Enhancing desertification risk mapping with entropy-based weighting and machine learning: Insights from Iraq

Desertification poses a significant environmental threat to arid and semi-arid regions, with Iraq experiencing escalating land degradation due to climate variability and human-induced pressures. This study refines desertification risk assessment by integrating an enhanced MEDALUS model with entropy-based weighting, machine learning, and uncertainty quantification. Unlike conventional fixed-weight indices, the entropy-weighted approach objectively determines the influence of key environmental and anthropogenic factors, improving classification accuracy and spatial differentiation. The analysis incorporates satellite-derived vegetation indices (NDVI, SAVI, FVC), field-measured soil properties, climatic parameters, and human activity indicators. A hybrid Desertification Risk Index (DRI) is computed using a combined geometric mean and entropy-based summation approach, ensuring a more adaptive and regionally specific risk classification. Results reveal that nearly half of Iraq's land area falls within moderate to extreme desertification risk zones, with climatic aridity and vegetation health emerging as the strongest predictors of land degradation. The machine learning-based classification using XGBoost refines risk categorization, achieving an overall accuracy of 94.46 % (κ = 0.9261), highlighting the limitations of traditional approaches. SHAP analysis provides a transparent interpretation of model predictions, confirming that climatic variability and vegetation stability exert the greatest influence on desertification processes. Monte Carlo simulations quantify classification uncertainty, particularly in transition zones where mixed land-use and environmental heterogeneity contribute to higher prediction variability. The entropy-weighted approach enhances spatial precision in desertification risk mapping, addressing methodological gaps of previous studies. This research integrates statistical weighting, remote sensing, and machine learning, providing a robust, data-driven framework for land management and policy to mitigate land degradation. • Enhanced desertification risk mapping using entropy-weighted MEDALUS model. • Integrated machine learning and remote sensing for improved risk classification. • XGBoost model achieved 94.46 % accuracy in identifying desertification zones. • SHAP analysis highlighted climatic variability and vegetation health as key risk drivers. • Monte Carlo simulations quantified uncertainty, improving spatial precision in risk mapping.

Перевод пока недоступен