295
Precipitation Estimate Using NEXRAD Ground-Based Radar Images
The precipitation maps estimated by the four prediction techniques on July 22,
2003 (Figure 12.7) all have similar spatial precipitation distribution patterns, with the
highest precipitation in the southeast and higher precipitation in the southwest than in
the north. Note that NEXRAD systematically underestimated the precipitation amount
with an EB value of 36.52%. Through the BA, the accuracy of NEXRAD was significantly improved. In terms of EB and EE, SKlm and RK outperformed the other techniques. The RVar values of SKlm and RK are also closer to 1 than the other methods.
12.5.4 coMPaRing the PeRfoRManceS of diffeRent caliBRation MethodS
12.5.4.1 Overall Performance Assessment
The BA, RK, and SKlm methods were applied to calibrate NEXRAD data using
the observed rain gauge data, and the evaluation coefficients for the three methods
were calculated (Table 12.7). The same pairs of data used to validate NEXRAD data
were used to calculate the coefficients by replacing the NEXRAD estimates with the
calibrated NEXRAD values from BA, RK, and SKlm using the leave-one-out crossvalidation method. After calibrating NEXRAD data using rain gauge data, performance measures D rain and D no-rain were slightly improved. For detection capability
TABLE 12.6
Comparison of the Performance of Different Methods on 3 Days
NEXRAD
BA
SKlm
RK
October 13, 2002
Mean
73.43
63.09
60.95
60.33
SDV
19.14
16.45
16.52
16.39
EB
16.39%
0.20%
1.29%
0.41%
R 2
0.32
0.27
0.67
0.66
EE
–0.26
0.08
0.65
0.68
RVar
1.07
0.95
0.77
0.73
July 22, 2003
Mean
22.38
35.26
37.15
37.09
SDV
7.41
11.68
17.73
16.82
EB
36.52%
0.32%
0.59%
0.24%
R 2
0.88
0.86
0.93
0.92
EE
–0.11
0.75
0.92
0.93
RVar
0.40
0.59
1.02
0.98
TABLE 12.7
Evaluation Coefficients of Different Calibration Methods
Evaluation Coefficients
Methods
D rain (%)
D no-rain (%)
D rain + D no-rain (%)
EB
r 2
EE
BA
82.66%
97.27%
179.94%
0.33%
0.83
0.85
RK
86.87%
96.01%
182.88%
–1.25%
0.90
0.92
SKlm
87.47%
96.05%
183.52%
–0.69
0.92
0.92
Precipitation Estimate Using NEXRAD Ground-Based Radar Images
The precipitation maps estimated by the four prediction techniques on July 22,
2003 (Figure 12.7) all have similar spatial precipitation distribution patterns, with the
highest precipitation in the southeast and higher precipitation in the southwest than in
the north. Note that NEXRAD systematically underestimated the precipitation amount
with an EB value of 36.52%. Through the BA, the accuracy of NEXRAD was significantly improved. In terms of EB and EE, SKlm and RK outperformed the other techniques. The RVar values of SKlm and RK are also closer to 1 than the other methods.
12.5.4 coMPaRing the PeRfoRManceS of diffeRent caliBRation MethodS
12.5.4.1 Overall Performance Assessment
The BA, RK, and SKlm methods were applied to calibrate NEXRAD data using
the observed rain gauge data, and the evaluation coefficients for the three methods
were calculated (Table 12.7). The same pairs of data used to validate NEXRAD data
were used to calculate the coefficients by replacing the NEXRAD estimates with the
calibrated NEXRAD values from BA, RK, and SKlm using the leave-one-out crossvalidation method. After calibrating NEXRAD data using rain gauge data, performance measures D rain and D no-rain were slightly improved. For detection capability
TABLE 12.6
Comparison of the Performance of Different Methods on 3 Days
NEXRAD
BA
SKlm
RK
October 13, 2002
Mean
73.43
63.09
60.95
60.33
SDV
19.14
16.45
16.52
16.39
EB
16.39%
0.20%
1.29%
0.41%
R 2
0.32
0.27
0.67
0.66
EE
–0.26
0.08
0.65
0.68
RVar
1.07
0.95
0.77
0.73
July 22, 2003
Mean
22.38
35.26
37.15
37.09
SDV
7.41
11.68
17.73
16.82
EB
36.52%
0.32%
0.59%
0.24%
R 2
0.88
0.86
0.93
0.92
EE
–0.11
0.75
0.92
0.93
RVar
0.40
0.59
1.02
0.98
TABLE 12.7
Evaluation Coefficients of Different Calibration Methods
Evaluation Coefficients
Methods
D rain (%)
D no-rain (%)
D rain + D no-rain (%)
EB
r 2
EE
BA
82.66%
97.27%
179.94%
0.33%
0.83
0.85
RK
86.87%
96.01%
182.88%
–1.25%
0.90
0.92
SKlm
87.47%
96.05%
183.52%
–0.69
0.92
0.92