4.2 Quantitative Evaluatiion of Rice-Planted Area Extraction by RADARSAT
It is quite difficult to perform quantitative evaluation on the accuracy of rice-planted area extraction by RADARSAT because the rice areas extracted by both of SPOT and NLDI-landuse are also not perfect. However, the rice areas by SPOT are considered to be better than those by NLDI because of its fine ground resolution and synchronization of observation by RADARSAT and SPOT, although some miss-classification by SPOT should be taken account.

We defined two indices, true production rate (TPR) and false production rate (FPR) for rice areas by RADARSAT. TPR is the coincidence rate of rice areas by RADARSAT within those by SPOT and FPR is the rate of non-rice areas by SPOT within rice area by RADARSAT. As the rice area images extracted by RADARSAT are still contaminated by speckle noises as shown in Figure 3, the majority filter with 7 by 7 window was applied once or twice to the rice extracted images by RADARSAT before evaluation. The rice extracted image by SPOT was also filtered once by the same majority operation as RADARSAT to make the ground resolution compatible each other.

Table 1 show the result of TPR and FPR by the thresholding method for the data pair Apr. 8 and may 26. As the threshold value, -2dB, -3db and -4dB were set. Table 1 shows almost equal rate for TPR and FPR in all cases, in which he case with -3 dB threshold and filtering twice is the best for the value of TPR-FPR. No other data combination (namely (May 26-June 19)) could obtain the better result for TPR-FPR.

Table 2 shows the result by ML classifier of multi data RADARSAT data. The results are slightly better than those in Table 1. The best result for TPR-FPR was obtained by using all three data and filtering twice, although there are only small differences between two dates and three dates.

Theshold Value	Filtering times	TPR (%)	FPR (%)	TPR -FPR
-2 dB	1	53.3	54.0	-0.7
	2	52.2	47.4	4.8
-3dB	1	42.2	39.9	2.3
	2	40.2	34.5	5.7
-4dB	1	31.3	31.6	-0.3
	2	28.8	28.1	0.7

Table 1 Result of rice area evaluation by RADARSAT (part-1).
(Thresholding of data on Apr.8 and May 26)

used data	Filering TPR times	TPR (%)	FPR (%)	TPR -FPR
Apr. & May	1	44.0	36.5	7.5
	2	42.0	31.8	10.2
Apr. & May & June	1	43.6	34.2	9.4
	2	41.6	30.0	11.6

Table 2 Result of rice area evaluation by RADARSAT (part-2).
(ML classifer of multi-data data)
In general, the rice area extraction by RADARSAT did not give much high coincidence rate as that by SPOT in the quantitative evaluation. Over 50 percent of rice areas by SPOT were not extracted by RADARSAT and about 30 percent of the areas by RADARSAT were outside areas of rice by SPOT. This result suggests that there might be some limitation for the performance of rice-planted area extraction only by RADARSAT. Therefore, to realize a practical rice area estimation, a combination of RADARSAT and other data sources, for example, optical sensor images or other thematic maps based on ground survey should be considered.

5. Conclusions
Rice-planted area extraction was attempted using multi-temporal RADARSAT data taken in an early stage of rice growing season. The overall rice distribution patterns extracted by RADARSAT showed a farily good coincidence with those by SPOT/HRV data taken in almost same period of RADARSAT observation. However, by a quantitative evaluation, the rice areas by RADARSAT resulted in poor coincidence rate with those by SPOT.

As the result of this study, the possibly of RADARSAT for ricer area estimation was verified, especially, the possibility for estimation in a very early stage, namely just after rice plating, was verified, However, in the same time, some limitation for precise area estimation by RADARSAT was also revealed. Therefore, the further study on the practical for rice area estimation by remote sensing should be continued.

Acknowledgements
We thank to National Space Development Agency of Japan (NASDA) for the corporation in RADARSAT and SPOT data acquisition.

References

• Suga Y., Y. Oguro, S.Takeuchi and K.Tsuchiya, 1999. COMPARISION OF VARIOUS SAR DATA FOR VEGETATION ANALYSIS OVER HIROSHIMA CITY, Adv. Space Res. 23(8), pp. 1509-1516.
• Ribbes, F. and T.Le Toan, 1999. Rice Field mapping and monitoring with RADARSAT data, INT.J.R. REMOTE SENSING, 20(4), pp. 745-765.
• Liew, S.C., P.Chen et al, 1999. Changes in Rice Cropping System Using Space-Borne SAR Imagery, IGARSA'99 proceedings, pp. 741-743.