The increase of unused agricultural land has become an issue in the Kushiro River watershed in eastern Hokkaido. Information on the spatial distribution of unused agricultural land is important so that we may understand the cause of its abandonment and plan its effective utilization. Satellite remote sensing is an effective approach for mapping unused agricultural land. We classified agricultural land into unused agricultural land, grassland, cultivated land, or forest using Landsat 8 operational land imager (OLI) surface reflectance products (1∼7 bands), the normalized difference vegetation index (NDVI) derived from those products, and polygons of agricultural field in Tsurui Village. Landsat 8 OLI surface reflectance products observed during the snowy season, shortly before harvest season, and late in harvest season were used for the classification. We applied two supervised classification methods: Maximum Likelihood and Random Forest. Accuracy validation was performed based on field survey and Google Earth aerial photography and street views at 30 or more random points for each classification class. The best classification was obtained by Random Forest with the data obtained shortly before harvest season and late in harvest season, which showed 0.92 for overall accuracy (OA), 0.79 for Kappa coefficient (κ), and 0.82 and 0.70 for producer’s (PA) and user’s (UA) accuracy, respectively, with regard to the assigned class of the unused agricultural land. Furthermore, we classified all agricultural land in the entire watershed of the Kushiro River using the best classification method in Tsurui Village (under which the OA, κ, and PA and UA of the unused agricultural land class were 0.89, 0.71, 0.81 and 0.54, respectively). Finally, we mapped the percentage of unused agricultural land and renewable unused agricultural land for each 150m×150m mesh.
