Integrating Machine Learning and Multi-Scale Landscape Analysis to Assess Ecosystem Service Thresholds: A Case Study of The Anzali Basin

Context: Forest-agricultural frontiers face accelerating transformation, yet frameworks integrating spatial processes with ecosystem thresholds remain limited. Objectives: We developed an ML-landscape ecology framework to: (۱) quantify multi-scale fragmentation processes (۲۰۰۰-۲۰۲۴), (۲) identify critical ecological thresholds, and (۳) project landscape connectivity under alternative scenarios to ۲۰۴۰. Methods: Ensemble machine learning (Random Forest, SVM, MLP) with ۱۵ biophysical and socioeconomic drivers applied to Landsat time-series in the Anzali basin, Iran-a biodiversity hotspot containing UNESCO Hyrcanian forests and Ramsar wetlands. Multi-scale analysis employed ۱۲ standardized metrics across patch, class, and landscape levels. Graph-theoretic connectivity modeling assessed structural and functional landscape connectivity. Results: Classification achieved ۹۷.۸% accuracy (Kappa=۰.۹۷), representing ۱۲.۴% improvement over traditional methods. Forest cover declined ۱۰.۵% (۲۰۰۰-۲۰۲۴), with deforestation rates accelerating from ۰.۳۱%/yr to ۰.۶۲%/yr. Multi-scale analysis revealed hierarchical processes: patch-level attrition (-۷۸.۷% mean patch size), class-level dissection (+۲۵۸.۸% patch number), and landscape-level connectivity collapse (-۶۳.۶%). Business-as-usual projections show forest reaching ۳۲.۶% by ۲۰۳۰-crossing the critical percolation threshold (۳۰%) and fragmenting into ۲-۳ isolated clusters. Ecosystem service degradation includes ۳۱% carbon storage loss and ۱۵۵% erosion risk increase. Conclusions: Current trajectories trigger irreversible connectivity breakdown by ۲۰۲۹-۲۰۳۰. The transferable ML-enhanced framework enables threshold-based early warning systems for proactive landscape management in global biodiversity hotspots.

کلیدواژه ها:

Landscape connectivity ، Forest fragmentation ، Percolation threshold ، Machine learning ، Ecosystem services ، Multi-scale analysis