A climate-smart suitability mapping framework for arid cities: Integrating machine learning and AHP to enhance vegetation resilience

Urban vegetation plays a critical role in enhancing environmental quality and climate resilience, particularly in arid regions facing extreme temperatures, water scarcity, and rapid urbanization. This study evaluates vegetation suitability in Al Ain City, United Arab Emirates (UAE), through an integrated Machine Learning–Analytical Hierarchy Process (ML–AHP) framework designed to assess both baseline and future climate–urban scenarios. Four machine learning algorithms, Categorical Boosting (CatBoost), Gradient Boosting Decision Trees (GBDT), Random Forest (RF), and Support Vector Machine (SVM), were trained using sixteen environmental and socio-economic variables derived from remote sensing datasets and CMIP6 climate projections. The outputs of the ML models for both baseline and CMIP6 SSP5–8.5 scenarios were integrated into an AHP-based multi-criteria assessment, where model performance metrics such as Area Under the Curve (AUC), F1-score, and True Skill Statistic (TSS) were used for pairwise comparisons on Saaty's 1–9 scale. A consistency check of the comparison matrix was performed using the Consistency Index (CI) and Consistency Ratio (CR), and the resulting composite weight, reflecting the relative predictive performance of each model, was applied to generate the final vegetation suitability maps. The results indicate substantial spatial variability and future decline in optimal vegetation conditions. Areas classified as Excellent suitability decreased from 11.6% to 10.3%, while Poor suitability zones expanded and high-quality areas became increasingly fragmented. Centroid analysis revealed a 5.02 km northeastward shift in optimal vegetation, driven by projected temperature increases (+2.6 °C to +4.0 °C), reductions in precipitation (−61 mm to −82 mm), and significant urban expansion (up to 187 km 2 ). Feature-importance analyses identified the Normalized Difference Water Index (NDWI), Land Surface Temperature (LST), and Land Use/Land Cover (LULC) as the most influential factors, with CatBoost achieving the highest accuracy (AUC = 0.93). The framework addresses the urgent need for robust spatial planning tools that can guide adaptive greening strategies in arid cities. It offers a transferable approach for evaluating vegetation resilience under climate and urbanization pressures, providing actionable insights for policymakers and urban planners to design sustainable greening strategies and enhance climate resilience.