Classifying Sub-Parcel Grassland Management Practices by Optical and Microwave Remote Sensing

Grassland management practices and intensities are key factors influencing the quality balance of their provisioning regulating ecosystem services. Most European temperate grasslands exploited through mowing, grazing, or a combination both in relatively small units. Grazing mowing can however not be considered equivalent because first is gradual selective second not. In this study, aim to differentiate terms retrieve homogeneous Grasslands classified hierarchically, pixel-based supervised classification grazed pastures from mown hay meadows then an object-based detection method timing frequency events. A large field dataset was used calibrate validate method. For classification, 18 different input feature combinations derived Sentinel-1 Sentinel-2 were tested for random forest classifier cross-validation scheme. The best results obtained based on Leaf Area Index (LAI) times series with cubic spline interpolation. differentiated (grazed) (mown) overall accuracy 88%. combined existing parcel delineation high-resolution ancillary data units, which coherence NDVI. performances increased thanks grassland mask, unit delineation, exclusion pastures, reaching precision 93% rate 82%. This hierarchical approach allowed three types grasslands, namely (including mixed practices) early event late event, 79%. could further by frequency, resulting five final classes.