-
Corresponding author address: Rafael Sánchez-Diezma, UPC-EHMA,
Jordi Girona, 1-3, D1, E-08034 Barcelona.e-mail:
[email protected]
P 7 . 5 AN IMPROVED METHODOLOGY FOR GROUND CLUTTER SUBSTITUTION
BASED ONA PRE-CLASSIFICATION OF PRECIPITATION TYPES
Rafael Sánchez-Diezma 1, Daniel Sempere-Torres 1, Guy Delrieu 2,
Isztar Zawadzki 31 Departament d’Enginyeria Hidràulica, Marítima i
Ambiental,Universitat Politècnica de Catalunya, Barcelona,
Spain.2 Laboratorie d’Étude des Transferts en Hydrologie et
Environnement, CNRS-INPG-UJF, Grenoble, France.3 J. S. Marshall
Radar Observatory, Department of Atmospheric and Oceanic Sciences,
McGill University,
Montréal, Québec, Canada.
1. Introduction
Radar data contamination due to ground echoesis a common problem
in any radar configuration, moreso in mountainous areas. Addressing
ground cluttercontamination can be seen as a double task:
firstdetection and elimination of contaminated areas, andsecond,
substitution of removed contaminated valuesfor suitable ones.
Different algorithms that usereflectivity information, or Doppler
data have beendeveloped to remove such contamination ([Lee et
al.,1995], [Bellon and Kilambi, 1999], [VanAndel andKessinger,
1999], see a revision of methodologies in[Steiner and Smith,
1997]).
The way of performing ground echoessubstitution can be either
vertical or horizontalinterpolation. The first one utilizes the
shape of thevertical profile of reflectivity (VPR) to extrapolate
non-contaminated values measured aloft, over the areaaffected by
ground clutter ([Joss and Lee, 1995],[Bellon and Kilambi, 1999]).
On the other hand, thehorizontal substitution interpolates
non-contaminatedneighboring values located at the same height
orelevation as the contaminated echoes ([Galli, 1984],[Bellon and
Kilambi, 1999]). These procedures arelimited by the size of ground
clutter regions and thespatial variability of the rainfall field.
It is difficult toassess which of the two options is the
mostsatisfactory.
The aim of this work is to propose a substitutionmethodology
that combines horizontal and verticalapproaches through considering
the type ofprecipitation. The idea of this approach is to
applyhorizontal interpolation in those areas of precipitationwith
low reflectivity gradients (generally consideredas stratiform
rain). On the contrary, verticalextrapolation will be applied in
areas with largereflectivity gradients (usually associated
toconvective precipitation). In this latter case the VPRtends to
have a uniform shape with height, that favorsvertical substitution,
and the relation between the rainat a point and its surroundings is
less clear (whichinvalidates the idea of horizontal interpolation).
Then,this methodology attempts to combine the horizontaland
vertical approaches considering the spatialvariability of the
rainfall field. The proposedmethodology uses a simplified approach
to identifythe convective regions, based in the use of anintensity
threshold level. This simplified approachintends to be a first
level of complexity solution,useful in an operational context.
2. Evaluated methodologies o fsubsti tut ion:
The methodology of substitution outlined aboveis compared with
four other substitution techniques inorder to verify its
performance, that are:
1) Pure vertical substitution: based on thesubstitution of
ground echo values by the first non-contaminated value located over
their vertical.2) Pure horizontal substitution: ground echoes
aresubstituted by interpolating neighbor values fromtheir same
elevation. The interpolation is applied
establishing a Delaunay triangulation for all non-contaminated
data. Then, each contaminated valueis substituted by means of a
distance–weigthedscheme that considers the reflectivity at
thevertexes of the triangle that contains the point to
besubstituted.3) The substitution methodology used by the
radarnetwork of the Spanish National Institute ofMeteorology (INM).
This methodology uses acombination of horizontal and vertical
substitutionthat is applied to polar radar data by means of
thefollowing criteria: for each identified ground clutter itlooks
for a substitute located at its same elevation,first considering
the azimutal neighbors (right andleft), and afterwards considering
the radial ones(front and rear). In the case that none of them
isfree of contamination the same procedure isrepeated in the next
elevation, employing then as afirst candidate the value located at
the vertical ofthe point that is been substituted. This procedure
isrepeated along the subsequent elevations until thatone
non-contaminated value is found. Notice thatthis method does not
determines a substitute byaveraging a certain number of values.4)
Finally the proposed method based on asimplified recognition of the
type of precipitation toperform an horizontal + vertical
substitution usesthe following scheme: as a first step the
purehorizontal substitution is applied to all ground echoregions.
Next, these substituted which exceed acertain threshold related to
convective precipitation(we used 45 dBZ) are substituted once again
butnow using the first non-contaminated value locatedat its
vertical. This latter value is accepted as afinal substitute if it
is greater than the one resultingfrom the horizontal substitution
(otherwise the firstsubstitute is preserved).
In our analysis all the methodologies are appliedover polar
radar data, and the triangulation isestablished from the cartesian
positions of the polardata in order to define the triangles
considering theactual positions of the valid points.
The different methodologies have been testedusing as a reference
the ground echoes measured bythe volumetric scanning C-band radar
of the INMlocated at Barcelona (0.9° 3-dB beamwidth, λ=5.6 cm,20
elevation angles). The Figure 1 depicts the radarlocation and the
average ground echoes patternobserved at the first radar elevation
(0.5°) derivedduring non-precipitating conditions and in absence
ofanomalous propagation. In order to generate a tri-dimensional
echoes pattern, a similar map wasobtained for the rest of
elevations affected by terrain-induced returns. Although our
analysis is limited tothe substitution of the ground echoes
registered atthe first radar elevation it is necessary to
determinethis tri-dimensional echoes pattern because some ofthe
tested methodologies look for non-contaminatedsubstitutes at higher
elevations.
In order to define a region where to apply thesubstitution
techniques, a threshold of 23 dBZ (1mm/h using the Z-R
Marshall-Palmer relation) is usedto establish the minimum intensity
of the groundclutter that should be substituted.
-
To test the different methodologies it isnecessary to have
measurements of reference tocompare them against the substituted
values, andasses the quality of the results. For this purpose
thethree-dimensional structure of the echoes pattern isrotated in
order to locate them in an area initially freeof problems. At this
new location the value ofreference is known (since now these data
is notcontaminated), and the substituted values are theones
resulting from applying the differentmethodologies to the ground
echoes pattern at thisnew location. The rotation angle used is
–67.57degree, corresponding with –80 positions in theazimutal
direction (radar maps are composed by 420values in the azimutal
sense and 120 in the radialsense, with a 2 km radial resolution).
After therotation, ground echoes that remain still overcontaminated
regions are not considered in theanalysis.
distance from the radar (km)
dist
ance
from
the
rada
r (k
m)
>
5
10
15
20
25
30
35
40
45
50
55
60
<
refle
ctiv
ity Z
(dB
Z)
-60 -40 -20 0 20 40 60-60
-40
-20
0
20
40
60
0
0
0
100
500 Radar
Figure 1. Ground echoes pattern up to 60 km from theradar for
the firs elevation (0.5°) of the INM radar atBarcelona.
3. Evaluation of the di f ferentmethodologies of
substitution:
The different methods have been applied over aevent registered
by the C-band radar of the INMlocated in Barcelona on June 10th
2000, with a timeduration of more than 24 hours. This event
wascharacterized by the mix of types of precipitation,with a first
markedly convective part, followed by astratiform period. The
comparison of reference andsubstituted values are shown in Figure 2
in terms ofinstantaneous rainfall intensities (i.e. reference
andsubstituted values for each radar map), hourly
rainfallaccumulations, and total rainfall accumulation for thewhole
event. The Marshall-Palmer Z = 200R1.6 relationis used to transform
the reflectivity into rain intensity.The results have been marked
with different symbolsas a function of its range of distances from
the radar.The performance of the different methods has
beenestimated by means of different statistics: the Nashefficiency,
the square correlation (r2) and the averageerror between the
substituted and reference values.
Figure 2, column 1 (instantaneous values)shows that the proposed
substitution methodprovides better results than the rest
ofmethodologies, basically as a result of the goodefficiency, and
r2, since all of them exhibit a similaraverage error, except for
the INM method whoseperformance is well below the other ones. The
resultsalso show that the horizontal substitution tends
tounderestimate the high and intermediate intensities.
For the hourly accumulations (see Figure 2column 2) once more
the proposed method shows thebest adjustment, although all the
methods provide
similar average errors. Moreover, notice that only theproposed
method and the vertical substitution correctthe high intensities in
a proper way. Nevertheless,this latter methodology and the INM
method exhibit aclear underestimation of a group of
intermediateintensities (from 0 to 20 mm), located between 40 and60
km from the radar. As before, the INM methodleads to a greater
dispersion between the referenceand the substituted values as
compared to the rest ofthe methods.
Finally, the total accumulation of the event(Figure 2 column 3)
shows similar results: theproposed substitution remarkably improves
theperformance of the rest of methods. The verticalsubstitution
clearly exhibits two tendencies, and twofamilies of points can be
observed: the first onelocated around the line of right adjustment
(1:1), andthe second one showing a clear underestimation dueto the
decreasing of precipitation with height in thestratiform periods of
the event. This twofold tendencyis also present in the results of
the INM methodologybut it is less severe. On the other hand,
thehorizontal substitution exhibits a clearunderestimation of the
medium to high rainfallaccumulations. It confirms that this
substitution donot enables a good correction of the
convectivevalues.
4. ConclusionsThis work analyzes a methodology to substitute
radar data contaminated by ground echoes for aproper
rain-related reflectivity value. The proposedmethodology intends to
combine the horizontal andvertical substitution methodologies
throughconsidering the most adequate approach as afunction of the
spatial variability of the rainfall field:horizontal for stratiform
rain and vertical forconvective.
The results for the analized event suggest that,among the four
analyzed methods, the proposedhorizontal + vertical substitution
provides the bestperformance. Future work should be done in order
toasses the factors that can affect the accuracy of theproposed
methodology (as the substitution threshold,or the convective
threshold, or the size of the groundecho regions), as well as to
analyze its performancein a wider number of cases.
4. Acknowledgements:This work has been done in the frame of the
R+D
CICYT project REN2000-17755-C01. Thanks are dueto the Spanish
Instituto Nacional de Meteorología forproviding radar data and
support.
ReferencesBellon, A., and A. Kilambi, Updates to the McGill
RAPID
(Radar Data Analysis, Procesing and Interactive Display)system,
in 29th Conference on radar meteorology, edited byAMS, pp. 121-124,
Montreal,Canada, 1999.
Galli, G., Generation of the Swiss radar composite, in22nd
Conference on radar meteorology, pp. 208-211, AMS,1984.
Joss, J., and R. Lee, The application of radar-gaugecomparisons
to operational precipitation profiles corrections, J.Appl.
Meteorol., 34, 2612-2630, 1995.
Lee, R., G. Della Bruna, and J. Joss, Intensity of groundclutter
and of echoes of anomalous propagation and itselimination, in 25th
Conference on radar meteorology, edited byAMS, pp. 651-652,
1995.
Steiner, M., and J.A. Smith, Anomalous propagation ofradar
signals-challenges with clutters, in 28th Conference onradar
meteorology, edited by AMS, pp. 501-503, Normal(Oklahoma), USA,
1997.
VanAndel, J., and C. Kessinger, APCAT: An AP clutteranalysis
tool, in 29th Conference on radar meteorology, editedby AMS, pp.
267-269, Montreal, Canada, 1999.
-
vertical substitution
0 50 100 150 200reference (mm/h)
0
50
100
150
200su
bstit
uted
(m
m/h
)
efficiency = 0.68r2 = 0.72average error = 0.98
INM method
0 50 100 150 200reference (mm/h)
0
50
100
150
200
subs
titut
ed (
mm
/h)
efficiency = 0.37r2 = 0.51average error = 1.16
horizontal substitution
0 50 100 150 200reference (mm/h)
0
50
100
150
200
subs
titut
ed (
mm
/h)
efficiency = 0.73r2 = 0.75average error = 1.02
horizontal+vertical substitution
0 50 100 150 200reference (mm/h)
0
50
100
150
200
subs
titut
ed (
mm
/h)
efficiency = 0.82r2 = 0.84average error = 1.04
vertical substitution
0 10 20 30 40 50 60reference (mm/h)
0
10
20
30
40
50
60
subs
titut
ed (
mm
/h)
efficiency = 0.62r2 = 0.67average error = 0.96
INM method
0 10 20 30 40 50 60reference (mm/h)
0
10
20
30
40
50
60
subs
titut
ed (
mm
/h)
efficiency = 0.57r2 = 0.65average error = 1.05
horizontal substitution
0 10 20 30 40 50 60reference (mm/h)
0
10
20
30
40
50
60
subs
titut
ed (
mm
/h)
efficiency = 0.81r2 = 0.83average error = 0.95
horizontal+vertical substitution
0 10 20 30 40 50 60reference (mm/h)
0
10
20
30
40
50
60
subs
titut
ed (
mm
/h)
efficiency = 0.90r2 = 0.91average error = 0.99
vertical substitution
0 20 40 60 80 100reference (mm)
0
20
40
60
80
100
subs
titut
ed (
mm
)
efficiency = -0.10r2 = 0.43average error = 0.75
INM method
0 20 40 60 80 100reference (mm)
0
20
40
60
80
100
subs
titut
ed (
mm
)
efficiency = 0.56r2 = 0.64average error = 1.00
horizontal substitution
0 20 40 60 80 100reference (mm)
0
20
40
60
80
100
subs
titut
ed (
mm
)efficiency = 0.80r2 = 0.86average error = 0.87
horizontal+vertical substitution
0 20 40 60 80 100reference (mm)
0
20
40
60
80
100
subs
titut
ed (
mm
)
efficiency = 0.92r2 = 0.92average error = 0.95
instantaneous acumulations hourly acumulations total
acumulations
0-30 km 30-60 km
60-90 km90-120 km
Figure 2. Evaluation of the different methodologies of
substitution for analyzed the event. The figures show thecomparison
of reference and substituted values for different accumulated
periods. The goodness of the differentmethods is tested through the
estimation of the Efficiency, the square correlation (r2), and the
average errorbetween the reference and substituted values.
