Data
This study used multi-temporal SPOT images to estimate the atmospheric aerosol optical depth. The data set includes 6 SPOT imageries acquired between April to August, 1998 covering Chung-Li area. The geometric information of the data set is shown in Table 1. Meanwhile, the sunphotometer data were collected at the same time. The aerosol depths derived from sunphotometer data were used to verify the accuracy of the method proposed in this study (Table 2).

Table 1. The geometric information of SPOT images used in this study

Date	Local Time	View Angle		Solar Angle
		Zenith	Azimuth	Zenith	Azimuth
1998/04/24	02:23	-29	193	24	117
1998/05/11	02:27	-23	193	20	107
1998/06/27	02:24	-30	193	21	89
1998/07/02	02:28	-24	193	20	90
1998/07/30	02:59	31	193	15	110
1998/08/21	02:36	-9	193	22	119

Table 2. Aerosol optical depth collected by sunphotometer.

Date	1020 nm	870 nm	670 nm	440 nm
1998/04/24	0.133	0.157	0.249	0.470
1998/05/11	0.272	0.326	0.499	0.860
1998/06/27	0.125	0.141	0.215	0.418
1998/07/02	0.088	0.092	0.151	0.296
1998/07/30	0.159	0.175	0.259	0.478
1998/08/21	0.050	0.038	0.059	0.105

Result and Analysis
The singular structure function pattern are depicted in the Figure 2(a). As mentioned previously, these poor structure functions are mainly caused by satellite observation geometry and complex landcover distribution. These patterns would induce further errors in the optical depth estimation. When multi-directional structure function is employed as illustrated in Fig 2(b), the improvement is evident.


Figure 2 (a) The variaties of the structure function with different distances of SPOT XS1 channel.



Figure 2 (b) Same as (a), except for multi-directional structure function.

Further analysis showed that the some abnormal patterns still existed in our approach. Therefore, the “optimal number” decision was added to determine the proper distance value, d and made sure that the abnormal structure function patterns can be removed.