Cropland Mapping Using Fusion of Multi-Sensor Data in a Complex Urban/Peri-Urban Area

Urban and Peri-urban Agriculture (UPA) has recently come into sharp focus as a valuable source of food for urban populations. High population density and competing land use demands lend a spatiotemporally dynamic and heterogeneous nature to urban and peri-urban croplands. For the provision of information to stakeholders in agriculture and urban planning and management, it is necessary to characterize UPA by means of regular mapping. In this study, partially cloudy, intermittent moderate resolution Landsat images were acquired for an area adjacent to the Tokyo Metropolis, and their Normalized Difference Vegetation Index (NDVI) was computed. Daily MODIS 250 m NDVI and intermittent Landsat NDVI images were then fused, to generate a high temporal frequency synthetic NDVI data set. The identification and distinction of upland croplands from other classes (including paddy rice fields), within the year, was evaluated on the temporally dense synthetic NDVI image time-series, using Random Forest classification. An overall classification accuracy of 91.7% was achieved, with user’s and producer’s accuracies of 86.4% and 79.8%, respectively, for the cropland class. Cropping patterns were also estimated, and classification of peanut cultivation based on post-harvest practices was assessed. Image spatiotemporal fusion provides a means for frequent mapping and continuous monitoring of complex UPA in a dynamic landscape.