Random forest-based approaches for mapping alpine peatlands in the italian alps: integrating remote sensing and environmental drivers

Alpine peatlands are important carbon reservoirs that provide essential ecosystem services and support endangered biodiversity. However, their spatial distribution remains insufficiently studied, particularly in the Italian Alps, where more than 6,000 peatland sites are typically smaller than 1 hectare. Their detection is challenging due to complex mountainous geomorphology and limitations of single-sensor or low-resolution remote sensing data. In addition, the relationship between peatland distribution and environmental factors in this region is not yet well understood. This study develops a mapping framework for alpine peatlands using multi-source satellite imagery and predicts peatland occurrence probability by analyzing environmental controls. The workflow integrates optical, radar, and topographic data, including Sentinel-2, Sentinel-1, and the Copernicus 30 m DEM. Data processing was conducted on the Google Earth Engine (GEE) platform, while DEM-derived variables were generated using SAGA software. A Random Forest algorithm was employed for peatland detection. The approach was first applied to the Avisio Basin (Trentino–Alto Adige), achieving 81.8% true positives and a 0.8% false-positive rate. Its transferability was further tested in the Orco Basin (Piemonte), where it successfully distinguished acid peatlands (>84% true positives) and base-rich peatlands (>90% true positives), with false-positive rates below 0.66% and 0.40%, respectively. A peatland occurrence probability map was then produced using environmental and bioclimatic variables at 100 m resolution. The Random Forest model achieved a high predictive performance (AUC = 0.98), with over 80% of validation points located in areas of high probability (>0.80). These results demonstrate the effectiveness of the proposed approach for mapping alpine peatlands and capturing their environmental drivers.