Towards Climate-Resilient Cities: Exploring Meteorological Drivers of Urban Heat Variability with Explainable Machine Learning

Urban areas are increasingly vulnerable to extreme thermal conditions due to rapid urbanization and climate change. The accurate prediction of ambient air temperature (AT) at fine temporal scales is essential for mitigating the impacts of urban heat waves, heat pockets, and heat islands. Despite ongoing research, limited interpretability of traditional AI models has constrained their utility in decision-making. This study aims to improve real-time temperature forecasting in the Central National Capital Region (Central-NCR) of India through explainable machine learning techniques. Hourly AT was modeled using four advanced machine learning algorithms—Random Forest (RF), Gradient Boosting (GB), XGBoost, and LightGBM (LGBM). A structured workflow was followed involving data preprocessing, hyperparameter tuning, cross-validation, model training, and evaluation. Model performance was compared using residual plots, validation curves, and statistical metrics including RMSE, MAE, MSE, R², MAPE, and Explained Variance Score (EVS). A Taylor dia-gram was used for holistic model comparison. Among all tested models, RF demonstrated the highest predictive accuracy, achieving an R² of 0.81 and the lowest RMSE of 3.36 during the test phase. Relative humidity (RH) and barometric pressure (BP) emerged as the most influential predictors. SHAP analysis further confirmed RH, BP, and solar radiation (SR) as key drivers of AT variability. Seasonal patterns indicated that increased RH during monsoon months reduced AT, while elevated SR levels during summer contributed to higher temperatures. Dependence and partial dependence plots revealed non-linear interactions: RH exhibited a strong inverse relationship with AT, SR drove exponential increases, and BP displayed oscillatory patterns reflective of atmospheric fluctuations. The integration of explainable AI techniques with meteorological data enables more accurate and interpretable urban temperature forecasting. These insights can support policymakers and urban planners in developing informed strategies for heat mitigation, regulatory compliance, and climate adaptation.