Mapping Woody Vegetation with the Compact High Resolution - ESA

Mapping Woody Vegetation with the Compact High Resolution 

Imaging Spectroradiometer on the Proba platform: 

A Monospectral, Multi-Angle Approach 

Mark Chopping 1 , Lihong Su 1 , Albert Rango 2 , Andrea Laliberte 2 , Debra P.C. Peters 2 , John V. 

Martonchik 3 , Naushad Kollikatharan 1 

1 2 3 

CHRIS / Proba 4th Principal Investigators! Workshop ESRIN Frascati, Italy, 19-21 September 2006

CHRIS/Proba Jornada Experiment: Advances in Multi-Angle Imaging 

Acknowledgments 

The data used were derived from the Compact High Resolution Imaging 

Spectrometer, developed by Sira Technology Ltd (formerly Sira Electro-Optics 

Ltd), mounted on board the European Space Agency!s PROBA-1 platform. The 

MISR data used were obtained from the NASA Langley Research Center 

Atmospheric Science Data Center. 

The Jornada Experimental Range is administered by the USDA Agricultural 

Research Service and is a Long Term Ecological Research site supported by 

the National Science Foundation. Forest Inventory Service data were provided 

by Gretchen Moisen, US Forest Service, Rocky Mountain Research Station, 

Ogden, UT. 

Thanks to Matt Smith and the Global Land Cover Facility, University of 

Maryland, College Park, MD, for providing the VCF and SDB data. 

The work described herein has been significantly enabled by NASA grant 

NNG04GK91G to Mark Chopping (EOS project EOS/03-0183-0465 “Quantifying 

Changes in Carbon Pools with Shrub Invasion of Desert Grasslands using 

Multi-Angular Data from EOS Terra and Aqua”, that also benefited from our 

work under the ESA CHRIS/Proba program. 



OUTLINE and SUMMARY 

• Goal: estimate sub-pixel woody shrub fractional cover at 

landscape scales using multi-angle RS techniques 

• Approach: evaluate the Simple Geometric Model (SGM) against 

the 631 nm directional signal from CHRIS; adjust against CHRIS 

(and MISR red band) multi-angle data 

• Major Challenge: to obtaining adequate estimates of 

background reflectance anisotropy at CHRIS/Proba (and 

MISR/Terra) acquisition angles 

• 1st order solution: exploit the isotropic, geometric, and volume 

scattering weights of Li-Ross model kernels 

• Results: mapped distributions of woody shrub cover are good 

with reference to estimates made using IKONOS 1 m 

panchromatic imagery 



Structure: Should be a Strength of M-angle Remote Sensing 

• Multi-angle sampling provides sensitivity to vegetation 

canopy structure through the physical phenomena of 

shadow-hiding and volume scattering by leaves 

• Multi-angle data are best interpreted through canopy 

reflectance models, which have greater explanatory power 

than empirical measures 

• However, interpretation is difficult because the real world is 

complex and complex models can be difficult, impractical, 

or even impossible to invert: We have to decide on the 

appropriate level of complexity, or devise appropriate 

inversion protocols (e.g., injection of a priori knowledge) 


Canopy architecture and spatial 

heterogeneity affect carbon 

storage, vegetation dynamics, 

and productivity of grasslands 

undergoing woody 

encroachment Extensive 

creosotebush 

shrublands in 

southern 

New Mexico 

We need maps of structural parameters for grasslands as well as 

savannas, woodlands and forests: shrubs now occupy 83% of 

current and former US grasslands (not to mention elsewhere) 


We have a lot of real 

estate occupied by 

non-forest woody 

plants (nowadays) 

AZ NM 

Mexico 

Forest 

Non-forest mask 

Lakes or reservoirs 

State boundaries 

Source: USDA, US Forest Service Forest Inventory 

Analysis, Forest Cover map for the Interior West 

CHRIS / Proba 4th Principal Investigators! Workshop ESRIN Frascati, Italy, 19-21 September 2006 

TX


In the 1980s we 

recognized large 

directional effects in 

AVHRR VNIR data 

resulting from 

variations in canopy 

structure; in the 

1990s we learned 

how to model these 

semi-empirically (Li- 

Ross & RPV models) 

Chihuahuan Desert 

northern boundary 

rivers and lakes 

USDA-ARS Jornada 

Experimental Range 

N 

100 km 

Multispectral image from AVHRR VIS, NIR, and Mid-IR channels ? …NO! 

RGB= iso, geo, vol kernel weights from Li-Ross BRDF model fitted to AVHRR VIS channel! 



Canopy Structure from M-Angle Observations 

Researchers have sought to obtain measures of 

canopy structure using NASA (MISR) & ESA (CHRIS) 

multi-angle observations by following a variety of 

approaches: 

A. Empirical / Data Mining / Synergistic 

B. Radiative Transfer modeling/LUTs 

C. Canopy Openness from MRPV-k parameter 

D. Clumping Index 

E. Structural Scattering Index (Li-Ross) 

F. Geometric-Optical and Hybrid Models 

All these avenues have proved worthwhile 

Applied with CHRIS Applied with other instruments (MISR, MODIS) 



study area 

USDA, ARS Jornada 

Experimental Range, 

~37 km N of Las Cruces, 

New Mexico 



State of New Mexico 

150 0 150 300 miles 

Sevilleta National 

Wildlife Refuge 

25 0 75 km 

Jornada 

Experimental 

Range 

MISR iso kernel weight image 



Vegetation Changes in the Last 150 yrs: space for time 

Typical Desert Grassland (Sevilleta) substitution 

Chihuahuan and Plains Grassland (Black 

Grama Grasslands with Blue Grama) 

Honey mesquite (Prosopis 

glandulosa) shrub-coppice dunes 

Desertified Grassland (Jornada) 


Black Grama Grassland 


Grassland-Shrub Transition 

(backscattering) 



(forward-scattering) 



with advanced mesquite encroachment 


Mesquite Shrub Coppice Dunes 



Chopping et al. performed GO modeling studies using 

DuncanTech, CHRIS, and MISR data in the SW US at the 

Plot scale* 25 m 2 2003-4 

Landscape scale 1 § > 27 km 2 2005 

Landscape scale "2 " # > 3,519 km2 2005-6 

Regional scale ~ 240,000 km 2 2006 

Continental/Global scale… 20xx? 

* DuncanTech Airborne Imaging Camera 

§ CHRIS/Proba (Sira/SSTL/ESA) 

# MISR (NASA/JPL) 



Chopping et al. performed GO modeling studies using 

DuncanTech, CHRIS, and MISR data in the SW US at the 

Plot scale* 25 m 2 field/air 

Landscape scale 1 § > 27 km 2 air/CHRIS 

Landscape scale "2 " # > 3,519 km2 MISR 

Regional scale ~ 240,000 km2 MISR 

Continental/Global scale… 20xx? 

* DuncanTech Airborne Imaging Camera 

§ CHRIS/Proba (Sira/SSTL/ESA) 

# MISR (NASA/JPL) 



Chopping, M. (2006a), Progress in retrieving canopy structure parameters from NASA multi-angle remote 

sensing, Terra/Salomonson session, Proc. IGARSS’06, Denver, CO, July 2006. 

Chopping, M., (2006b) Multi-angle remote sensing and applications, in Advances in Land Remote Sensing: 

System, Modeling, Inversion and Applications, S. Liang (ed), in revision. 

Chopping, M., Su, L., Rango, A., Martonchik, J.V., Peters, D.P.C., and Laliberte, A. (2006a), Remote 

sensing of woody shrub cover in desert grasslands using MISR with a geometric-optical 

canopy reflectance model, Remote Sens. Environ., accepted, 07/06. 

Chopping, M., Su, L., Laliberte, A., Rango, A., Peters, D.P.C., and Kollikkathara, N. (2006b), Mapping 

shrub abundance in desert grasslands using geometric-optical modeling and multiangle remote 

sensing with CHRIS/Proba. Remote Sens. Environ. 104 (2006) 62 – 73. 

Chopping, M., Su, L., Laliberte, A., Rango, A., Peters, D.P.C., and Martonchik, J.V. (2006c) Mapping 

woody plant cover in desert grasslands using canopy reflectance modeling and MISR data, 

Geophysical Research Letters, Vol. 33, L17402, doi:10.1029/2006GL027148. 

Chopping, M., J.V. Martonchik, et al. (2005), Geometric-optical modeling of desert grassland canopy 

structure with MISR, Proc. ISPMSRS 2005, I: 141-143 

Chopping, M.J., Su, L., Rango, A., and Maxwell, C. (2004). Modelling the reflectance anisotropy of 

Chihuahuan Desert grass-shrub transition canopy-soil complexes, IJRS 25: 2725–2745. 

Chopping, M., Laliberte, A., and Rango, A. (2004), Exploitation of multi-angle data from CHRIS on 

Proba: First results from the Jornada Experimental Range, European Space Research Institute 

(ESRIN), Frascati, Italy, ESA Special Publication SP-578,109-117. 

Chopping M.J., Rango, A., Havstad, K.M., Schiebe, F.R., Ritchie, J.C., Schmugge, T.J., French, A., Su, L., 

McKee, L., and Davis, R.M. (2003), Canopy attributes of Chihuahuan Desert grassland and 

transition communities derived from multi-angular airborne imagery, Remote Sens. Environ. 

85(3): 339-354. 



Geometric-Optical 

(Li-Strahler) models 

provide one way to 

interpret multi-angle data 

• August 22, 2003 CHRIS data 

set (Mode 3: ~18 m GIFOVs) 

• 4 overlapping looks 

• Geometry from Xephem/TLEs 

• Geometric transformation 

w/poly2/3 and rubbersheeting 

(>60 GCPs); 

resampled to 25m 

• Atmospheric correction using 

6S/visibility estimates 

• No correction for striping 

• Only the 631 nm (red) data 

were used in modeling (KISS 

principle) 



Geometric-Optical Models predict bidirectional reflectance at specified 

illumination and viewing directions as a function of canopy structure 

parameters of practical and ecological significance: 

• mean crown shape (b/r) 

• mean crown radius (r) 

• plant number density (!) 

• mean canopy height (h) 

r 

• background brightness 

and anisotropy (Walthall 

model (a function of 

of understory density) 

b 

R = G . k G + C . k C + T . k T + Z . k Z 

G, C, T, Z can be 

assumed Lambertian or 

may have defined 

reflectance anisotropies 

sunlit crown shadowed crown 

sunlit ground shadowed ground 


h


Knowledge of the background contribution is critical if the background is highly 

exposed, as in this shrub-grass transition zone in the USDA, ARS Jornada 

Experimental Range in southern New Mexico 

CHRIS CHRIS / Proba / Proba 4th 4th Principal Investigators! Workshop ESRIN ESRIN Frascati, Italy, Italy, 19-21 19-21 September 2006 2006

5 

1 

4 

3 

IKONOS Panchromatic Image 

05/23/01 Sites are indicated in RED 

6 

2 


600 m 

N 

The problem: 

understory 

We looked at this by 

studying the soilunderstory 

behaviour 

at a number of 

Selected Sites: 

1. grama grass with some 

PRGL (mesquite) 

2. large PRGL on sand 

3. small PRGL on dense 

understory 

4. small PRGL on sparse 

understory 

5 small PRGL on dense 

understory(2) 

6. mixed area near 

WestWell

250 meters 

250 meters 

IKONOS 1 meter pan images for 6 contrasting sites 

Used as proxy for field measurements & cal/validation 



How to Obtain the Background Signal? 

The ability to estimate the 

brightness and anisotropy of the 

background is a major 

prerequisite for the use of GO 

models in many places 

R = G . k G + C . k C + T . k T + Z . k Z 

i.e., G = ?? 

Can G be Lambertian? 

NO 

Can G have static anisotropy? 

NO 

We have a problem 

Can we make this work? 

k G >= 0.75 

Photo: Scott Bauer, USDA, ARS Photo Unit 



1st Order: The Li-Ross vol kernel weight is sensitive to the understory 

E.g., MISR vol kernel weight Ikonos pan image 

+shrub 

-vol scattering 

high low 

+u.story 

+vol scattering 


Simple 

Geometricoptical 

Model 

Inversion 

Protocol 

The Walthall 

model (W) is used 

to represent the 

background and a 

Ross function 

allows for withincrown 

volume 

scattering; G is 

replaced with W 

and C with Ross, 

so that 

R = 

G W.k G +C Ross.k C 

T and Z are 

considered black 

Exposed 

soil 

invert the 

LiSparse-RossThin 

kernel-driven model against 

Red band MISR CHRIS data in 9 4 views 

obtain relationships between the 

kernel weights and the background 

BRDF in the MISR CHRISplane 

by fixing 

shrub statistics and adjusting the 

Walthall model parameters 

obtain estimates of woody shrub 

mean radiius and number density 

via thresholding 1 m IKONOS 

panchromatic imagery 

CHRIS MISR 

Red Band 

BRFs 

in Nine Cameras 

4 looks 

invert the SGM model 

using the estimated 

background and fixing 

all parameters except 

mean shrub radius; 

adjust the model 

against MISR CHRIS data 

calculate fractional 

shrub cover from mean 

shrub density 

CHRIS / Proba 4th Principal Investigators! Mesquite shrubs Workshop Black grama ESRIN grass Frascati, Italy, 19-21 September 2006


Estimated surface 

bidirectional reflectance vs. 

surface bidirectional 

reflectance at 631 nm 

modeled using the Simple 

Geometric Model (SGM) 

when driven with IKONOS- 

derived shrub statistics and 

backgrounds estimated a 

priori from isotropic- 

LiSparse-RossThin LiSK 

model kernel weights 

(represented by the Walthall 

model). R 2 =0.75 N= 38,934. 

CHRIS Reflectance at 631 nm 

Noisy! 

Too Noisy? 

(no!) 



Modeled vs. measured woody shrub 

fractional cover, the latter estimated 

from IKONOS 1 m panchromatic 

imagery. These results were obtained 

by adjusting the Walthall model 

coefficients so that the lowest 

absolute RMSE was obtained 

between the Simple Geometric Model 

(SGM) and the corresponding 

CHRIS/Proba 631 nm multi-angle 

reflectance data. Error in retrieving 

shrub cover using this method is 

almost entirely owing to inaccuracy 

in estimating the background signal. 

y = 0.8746x + 0.0055 

R 2 = 0.9987 


Modeled Fractional Shrub Cover 

0.35 

0.30 

0.25 

0.20 

0.15 

0.10 

0.05 

0.00 

1:1 

line 

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 

Measured Fractional Shrub Cover


Multi-angle false colour 

composites: 

(a) RGB = band 421 composite 

(b) RGB = band 421 simulation 

(c) RGB = band 431 composite 

(d) RGB = band 431 simulation 

These composites show strong 

agreement between modeled and 

observed CHRIS/Proba data in 

different views and also provide 

an indication of the information 

content of multi-angle imagery. 

(a) 

(c) 


(b) 

(d)


Estimated 

with 

IKONOS 

500 m 

Estimated 

with 

CHRIS/SGM 

Mapped values of retrieved fractional woody shrub cover match the 

distribution of the IKONOS-estimated values: (a) fractional woody shrub 

cover estimated using IKONOS 1 m panchromatic imagery (b) woody shrub 

cover retrieved by adjusting the SGM against CHRIS/Proba multi-angle data. 



Frequency distributions of 

absolute deviation between 

retrieved and measured 

fractional woody shrub cover 

(retrieved using SGM with 

calibrated Walthall soil- 

understory response). Almost 

90% of estimates are 

associated with a deviation 

from the IKONOS-estimated 

value of less than 0.08. 

Frequency 

700 

600 

500 

400 

300 

200 

100 

0 

0.001 0.076 0.151 0.226 0.301 0.376 0.451 

Absolute deviation 

Frequency 

Cumulative % 

120.0% 

100.0% 


80.0% 

60.0% 

40.0% 

20.0% 

.0%


Note: there is error in the 

IKONOS-derived 

estimates: Fractional 

woody shrub cover from 1 

m IKONOS pan imagery 

vs. independently-derived 

estimates from 

segmentation of 0.5 m 

QuickBird pan imagery, 

using eCognition (random 

sample) 



Conclusions -- Part 1 

The multi-angle remote sensing signal from 

CHRIS/Proba at 631 nm can be explained in terms of 

a combined soil-understory background response 

and woody shrub cover and exploited to map woody 

shrub cover in desert grasslands at landscape 

scales. 

The modeling framework developed allows retrieval 

of shrub cover for both canopies with an important 

understory component as well as those with sparse 

understories and a high proportion of exposed soil. 



Can we apply similar techniques over larger 

areas -- and with a higher revisit rate -- using 

NASA!s Multi-angle Imaging Spectro- 

Radiometer (MISR) on the Terra satellite? 

YES! 

The following slides show the results from 

the first run… 


Comparisons of Retrievals with VCF % Tree Cover Map 

0.89 

tree 

0.30 

0.29 

shrub 

0.00 

Fractional 

Cover 

Rio Grande 

riparian zone 

Left: MISR/GO Woody Plant Cover Right: MODIS Vegetation Continuous 

Fields % Tree Cover for the USDA, ARS Jornada Experimental Range 


10 km 

Fractional 

Cover 

Summerford San Andres Mountains 

Mountain 

0.27 

0.02 

0.01

Comparisons of Retrievals with VCF % Tree Cover Map 

Fractional Cover 

0.89 

tree 

0.30 

0.27 

0.02 

0.29 

shrub 

0.00 

0.01 

10 km 

--10 km-- 

Rio Grande 

riparian 

zone 

San Andres 

Mountains 

Top: MISR/GO 

Woody Plant Cover 

Bottom: MODIS 

Vegetation 

Continuous Fields 

% Tree Cover for 

the Sevilleta 

National Wildlife 

Refuge near 

Socorro, NM 


MISR/GO Regional Woody Plant Canopy Cover 

white=cloud 

cover and 

White Sands 

National 

Monument 

AZ NM 

Mexico 

MISR/GO Canopy Cover Forest Inventory Analysis Canopy Cover 


TX

MISR/GO Regional Woody Plant Canopy Height 

AZ NM 

Mexico 

MISR/GO Mean Canopy Height Forest Inventory Analysis Canopy Height 


TX

MISR/GO Regional Woody Plant Canopy Cover 


MISR/GO Regional Woody Plant Canopy Height 


MISR/GO Regional Crown Cover x Canopy Height 


MISR / SGM work: Caveats Etc. 

• The Li-Ross background relationship was 

calibrated on only 19 grass-shrub sites in the W. 

part of the USDA Jornada Experimental Range 

• The results from this 1st run should be much 

worse: crown LAI was fixed over the entire area, 

which is insane (but existing LAI products are not 

crown LAI, so cannot be used directly) 

• These results should be classified as “very 

interesting” at this stage: validation is required. On 

the other hand, they are to our knowledge unique in 

the moderate resolution RS world 



Conclusions -- Part 2 

• CHRIS is an unique instrument providing data for the testing 

of algorithms that have important applications in many 

disciplines (including mapping canopy structural parameters) 

• We have made significant progress towards exploiting the 

structural information content of multi-angle data (empirical 

methods currently give the best results; models are gaining…) 

• Multi-angle data provide a way to separate the background 

and upper canopy contributions, enabling the inversion of GO 

models for crown cover and mean canopy height over large 

areas (validation must be pursued) 

• Retrieval of canopy structure measures is more challenging 

over shrubs than over trees (no surprise) 

• Working with CHRIS and MISR multi-angle data has benefitted 

our research in both EO programs. 



Publications: 



Publications: 



Publications: 



Publications: 


Thank You! 

Questions? 

Google: 

Chopping CHRIS Jornada 

Chopping MISR Jornada

Mapping Woody Vegetation with the Compact High Resolution - ESA

Create successful ePaper yourself

Delete template?

Save as template?