Reference-anchored envelope modelling for predicting tree mortality hotspots driven by drought

Kurzfassung

Predicting tree mortality hotspots driven by drought remains challenging, since vulnerability to drought is strongly influenced by climate, landform, and substrate. We propose a novel reference-anchored, quantile-envelope susceptibility model-implemented through a reproducible workflow- that exports mortality-resemblance (high-risk) environments from three confirmed tree-mortality footprints (training AOIs), based on semi-arid conifer and evergreen sclerophyllous forests in Cyprus, to produce wide screening layers. The approach combines (i) physiographic similarity derived from AOI-calibrated envelopes for slope, multi-scale topographic position and curvature (broad quantiles; majority-rule), (ii) exposure constraints via circular aspect windows, warm-season solar loading (April-September) and substrate (lithology) filtering, (iii) a drought severity gate (SPEI-12 (Standardized Precipitation-Evapotranspiration Index-12) ≤ -1.4), and (iv) Landsat event-year spectral collapse relative to a fixed 3-year pre-event baseline, coupled with a year + 1 persistence rule to suppress transient stress. Our within-AOI evaluation (forest-only) showed substantial representativeness (coverage 70.49–85.35%), while recovering most reference mortality pixels (recall 67.62–100%). In 2025, two independent mortality sites were identified in Cyprus, coinciding with exceptional drought (SPEI < -2) and a sharp decline in the Normalized Burn Ratio (NBR) consistent with foliage browning, and occurring within high-risk zones as mapped in the final risk map. As a plausibility check, both 2025 sites fell within pre-mapped high-risk zones, with 83.50% and 82.69% of their forested area exceeding HEAT ≥ 0.60 and mean risk scores of 84.91% and 75.55%, respectively. With 18.88% of Cyprus’s total forest flagged as high-risk, the probability of this co-occurrence being random is p ≈ 0.036 (binomial test). Anchoring calibration to local mortality footprints makes the approach readily transferable to other regions with documented events.