Preparation of land subsidence susceptibility map using machine learning methods based on decision tree (case study: Isfahan–Borkhar)[تهیة نقشة حساسیت فرونشست زمین با رو شهای یادگیری ماشین مبتنی بر درخت تصمیم )مطالعة موردی: منطقة اصفهان-برخوار(]

Abstract

Land subsidence represents a severe environmental hazard, causing significant infrastructure damage and threatening cultural heritage sites. The Isfahan–Borkhar region of central Iran, with its dry climate and diverse topographical conditions, has been highly susceptible to this phenomenon. Using remote sensing techniques, particularly radar interferometry (InSAR), this study investigates subsidence rates over the 2019–2023 period. Advanced machine learning methods, namely Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) are employed to develop a susceptibility map divided into five probability classes: very high, high, medium, low, and very low. The analysis incorporates 145 Sentinel-1 radar satellite images and factors such as elevation, groundwater levels, rock composition, vegetation cover and fault proximity. Among these, RF emerges as the most effective algorithm, achieving a classification accuracy of 95.63%, while XGBoost proved inefficient for certain critical subsidence zones. Results reveal that subsidence risk is concentrated in the central and eastern parts of the region due to excessive groundwater extraction and geological vulnerabilities. Conversely, the western and northwestern areas exhibit lower risk due to stable geological formations and controlled groundwater usage. These findings aim to inform regional planning and subsidence mitigation strategies. © 2025 University of Isfahan.