Applications of Satellite Data to the Studies of Agricultural Meteorology

(3) District Classification and Local Temperature Estimation by GMS IR Data

Abstract

In estimating air temperature using ground temperature from GMS IR data, ground effect corrections are equally important as atmospheric effect corrections. However, ground effect corrections are very difficult to conduct, requiring a great amount of analyses because of complicated relationship between air temperatures and ground temperatures.
In our previous study (Tani et al, 1984), it was found that the classification of AMeDAS meteorological sites using deviation between temperatures estimated from the GMS IR data and those obtained from AMeDAS data, indicated that meteorological sites of the same type tend to form groups. This shows that the accuracy of temperature estimations increases when the estimation is carried out by small districts. The classification of AMeDAS meteorological sites in Hokkaido was conducted by cluster analysis using temperature deviations. The following two kinds of data were used for the cluster analysis: sixty-one GMS IR data and AMeDAS data collected from 1979 through 1984. Mean deviations between the temperatures estimated from GMS IR data obtained at three hour intervals and those obtained from AMeDAS data were calculated. Using these deviations, cluster analyses of AMeDAS meteorological sites were made. The results are shown in Fig. 2. Furthermore, AMeDAS meteorological sites were classified based on the deviations between the time average temperatures of AMeDAS and the time temperature of AMeDAS. The results are shown in Fig. 3.
Using the results of this classification, temperatures of the five districts were estimated, shown in Figs. 2 and 3. The temperature estimations of the five districts were conducted using four methods (six different calculation methods) and the accuracies clarified.
1) The regression equation of surface temperature from GMS data and AMeDAS temperature was calculated throughout Hokkaido island. The regression equation was applied to five districts. (named One-equation method)
2) The regression equations were calculated for each cluster. The regression equations were applied to the clustered meteorological sites in five districts. The following two kinds of data were used for the cluster analysis.
a) GMS IR data.
b) AMeDAS data.
(named Cluster method)
3) The regression equations were calculated for each five district. (named District method)
4) AMeDAS meteorological sites were selected based on the classification results of Figs. 2 and 3 for each five district. The regression equations were calculated for those groups. The classifications in Figs. 2 and 3 were calculated using following two data.
a) GMS IR data.
b) AMeDAS data.
(named District-cluster method)
Table 1 shows the estimation accuracy of different calculation methods. The calculation results of method 4 (District-cluster method) were the most accurate with an error of 1.0±0.1 (K).