Detectability of Forest Peatlands in the Temperate Zone Using Random Forest and Open Remote Sensing Data – A Case Study of Northeast Brandenburg, Germany

Environmental services of mires and peatlands and negative impacts of their alteration are comprehensively documented. A spatial detection of these organic soils is therefore essential. This paper examines the detectability of organic soils in forests using open ge-ospatial and remote sensing data combined with mapped soil and water level infor-mation in two Random Forest (RF) approaches. Either surrounded in or covered by for-est, organic soils of the study region exhibit elevated soil water content reaching satura-tion during the hydrological winter. Consequently, terrain indices from Digital Elevation Models (DEM) and soil moisture from L-band ALOS PALSAR signals are used as pre-dictors in RF algorithms. CORINE Land Cover data help assess how different forest cov-er types (FCT) influence RF models. Substantial agreement is reached in the target classi-fication when FCT is included and when using the higher spatial resolution DEM. The Boolean approach is less affected by different compositions of predictor variables, but is more sensitive to the level of imbalance in the reference data. This becomes evident when comparing the “event error” and “no event error”. In all RF models, the variable importance of soil moisture pixel values retrieved from L-band ALOS PALSAR is the highest when FCT is not in-cluded.