Preprint_Alpine peatlands: a multisensory pipeline allows for harmonised detection and reveals widespread degradation in the European Alps

Alpine peatlands are critical, disproportionate carbon sinks and biodiversity reservoirs, yet their small size, rugged terrain, and rapid change hamper consistent mapping and monitoring. We developed a cross-country, multi-sensor workflow for the Central European Alps, to detect peatlands and diagnose where peatlands may be degraded. More than 600 ground-truth polygons were selected from >1,500 mapped, multinational sites, after applying a minimum-mapping unit (4 pixels ≈ 1,700 m²). We investigated the best combination of Sentinel-2 optical features, Sentinel-1 backscatter, Landsat-8 surface temperature, and terrain metrics using a ridge regression model in a k-fold cross-validation, noting Fβ-scores, alongside false-negative (FNR) and false-positive (FPR) rates. The best configuration (Sentinel-1 + Sentinel-2 + terrain) achieved a mean F₀.₅ of 0.415 ± 0.290; with optical features dominating the model. Mean FNR (0.575 ± 0.311) substantially exceeded FPR (0.0369 ± 0.0332), indicating that the model misses peatland pixels from the ground-truth—especially where orthophotos reveal disturbances such as forest encroachment or agricultural use. Apparent “false positives” often coincide with plausible unmapped peatlands. Aggregating divergencies between predictions and ground-truth in ~250 km² hexagons produced a hotspot layer that prioritizes re-monitoring, boundary refinement. This scalable, open workflow enables harmonised, cross-border peatland updates in complex mountain terrain and provides policy-ready products to target conservation and restoration where high-elevation carbon sinks and biodiversity hotspots are most at risk.