Use of TanDEM-X and Sentinel Products to Derive Gully Activity Maps in Kunene Region (Namibia) Based on Automatic Iterative Random Forest Approach

Gullies are landforms with specific patterns of shape, topography, hydrology, vegetation, and soil characteristics. Remote sensing products (TanDEM-X, Sentinel-1, Sentinel-2) serve as inputs into an iterative algorithm, initialized using a micro-mapping simulation training data, to map gullies in the northwestern Namibia. A Random Forest Classifier examines pixels similar characteristics pool unlabeled gully objects detected where high densities enclosed by alpha shape. Gully used subsequent iterations following mechanism algorithm uses most reliable samples. The class continuously grows until optimal scenario terms accuracy is achieved. Results benchmarked manually tagged (initial labeled area <; 0.3% total study area) two different watersheds (408 302 km 2 , respectively) yielding accuracies >98%, 60% class, Cohen Kappa >0.5, Matthews Correlation Coefficient receiver operating characteristic Area Under Curve >0.89. Hence, our method outlines keeping low false-positive rates while classification quality has good balance for classes (gully/no gully). show significant descriptors temporal radar signal coherence (22.4%) variability Normalized Difference Vegetation Index (21.8%). This research builds on previous studies face challenge identifying outlining gully-affected areas shortage data global datasets, which then transferable other large (semi-) arid regions.