Mapping semi-natural grassland communities using multi-temporal RapidEye remote sensing data

Mapping semi-natural grassland has become increasingly important with regard to climate variability, invasive species, and the intensification of land use. At the same time, adequate field data collection is of pivotal importance for national and international reporting obligations, such as the European Habitats Directive. We present a remote-sensing-based monitoring framework for a Natura 2000 site with a heterogeneous composition of different grassland communities, using the Random Forest algorithm. Automated training data selection was successfully implemented based on the Random Forest proximity measure (Overall Accuracy ranging from 77.5–86.5%). RapidEye acquisitions originating from the onset of vegetation (prespring and first spring) and senescence (late summer and first autumn) were identified as important phenological phases for mapping semi-natural grassland communities. The derived probability maps of occurrences for each grassland class captured transitions between grassland communities and are therefore a better approximation of real-world conditions compared to classical, discrete maps.