Responding to Large-Scale Forest Damage in an Alpine Environment with Remote Sensing, Machine Learning, and Web-GIS

This paper reports a semi-automated workflow for detection and quantification of forest damage from windthrow in an Alpine region, particular the Vaia storm October 2018. A web-GIS platform allows to select damaged area by drawing polygons; several vegetation indices (VIs) are automatically calculated using remote sensing data (Sentinel-2A) tested identify more suitable ones quantifying cross-validation with ground-truth data. Results show that mean value NDVI NDMI decreased areas, have strong negative correlation severity. RGI has opposite behavior contrast NDMI, as it highlights red component land surface. In all cases, variance VI increases after event between 0.03 0.15. Understorey not windthrow, if consisting 40% or total cover area, undermines significantly sensibility VIs detecting predicting Using aggregational statistics (average standard deviation) over polygons input machine learning algorithm, i.e., Random Forest, results severity prediction regression reaching root square error (RMSE) 9.96, on scale 0–100, ensemble averages deviations NDVI, indices. The combining than one can improve estimation severity, tools support decisions selected VIs. reported prove Sentinel-2 imagery be deployed analysed via web-tools estimate used damage, careful evaluation effect understorey each situation.