Quantitative Precipitation Estimation from Radar Data â A ... - einfalt.de
Quantitative Precipitation Estimation from Radar Data â A ... - einfalt.de
Quantitative Precipitation Estimation from Radar Data â A ... - einfalt.de
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
MUSIC – Multiple-Sensor <strong>Precipitation</strong> Measurements,<br />
Integration, Calibration and Flood Forecasting<br />
A Project supported by the European Commission<br />
un<strong>de</strong>r Contract N o EVK1-CT-2000-00058<br />
Published is here the Deliverable 4.1<br />
titled<br />
<strong>Quantitative</strong> <strong>Precipitation</strong> <strong>Estimation</strong><br />
<strong>from</strong> <strong>Radar</strong> <strong>Data</strong> –<br />
A Review of Current Methodologies<br />
by Ronald Hannesen<br />
Gematronik GmbH<br />
resulting <strong>from</strong><br />
WP 4<br />
Gematronik GmbH<br />
Raiffeisenstr. 10<br />
41470 Neuss<br />
Germany<br />
Tel.: (+49) 2137 782 0<br />
Fax: (+49) 2137 782 11<br />
EMail: Info@Gematronik.com<br />
Web: www.gematronik.com<br />
Assessment of presently available radar estimation techniques and<br />
implementation of improved techniques for radar rainfall estimates
Contents<br />
Contents ..................................................................................................................... 2<br />
1 Scope of this paper ............................................................................................. 3<br />
2 Basic Principles of <strong>Radar</strong> Meteorology................................................................ 5<br />
3 <strong>Quantitative</strong> <strong>Precipitation</strong> <strong>Estimation</strong> I – Aspects of Rainfall Rate Derivation...... 7<br />
3.1 Drop Size Distributions and Z-R-Relations ...................................................... 7<br />
3.2 Clutter Filtering and Speckle Removing........................................................... 9<br />
3.3 Attenuation..................................................................................................... 11<br />
3.4 Vertical Profiles and Beam Blocking Corrections........................................... 13<br />
3.5 Orographic Enhancement.............................................................................. 14<br />
3.6 Bright Band, Snow and Hail ........................................................................... 14<br />
3.7 Stratiform and Convective <strong>Precipitation</strong>......................................................... 16<br />
3.8 Dual Polarisation <strong>Radar</strong>s ............................................................................... 17<br />
3.9 Dual Wavelength <strong>Radar</strong>s............................................................................... 20<br />
3.10 Scan Strategy............................................................................................. 21<br />
3.11 Operational Applicability............................................................................. 22<br />
4 <strong>Quantitative</strong> <strong>Precipitation</strong> <strong>Estimation</strong> II – Accumulated Rain: Aspects of<br />
Integration in Time.................................................................................................... 24<br />
4.1 Scan Strategy and Time Steps between Single Scans.................................. 24<br />
4.2 Speckle Filtering ............................................................................................ 27<br />
4.3 Handling of Missing Scans............................................................................. 27<br />
5 Outlook: Future Work in the MUSIC Project...................................................... 28<br />
References ............................................................................................................... 29<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 2
1 Scope of this paper<br />
Introduction<br />
1 · Scope of this paper<br />
<strong>Precipitation</strong> measurement is an essential task for many purposes. The amount and<br />
the horizontal distribution of precipitation strongly influence evaporation <strong>from</strong> the<br />
earth’s surface, which triggers the global atmospheric circulation. For example<br />
farmers and forest authorities need precipitation information to control irrigation<br />
<strong>de</strong>vices for optimal agricultural use. Hydrologists need precipitation data as input for<br />
river stage forecasts, flood warnings or waste water flow regulation.<br />
If much rain falls in a short time, the soil will not allow it all to infiltrate. The resulting<br />
surface runoff may result in floods causing a large amount of damage. The extent of<br />
a flood <strong>de</strong>pends on the precipitation type. Convective precipitation systems, which<br />
seldom last longer than a few hours, can have very intense small-scale rain events<br />
causing hazardous flash floods in small subcatchments. The 1993 Brig flash flood is<br />
an example (e.g. Benoit and Desgagné, 1996).<br />
To recognise the danger of a potential flood, operational weather forecasts are<br />
necessary as well as sufficiently <strong>de</strong>nse networks of precipitation measuring stations.<br />
Denser networks are required to observe rain events with higher intensity and<br />
smaller spatial extent. The observation of strong convective precipitation in large<br />
areas might require too much gauges, and therefore, use of weather radar data is<br />
essential.<br />
A weather radar provi<strong>de</strong>s information about the intensity of precipitation with a spatial<br />
resolution of less than a kilometer and a temporal resolution of about one minute. An<br />
area of several hundred kilometers can be observed with one <strong>de</strong>vice. A radar gives<br />
the chance to i<strong>de</strong>ntify dangerous precipitation regions before they appear at a<br />
specific site. The three-dimensional data sets allow the investigation of vertical<br />
structures and of dynamics of precipitation systems. The problem, that the rain<br />
intensity itself cannot be measured directly, has resulted in a wi<strong>de</strong> range of research<br />
fields with the aim of making precipitation estimates <strong>from</strong> weather radars as good as<br />
possible (Atlas, 1990).<br />
The MUSIC Project<br />
The main goal of the MUSIC (Multiple-Sensor <strong>Precipitation</strong> Measurements,<br />
Integration, Calibration and Flood Forecasting) Project is to <strong>de</strong>velop an innovative<br />
technique to improve weather radar, weather satellite and rain gauge <strong>de</strong>rived<br />
precipitation data, resulting in an integrated prototype flood forecasting system. The<br />
Project is subdivi<strong>de</strong>d into different work packages (WP). One of these, WP 4, has the<br />
aim of assessing available methodologies and of <strong>de</strong>veloping enhanced weather radar<br />
precipitation estimates. The present paper summarises the results of the first part of<br />
this package, namely a review of currently available methods in <strong>Quantitative</strong><br />
<strong>Precipitation</strong> <strong>Estimation</strong> (QPE) <strong>from</strong> weather radar data.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 3
1 · Scope of this paper<br />
Very often radar precipitation estimates are adjusted by other-sensor data like rain<br />
gauge data, or are combined with numerical mo<strong>de</strong>ls. These methods will not be<br />
<strong>de</strong>scribed in this paper. It is the task of other work packages within the Project to<br />
combine weather radar, weather satellite and rain gauge data by applying the Block<br />
Kriging and Bayesian combination techniques. Thus the radar data must not be<br />
adjusted by rain gauges or other-sensor data when using these methods.<br />
This paper presents an overview of currently used methods in <strong>Quantitative</strong><br />
<strong>Precipitation</strong> <strong>Estimation</strong> (QPE) <strong>from</strong> weather radar data alone. Chapter 2 gives a<br />
basic introduction to radar meteorology. QPE can be divi<strong>de</strong>d into two steps: i)<br />
<strong>de</strong>rivation of rainfall intensity <strong>from</strong> single- or multiple-parameter radar data; and ii)<br />
accumulation, i.e. integration in time of rainfall intensities. Chapter 3 reviews current<br />
techniques with respect to the first step, i.e. <strong>de</strong>rivation of rainfall intensities <strong>from</strong> radar<br />
data. Chapter 4 lists aspects of the second step, namely the integration in time of<br />
rainfall intensity data. The last Chapter presents an outlook to future work within the<br />
MUSIC Project.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 4
2 · Basic Principles of <strong>Radar</strong> Meteorology<br />
2 Basic Principles of <strong>Radar</strong> Meteorology<br />
The basic principle of a weather radar is its ability to <strong>de</strong>rive quantitative information of<br />
precipitation particles <strong>from</strong> the characteristics of electromagnetic radiation which is<br />
transmitted and received by such a <strong>de</strong>vice. The intensity of precipitation can be<br />
<strong>de</strong>rived <strong>from</strong> the amplitu<strong>de</strong> of the received radiation. Doppler radars measure the<br />
phase shift of the electromagnetic wave, which is used to obtain information about<br />
the atmospheric circulation. Some radars can measure polarimetric parameters,<br />
which give hints on the type and shape of the hydrometeors. A <strong>de</strong>tailed review of the<br />
<strong>de</strong>velopment in radar meteorology can be found in the articles collected in Atlas<br />
(1990). Further <strong>de</strong>tails of the theory of radar meteorology can be found in e.g. Battan<br />
(1973), Rinehart (1991), Sauvageot (1992) or Doviak and Zrnic (1993).<br />
This chapter will present the most important equations for the measurement of the<br />
reflectivity. The <strong>de</strong>rivation of rainfall parameters will be handled in the next chapter<br />
(section 3.1), which gives an overview of techniques used therefor. The theory of<br />
polarimetric measurements will also be introduced in the next chapter (section 3.8).<br />
With a pulsed weather radar, the atmosphere – containing many scattering<br />
particles – is “illuminated” with a short pulse of electromagnetic radiation. The <strong>Radar</strong><br />
Equation gives the relation between the transmitted power Pt and the received power<br />
Pr (after Sauvageot, 1992):<br />
2<br />
2<br />
2<br />
Pt G λ L c τ η 4<br />
Pr = ∫ f ( θ,<br />
ϕ)<br />
dΩ<br />
. (2.1)<br />
3<br />
2<br />
( 4π)<br />
2 r Ω<br />
Here G is the so-called antenna gain, which <strong>de</strong>scribes the ratio between the radiation<br />
intensity of a beam collected by the antenna reflector and an isotropic transmission of<br />
radiation. λ is the wavelength of the radiation. 1–L is the fraction of radiation lost by<br />
extinction (mostly called attenuation) on a single trip between the antenna and the<br />
scattering particles. c is the speed of light and τ the pulse duration. η is the volume<br />
specific backscattering cross section of the particles and is called radar reflectivity.<br />
r is the distance to the radar, which results <strong>from</strong> the time since pulse transmission.<br />
f 2 (θ,φ) is the normalised radiation intensity at the azimuth angle θ and the zenith<br />
angle φ (given relative to the beam axis, thus f 2 (0,0) = 1). The right-hand integral of<br />
eq. (2.1) extends over the complete solid angle Ω of the pulse volume, assuming<br />
homogeneous distribution of the scatterers.<br />
For spherical water droplets with a diameter D « λ, the backscattering cross section σ<br />
of the individual drops can be calculated using the Rayleigh approximation. We then<br />
find for the radar reflectivity<br />
Dmax<br />
η = ∫ σ ( D)<br />
n(<br />
D)<br />
dD ≈ dD ) D ( n D | K |<br />
π 2 6<br />
4 ∫<br />
, (2.2)<br />
λ<br />
0<br />
5<br />
where Dmax is the maximum diameter of the droplets. n(D) is the drop size distribution<br />
and |K| 2 the dielectric coefficient (|K| 2 ≈ 0,93 for water and |K| 2 ≈ 0,18 for ice). The<br />
Rayleigh approximation is valid in most cases, as weather radars typically operate at<br />
X-Band (≈3 cm wavelength), C-Band (≈5 cm) or S-Band (≈10 cm).<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 5<br />
Dmax<br />
0
2 · Basic Principles of <strong>Radar</strong> Meteorology<br />
Eq. (2.2) in combination with eq. (2.1) implies the <strong>de</strong>finition of the radar reflectivity<br />
factor Z (which is usually shortly expressed as reflectivity):<br />
Dmax<br />
D<br />
6<br />
Z = ∫ n(<br />
D)<br />
dD.<br />
(2.3)<br />
0<br />
The real reflectivity Z of the particles has to be distinguished <strong>from</strong> the measured Zm,<br />
which can be calculated <strong>from</strong> the received power Pr and the distance r <strong>from</strong> eq. (2.1):<br />
1 2<br />
Zm = Pr r , (2.4)<br />
C<br />
with C being the radar constant:<br />
5<br />
2<br />
2<br />
2<br />
2 π Pt<br />
G L λ c τ 4<br />
C = | K|<br />
∫ f ( θ,<br />
φ)<br />
dΩ<br />
. (2.5)<br />
4<br />
3<br />
λ ( 4π)<br />
2 Ω<br />
The radar constant is mainly a hardware-<strong>de</strong>pen<strong>de</strong>nt quantity. For the calculation of<br />
the measured reflectivity Zm according to eq. (2.4), usually the dielectric coefficient<br />
for water is used, and the attenuation losses are assumed by standard gaseous<br />
attenuation.<br />
The values of the reflectivity Z extends over several or<strong>de</strong>rs of magnitu<strong>de</strong>. Thus it is<br />
used usually on a logarithmic scale:<br />
⎛ Z ⎞<br />
DBZ = 10 log ⎜ ⎟<br />
10⎜<br />
6 −3 ⎟<br />
(2.6)<br />
⎝ mm m ⎠<br />
This dimension-less quantity is usually called reflectivity as well. The statement that<br />
“reflectivity is 1000 mm 6 m –3 “ is synonymous to “reflectivity is 30 dBZ”. Both means:<br />
Z = 1000 mm 6 m –3 ⇔ DBZ = 30 = 30 dBZ. (Note that dBZ is a dimension-less “unit”).<br />
The measured reflectivity Zm <strong>de</strong>rived in such way is not necessarily equal to the real<br />
reflectivity Z. The simplification used in eqs. (2.2) to (2.5) can result in some<br />
problems, e.g. for non-liquid precipitation or strong attenuation. The consequences of<br />
such problems will be discussed in the next chapter. At the beginning of that<br />
chapter, the <strong>de</strong>rivation of rainfall data will be discussed. Later in that chapter, basic<br />
principles of polarimetric measurements will be presented.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 6
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
3 <strong>Quantitative</strong> <strong>Precipitation</strong> <strong>Estimation</strong> I – Aspects<br />
of Rainfall Rate Derivation<br />
In the last chapter, the basic principles of reflectivity measurements have been<br />
introduced. Several steps are only valid if certain assumptions are true. This present<br />
chapter will <strong>de</strong>al with the consequences of situations where the assumptions are not<br />
true in reality.<br />
The <strong>de</strong>rivation of rainfall data <strong>from</strong> reflectivity measurements has not been presented<br />
in the previous chapter, even though it is an important point of radar meteorology.<br />
This topic is <strong>de</strong>alt with in this chapter for the reason that the assumptions necessary<br />
for the <strong>de</strong>rivation of rainfall rate are never given exactly. The reasons and<br />
consequences of this are given in the same section as the <strong>de</strong>rivation of rainfall data<br />
(section 3.1).<br />
As already stated in the first chapter, the <strong>de</strong>rivation of precipitation amounts <strong>from</strong><br />
radar data can be divi<strong>de</strong>d into two steps<br />
i) the <strong>de</strong>rivation of rainfall intensity data (valid for a certain time point); and<br />
ii) the accumulation of rainfall intensity data, i.e. integration in time.<br />
This chapter presents different possible error sources and issues that have to be<br />
consi<strong>de</strong>red during the <strong>de</strong>rivation of rainfall intensity data. Available algorithms to<br />
overcome such errors are listed. One section <strong>de</strong>als with the problem of operability,<br />
i.e. real-time application of algorithms. Problems that arise with the accumulation in<br />
time will be discussed in the following chapter. Algorithms to adjust radar data by<br />
other-sensor data will not be presented at all, as they are not appropriate to WP 4.<br />
3.1 Drop Size Distributions and Z-R-Relations<br />
In the last chapter, the <strong>de</strong>finition of the reflectivity Z was given in eq. (2.3); and the<br />
way how to measure it (or, better, how to <strong>de</strong>rive the measured reflectivity Zm) was<br />
given in eq. (2.4). Assuming that the measured reflectivity is same than the real, the<br />
rainfall intensity R may be <strong>de</strong>rived. R is <strong>de</strong>fined as<br />
∞<br />
6 ∫<br />
0<br />
π 3<br />
R = D v(<br />
D)<br />
n(<br />
D)<br />
dD , (3.1)<br />
where v(D) <strong>de</strong>notes the fall velocity of a drop of the diameter D. To calculate the<br />
rainfall rate R <strong>from</strong> reflectivity Z, the fall velocity has to be known as well as the drop<br />
size distribution n(D). Of course these are given only in seldom cases. But lots of<br />
measurements have shown that the drop size distribution roughly follows an<br />
exponential law: n(D) = N0e -ΛD (e.g. Marshall and Palmer, 1948). If the rain drop fall<br />
velocity is expressed as v(D) = v0·(D/D0) P , for reflectivity Z and rainfall intensity R<br />
follows:<br />
Z = 6! N0 Λ –7 , (3.1a)<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 7
R =<br />
π v<br />
6 D<br />
0<br />
P<br />
0<br />
N<br />
0<br />
Γ(<br />
4 + P)<br />
Λ<br />
−4−P<br />
where Γ is the gamma function.<br />
Marshall and Palmer (1948) found<br />
Λ = 4,1 mm –1 (R / mm·h –1 ) –0,21<br />
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
, (3.1b)<br />
and N0 = 8000 mm –1 m –3 ,<br />
which gives <strong>from</strong> eq. (3.1a)<br />
Z = 296 mm 6 m –3 (R / mm·h –1 ) 1,47 (3.2)<br />
Substituting Λ <strong>from</strong> eq. (3.1b) with the result <strong>from</strong> Marshall and Palmer (1948) gives<br />
for the fall velocity relation P = 0,76 and v0 = 3,34 m/s (for D0 = 1 mm). This fall<br />
velocity relation is nearly i<strong>de</strong>ntical to that one <strong>de</strong>rived by Liu and Orville (1969) with<br />
direct measurements; they found P = 0,8 and v0 = 3,35 m/s for D0 = 1 mm. For very<br />
large drops (with diameters above about 4 mm) this relation gives too high velocities.<br />
The equation <strong>from</strong> Atlas et al. (1973) give better results then. The found a fall velocity<br />
relation of the form v(D) = v0 – v1·e –(0,6 D/mm) with v0 = 9,65 m/s and v1 = 10,3 m/s<br />
(note that this relation gives erroneous results for small drops).<br />
It is a common way in radar meteorology to express the relation between reflectivity<br />
and rainfall rate in a Z-R-relation of the form Z = a·R b (cf. eq. (3.2)), where reflectivity<br />
Z is expressed in mm 6 m –3 and rainfall rate R in mm h –1 . Battan (1973) lists several<br />
dozen Z-R-relations. The application of a certain Z-R-relation always implies a certain<br />
drop size distribution and a certain fall velocity law. The <strong>de</strong>viation of the reality <strong>from</strong><br />
these assumptions may result in large errors of R.<br />
Such errors can be reduced, if the precipitation type is classified e.g. as<br />
“thun<strong>de</strong>rstorm” or “drizzle” (Fiser, 2001) and then applying different Z-R-relations<br />
(see also section 3.7 for that topic). Austin (1987) found that the Z-R-relation has only<br />
small variations within one precipitation event of the same genesis, even if R and Z<br />
vary over a wi<strong>de</strong> range. But this method may sometimes lead to wrong results:<br />
Kreuels (1988) analysed Z-R-relations of more than ten years continuous<br />
measurements with a Joss-Waldvogel distrometer and found no correlation between<br />
precipitation type and Z-R-relation nor between season and Z-R-relation.<br />
For this reason, individual drop size distributions measured by distrometers may be<br />
taken as input for improved, site- and time-<strong>de</strong>pen<strong>de</strong>nt Z-R-relations. This procedure<br />
must not be confused with adjustment: adjustment means that radar <strong>de</strong>rived R data<br />
are corrected in a post-processing step, whereas selections of distrometer-<strong>de</strong>rived Z-<br />
R-relations do not care about the absolute values of R <strong>from</strong> the distrometer and the<br />
corresponding radar data.<br />
Another problem arises with inhomogeneous beam filling and with data interpolation.<br />
If the scatterers do not fill the pulse volume homogeneously, the radar will measure a<br />
mean reflectivity factor. Due to the non-linearity of the Z-R-relation, the rainfall<br />
intensity <strong>de</strong>rived <strong>from</strong> that mean reflectivity will be different <strong>from</strong> a mean rainfall rate<br />
<strong>de</strong>rived <strong>from</strong> the Z-distribution within the pulse volume (if the latter could be<br />
measured at all).<br />
In a similar way data interpolation influences the results. Whereas the measured data<br />
are based on a polar grid, rainfall data mostly have to be <strong>de</strong>rived on a cartesian<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 8
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
system. If a cartesian grid point is interpolated <strong>from</strong> several polar co-ordinates, the<br />
interpolation could be done in the Z data, or in the DBZ data, or in the R(Z) data.<br />
Again, three different results will occur and the interpolation in R does not always<br />
give the best results.<br />
3.2 Clutter Filtering and Speckle Removing<br />
Clutter Filtering<br />
Clutter refers to all non-meteorological echoes that influence the radar data quality.<br />
Very often clutter is used as a synonym for ground clutter, meaning all returns <strong>from</strong><br />
the earth’s surface. The echoes <strong>from</strong> the ground are usually the strongest of all<br />
echoes, at least close to the radar. This makes removal or at least reduction of clutter<br />
necessary.<br />
Different techniques of clutter filtering may be used:<br />
• Doppler filter or statistical filter (applied in the signal processor): In the first case,<br />
the Doppler spectrum of the received signal is analysed and a narrow band width<br />
around zero velocity is removed or interpolated. Doppler filter require a Doppler<br />
radar, which fortunately has become standard in recent years. In the latter case,<br />
the long correlation time of ground clutter is used to i<strong>de</strong>ntify clutter by the pulseto-pulse<br />
changes of reflectivity samples.<br />
• Clutter maps (usually applied as a post-processing after the signal processor):<br />
The reflectivity values sampled at fair-weather conditions are store in a file. For all<br />
subsequent scans, the clutter map amount is subtracted <strong>from</strong> the measured<br />
reflectivity.<br />
• Sophisticated clutter suppression schemes which are a combination of the above<br />
mentioned and may take into account other information like terrain data. An<br />
example is <strong>de</strong>scribed in Lee et al. (1995).<br />
It must be noted that no clutter filter will work perfectly. Of course the solid earth is<br />
not moving, but the radar antenna is, and so are plants blown by the wind. Thus<br />
clutter signals have a velocity and spectrum width differing <strong>from</strong> zero and may<br />
sometimes not be separable <strong>from</strong> a weather echo. The clutter intensity is not<br />
constant over time, but changing e.g. due to a wet coating. A clutter filter may also<br />
remove large parts of the weather signal, often if the radial velocity of the weather is<br />
close to zero and if the spectrum of the weather signal is narrow (as e.g. in stratiform<br />
snow).<br />
Sea clutter caused by waves is a severe problem, because it has velocities<br />
significantly differing <strong>from</strong> zero (see figure 3.1). But as in all ground clutter, it is<br />
strongest in the lowest elevation scans and always related to the same positions.<br />
Thus it can be i<strong>de</strong>ntified by special algorithms.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 9
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
Figure 3.1: Examples of sea clutter. In the upper image showing reflectivity data <strong>from</strong> a Norwegian<br />
radar, the sea clutter is visible only in those sectors, where the valleys in the vicinity of the radar allow<br />
direct sight to the sea. The data of the lower images were obtained <strong>from</strong> a radar in Taiwan located at<br />
the coast line. Thus the sea clutter covers a wi<strong>de</strong> range up to about 50 km distance, beyond which the<br />
earth curvature became effective. The velocity data (lower right image) illustrate the wave motion of<br />
about 4 m/s <strong>from</strong> the Northeast.<br />
Other clutter may appear <strong>from</strong> time to time at random locations and thus is not<br />
removable by real-time algorithms. Such clutter types are<br />
• Echoes <strong>from</strong> birds or insects<br />
• Echoes <strong>from</strong> aircraft, balloons, ships or trains<br />
• Echoes <strong>from</strong> artificial atmospheric tracers (chaff)<br />
But even the normal clutter of a given radar site may sometimes be enlarged<br />
significantly: In cases of anomalous beam propagation (refraction), caused e.g. <strong>from</strong><br />
low-level temperature inversions, the radar beam may hit the ground at places were<br />
this is not possible un<strong>de</strong>r normal conditions. In such cases, application of clutter<br />
maps may fail completely. Nevertheless algorithms have been <strong>de</strong>veloped to i<strong>de</strong>ntify<br />
such conditions and to reduce the clutter influence (Harrison et al., 2000; Kessinger<br />
et al., 2001).<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 10
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
In general it can be stated that the clutter contamination is less for higher elevation<br />
angles. Thus the use of higher elevation angles may be applicable for some<br />
purposes. See section 3.10 (scan strategy) for more aspects on this topic.<br />
Speckle removing<br />
As even the best clutter removal techniques may fail in some conditions, clutter<br />
contaminated data may remain. Such data may come up as isolated speckles or as<br />
points with abnormally large magnitu<strong>de</strong> and can be interpolated by surrounding<br />
measurements. Receiver noise speckles can be eliminated in the same way. Speckle<br />
removing can be set up in the signal processor or can be applied as a postprocessing<br />
step (e.g. Fulton et al., 1998).<br />
3.3 Attenuation<br />
Attenuation is a serious problem in radar meteorology. Every material on the way<br />
between antenna and target interfere with the radiation. Some part of the radiation is<br />
lost due to this interference. Attenuation consists of absorption and scattering. The<br />
atmospheric gases cause attenuation as well as the radome used for protection of<br />
radars. The scattering of radiation in atmospheric particles is used to measure<br />
reflectivity; this implies that meteorological targets cause attenuation as well. Some<br />
attenuators have effects which are more or less constant with time and thus can be<br />
corrected quite easy. Attenuation is <strong>de</strong>pen<strong>de</strong>nt of the radar’s wavelength; for<br />
precipitation, it is strong for X-Band radars and weak for S-band radars. The most<br />
important attenuation effects are listed in this section.<br />
Radome Attenuation<br />
The attenuation caused by the a dry radome <strong>de</strong>pends only weakly <strong>from</strong> antenna<br />
direction and thus can be consi<strong>de</strong>red as constant. It can be corrected in the radar<br />
calibration itself. Nevertheless caution must be taken at radome <strong>de</strong>sign for the<br />
arrangement of radome joints; they shall produce as least scatter as possible (Manz<br />
et al., 1998).<br />
For precipitation estimates, the change of radome attenuation due to water on its<br />
surface may become quite large: one-way attenuation of several dB even for<br />
mo<strong>de</strong>rate rain has been reported (Manz et al., 1998; Löffler-Mang and Gysi, 1998);<br />
for strong rain, this might make a quantitative rainfall estimation impossible. For such<br />
reasons, radome surfaces usually have hydrophobic coating. Radomes should be<br />
cleaned <strong>from</strong> time to time, because dirt could eliminate the hydrophobic coating<br />
effect completely.<br />
Attenuation in Gases (Air)<br />
The attenuation in atmospheric gases is <strong>de</strong>pending on air <strong>de</strong>nsity which mainly is a<br />
function of height. Additionally the composition of air, mainly the amount of water<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 11
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
vapour, influences the gaseous attenuation. Typically values of two-way gas<br />
attenuation lie around 1dB / 100km (for S-Band radars). For this values being small<br />
compared to possible attenuation in rain and because the variations are not very<br />
large, the gas attenuation is usually corrected in the signal processor, where the<br />
product of a hardware-specific constant and the distance is ad<strong>de</strong>d to the measured<br />
values.<br />
Attenuation in <strong>Precipitation</strong><br />
If the precipitation particles consist of drops being small compared to the radar<br />
wavelength, the Rayleigh approximation can be applied. Then the absorption is<br />
proportional to the integral over the drops’ diameter to the power of 3, and the<br />
scattering is proportional to the integral over the drops’ diameter to the power of 6.<br />
Thus for the attenuation, which is the sum of absorption and scattering, the extinction<br />
coefficient σE can be approximated by<br />
∞<br />
∫<br />
0<br />
α<br />
σE = c D n(<br />
D)<br />
dD , (3.3)<br />
where c is a constant and α an exponent between 3 and 6 (in most cases between<br />
3.5 and 4.0), <strong>de</strong>pending <strong>from</strong> the wavelength. From the Marshall and Palmer (1948)<br />
results, the rainfall rate can be written as<br />
∞<br />
2 ∫<br />
0<br />
3.<br />
76<br />
R = c D n(<br />
D)<br />
dD,<br />
(3.4)<br />
(cf. the discussion of the eqs.(3.1) to (3.2)). This indicates, that for typical weather<br />
radars, the absorption might be directly proportional to the rainfall rate and<br />
in<strong>de</strong>pen<strong>de</strong>nt of the drop size distribution, namely if α = 3.76. This statement is best<br />
fulfilled for K-band radars (≈1 cm wavelength). For other wavelengths, attenuation<br />
relations of the form σE = c3·R β can be applied (with β around 1.0). See Doviak and<br />
Zrnic (1993) for further <strong>de</strong>tails.<br />
Such formulas can be used to calculate the attenuation <strong>from</strong> rainfall rate, which is<br />
calculated <strong>from</strong> the measured reflectivity. With this method, each ray of radar data<br />
can be corrected step by step.<br />
It has to be noted that the above consi<strong>de</strong>rations are only valid for small raindrops. In<br />
case of large raindrops, when the Rayleigh approximation becomes invalid, or in<br />
presence of snow or hail, attenuation correction formula may give wrong results. For<br />
C-Band radars, the attenuation <strong>from</strong> light rain can be neglected, for S-Band radar<br />
even <strong>from</strong> mo<strong>de</strong>rate to strong rain. In such cases, only strong rain cells give<br />
significant attenuation in relatively small sectors. The echoes <strong>from</strong> behind the cell in<br />
this sectors can be compared with neighboured, not attenuated rays to <strong>de</strong>rive the<br />
total attenuation <strong>from</strong> one rain cell (Upton and Fernán<strong>de</strong>z-Durán, 1998). However,<br />
such method is difficult to implement on real-time processing.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 12
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
3.4 Vertical Profiles and Beam Blocking Corrections<br />
<strong>Radar</strong>s which are located in mountain regions have to <strong>de</strong>al with the difficulty that<br />
several valleys cannot be “seen” by the radar. Thus information <strong>from</strong> upper-level<br />
elevations have to be extrapolated to the ground. A first attempt is to assume<br />
constant reflectivity. This gives good results only if no major microphysical processes<br />
take place in the atmospheric layers below the lowest accessible elevation. But<br />
several processes may change the reflectivity Z or the rain rate R during the falling of<br />
hydrometeors, which gives different vertical profiles (e.g. Huggel et al., 1996):<br />
• Evaporation <strong>de</strong>creases Z and R<br />
• Con<strong>de</strong>nsation increases Z and R<br />
• Particle type conversion (e.g. melting snow) changes Z but not R<br />
• Particle coagulation or disruption changes Z but not R<br />
• Increasing friction through <strong>de</strong>nser air increases Z but does not change R<br />
Usually several of this processes will happen simultaneously. For such effects, a<br />
vertical profile correction which is <strong>de</strong>pending on location, time, season or weather<br />
condition may give better results than assuming constant reflectivity (Germann, 1998;<br />
Germann and Joss, 2001; Germann and Joss, 2000). Such profiles can be <strong>de</strong>rived<br />
<strong>from</strong> radar volume data or may be set up by the radar operator.<br />
If a radar is surroun<strong>de</strong>d by mountains, several rays at lower elevations will be<br />
blocked totally by mountains as <strong>de</strong>scribed above. But a lot of beams will be blocked<br />
only partially. If the blocked part is not too large, the corresponding reflectivity data<br />
can be corrected <strong>from</strong> geometric consi<strong>de</strong>ration of the beam using high-precision<br />
digital terrain data. For this correction, inhomogeneous scatterer distribution may also<br />
be consi<strong>de</strong>red (Hannesen, 1998, chapter 3; Hannesen and Löffler-Mang, 1998). See<br />
figure 3.2 for an example.<br />
a)<br />
<strong>Precipitation</strong> im mm <strong>de</strong>rived <strong>from</strong><br />
non-corrected radar data<br />
25<br />
20<br />
15<br />
10<br />
5<br />
Distance smaller than 60 km<br />
Distance between 60 and 90 km<br />
Distance larger than 90 km<br />
0<br />
0 5 10 15 20 25<br />
<strong>Precipitation</strong> in mm at ground<br />
b)<br />
<strong>Precipitation</strong> in mm <strong>de</strong>rived <strong>from</strong><br />
corrected radar data<br />
0<br />
0 5 10 15 20 25<br />
<strong>Precipitation</strong> in mm at ground<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 13<br />
25<br />
20<br />
15<br />
10<br />
5<br />
Distance smaller than 60 km<br />
Distance between 60 and 90 km<br />
Distance larger than 90 km<br />
Figure 3.2: Comparison of 24h radar <strong>de</strong>rived rainfall data (left axis) with rain gauge data (abscissa)<br />
(<strong>from</strong> Hannesen, 1998, chapter 3). The left image shows uncorrected radar data. For the right image,<br />
the radar date were corrected for partial beam filling. Solid squares indicate gauges closer than 60 km<br />
to the radar, open squares between 60 and 90 km, and crosses more than 90 km away <strong>from</strong> the radar.
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
It should be noted, that even in flat terrain a vertical profile correction may be<br />
necessary. If the lowest elevation angle is set e.g. to 0.5 <strong>de</strong>gree, this results in an<br />
altitu<strong>de</strong> of about 1,5 km above the radar in 100 km distance, and of about 4 km in<br />
200 km distance. If the area of investigation is very small and close to the radar, the<br />
assumption of constant reflectivity may give sufficient results.<br />
3.5 Orographic Enhancement<br />
If warm, moist air is flowing over hills, con<strong>de</strong>nsation processes can occur. Sometimes<br />
this con<strong>de</strong>nsation increases significantly already falling precipitation <strong>from</strong> upper<br />
altitu<strong>de</strong>s. This additional con<strong>de</strong>nsation is very often limited to a shallow layer directly<br />
above the surface and is called orographic enhancement. Due to the con<strong>de</strong>nsation<br />
processes, the drop size distribution is changed significantly and thus also reflectivity<br />
Z, rainfall rate R and the Z-R-relation itself. These phenomena are related to certain<br />
weather conditions as the warm sector of extra-tropical cyclones (Kitchen et al.,<br />
1994; Neimann et al., 2001; White et al., 2001).<br />
Due to their small height, areas of orographic enhancement are difficult to <strong>de</strong>tect.<br />
Furthermore, the corresponding radar data may be contaminated by ground clutter<br />
<strong>from</strong> the mountains that cause the enhancement. As orographic enhancement is<br />
correlated to special weather conditions, algorithms for their correction may need<br />
other information besi<strong>de</strong>s radar data alone. The correction algorithms have to take<br />
into account the specific site conditions, mainly orography, as well. Nevertheless,<br />
some real-time correction schemes exist (e.g. Kitchen et al., 1994).<br />
3.6 Bright Band, Snow and Hail<br />
In the previous sections, the assumption of small liquid water droplets has often been<br />
ma<strong>de</strong> to <strong>de</strong>scribe phenomena or correction schemes. In reality, non-liquid<br />
precipitation particles like snow or hail may occur in those layers which are taken for<br />
the rainfall rate estimation <strong>from</strong> reflectivity data. In such cases, the simple equations<br />
presented in the preceding sections may be not valid. The consequences will be<br />
discussed in this section.<br />
The Bright Band<br />
When snow flakes or ice crystals begin to melt, the cover with a thin water film. Due<br />
to the fact that the dielectric coefficient |K| 2 for water is about five times higher than<br />
for ice, the reflectivity increases. Theoretical calculations have shown that a small<br />
water film may result in nearly the same backscattering cross section of the particle<br />
as if it was completely liquid. Snow flakes contain large parts of air, but regarding the<br />
backscattering cross section these parts have no large effect: they appear as if they<br />
were completely filled with ice. During the melting phase, the particles tend to<br />
accumulate together which increases Z further.<br />
When the particles continue to fall down, they become more and more liquid. Thus<br />
their volume <strong>de</strong>creases and their fall speed increases until they have become liquid<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 14
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
droplets. Both effects reduce Z significantly: the volume loss because of the D 6<br />
<strong>de</strong>pen<strong>de</strong>nce, and the increasing fall velocity because the volume-specific particle<br />
<strong>de</strong>nsity is <strong>de</strong>creased.<br />
The above <strong>de</strong>scribed effects result in a horizontal layer of increased reflectivity (e.g.<br />
Austin and Bemis, 1950), which is called the Bright Band. This name comes <strong>from</strong> the<br />
early time of radar meteorology with analogue displays, where the brightness<br />
represented the reflectivity. The bright band appeared in vertical cross sections as a<br />
thin band of very bright echo display. Fig. 3.3 shows an example of a bright band.<br />
Fig. 3.3: Vertical cross section through a bright band. The used grey scale would make the name<br />
“Dark Band” more appropriate here (<strong>from</strong> Hannesen, 1998).<br />
The above mentioned <strong>de</strong>crease of reflectivity due to volume reduction and increased<br />
fall speed at the final stages of the melting process are stronger pronounced in<br />
stratiform precipitation, which mainly consists of slow falling ice crystals and snow<br />
flakes above the freezing level, than in convective precipitation, which mainly contain<br />
faster falling grauple-like particles in the ice phase. These different precipitation types<br />
cause different microphysical processes below the bright band; thus typical bright<br />
band profiles for different precipitation types can be obtained (Huggel et al., 1996).<br />
If radar data are not corrected for the effects of bright band, too high rainfall<br />
estimates will be the result in those measurement points that are affected. Thus<br />
several bright band i<strong>de</strong>ntification and correction algorithms have been <strong>de</strong>veloped<br />
(e.g. Kitchen et al., 1994; Gysi et al., 1997), which provi<strong>de</strong> significant improvements<br />
in radar rainfall estimates. Such algorithms also have to care about far-distance<br />
measurements, where the vertical extent of the radar beam is much larger than the<br />
bright band thickness, and for the data <strong>from</strong> above the bright band with ice particles,<br />
where snow Z-R-relations should be applied (see also below).<br />
Bright band <strong>de</strong>tection algorithms may be used as an estimator for the freezing level<br />
and provi<strong>de</strong> information for forecasters. On the other hand, if the freezing level is<br />
known and given as information supplement for a bright band <strong>de</strong>tection algorithm, the<br />
algorithm will have more reliable results than if it would only perform on radar data.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 15
Snow<br />
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
Snow flakes and ice crystals cannot be <strong>de</strong>scribed by a simple radar equation<br />
because of their irregular shape and orientation. The effect of the different dielectric<br />
constant may easily be taken into account, but snow flakes can become as large as<br />
5 cm, whereas ice crystals typically have sizes of about one mm. As mentioned<br />
above, the large amounts of air insi<strong>de</strong> snow flakes do not allow a common relation<br />
between the particle diameter and its backscattering cross section. Thus a reflectivity<br />
as in eq. (2.3) makes no sense for snow flakes. So for ice particles, reflectivity<br />
usually means measured reflectivity. Instead of rainfall rate, the equivalent rain rate is<br />
taken, which basically means the amount of liquid water if the ice was melted.<br />
Due to the variations of snow shape, orientation and size, the parameters of<br />
applicable Z-R-relations cover a wi<strong>de</strong> range. Several Z-R-relations for different snow<br />
types have been obtained and are given in the literature (e.g. Battan, 1973). But it<br />
must be noted that the errors affecting precipitation estimates <strong>from</strong> radar <strong>de</strong>rived<br />
snow reflectivity data usually are larger than for liquid rain.<br />
Hail<br />
Hail can cause strong damage on agricultural plants and on infrastructure. It can<br />
occur in all heights of the troposphere and is thus difficult to <strong>de</strong>tect by singleparameter<br />
radar. Probabilities of hail occurrence can be <strong>de</strong>rived regarding the<br />
following theoretical assumption: Above the freezing level, besi<strong>de</strong>s hail only supercooled<br />
water droplets and ice crystals and snow flakes can occur. Super-cooled<br />
water droplets are very small and thus have negligible reflectivity. Snow flakes may<br />
have large sizes (and thus large reflectivity values) only around the melting layer,<br />
where coagulation is very likely. Thus if strong echoes appear somewhat above the<br />
freezing level, the occurrence of hail is very likely. The hail probability is larger, the<br />
larger the vertical extent of such data above the freezing level is (Waldvogel et al.,<br />
1979). It must be kept in mind, that for typical summer thun<strong>de</strong>rstorms, the hail will<br />
melt in most cases on its way down to the earth’s surface.<br />
Hail has no large direct influence on hydrology, but it affects the Z-R-relation: A few<br />
large hailstones will result in very large values of Z, but the corresponding R remains<br />
too weak; thus R may be over-estimated. Large hail does no longer fulfil the Rayleigh<br />
approximation, even for S-Band radar. As a consequence, not only the R estimation<br />
may be erroneous in case of hail, but also attenuation correction algorithms may fail<br />
completely. These errors can be reduced if the presence of hail can be i<strong>de</strong>ntified with<br />
sufficient accuracy; some hail Z-R-relations exist in literature. As <strong>de</strong>scribed above, a<br />
single-parameter radar has only limited chances to <strong>de</strong>tect hail. Much better results<br />
are obtained using polarimetric radars (see section 3.8)<br />
3.7 Stratiform and Convective <strong>Precipitation</strong><br />
The meteorological conditions and thus microphysical processes are different for<br />
stratiform and convective precipitation. The main difference is the vertical air velocity,<br />
which is much larger in convective cases (a few m/s compared to several cm/s). The<br />
microphysical processes result in different drop size distributions for stratiform and<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 16
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
convective precipitation and thus in different Z-R-relations. Battan (1973) lists Z-Rrelations<br />
for various weather types. But it must be kept in mind that sometimes the<br />
correlation between the weather type and the Z-R-relation is poor (Kreuels, 1988;<br />
see section 3.1).<br />
Due to the meteorological forcing, stratiform precipitation events exhibit different<br />
patterns than convective precipitation. In the first case, horizontal variations of the<br />
reflectivity are small, and a bright band is likely to appear (if the clouds top over the<br />
freezing level). In the latter case, the horizontal fluctuations are quite large, whereas<br />
vertical reflectivity gradients are weaker, at least in the lowest few kilometers. So<br />
three-dimensional properties of the reflectivity data can be used so separate<br />
convective <strong>from</strong> stratiform precipitation (Rosenfeld et al., 1995; Hannesen, 1998,<br />
chapter 4; Germann and Joss, 2001). The results can be used to apply for different<br />
Z-R-relations or for different vertical profile corrections in case of blocked lowelevation<br />
beams.<br />
3.8 Dual Polarisation <strong>Radar</strong>s<br />
Conventional weather radars transmit a horizontally polarised electromagnetic wave.<br />
Some type of radars use dual polarisation techniques: they have the capability to<br />
transmit and receive horizontally and vertically polarised radiation. This allows the<br />
measurement of several polarimetric quantities (e.g. Doviak and Zrnic, 1993; Zrnic<br />
and Ryzhkov, 1999):<br />
The differential reflectivity ZDR is <strong>de</strong>fined as<br />
10 h Zv<br />
= ⎜<br />
⎛ 2<br />
2<br />
10 log s<br />
⎟<br />
⎞<br />
hh svv<br />
, (3.5)<br />
⎝<br />
⎠<br />
where Zh is the reflectivity <strong>from</strong> horizontal polarisation, and Zv <strong>from</strong> vertical<br />
polarisation. sij are members of the backscattering matrix of the particles (i,j = h,v).<br />
Spherical particles like small drops give Zh = Zv, thus ZDR = 0. Large drops are oblate<br />
and thus give positive values of ZDR.<br />
ZDR = log ( Z )<br />
The linear <strong>de</strong>polarisation ratio LDR is <strong>de</strong>fined as<br />
10 hv Zh<br />
= ⎜<br />
⎛ 2<br />
2<br />
10 log s<br />
⎟<br />
⎞<br />
hv shh<br />
, (3.6)<br />
⎝<br />
⎠<br />
where Zhv is the reflectivity received on the vertical channel at transmission on the<br />
horizontal channel. For spherical particles as well as for oblate particles with<br />
horizontal axis, Zhv vanishes and thus LDR becomes negative infinite. But if oblate<br />
particles are tumbling and thus have a tilted axis, Zhv becomes different <strong>from</strong> zero. It<br />
is typically some or<strong>de</strong>rs of magnitu<strong>de</strong> smaller than Zh, thus LDR lies somewhere<br />
between -40 dB and -20 dB.<br />
LDR = log ( Z )<br />
The differential phase ΦDP is <strong>de</strong>fined as<br />
ΦDP = Φhh – Φvv, (3.7)<br />
where Φhh and Φvv are the integrated phase shift of radiation along the transmission<br />
path for horizontal and vertical polarisation, respectively.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 17
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
The specific differential phase KDP is <strong>de</strong>fined as<br />
∆ ΦDP<br />
KDP = . (3.8)<br />
2 ∆r<br />
It is the change of the differential phase with range. From theoretical consi<strong>de</strong>rations,<br />
KDP is almost proportional to the rain rate and only little <strong>de</strong>pen<strong>de</strong>nt of the drop size<br />
distribution.<br />
The correlation coefficient (at zero lag) ρHV is <strong>de</strong>fined as<br />
ρHV = s vvs*<br />
hh<br />
⎡<br />
⎢<br />
⎣<br />
2<br />
shh<br />
1/<br />
2<br />
svv<br />
2<br />
1/<br />
2 ⎤<br />
⎥ ,<br />
⎦<br />
(3.9)<br />
where s*hh is a member of the backscattering covariance matrix.<br />
The rea<strong>de</strong>r is referred to Doviak and Zrnic (1993), Chapter 8, for further theoretical<br />
background and <strong>de</strong>tails of these quantities.<br />
<strong>Quantitative</strong> precipitation estimation can be improved significantly, if polarimetric<br />
measurements are taken into account. To illustrate this, let us consi<strong>de</strong>r the following<br />
example: Two drop size distributions may be given which result in the same rainfall<br />
rate, but the first one contains some large drops and relatively few small drops (like in<br />
a light shower), whereas the second distribution consists of many small drops and no<br />
large drops (as in a <strong>de</strong>nse drizzle). Due to the power-of-6 law, the reflectivity of the<br />
first distribution will be larger than for the second. With Z-R-relations, this problem<br />
cannot be solved. Now the large drops are oblate, thus the first distribution will give a<br />
higher differential reflectivity ZDR than the second one. This means that taking ZDR<br />
data into account, <strong>de</strong>viations <strong>from</strong> a standard drop size distribution may be resolved<br />
at the rainfall estimation and give better results than using Z-R relations (e.g.<br />
Ryzhkov et al., 2001).<br />
Several Z-ZDR-R-relations have been <strong>de</strong>rived <strong>from</strong> theoretical calculations and<br />
measurements. Table 3.1 shows an example which also illustrates our above<br />
example consi<strong>de</strong>rations (using R = 0.0076 Z 0.93 10 –0.281 ZDR ; <strong>from</strong> Gorgucci et al.,<br />
1994): Same R values result <strong>from</strong> higher reflectivity, if ZDR increases. The table also<br />
contains R data obtained <strong>from</strong> a standard Z-R-relation (Z = 200 R 1.6 ). These values<br />
are <strong>de</strong>noted by the thin green line across the Z-ZDR-table. A look at the<br />
corresponding ZDR scale shows, that these standard Z-R-relation implies differential<br />
reflectivity around zero (being typical for small drops), if the reflectivity is weak,<br />
whereas ZDR rises up to a few dB for high reflectivity values (caused by several large,<br />
oblate drops) according to a standard Z-R-relation based on a standard exponential<br />
drop size distribution.<br />
Table 3.1 gives a hint to the danger that lies in the application of such Z-ZDR-Rrelations:<br />
by the red lines, the bor<strong>de</strong>r of rain rate values within a factor of five around<br />
the standard case are indicated (for constant DBZ). One can see that for a given<br />
reflectivity value, a change of ZDR by about 2 dB causes a factor of five in rainfall rate.<br />
This must be compared with the necessary change in reflectivity: for DBZ, a change<br />
of 7 dB is necessary (for constant ZDR) to change the rain rate by a factor of five, and<br />
using standard Z-R-relations, DBZ needs a change of about 10 dB to bias R by such<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 18
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
Standard Z-R a = 200 b = 1.6<br />
R (mm/h) 0,2 0,3 0,6 1,3 2,7 5,6 11,5 23,7 48,6 99,9 205,0 421,1 864,7<br />
R (mm/h) dBZ<br />
ZDR (dB) 10 15 20 25 30 35 40 45 50 55 60 65 70<br />
5 0,0 0,0 0,0 0,1 0,2 0,5 1,6 4,6 13,4 39,0 113,7 331,8 967,9<br />
4,75 0,0 0,0 0,0 0,1 0,2 0,6 1,8 5,4 15,7 45,8 133,7 390,0 1137,8<br />
4,5 0,0 0,0 0,0 0,1 0,3 0,7 2,2 6,3 18,5 53,9 157,1 458,5 1337,6<br />
4,25 0,0 0,0 0,0 0,1 0,3 0,9 2,6 7,4 21,7 63,3 184,7 539,0 1572,4<br />
4 0,0 0,0 0,0 0,1 0,4 1,0 3,0 8,7 25,5 74,4 217,2 633,6 1848,5<br />
3,75 0,0 0,0 0,0 0,1 0,4 1,2 3,5 10,3 30,0 87,5 255,3 744,8 2173,0<br />
3,5 0,0 0,0 0,1 0,2 0,5 1,4 4,1 12,1 35,3 102,9 300,1 875,6 2554,5<br />
3,25 0,0 0,0 0,1 0,2 0,6 1,7 4,9 14,2 41,5 120,9 352,8 1029,4 3003,1<br />
3 0,0 0,0 0,1 0,2 0,7 2,0 5,7 16,7 48,7 142,2 414,8 1210,1 3530,3<br />
2,75 0,0 0,0 0,1 0,3 0,8 2,3 6,7 19,6 57,3 167,1 487,6 1422,5 4150,1<br />
2,5 0,0 0,0 0,1 0,3 0,9 2,7 7,9 23,1 67,3 196,5 573,2 1672,3 4878,8<br />
2,25 0,0 0,0 0,1 0,4 1,1 3,2 9,3 27,1 79,2 231,0 673,9 1965,9 5735,4<br />
2 0,0 0,1 0,2 0,4 1,3 3,7 10,9 31,9 93,1 271,5 792,2 2311,1 6742,4<br />
1,75 0,0 0,1 0,2 0,5 1,5 4,4 12,9 37,5 109,4 319,2 931,2 2716,8 7926,2<br />
1,5 0,0 0,1 0,2 0,6 1,8 5,2 15,1 44,1 128,6 375,2 1094,7 3193,8 9317,8<br />
1,25 0,0 0,1 0,2 0,7 2,1 6,1 17,8 51,8 151,2 441,1 1287,0 3754,6 ######<br />
1 0,0 0,1 0,3 0,8 2,5 7,2 20,9 60,9 177,8 518,6 1512,9 4413,8 ######<br />
0,75 0,0 0,1 0,3 1,0 2,9 8,4 24,6 71,6 209,0 609,6 1778,5 5188,8 ######<br />
0,5 0,0 0,1 0,4 1,2 3,4 9,9 28,9 84,2 245,6 716,7 2090,8 6099,8 ######<br />
0,25 0,1 0,2 0,5 1,4 4,0 11,6 33,9 99,0 288,8 842,5 2457,9 7170,7 ######<br />
0 0,1 0,2 0,6 1,6 4,7 13,7 39,9 116,4 339,5 990,4 2889,4 8429,7 ######<br />
-0,25 0,1 0,2 0,6 1,9 5,5 16,1 46,9 136,8 399,1 1164,3 3396,8 9909,8 ######<br />
-0,5 0,1 0,3 0,8 2,2 6,5 18,9 55,1 160,8 469,2 1368,7 3993,1 ###### ######<br />
-0,75 0,1 0,3 0,9 2,6 7,6 22,2 64,8 189,0 551,5 1609,0 4694,2 ###### ######<br />
-1 0,1 0,4 1,1 3,1 8,9 26,1 76,2 222,2 648,4 1891,5 5518,4 ###### ######<br />
Table 3.1: Rainfall rates R <strong>de</strong>rived <strong>from</strong> Z and ZDR. R values <strong>from</strong> a standard Z-R-relation are given at<br />
the top, the corresponding values are indicated by the green line in the Z-ZDR-table. The red lines<br />
illustrate the bor<strong>de</strong>r of all R data within a factor of five around the standard Z-R data.<br />
amount. This illustrates that the ZDR measurement and the corresponding calibration<br />
must be done very precisely (better than 0.2 dB) to obtain reliable results.<br />
But even with such precise equipment, Z-ZDR-R-relations give erroneous results in<br />
case of hail or snow. If for example a strong thun<strong>de</strong>rstorm contains a few large hail<br />
stones, these hail causes very high reflectivity. Compared to standard Z-R-relations,<br />
the real rainfall rate is then smaller, thus too high rain rates would be obtained. But<br />
applying a Z-ZDR-R-relations as in table 3.1 would make this estimation much worse:<br />
hail stones are random oriented and thus ZDR will be around zero. For high<br />
reflectivity, the corresponding rainfall rate may be more than one or<strong>de</strong>r of magnitu<strong>de</strong><br />
higher than R <strong>de</strong>rived by a standard Z-R-relation <strong>from</strong> reflectivity data alone. And<br />
even this R is too high compared to reality.<br />
For C-Band radars, attenuation becomes a severe problem, because the attenuation<br />
is stronger for horizontal reflectivity than for vertical in the case of large, oblate drops;<br />
thus ZDR is biased to negative values.<br />
According to theoretical consi<strong>de</strong>rations, KDP is almost proportional to the rainfall rate<br />
and only weakly <strong>de</strong>pen<strong>de</strong>nt of the drop size distribution. Thus <strong>de</strong>riving R <strong>from</strong> KDP is<br />
more reliable than using Z-R-relations or Z-ZDR-R-relations. Furthermore KDP<br />
measurements are very robust to partial beam blocking and attenuation effects.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 19
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
Reliable results have also been obtained using ZDR-KDP-R-relations, i.e. <strong>de</strong>riving<br />
rainfall rate <strong>from</strong> differential reflectivity and specific differential phase data. This has<br />
been proven in several measurements (e.g. Ryzhkov et al., 2001).<br />
KDP-R-processing has one major disadvantage: As KDP is a <strong>de</strong>rivative, noise may<br />
have significant influence and could cause negative KDP data. This makes it<br />
necessary to average the measurements in radial direction, losing the advantage of<br />
high resolution (Illingworth, 2001). Furthermore, anisotropic scatterers or non-<br />
Rayleigh conditions will bias KDP-R- or ZDR-KDP-R-relations.<br />
Besi<strong>de</strong>s more accurate rainfall estimates, dual polarisation data offer the possibility to<br />
i<strong>de</strong>ntify the main hydrometeor type within a radar pulse volume. Hail can be <strong>de</strong>tected<br />
<strong>from</strong> dual polarisation radars (e.g. Nanni et al., 1998). Sophisticated precipitation<br />
type <strong>de</strong>termination algorithms use all available polarimetric quantities to distinguish<br />
between rain, drizzle, hail, dry snow, wet snow and so on (e.g. Lim et al., 2001).<br />
Until now, only few dual polarisation radars are available, but polarimetric<br />
measurements cover a wi<strong>de</strong> range of present research. So in future possibly dual<br />
polarisation radar become available also for operational observations.<br />
3.9 Dual Wavelength <strong>Radar</strong>s<br />
As discussed in section 3.3, attenuation due to precipitation is stronger for X-Band<br />
than for S-Band radars. For S-Band radars it can be neglected for all cases except<br />
very intense rain cells. This circumstances can be used for the application of dual<br />
wavelength radars: If a radar has for example the possibility to <strong>de</strong>tect reflectivity data<br />
<strong>from</strong> and S-Band and X-Band system simultaneously, the total attenuation of the X-<br />
Band data can be calculated for each position in each ray. Differentiating with respect<br />
to the range, the attenuation coefficient for the X-Band can be <strong>de</strong>rived for each<br />
measurement point. According to eqs. (3.3) and (3.4), this attenuation can be<br />
calculated into rainfall rate R with very robust relations, which are almost<br />
in<strong>de</strong>pen<strong>de</strong>nt of the drop size distributions (Doviak and Zrnic, 1993).<br />
The rainfall rate is <strong>de</strong>rived <strong>from</strong> a differential, thus even weak noise in the reflectivity<br />
data might cause significant errors, maybe even negative rain rates. Some radial<br />
averaging may become necessary to overcome these errors. But this means a loss in<br />
one of the major advantages of the radar: its high spatial resolution. Fortunately the<br />
noise effects are negligible in case of strong precipitation, correlated to strong<br />
differential attenuation. And these are the important cases for hydrological purposes,<br />
where the robustness of this method has its advantages compared to rainfall<br />
estimation <strong>from</strong> single parameter radars. Unfortunately only few dual wavelength<br />
radars are available.<br />
Again it should be noted that in some weather conditions this method may become<br />
erroneous: The attenuation-rainfall-relation is based on the assumption of liquid<br />
water, thus snow, hail or melting particles do not allow for such kind of precipitation<br />
estimation.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 20
3.10 Scan Strategy<br />
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
Reflectivity data for Z-R conversion should be measured close to ground, but they<br />
should not be contaminated by ground clutter, nor should they be blocked<br />
completely, to avoid as much errors as possible which might bias the data by<br />
application of different correction schemes. Very often the scan strategy is not only<br />
<strong>de</strong>termined by the need to obtain quantitative precipitation estimates, but also other<br />
meteorological phenomena may be essential to be <strong>de</strong>tected. Rainfall intensity data<br />
can be obtained <strong>from</strong> three different type of scans:<br />
• PPI-scan, i.e. data <strong>from</strong> one elevation angle only are consi<strong>de</strong>red<br />
• Pseudo-PPI-scan: one elevation with azimuth-<strong>de</strong>pen<strong>de</strong>nt tilting elevation angle<br />
• Volume-scan, i.e. a scan with several subsequent elevations<br />
The PPI-scan with one elevation only is a simple and rapid possibility to measure a<br />
horizontal distribution of reflectivity data. It can be repeated within a very short time<br />
and thus provi<strong>de</strong>s the opportunity to observe even rapidly changing weather<br />
phenomena. But taking only PPI-scans brings up some major disadvantages: The<br />
height above ground is varying strongly with range, and thus e.g. clutter<br />
contamination is much larger close to the radar than far away. In mountain regions,<br />
either areas with beam blocking will appear (if the elevation angle is set low) or there<br />
are areas, where the measurement position is too high above the ground (if the<br />
elevation angle is set high) or even both. If only one PPI angle is used, individual<br />
vertical profiles cannot be obtained <strong>from</strong> the radar data, and the application of bright<br />
band <strong>de</strong>tection algorithms becomes very difficult or impossible.<br />
One disadvantage of the single-PPI-scan with fixed elevation angle can be overcome<br />
by using a varying elevation angle: into directions with flat terrain, a low elevation<br />
angle is used. This angle is shifted slightly upward when the antenna turns to<br />
directions with mountain areas. This procedure can give a good compromise<br />
between the need of being close to the ground and not to block the antenna beam.<br />
Such scan type is used by the German Weather Service with elevation angles<br />
between 0.5 and 1.8 <strong>de</strong>grees for precipitation estimates (Schreiber, 1998). The main<br />
disadvantage of such scan compared to the previous type is that any correction<br />
algorithms become more complicated due to the azimuth-<strong>de</strong>pen<strong>de</strong>nce of the<br />
elevation angle (e.g. for bright band <strong>de</strong>tection or anomalous propagation correction),<br />
which means that more CPU time is nee<strong>de</strong>d (with possible negative impact on<br />
operational applicability).<br />
Finally precipitation data may be <strong>de</strong>rived using volume-scans, i.e. scans with multiple<br />
elevations. Such three-dimensional data sets allow to avoid clutter contamination<br />
close to the radar, because data <strong>from</strong> upper elevation angles can be obtained there.<br />
Usually the data are taken <strong>from</strong> different elevation angles, which are <strong>de</strong>termined by<br />
the horizontal distribution of the terrain height. A best fit between the need to have<br />
data close to the ground, but not clutter-contaminated nor totally beam-blocked can<br />
always be obtained. This type of precipitation estimation is used by the NEXRAD<br />
system (Fulton et al., 1998) and the SRI product (Surface Rainfall Intensity) of<br />
Rainbow ® . Volume scans provi<strong>de</strong> much better chances for bright band <strong>de</strong>tection<br />
algorithms and for vertical profile analysis and correction; they are essential for some<br />
sophisticated wind retrieval or phenomena <strong>de</strong>tection algorithms which often have to<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 21
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
be performed on the same radar data as the rainfall estimation algorithms. Volume<br />
scans have one severe disadvantage compared to the others: they require much<br />
more time. This may prevent the <strong>de</strong>tection and analysis of fast-<strong>de</strong>veloping<br />
phenomena.<br />
Mo<strong>de</strong>rn radar control and data evaluation software allows interlaced scan strategy: a<br />
volume scan with several elevations for three-dimensional analysis (like vertical<br />
profile, wind algorithm, phenomena <strong>de</strong>tection) is interrupted several times by a lowlevel<br />
scan with one (or a few) elevations for precipitation estimation. This allows a<br />
compromise between the need of small time steps between precipitation scans and<br />
the need of three-dimensional data. The first chapter of the next section will focus on<br />
the effects of different time steps between precipitation scans in more <strong>de</strong>tail.<br />
3.11 Operational Applicability<br />
In the previous sections, many concepts for the improvement of rainfall intensity<br />
estimation have been presented. The most important constraints were listed. Of<br />
course this review is far away <strong>from</strong> being complete. Details about the correction<br />
schemes can be found in the referenced literature. One final point should given<br />
attention here: the applicability of rainfall estimation on real-time.<br />
The possibility or the need to use certain correction steps is first of all limited by the<br />
radar hardware and the signal processor: For example, dual-polarisation techniques<br />
cannot be applied for the most operational radars. A Doppler clutter filter requires a<br />
Doppler radar; many radars have to use statistical filters and clutter maps. S-band<br />
radars are affected by attenuation through precipitation only very weakly;<br />
corresponding correction algorithms can be omitted without severe loss of data<br />
quality.<br />
Some correction steps are necessary only at specific site conditions:<br />
• Climate: In the tropics is no need for bright band correction or snow Z-R-relations,<br />
if the area of interest is not too large (thus elevation heights remaining small).<br />
Snow will seldom fall below 4 km above MSL in the tropics.<br />
• Orography: In flat terrain is minor need to extrapolate radar data <strong>from</strong> upper levels<br />
to the surface than in mountain areas with the problems of beam blocking. Thus<br />
less sophisticated vertical profile corrections may be applied or can be omitted<br />
totally. This could even mean that no three-dimensional scans are necessary<br />
which gives the chance of rapid repetition.<br />
• Coastal sites: If a radar is located close to an ocean or large lake, algorithms for<br />
sea-clutter reductions are necessary. Some oceans favour the appearance of<br />
low-level temperature inversions; thus increased problems with anomalous<br />
propagation may occur.<br />
These circumstances require that operational rainfall estimating algorithms should be<br />
constructed modular, it means that they should have the possibility to apply or omit<br />
different correction steps in<strong>de</strong>pen<strong>de</strong>ntly.<br />
Usually other real-time radar data processing is running simultaneously on the same<br />
platform, and other processing require correction steps as well. This gives some<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 22
3 · QPE I – Aspects of Rainfall Rate Derivation<br />
limitation for the CPU time consumption of all algorithms for radar data processing.<br />
For this reason it is sometimes necessary to skip a highly sophisticated correction<br />
step and use a simpler scheme instead, whose resulting data quality may be a few<br />
percent less than <strong>from</strong> the sophisticated algorithm.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 23
4 · QPE II – Accumulated Rain: Aspects of Integration in Time<br />
4 <strong>Quantitative</strong> <strong>Precipitation</strong> <strong>Estimation</strong> II –<br />
Accumulated Rain: Aspects of Integration in<br />
Time<br />
In the previous chapter, methods currently available to estimate quantitative rainfall<br />
rate <strong>from</strong> weather radar data have been presented. Important sources of possible<br />
errors have been discussed, possible solutions were shown. The effects of used<br />
radar hardware were discussed as well: influence of different wavelengths as well as<br />
the ability of Doppler measurements or of obtaining polarimetric quantities.<br />
If all necessary correction steps have been applied to <strong>de</strong>rive the best rainfall rate<br />
data <strong>from</strong> radar measurements, an important step towards hydrological application of<br />
such data has been ma<strong>de</strong>. But there remains another problem, which will be<br />
discussed in this chapter: the integration of rainfall data in time. A rain gauge is<br />
measuring continuously, thus instantaneous rainfall rates as well as accumulations<br />
over selectable time intervals can be available. <strong>Radar</strong> <strong>de</strong>rived rainfall intensities exist<br />
only in discrete time steps. Information about the <strong>de</strong>velopment between these steps<br />
is not directly available.<br />
The present chapter <strong>de</strong>als with different ways of filling this gap of information. Some<br />
aspects of data quality control will be given as well.<br />
4.1 Scan Strategy and Time Steps between Single Scans<br />
To obtain accumulated precipitation amounts <strong>from</strong> radar <strong>de</strong>rived rainfall rates R, the<br />
R data have to be integrated in time. A common way to do this is to multiply the<br />
instantaneous R data with the time interval between the scans. This method is simple<br />
and consumes only little CPU time, thus real-time application is no problem.<br />
However, the time ∆t between the scans must be small enough. A first estimate is<br />
that<br />
∆t < X / V (4.1)<br />
will give sufficient accuracy, with X being a measure for the horizontal extent of<br />
precipitation patterns and V being their propagating speed. In stratiform precipitation<br />
events, X is of the or<strong>de</strong>r of 10 km, whereas rain cells can be as small as 1 km or<br />
even less. Consi<strong>de</strong>ring a propagation speed of 10 m/s, this would result in a time<br />
step of about 15 minutes or less for the first case, and about one and a half minute or<br />
less for the second case. The smaller the time step ∆t, the better the resulting<br />
accumulates will be.<br />
If the time step between scans is much longer, the accumulated data will show some<br />
discrete patterns in the horizontal distribution. If for example a small rain cell<br />
propagates across the radar coverage, those places where the cell was at each scan<br />
time will experience too high rain amounts, whereas the places between will miss a<br />
large portion of the data. As a result not only hydrological use of such accumulated<br />
data will give wrong results, but also any radar-raingauge comparison must fail.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 24
4 · QPE II – Accumulated Rain: Aspects of Integration in Time<br />
Fig. 4.1a: 75 minute precipitation accumulation with a time step of 15 minutes (for <strong>de</strong>tails see text).<br />
Fig. 4.1b: As fig. 4.1a, but with a time step of 6 minutes.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 25
4 · QPE II – Accumulated Rain: Aspects of Integration in Time<br />
Fig. 4.1c: As fig. 4.1a, but with a time step of 1:30 minutes.<br />
The effect of too large time steps is illustrated in the figures 4.1a to 4.1c: each image<br />
shows the accumulated rainfall data of 75 minutes of observation. The individual<br />
rainfall rate data were multiplied by the time step between the scans. In the first case<br />
(fig. 4.1a), only one scan was taken every fifteen minutes. Random distributed areas<br />
with high precipitation amounts are the consequence. For the second image, more<br />
than twice the scans were taken with a time step of 6 minutes. Now the propagation<br />
path of different rain cells can be seen (<strong>from</strong> Southwest to Northeast), but several<br />
regions with high and low precipitation alternate along the paths. In the last case (fig.<br />
4.1c), the time between scans was only one and a half minutes. The result is a good<br />
reflection of the real precipitation swaths.<br />
Very fast repetition of scans (one minute or less) is only possible, if just one or a few<br />
elevations are sampled in each scan with a high antenna speed. But due to other<br />
requirements this may not be possible. In such cases, other methods have to be<br />
applied to avoid unrealistic peaks <strong>from</strong> small-scale cells as in figure 4.1a. The time<br />
step between scans can be reduced artificially by <strong>de</strong>riving inter-scan images or,<br />
equivalently, by tracking the individual precipitation patterns <strong>from</strong> one image to the<br />
next and consi<strong>de</strong>ring their movement and intensity changes in the time integration.<br />
Fabry (1994, chapter V) reported very promising results. However, tracking of<br />
precipitation patterns, especially by cross-correlation methods, is a very CPUintensive<br />
task and thus may face limitations for operational application. As a simpler<br />
and much faster method, the precipitation pattern propagation might be<br />
approximated by a mid-tropospheric mean wind vector <strong>de</strong>rived <strong>from</strong> Doppler velocity<br />
data. It will be a main task of the future work within the MUSIC Project to investigate<br />
such methods with respect to the quality of <strong>de</strong>rived precipitation accumulation and<br />
with respect to the operational applicability regarding the CPU time consumption. The<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 26
4 · QPE II – Accumulated Rain: Aspects of Integration in Time<br />
best precipitation estimating algorithm coming out <strong>from</strong> these investigations will be<br />
used for radar <strong>de</strong>rived rainfall data within the Project.<br />
4.2 Speckle Filtering<br />
Speckle filtering was a task carried out during the <strong>de</strong>rivation of rainfall rates (see<br />
section 3.2), but it might be necessary to apply some speckle filtering again on the<br />
accumulated data. Even the best clutter filter will not work perfectly, thus small parts<br />
of clutter contamination may remain in the reflectivity data and pass all other filtering.<br />
Accumulated over long time, these small parts may come out as significant<br />
precipitation amounts. If the real precipitation was very weak, the clutteraccumulation<br />
may be a multiple of it and appear as speckles of very high<br />
precipitation amount. Such speckles can be threshol<strong>de</strong>d down to an upper limit which<br />
<strong>de</strong>pends on the site, season and integration time, or may be interpolated by mean<br />
values of the surrounding.<br />
4.3 Handling of Missing Scans<br />
If the radar operation is interrupted for some reasons and thus the time interval<br />
between the last step before and the first one after interruption becomes too large<br />
(e.g. more than half an hour), no accumulation technique can be applied for the<br />
corresponding time. Thus the precipitation accumulation must be stopped at the<br />
beginning of the interruption and restarted when the radar starts its operation again.<br />
For several purposes, precipitation accumulation with varying observation time are<br />
not useful: typically hourly or daily precipitation amounts are required for further<br />
hydrological processing. For this purposes, any information about missing scans<br />
should be given in the precipitation accumulation data set. This could be done by<br />
giving the loss time as percentage of the total observation time, or by noting the<br />
missing times explicitly. For further hydrological mo<strong>de</strong>lling, the radar <strong>de</strong>rived<br />
precipitation amounts might be multiplied with a correction factor <strong>de</strong>pending on the<br />
missing time percentage, or if other sensor data are taken as well and if the missing<br />
times are given explicitly, other interpolation techniques may be used. However, such<br />
is no task of radar rainfall estimation in the context of this paper.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 27
5 · Outlook: Future Work in the MUSIC Project<br />
5 Outlook: Future Work in the MUSIC Project<br />
The previous chapters gave an overview of problems that arise when radar data has<br />
to be transformed to rainfall data. Several algorithms have been presented, which<br />
reduce the errors arising <strong>from</strong> such problems. Gematronik’s radar data calculation<br />
and visualisation software Rainbow ® is able to apply most of these correction steps in<br />
real-time. This software is going to be implemented in the MUSIC Project.<br />
Some steps in rainfall data <strong>de</strong>rivation will be improved using enhanced algorithms<br />
that will be <strong>de</strong>veloped in the Project within the next months. The beam blockage<br />
correction for example can be automated and gives better results, when highresolution<br />
digital terrain data are used <strong>from</strong> WP 3 (data bank).<br />
The most significant improvement of radar rainfall estimation will be achieved using<br />
new schemes for integration of radar rainfall data in time (see also section 4.1):<br />
Tracking of rain cells provi<strong>de</strong>s vector information that will be used to obtain better<br />
accumulated data (following the instructions of Fabry (1994)). This step will be<br />
performed in collaboration with WP 5 (UniNEW), who <strong>de</strong>rive storm, rain cell and rain<br />
band properties interactively using automated stochastic mo<strong>de</strong>ls .<br />
All new algorithms will have a special focus on their operational applicability; they<br />
must be able to run in real-time on quite large data sets.<br />
Finally these new algorithms will be provi<strong>de</strong>d to the MUSIC users on a UNIX<br />
workstation (SUN) that will also contain the implementation of the improved rainfall<br />
estimation estimators into the Rainbow ® radar data visualisation software, together<br />
with a standard set of conventional meteorological radar products. The<br />
corresponding radar data will be supplied in the file format specified by WP 2<br />
(“Hydro-Meteorological <strong>Data</strong> Collection and Supply”). This format will also provi<strong>de</strong> the<br />
interface to the work packages WP 3 (<strong>Data</strong> Bank), WP 7 (Block Kriging, Bayesian<br />
combination) and WP 9 (3D Visualisation).<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 28
References<br />
6 · References<br />
Atlas, D. (Ed.) (1990): <strong>Radar</strong> in meteorology. Amer. Meteor. Soc., Boston, 806 p.<br />
Atlas, D., R.C. Srivastava and R.S. Sekhon (1973): Doppler radar characteristics of<br />
precipitation at vertical evi<strong>de</strong>nce. Rev. Geophys Space Phys. 11, 1–35.<br />
Austin, P.M. (1987): Relation between measured radar reflectivity and surface<br />
rainfall. Mon. Wea. Rev. 115, 1053–1070.<br />
Austin, P.M. and A.C. Bemis (1950): A quantitative study of the “bright band” in radar<br />
precipitation echoes. J. Meteorol. 7, 145–151.<br />
Battan, L.J. (1973): <strong>Radar</strong> observations of the atmosphere. Univ. of Chicago Press,<br />
Chicago, 323 p.<br />
Benoit, R. and M. Desgagné (1996): Further non-hydrostatic mo<strong>de</strong>lling of the Brig<br />
1993 flash flood event. MAP Newsletter 5, SMA, Zürich, 36–37.<br />
Doviak, R.J. and D.S. Zrnic (1993): Doppler radar and weather observations.<br />
Aca<strong>de</strong>mic Press, New York, 562 p.<br />
Fabry, F. (1994): Observations and uses of high-resolution radar data <strong>from</strong><br />
precipitation. PhD thesis, McGill University, Montreal.<br />
Fiser, O. (2001): On impact of drop size distribution mo<strong>de</strong>ls on radar measurement.<br />
Proc. 30 th Int. Conf. on radar Meteor., Munich, 19 to 24 July 2001, 556–558.<br />
Fulton, R.A., J.P. Brei<strong>de</strong>nbach, D.-J. Seo and D.A. Miller (1998): The WSR-88D<br />
rainfall algorithm. Wea. and Forecasting 13, No. 2, 377–395.<br />
Germann, U. (1998): A concept for estimating the local vertical reflectivity profile for<br />
precipitation extrapolation. Proc. COST75 Int. seminar, Locarno, 23 to 27 March<br />
1998, 485–492.<br />
Germann, U. and J. Joss (2000): Meso-beta profiles to extrapolate radar precipitation<br />
measurements above the Alps to the ground. Submitted to J. Appl. Meteor.<br />
Germann, U. and J. Joss (2001): On the use of meso-β profiles and reflectivity<br />
variograms to better <strong>de</strong>scribe precipitation in complex orography. Proc. 30 th Int.<br />
Conf. on <strong>Radar</strong> Meteor., Munich, 19 to 24 July 2001, 518–519.<br />
Gorgucci, E., G. Scarchilli and V. Chandrasekar (1994): A robust estimator of rainfall<br />
rate using differential reflectivity. J. Atm. Ocean. Technol. 11, 586–592.<br />
Gysi, H., R. Hannesen and K.D. Beheng (1997): A method for bright band correction<br />
in horizontal rain intensity distributions. Proc. 28 th Conf. on <strong>Radar</strong> Meteor.,<br />
Austin, 7 to 12 Sept. 1997, 214–215.<br />
Hannesen, R. (1998): Analyse konvektiver Nie<strong>de</strong>rschlagssysteme mit einem C-Band<br />
Dopplerradar in orographisch geglie<strong>de</strong>rtem Gelän<strong>de</strong> (Analysis of convective<br />
precipitation systems with a C-band Doppler radar in orographic terrain, in<br />
German language). PhD thesis, Univ. Karlsruhe, 119 p.<br />
Hannesen, R. and M. Löffler-Mang (1998): Improvement of quantitative rain<br />
measurements with a C-band Doppler radar through consi<strong>de</strong>ration of<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 29
6 · References<br />
orographically induced partial beam screening. Proc. COST75 Int. seminar,<br />
Locarno, 23 to 27 March 1998, 511–519.<br />
Harrison, D.L., S.J. Driscoll and M. Kitchen (2000): Improving precipitation estimates<br />
<strong>from</strong> weather radar using quality control and correction techniques. Meteorol.<br />
Appl. 6, 135–144.<br />
Huggel, A., W. Schmid and A. Waldvogel (1996): Raindrop size distributions and the<br />
radar bright band. J. Appl. Meteor. 35, No. 10, 1688–1701.<br />
Illingworth, A.J.. (2001): Potential operational performance of rainfall algorithms using<br />
polarisation radar. Proc. 30 th Int. Conf. on <strong>Radar</strong> Meteor., Munich, 19 to 24 July<br />
2001, 615–617.<br />
Kessinger, C., S. Ellis and J. Van An<strong>de</strong>l (2001): NEXRAD data quality: The AP clutter<br />
mitigation scheme. Proc. 30 th Int. Conf. on <strong>Radar</strong> Meteor., Munich, 19 to 24 July<br />
2001, 707–709.<br />
Kitchen, M., R. Brown and A.G. Davies (1994): Real-time correction of weather radar<br />
data for the effects of bright band, range and orographic growth in wi<strong>de</strong>spread<br />
precipitation. Quart. J. Roy. Met. Soc. 120, 1231–1254.<br />
Kreuels, R.K. (1988): Repräsentativität und Genauigkeit von Regenmeßsystemen<br />
(Representativity and accuracy of rain measuring systems, in German<br />
language). Zeitschr. Stadtentwäss. Gewässerschutz 4, 39ff.<br />
Lee, R., G. Della Bruna and J. Joss (1995): Intensity of ground clutter and of echoes<br />
of anomalous propagation and its elimination. Proc. 27 th Conf. on <strong>Radar</strong><br />
Meteor., Vail, 9 to 13 October 1995, 651–652.<br />
Lim, S., V. Chandrasekar, V.N. Bringi, W. Li and A. Al-Zaben (2001): Hydrometeor<br />
classification <strong>from</strong> polarimetric radar measurements during STEPS. Proc. 30 th<br />
Int. Conf. on <strong>Radar</strong> Meteor., Munich, 19 to 24 July 2001, 426–428.<br />
Liu, J.Y. and H.D. Orville (1969): Numerical mo<strong>de</strong>lling of precipitation and cloud<br />
shadow effects on mountain induced cumuli. J. Atm. Sci. 26, 1283–1289.<br />
Löffler-Mang, M. and H. Gysi (1998): Radome attenuation of C-band radar as a<br />
function of rain characteristics. Proc. COST75 Int. seminar, Locarno, 23 to 27<br />
March 1998, 520–526.<br />
Manz, M., T. Monk and J. Sangiolo (1998): Radome effects on weather radar<br />
systems. Proc. COST75 Int. seminar, Locarno, 23 to 27 March 1998, 467–478.<br />
Marshall, J.S and W. McK. Palmer (1948): The distribution of raindrops with size. J.<br />
Meteor. 5, 165–166.<br />
Nanni, S., P.P. Alberoni and P. Mezzasalma (1998): I<strong>de</strong>ntification of hail by means of<br />
polarimetric radar: results <strong>from</strong> some cases. Proc. COST75 Int. seminar,<br />
Locarno, 23 to 27 March 1998, 738–746.<br />
Neimann, P.J., F.M. Ralph, A.B. White, D.E. Kingsmill and P.O.G. Persson (2001):<br />
Using radar wind profilers to document orographic precipitation enhancement<br />
during the CALJET field experiment. Proc. 30 th Int. Conf. on <strong>Radar</strong> Meteor.,<br />
Munich, 19 to 24 July 2001, 509–511.<br />
Rinehart, R.E. (1991): <strong>Radar</strong> for meteorologists. Dep. of Atm. Sci., Univ. of North<br />
Dakota. 224 p.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 30
6 · References<br />
Rosenfeld, D., E. Amitai and D.B. Wolff (1995): Classification of rain regimes by<br />
three-dimensional properties of reflectivity fields. J. Appl. Meteor. 34, 198–211.<br />
Ryzhkov, A.V., T.J. Schuur and D.S. Zrnic (2001): <strong>Radar</strong> rainfall estimation using<br />
different polarimetric algorithms. Proc. 30 th Int. Conf. on <strong>Radar</strong> Meteor., Munich,<br />
19 to 24 July 2001, 641–643.<br />
Sauvageot, H. (1992): <strong>Radar</strong> meteorology. Artech House, Boston, 366 p.<br />
Schreiber, K.-J. (1998): Der <strong>Radar</strong>verbund <strong>de</strong>s Deutschen Wetterdienstes (The radar<br />
network of the German Weather Service, in German language). Annln. Meteor.<br />
38, 47–64.<br />
Upton, G. and J.-J. Fernán<strong>de</strong>z-Durán (1998): Statistical techniques for clutter<br />
removal and attenuation correction in radar reflectivity images. Proc. COST75<br />
Int. seminar, Locarno, 23 to 27 March 1998, 747–757.<br />
Waldvogel, A., B. Fe<strong>de</strong>rer and P. Grimm (1979): Criteria for the <strong>de</strong>tection of hail<br />
cells. J. Appl. Meteor. 18, 1521–1525.<br />
White, A.B, J.R. Jordan, F.M. Ralph, P.J. Neimann, D.J. Gottas, D.E. Kingsmill and<br />
P.O.G. Persson (2001): S-band radar observations of coastal orographic rain.<br />
Proc. 30 th Int. Conf. on <strong>Radar</strong> Meteor., Munich, 19 to 24 July 2001, 512–514.<br />
Zrnic, D.S and A.V. Ryzhkov (1999): Polarimetry for weather surveillance radars.<br />
Bull. Amer. Meteor. Soc. 80, No. 3, 389–406.<br />
MUSIC · Deliverable 4.1 · 8 th Oct. 2001 Page 31