMA) Geopotential Model EGM96, NASA/TP-1998-206861, NASA<br />
Technical Paper, USA, 575 pp.<br />
Le Traon, P. Y. and Morrow, R. (2001) Ocean Currents and Eddies, in: Satellite<br />
Altimetry and Earth Sciences: A Handbook of Techniques and Applications, Fu,<br />
L.-L. and Cazenave, A. (eds.), Academic Press, San Diego, pp. 171-216.<br />
Levitus, S., Monterey, G. I. and Boyer, T. (1997) Seasonal Variability of Dynamic<br />
Height and Its Fourier Analysis, NOAA NESDIS Atlas 15, Government Printing<br />
Office, Washington, D.C., USA.<br />
Levitus, S., Monterey, G. I. and Boyer, T. (2002) Seasonal Variability of Dynamic<br />
Height and Its Fourier Analysis, http:/www.nodc.noaa.gov/OC5/dyndoc.html<br />
[last accessed: 8 March 2004].<br />
Li, D. R. (1988) The Error and Reliability Theory, Surveying and Mapping Publisher,<br />
Beijing, China, 331 pp.<br />
Lipa, B. J. and Barrick, D. E. (1981) Ocean Surface Height-Slope Probability<br />
Density Function from SEASAT Altimeter Echo, Journal of Geophysical<br />
Research, Vol. 86, No. C11, pp. 10921-10930.<br />
McAdoo, D. C. (1990) Gravity Field of the Southeast Central Pacific from Geosat<br />
Exact Repeat Mission Data, Journal of Geophysical Research, Vol. 97, No. B3,<br />
pp. 3274-3260.<br />
Mantripp, D. (1996) Radar Altimetry, in: The Determination of Geophysical<br />
Parameters from Space, Fancey, N. E., Gardiner, I.D. and Gardiner, R.A.<br />
(eds.), Institute of Physics Publishing, London, UK, pp. 119-171.<br />
Marth, P. C., Jensen, J. R., Kilgus, C. C., Perschy, J. A., MacArthur, J. L., Hancock,<br />
D. W., Hayne, G. S., Purdy, C. L., Rossi, L. C. and Koblinsky, C. J. (1993)<br />
Prelaunch Performance of the NASA Altimeter for the TOPEX/POSEIDON<br />
Project, IEEE Transactions on Geoscience and Remote Sensing, Vol. 31, No. 2,<br />
pp. 315-332.
245<br />
Martin, T. V., Zwally, H. J., Brenner, A. C. and Bindschadler, R. A. (1983) Analysis<br />
and Retracking of Continental Ice Sheet Radar Altimeter Waveforms, Journal<br />
of Geophysical Research, Vol. 88, No. C3, pp. 1608-1616.<br />
Menard, Y. and Fu, L. L. (2000) Jason-1, on the Tracks of TOPEX/POSEIDON,<br />
AVISO Altimetry Newsletter, Vol. 7, pp. 3-8.<br />
Moritz, H. (1980) Advanced Physical Geodesy, Wichmann, Karlsruhe, 500 pp.<br />
Olgiati, A., Balmino, G., Sarrailh, M. and Green, C. M. (1995) Gravity Anomalies<br />
from Satellite Altimetry: Comparison between Computation via Geoid Heights<br />
and via Deflections of the Vertical, Bulletin Geodesique, Vol. 69, No. 4, pp.<br />
252-260.<br />
Nerem, R. S. (1995) Measuring Global Mean Sea Level Variations Using<br />
TOPEX/POSEIDON Altimeter Data, Journal of Geophysical Research, Vol.<br />
100, No. C12, pp. 25135-25151.<br />
NRSC (1995) Altimeter Waveform Product ALT.WAP Compact User Guide, PF-<br />
UG-NRL-AL-0001, The United Kingdom Processing and Archiving Facility,<br />
UK, 32 pp.<br />
Parsons, C. L. (1979) An Assessment of GEOS-3 Waveheight Measurements, in:<br />
Ocean Wave Climate, Earle, M. D. and Malahoff, A. (eds.), Plenum Press,<br />
New York, USA, pp. 235-251.<br />
Partington, K. C., Ridley, J. K. and Rapley, C. G. (1989) Observations of Surface<br />
Properties of the Ice Sheets by Satellite Radar Altimetry, Journal of Glaciology,<br />
Vol. 35, No. 120, pp. 267-275.<br />
Partington, K. C., Cudlip, W. and Rapley, C. G. (1991) An Assessment of the<br />
Capability of the Satellite Radar Altimeter for Measuring Ice Sheet<br />
Topographic Change, International Journal of Remote Sensing, Vol. 12, No. 3,<br />
pp. 585-609.<br />
Pope, A. J. (1976) The Statistics of Residual and the Detection of Outliers, Technical<br />
Report, TR-NOS-65-NGS-1, National Ocean Survey, Rockville.<br />
Priester, R. W. and Miller, L. S. (1979) Estimation of Significant Wave Height and<br />
Wave Height Density Function Using Satellite Altimeter Data, Journal of<br />
Geophysical Research, Vol. 84, No. B8, pp. 4021-4026.<br />
Quartly, G. D., Srokosz, M. A. and Guymer, T. H. (1998) Understanding the Effects<br />
of Rain on Radar Altimeter Waveforms, Journal of Geophysical Research, Vol.<br />
22, No. 11, pp. 1567-1570.<br />
Quartly, G. D. and Srokosz, M. A. (2001) Analyzing Altimeter Artifacts: Statistical<br />
Properties of Ocean Waveforms, Journal of Atmospheric and Oceanic<br />
Technology, Vol. 18, pp. 2074-2091.
246<br />
Rapp, R. H. and Yi, Y. (1997) Role of Ocean Variability and Dynamic Ocean<br />
Topography in the Recovery of the Mean Sea Surface and Gravity Anomalies<br />
from Satellite Altimeter Data, Journal of Geodesy, Vol. 71, pp. 617-629.<br />
Remy, F. and Minster, J. F. (1993) Precise Altimetric Topography in Ice-sheet Flow<br />
Studies, Annals of Glaciology, Vol. 17, pp. 195-200.<br />
Remy, F., Mazzega, P., Houry, S., Brossier, C. and Minster, J. F. (1989) Mapping of<br />
the Topography of Continental Ice by Inversion of Satellite-Altimeter Data,<br />
Journal of Glaciology, Vol. 35, No. 119, pp. 98-107.<br />
Resource Assessment Commission (1993) Coastal Zone Inquiry: Final Report,<br />
Australian Government Publishing Service, Canberra, Australia, 635 pp.<br />
Ridley, J. K. and Partington, K. C. (1988) A Model of Satellite Radar Altimeter<br />
Return from Ice Sheets, International Journal of Remote Sensing, Vol. 9, No. 4,<br />
pp. 601-624.<br />
Rodriguez, E. (1988) Altimetry for Non-Gaussian Oceans: Height Biases and<br />
Estimation of Parameters, Journal of Geophysical Research, Vol. 93, No. C11,<br />
pp. 14107-14120.<br />
Rodriguez, E. and Chapman, B. (1989) Extracting Ocean Surface Information from<br />
Altimeter Returns: The Deconvolution Method, Journal of Geophysical<br />
Research, Vol. 94, No. C7, pp. 9761-9778.<br />
Rodriguez, E. and Martin, M. (1994a) Assessment of the TOPEX Altimeter<br />
Performance Using Waveform Retracking, Journal of Geophysical Research,<br />
Vol. 99, No. C12, pp. 24957-24969.<br />
Rodriguez, E. and Martin, M. (1994b) Estimation of the Electromagnetic Bias from<br />
Retracked TOPEX Data, Journal of Geophysical Research, Vol. 99, No. C12,<br />
pp. 24971-24979.<br />
Sandwell, D. T. (1984) A Detailed View of the South Pacific Geoid from Satellite<br />
Altimetry, Journal of Geophysical Research, Vol. 102, pp. 10039-10054.<br />
Sandwell, D. T. (1992) Antactic Marine Gravity Field from High-Density Satellite<br />
Alitmetry, Geophysical Journal International, Vol. 109, pp. 437-448.<br />
Sandwell, D. T. and Smith, W. H. F. (1995) Gravity Anomaly Profiles from ERS-1,<br />
Topex and Geosat Altimetry, EOS, Transaction of the American Geophysical<br />
Union, Vol. 76, No. 17, pp. S89.<br />
Sandwell, D. T. and Smith, W. H. F. (1997) Marine Gravity Anomaly from Geosat<br />
and ERS-1 Satellite Altimetry, Journal of Geophysical Research, Vol. 102, pp.<br />
10039-10054.
247<br />
Sandwell, D.T. (2003) Significant Improvement In Global Marine Gravity By Re-<br />
Tracking Altimeter Waveforms, paper presented to the IUGG General<br />
Assembly, Sapporo, Japan, July [CD-ROM].<br />
Scott, R. F., Baker, S. G., Birkett, C. M., Cudlip, W., Laxon, S. W., Mantripp, D. R.,<br />
Mansley, J. I., Morley, J. G., Rapley, C. G., Ridley, J. K., Strawbridge, F. and<br />
Wingham, D. J. (1994) A Comparison of The Performance of the Ice and<br />
Ocean Tracking Modes of the ERS-1 Radar Altimeter over Non-Ocean Surface,<br />
Geophysical Research Letters, Vol. 21, No. 7, pp. 553-556.<br />
Shum, C. K., Woodworth, P., Andersen, O., Egbert, G., Francis, O., King, C., Klosko,<br />
S., Provost, Le. C., Li, X., Molines, J., Parke, M., Ray, R., Schlax, M.,<br />
Stammer, D., Tierney, C., Vincent, P. and Wunsch, C. (1997) Accuracy<br />
Assessment of Recent Ocean Tide Models, Journal of Geophysical Research,<br />
Vol. 102, No. C11, pp. 25173-25194.<br />
Shum, C. K., Parke, M. E., Schutz, B. E., Abusali, P. A. M., Gutierrez, R., Pekker, T.,<br />
Tapley, B. D., Martinez Benjamin, J. J., Blaha, J., Jacobs, G., Jeffress, G. A.,<br />
Schaudt, K. J. and Wingham, D. J. (1998) Improvement of<br />
TOPEX/POSEIDON Altimeter Data for Global Change Studies and Coastal<br />
Applications, AVISO Altimetry Newsletter, Vol. 6, pp. 102-103.<br />
Soussi, B. and Zanife, O.-Z. (1999) ERS-2 Radar Altimeter off Pointing Impact Study<br />
Task1 - Impact of the Mispointing on the Altimeters measurements by<br />
Simulation,<br />
http://www.ifremer.fr/cersat/FICHES/DIVERS/CLS/PDF/E_002F1.PDF<br />
[last accessed: 8 March 2004].<br />
Srokosz, M. A. (1986) On the Joint Distribution of Surface Elevation and Slopes for<br />
a Nonlinear Random Sea, with an Application to Radar Altimetry, Journal of<br />
Geophysical Research, Vol. 91, No. C1, pp. 995-1006.<br />
Tapley, B. D. and Kim, M. C. (2001) Applications to Geodesy, in: Satellite Altimetry<br />
and Earth Sciences, Fu, L.-L. and Cazenave, A. (eds.), Academic Press, New<br />
York, USA, pp. 371-406.<br />
Thom, B. G. (1984) Geomorphic Research on the Coast of Australia: A Preview, in:<br />
Coastal Geomorphology in Australia, Thom, B. G. (eds.), Academic Press<br />
Australia, Sydney, pp. 1-15.<br />
Tierney, C., Parke, M. E. and Born, G. H. (1998) An Investigation of Ocean Tides<br />
Derived from Along-Track Altimetry, Journal of Geophysical Research, Vol.<br />
103, No. C5, pp. 10273-10287.<br />
Tokmakian, R. T., Challenor, P. G., Guymer, T. H. and Srokosz, M. A. (1994) The<br />
U.K. EODC ERS-1 Altimeter Oceans Processing Scheme, International<br />
Journal of Remote Sensing, Vol. 15, No. 4, pp. 939-962.
248<br />
Tziavos, I. N., Forsberg, R. and Sideris, M. G. (1998) Marine Gravity Field<br />
Modelling Using Ship-Borne Gravity and Geodetic Mission Altimetry Data,<br />
Geomatics Research Australasia, Vol. 69, pp. 1-18.<br />
Wang, Y.-M. and Rapp, Richard H. (1991) Geoid Gradients for Geosat and<br />
Topex/Poseidon Repeat Ground Tracks, 408, The Ohio State <strong>University</strong>,<br />
Columbus, USA, 26 pp.<br />
Wessel, P. and Smith, W. H. F. (1996) A Global, Self-Consistent, Hierarchical,<br />
High-Resolution Shoreline Database, Journal of Geophysical Research, Vol.<br />
101, No. B4, pp. 8741-8743.<br />
Wessel, P. (2000) GMT - Generic Mapping Tools, http://gmt.soest.hawaii.edu/ [last<br />
accessed: 8 March 2004].<br />
Wingham, D. J., Rapley, C. G. and Griffiths, H. (1986) New Techniques in Satellite<br />
Tracking Systems, Proceedings of IGARSS' 88 Symposium, Zurich,<br />
Switzerland, September, pp. 1339-1344.<br />
Wingham, D. J., Rapley, C. G. and Morley, J. G. (1993) Improved Resolution Ice<br />
Sheet Mapping with Satellite Altimeters, EOS, Transaction of the American<br />
Geophysical Union, Vol. 74, No. 10, pp. 113-116.<br />
Wingham, D. J. (1995a) The Limiting Resolution of Ice-Sheet Elevations Derived<br />
from Pulse-Limited Satellite Altimetry, Journal of Glaciology, Vol. 41, No.<br />
138, pp. 413-422.<br />
Wingham, D. J. (1995b) A Method for Determining the Average Height of a Large<br />
Topographic Ice Sheet from Observations of the Echo Received by a Satellite<br />
Altimeter, Journal of Glaciology, Vol. 41, No. 137, pp. 125-141.<br />
Wingham, D. J., Ridout, A.J., Scharroo, R., Arthern, R. and Shum, C. K. (1998)<br />
Antarctic Elevation Change from 1992 to 1996, Science, Vol. 282, No. 5388,<br />
pp. 369-580.<br />
Zwally, H. J. (1996) GSFC Retracking Algorithms, http://icesat4.gsfc.nasa.gov/ [last<br />
accessed: 8 March 2004].<br />
Zwally, H. J. and Brenner, A. C. (2001) Ice Sheet Dynamics and Mass Balance, in:<br />
Satellite Altimetry and Earth Sciences: A Handbook of Techniques and<br />
Applications, Fu, L.-L. and Cazenave, A. (eds.), Academic Press, San Diego,<br />
USA, pp. 351-370.