Feature Fusion Approach for Temporal Land Use Mapping in Complex Agricultural Areas

Accurate temporal land use mapping provides important and timely information for decision making large-scale management of crop production. At present, cover classifications within a study area have neglected the differences between subregions. In this paper, we propose classification rule by integrating terrain, time series characteristics, priority, seasonality (TTPSR) with Sentinel-2 satellite imagery. Based on Normalized Difference Water Index (NDWI) Vegetation (NDVI), dynamic tree forests, cultivation, urban, water was created in Google Earth Engine (GEE) each subregion to extract cultivated land. Then, or without mask data, original results were completed based composite image acquisition five vegetation indices using Random Forest. During post-reclassification process, 4-bit coding type, seasonal rhythm, priority generated analyzing characteristics results. Finally, statistical processed. The showed that feature importance dominated B2, NDWI, RENDVI, B11, B12 over winter, B12, NDBI, B8A summer. Meanwhile, improved overall accuracy multicategories (seven eight nine 13 during winter summer, respectively) subregion, average ranges summer 0.857–0.935 0.873–0.963, respectively, kappa coefficients 0.803–0.902 0.835–0.950, respectively. analysis above comparison resampling plots identified various sources error accuracy, including spectral differences, degree field fragmentation, planting complexity. demonstrated capability TTPSR mapping, especially regard complex crops automated post-processing, thereby providing viable option mapping.