UHI-Bench Benchmarks Dual-Source Urban Heat Island Modeling

The intensifying urban heat island (UHI) effect, driven by climate change, poses significant thermal exposure risks. Current research often relies on single-source UHI observations (either land surface temperature or near-surface air temperature), which can lead to biased assessments of human heat exposure due to their distinct physical characteristics. Furthermore, integrating dynamic meteorological data with static urban morphology features is challenging due to spatiotemporal incompatibilities and data gaps. To address these limitations, researchers have introduced UHI-Bench, the first comprehensive benchmark specifically designed for dual-source UHI modeling. This benchmark integrates both dynamic and static environmental contexts, evaluating over 20 baseline models from four families across five tasks. The study spans 20 cities located in nine different Köppen climate classes, providing a diverse and robust testing ground. Key findings from UHI-Bench indicate that no single model consistently outperforms all others, though foundation models demonstrate competitive and stable performance. Environmental covariates generally improve modeling accuracy, but their utility varies depending on the UHI source and task. Crucially, cross-city transferability of UHI models is better explained by the overlap in UHI regimes rather than simple climate zone similarity. This work, along with its standardized dataset and pipeline, offers practical guidance for urban heat modeling and promotes climate data equity.