Satellite-guided, AI-enhanced framework for assessing UHI-driven cooling and heating loads

Quantifying how urban heat islands (UHIs) influence regional electricity consumption remains challenging because station-based indicators and prescribed heating/cooling degree-day (HDD/CDD) thresholds often fail to capture intra-urban thermal heterogeneity and nonlinear demand responses during extremes. This study addresses these limitations by combining machine learning with MODIS thermal remote sensing to reconstruct gap-free daily land surface temperature (LST) fields across Local Climate Zones (LCZs) in the Swiss cantons of Vaud and Geneva (2015–2024) and by deriving demand-relevant temperature thresholds directly from observations. MODIS Terra and Aqua LST are integrated with in-situ meteorological observations, sub-pixel LCZ fractions, and urban morphological predictors to represent within-pixel thermal variability associated with urban form. Multiple machine-learning regressors are benchmarked for LST reconstruction, with XGBoost consistently achieving the highest performance across cantons, satellite overpasses, and independent field stations. Reconstructed LST fields are then linked to decade-long electricity demand records using an interpretable piecewise-linear hinge model, which infers empirical heating and cooling transition temperatures (∼19.5 °C and ∼30.3 °C) from observed demand–LST relationships rather than prescribing them. These thresholds enable LST-based degree-day metrics for mapping exceedances and estimating LCZ-resolved demand sensitivities. Heating sensitivity (demand–temperature slope) is relatively uniform across LCZs, whereas cooling sensitivity varies strongly with urban form—being highest in compact and open mid-rise LCZs and lower in sparsely built and industrial zones. Cooling sensitivity is pronounced on working days but negligible on weekends/holidays, indicating a predominantly non-residential cooling signal, consistent with limited residential air-conditioning adoption in this temperate region. Overall, the framework provides a transferable, spatially explicit basis for quantifying LCZ-specific links between surface-UHI exposure and electricity demand and for identifying urban forms most likely to amplify peak-load risk under intensifying heat extremes.