Assessing Urban Heat Island Impact on Extreme Rainfall Events in Nanjing Using Machine Learning

Abstract

With accelerating global urbanization, the impact of urban heat island (UHI) effects on extreme rainfall events has become increasingly significant, posing severe challenges to urban climate resilience. This study investigates the spatiotemporal coupling relationship between UHI effects and extreme precipitation events in Nanjing, China, from 2013 to 2022, using multi-source remote sensing data and machine learning approaches. We integrated MODIS land surface temperature data, meteorological observations from 35 national stations and 255 automatic weather stations, along with high-resolution land use datasets to comprehensively analyze UHI-precipitation interactions. XGBoost model was used to decode nonlinear coupling mechanisms between thermal anomalies and extreme rainfall events. Results reveal distinct seasonal variations in UHI intensity, with summer showing the strongest effects (average 18.16°C temperature difference) concentrated in the Qinhuai district. Spatial autocorrelation analysis demonstrates significant clustering patterns (Moran's I = 0.075-0.088, p < 0.01) with consistent hotspot agglomerations. The UHI centroid exhibited systematic northward migration during summer (14.5 km over the decade), aligning with urban expansion patterns. Urban areas showed 16.2% higher extreme precipitation frequency during summer compared to rural areas, with strong positive correlations between UHI intensity and rainfall (r = 0.55 in 2016). The XGBoost model achieved moderate predictive performance (R² = 0.508, RMSE = 0.852) for UHI-precipitation relationships. These findings provide scientific foundations for climate-adaptive urban planning and highlight the need for spatially-targeted flood control strategies in rapidly urbanizing regions.