Transcript Some experiences on satellite rainfall estimation over South America

SOME EXPERIENCES ON SATELLITE
RAINFALL ESTIMATION OVER SOUTH
AMERICA.
Daniel
1
1
Vila ,
Inés
2
Velasco
Sistema de Alerta Hidrológico - Instituto Nacional del Agua y de Ambiente
Autopista Ezeiza - Cañuelas km 1.60 - (1402) Ezeiza - Buenos Aires - Argentina
TE/FAX: +54 -11 - 4480 - 9174
2
Universidad de Buenos Aires
Photo:Iguazu Falls
OVERVIEW
• Results of the study of the South American version of
NOAA/NESDIS “Hydro-Estimator” satellite rainfall estimation
technique in selected regions of the Del Plata River basin.
• Brief algorithm description and correction methodologies:
constant rate integration and local bias correction.
• Verification methods.
• Case studies: Salado River Basin (Pcia de Buenos Aires,
Argentina) and Uruguay River subcatchment (Argentina, Brazil and
Uruguay.
• Conclusions.
• Some results and research activities in progress.
ALGORITHM DESCRIPTION
• This is a fully automated method using an empirical power-law function that
generates rainfall rates (mm/h) based on GOES-8 channel 4 brightness
temperature
• Moisture correction factor (PWRH) defined as the product of precipitable water
(PW) (integrated over the layer from surface to 500 hPa) times the relative
humidity (RH) (mean value between surface and 500 hPa., in percentage) is
applied to decrease rainfall rates in dry environments and increases them in the
moist ones.
• New screening method: This technique assumes that raining pixels are colder
than the mean of the surrounding pixels.
• Standardized temperature is defined as:
Ts =
T - Tave
s
ALGORITHM DESCRIPTION
• Tơ = 0
• Tơ < -1.5
Stratiform precipitation: whose
Convective precipitation:
maximum value cannot exceed 12mmh-1
and must be less than the fifth part of the
convective rainfall for a given pixel
defined essentially by the
empirical power-law function
corrected by PWRH.
• –1.5 < Tơ < 0
Entirely
Convective
• Tơ > 0  pp = 0
Standarized Tem perature
0.0
-1.0
-1.5
Entirely
Stratiform
CORRECTION METHODOLOGIES
GOES 8 - Ch4 - Image availability for southern hemisphere sector from 20 May 12 Z to 22 June 12 Z (open circles). The time difference (in hours) between
consecutive images are plotted in blue (left axis).
CORRECTION METHODOLOGIES
· CONSTANT RATE INTEGRATION
• Rain rate remains constant between images …
· CONSTANT RATE INTEGRATION
• but something better may be made …
· LOCAL BIAS CORRECTION
• This algorithm takes into account the difference between rain
gauges and the HE estimation for a given rain gauge network
Satellite Estimation
Nine pixel compar ison kernel
Rain Data Gauge
(ground truth)
Schematic procedure of the
best adjusted value (MVE).
Rainfall data is compared
with a nine pixels kernel
centered in the rain gauge
location
· LOCAL BIAS CORRECTION
BASIN
LIMITS
ARGENTINA
BRAZIL
URUGUAY
ATLANTIC OCEAN
5
50
100
150
200
250
24-hour estimated rainfall: 21 – Aug -2002
· CASE STUDY : SALADO RIVER
• LOCAL BIAS CORRECTION
-30
-34
Buenos Aires Province,
Argentina
-32
Río
-35
de
la
Pla
ta
URUGUAY
-34
-36
Ri
o
de
l
aP
LATITUD
LATITUDE
ARGENTINA
la t
a
-36
-37
-38
a
é
c
O
-40
-65
-63
-61
-59
LONGITUDE
o
n
la
t
A
o
c
i
t
n
-38
c
O
-57
-55
-39
-61
-60
-59
-58
a
é
o
n
-57
lt a
A
o
c
i
t
n
-56
LONGITUD
The 10º x 10º box used to evaluate the technique. Dashed area belongs to the Salado River catchment. Solid
triangles show the location of rain gauges used for the local bias correction. Right: Geographical distribution
of rain gauges used to validate the technique.
· CASE STUDY : SALADO RIVER
Observed vs. estimated values for the 23-24 September 2001 event.
Straight line represents the ideal estimation
· CASE STUDY : SALADO RIVER
VALIDATION STATISTICAL PARAMETERS
Table I
THRESH
NUM
BIAS
CORR
RMSE
POD
FAR
SKILL
0.0
216
3.98
0.72
15.76
0.98
0.01
0.61
7.0
186
4.63
0.67
16.93
0.87
0.12
0.67
26.0
104
7.97
0.46
21.21
0.78
0.63
0.72
52.0
34
5.65
0.10
24.28
0.65
0.70
0.79
WTAVG
1276
5.82
0.49
19.85
0.81
0.39
0.70
• Overestimation are present in all intervals.
• Weighted averaged bias of 5.8 mm represents a positive difference of
around 27% between estimated and observed values.
• While for the first rows POD and FAR appear close to ideal, for the higher
intervals (26 and 52 mm) high values of FAR and lower of POD are present
· CASE STUDY : URUGUAY RIVER
• CONSTANT RAIN RATE
-28
Sub-Catchment limit
ARGENTINA
LATITUD
Ur
ug
ua
y
Ri
ve
r
-29
-30
Qu
ar
eim
BRAZIL
Ri
ve
r
-31
Salto Grande Dam
URUGUAY
Country borders
-32
-59
-58
-57
-56
-55
LONGITUDE
Geographical position of rain gauges
used for evaluation purposes. Dashed
area belongs to the Salto Grande Dam
Immediate catchment
Satellite rainfall estimation for Salto
Grande Dam region - 31 May/ 1 June
2001
· CASE STUDY : SALADO RIVER
Observed vs. estimated values for the 31 May –1 June, 2001 event.
Straight line represents the ideal estimation
· CASE STUDY : URUGUAY RIVER
VALIDATION STATISTICAL PARAMETERS
TABLE II
THRESH
NUM
BIAS
CORR
RMSE
POD
FAR
SKILL
0.0
321
-9.41
0.93
24.10
1.00
0.03
0.84
7.0
313
-9.49
0.92
24.40
0.92
0.06
0.86
26.0
210
-14.38
0.87
29.64
0.89
0.06
0.86
52.0
145
-17.25
0.76
34.47
0.93
0.08
0.88
WTAVG
2947
-13.91
0.85
29.82
0.93
0.07
0.86
• Underestimation are present in all intervals.
• Weighted bias represents only 15% of underestimation and the RMSE is
around 30%.
• The probability of detection (POD) and False alarm ratio (FAR) exhibit
very good values near 1 and 0 respectively.
· CONCLUSIONS
• The main purpose of this work is to present the recent
improvements of the Auto-Estimator Algorithm and the
application of this technique in two flash flood events in Del Plata
basin in South America.
• The main difference between the South American model and
the one for North America is the image availability. Gaps up to
three hours in South America imagery may be a very important
factor in the accuracy of the estimations.
• The errors involved in these kind of techniques were evaluated
in the cases study presented.
• Future efforts should include a detailed validation and
statistical analysis of a reasonable number of cases
· OPERATIVE RESEARCH
-27
Rio
-28
u
r
U
Areal rainfall estimation
y
a
gu
• 15-Feb / 15 Mar2002 – 24
hours rainfall estimation and mean
river level at Paso Mariano Pinto.
-29
• Local bias correction applied
-30
Rio Ibicuy Basin
Aproximated area: 46.500 km2
-31
-58
-57
-56
-55
-54
-53
-52
-51
-50
-49
· OPERATIVE RESEARCH
70
300
Paso Mariano Pinto (Brazil)
60
10 days
250
50
40
150
30
100
20
50
10
0
20/02/2002
0
25/02/2002
02/03/2002
07/03/2002
date
12/03/2002
17/03/2002
height (cm)
pp (mm)
200
· OPERATIVE RESEARCH
70
40
IBICUY RIVER BASIN: ESTIMATED
& CORRECTED RAINFALL
35
60
30
50
25
40
20
30
Mean Estimated rainfall: 9,6 mm
15
Mean Corrected rainfall: 10,4 mm
20
Number of images (ave): 29 /48 (ideal)
10
10
0
20/02/2002
5
0
25/02/2002
Corrected
02/03/2002
07/03/2002
Estimated
12/03/2002
Nr of Images