Land Cover and Crop Type Classification along the Season Based on Biophysical Variables Retrieved from Multi-Sensor High-Resolution Time Series

摘要: With the ever-increasing number of satellites and availability data free charge, integration multi-sensor images in coherent time series offers new opportunities for land cover crop type classification. This article investigates potential structural biophysical variables as common parameters to consistently combine exploit them land/crop Artificial neural networks were trained based on a radiative transfer model order retrieve high resolution LAI, FAPAR FCOVER from Landsat-8 SPOT-4. The correlation coefficients between field measurements retrieved 0.83, 0.85 0.79 FCOVER, respectively. variables’ displayed consistent average temporal trajectories, even though class variability signal-to-noise ratio increased compared NDVI. Six random forest classifiers applied along season with different inputs: spectral bands, NDVI, well FAPAR, LAI separately jointly. Classifications reached end-of-season overall accuracies ranging 73%–76% when used alone 77% corresponds 90% 95% accuracy level achieved bands appears be most promising variable When assuming that cropland extent is known, classification reaches 89% information, 87% NDVI 81%–84% variables.