Identifying areas of homogeneous grassland management based on iterative segmentation of Sentinel-1 and Sentinel-2 data

Remote sensing data in combination with image segmentation approaches have been shown to be valuable for identifying homogeneous areas such as agricultural land parcels. However, these approaches have not been widely explored for their usefulness in identifying homogeneous areas of grassland management under different management intensity levels. We present an unsupervised Bayesian segmentation approach for the combined analysis of Sentinel-1 and Sentinel-2 monthly composite data aiming to identify homogeneously managed grassland parcels, which was tested on an area of about 50 ha. We applied a segmentation refinement procedure with Sentinel-1 data in the first iteration and used Sentinel-2 data in the second iteration to delineate smaller areas within the identified segments. The approach led to promising results in intensively managed grassland areas with many mowing events. The results were also plausible under agriculturally extensive use. However, in semi- natural managed areas and in regions with varying environmental influences the identification of segments was prone to errors. The results show the potential of image segmentation in a grassland context, especially when management data are not available but needed. Our approach, although tested on a small scale, is applicable to larger regions.