Abstract
Agricultural statistics is fundamental reference for evaluating damage of a natural disaster, estimating food supply and demand, and framing a policy. The remote sensing technique is necessary for estimating the area of agricultural land in a large region or a difficult area for an on-site survey. The authors developed a method to classify land-use/land-cover, especially paddy field, using the modified normalized difference water index (MNDWI) and the normalized difference vegetation index (NDVI) derived from satellite optical sensor data. This method was applied to the Landsat TM (Thematic Mapper)/ETM+ (Enhanced Thematic Mapper Plus) data to detect paddy fields in Aomori Prefecture. The area of paddy field in each district in 2002 was estimated from the area ratio of paddy field in each mixel consisting of paddy field and other categories. The area ratio 100% of paddy field in a mixel was determined to be an average MNDWI - 3σ of water areas in the rice-planting season (0.15 for the Path 107 and 0.10 for the Path 108); the area ratio 0% of paddy field, an average MNDWI + 2σ of ground and built-up (-0.17). As the result, the total area of paddy field in Aomori Prefecture in 2002 was estimated to be 51,283 ha, which was 97.5% of the statistics (52,597 ha). The accuracy of paddy field classification was 93.0-97.7% and that of paddy field detection was 85.0-97.0%. It was shown that the simple classification method developed in this study could reduce working time for the land-use/land-cover classification, comparing with unsupervised or supervised classification such as the clustering or the most likelihood method.
