Verlag der Österreichischen Akademie der Wissenschaften

Abstract:
This study investigates the combination of images by both radar and optical sensors for environmental modelling in dynamic landscapes. The study area is located in Eastern Kenya and contains three of the largest refugee camps in the world. The fast provision of reliable information about the environmental conditions is needed by humanitarian organizations in order to coordinate their work and help the displaced people. We conducted a land cover classification based on RapidEye, Sentinel-1 and SRTM data using a Random Forest approach. Different combinations of input data were compared in order to achieve best results. The output accuracy was 84 % based on 17 predictor layers. These were generated using second order SAR texture filters and multi-scale topographic indices. Especially a combination of all three input types led to promising results.