19.4.3 Simulation Analyses
One season (summer) has been tested for nine previously published split-window-type
algorithms and their modified (with zenith angle correction term) forms, totally 18 SW
algorithms and two proposed algorithms for GOES M (12)-Q.
For each of the tested algorithms, we calculated the bias and standard deviation
of the regressions. Due to the high water vapor amount during summer, the LST
retrieval errors are usually larger than in other seasons. As shown in Fig. 19.9, it is
found that the largest errors always appear at warm surface temperature above
280 K, and viewing zenith angle larger than 4
or satellite zenith angle greater than
41.75
, and the following algorithms gave better performance than other splitwindow algorithms:
The modified Becker and Li (1990) algorithm, which was a local split-window
algorithm, and later modified by Wan and Dozier (1996) to make the coefficients
varying with different conditions as the generalized split-window algorithm. The
maximum standard deviation is only 0.75 K for this algorithm.
The modified Vidal 1991 algorithm (Vidal 1991; Yu et al. 2008). The maximum
standard deviation is less than 1.0 K for this algorithm.
The modified Sobrino 1993 algorithm (Sobrino et al. 1993; Yu et al. 2008). The
maximum standard deviation is only 0.75 K for this algorithm.
The modified Sobrino 1994 algorithm (Sobrino 1994; Yu et al. 2008). The
maximum standard deviation is less than 1.0 K for this algorithm.
Meanwhile, it is found that the modified Sobrino et al. (1993) algorithm with
nonlinear term gave better performance than the modified Sobrino (1994) algorithm
without nonlinear term.
Nevertheless, Yu et al. (2008) found that the modified Ulivieri-1985 algorithm
showed the least sensitivity to the emissivity variation, so they suggest this algorithm as the baseline GOES-R LST algorithm (Yu et al. 2010).
Input (looping)
Parameters
Algorithms
Algorithms
Sensor Spectral
Response Fcns.
Sensor
Brightness
Temperature
Calculation
Temperature
Algorithm
Coeffs.
STD Error
Of
TOA
Radianc
Sensor
Brightness
Regression
Of LST
Filter of
Data
Distribution
Fig. 19.8 Procedure of the algorithm regression analyses
318
D. Sun and Y. Yu
One season (summer) has been tested for nine previously published split-window-type
algorithms and their modified (with zenith angle correction term) forms, totally 18 SW
algorithms and two proposed algorithms for GOES M (12)-Q.
For each of the tested algorithms, we calculated the bias and standard deviation
of the regressions. Due to the high water vapor amount during summer, the LST
retrieval errors are usually larger than in other seasons. As shown in Fig. 19.9, it is
found that the largest errors always appear at warm surface temperature above
280 K, and viewing zenith angle larger than 4
or satellite zenith angle greater than
41.75
, and the following algorithms gave better performance than other splitwindow algorithms:
The modified Becker and Li (1990) algorithm, which was a local split-window
algorithm, and later modified by Wan and Dozier (1996) to make the coefficients
varying with different conditions as the generalized split-window algorithm. The
maximum standard deviation is only 0.75 K for this algorithm.
The modified Vidal 1991 algorithm (Vidal 1991; Yu et al. 2008). The maximum
standard deviation is less than 1.0 K for this algorithm.
The modified Sobrino 1993 algorithm (Sobrino et al. 1993; Yu et al. 2008). The
maximum standard deviation is only 0.75 K for this algorithm.
The modified Sobrino 1994 algorithm (Sobrino 1994; Yu et al. 2008). The
maximum standard deviation is less than 1.0 K for this algorithm.
Meanwhile, it is found that the modified Sobrino et al. (1993) algorithm with
nonlinear term gave better performance than the modified Sobrino (1994) algorithm
without nonlinear term.
Nevertheless, Yu et al. (2008) found that the modified Ulivieri-1985 algorithm
showed the least sensitivity to the emissivity variation, so they suggest this algorithm as the baseline GOES-R LST algorithm (Yu et al. 2010).
Input (looping)
Parameters
Algorithms
Algorithms
Sensor Spectral
Response Fcns.
Sensor
Brightness
Temperature
Calculation
Temperature
Algorithm
Coeffs.
STD Error
Of
TOA
Radianc
Sensor
Brightness
Regression
Of LST
Filter of
Data
Distribution
Fig. 19.8 Procedure of the algorithm regression analyses
318
D. Sun and Y. Yu