Abstract
We propose a method to distinguish the land cover of paddy fields with high resolution satellite data. The method is a combination of image classification and visual interpretation. First, pixel-based image classification of land cover is performed using a supervised classification method. Next, land cover of paddy fields is distinguished by overlaying the resulting images with parcel boundary data of paddy fields, and consolidating the classification results for each parcel (that is, extracting the class with the majority of classified pixels in each parcel). Last, confirmation and corrections of the primary distinction results of all paddy fields is performed by visual interpretation of the satellite image. We distinguished land cover of paddy fields in flat paddy field areas using WorldView-2 satellite data (8-band pan-sharpened image with a resolution of 50 cm) by the proposed method, and verified the distinction accuracy using on-site survey data of paddy field land cover. In the distinction using satellite data acquired on July 21, the number of parcels where land cover was correctly distinguished for 1,527 parcels was 1,517 with a distinction accuracy of 99.3%. In the distinction using satellite data acquired on August 23, the number of parcels where land cover was correctly distinguished for 1,158 parcels was 1,143 with a distinction accuracy of 98.7%.
