Modeling downwind, urban heat island and rainfall, using a machine learning–integrated spatiotemporal framework

Abstract

This study investigates the day-to-day relationship between the Urban Heat Island (UHI) effect and precipitation in Houston, Texas, using a data-driven, multi-layer spatiotemporal framework. By integrating satellite observations, reanalysis datasets, and wind-field information with machine learning based reconstruction of missing thermal data, the approach overcomes observational gaps and identifies atmospheric conditions favorable for UHI–precipitation interactions. Results show that approximately 290 out of 800 critical days exhibit strong alignment between UHI intensity, prevailing wind direction, and rainfall occurrence, confirming that UHI-induced precipitation enhancement is highly conditional and direction-dependent. Thermo-hygrometric clustering further reveals four atmospheric regimes linking temperature, humidity, and UHI intensity to rainfall patterns. Notably, nearly 48% of all events occur within a warm–humid regime characterized by mean temperature of 28.3°C, mean relative humidity of 83.6%, and mean UHI intensity of 5.8°C, conditions that promote convective uplift and moisture convergence. The findings demonstrate that static or mean-based analyses obscure transient feedback between urban heating and atmospheric dynamics. The proposed framework highlights the necessity of combining machine learning, downwind diagnostics, and high-resolution temporal coupling to better understand complex UHI–rainfall interactions, offering actionable insights for urban planning, hazard mitigation, and climate-resilient infrastructure design.