Is satellite land surface temperature an appropriate proxy for intra-urban variability of daytime heat stress?

Abstract

Adaptation of urban areas to heat extremes requires adequate information on intra-urban variability patterns of outdoor thermal comfort (OTC). Remotely sensed Land Surface Temperatures (LST) are often used to map heat hotspots in urban areas. However, this approach has limitations as LST and OTC are influenced by different physical processes. This study investigates the relationship between satellite-derived Landsat Level-2 LST data and pedestrian-level Universal Thermal Climate Index (UTCI) predictions from a microscale thermal comfort model across Freiburg, Germany. A cluster analysis of the differences is performed, and multiple random forest models are trained using different combinations of LST, ERA5-Land reanalysis, and local-specific urban morphology and land cover data as predictors.

While a linear relationship between LST and UTCI exists under non-heat stress conditions (UTCI <26 °C) and in vegetated or open areas, this becomes non-linear and spatially inconsistent under heat stress, particularly in compact urban environments. The growing divergence between LST and UTCI along an urbanization gradient ranging from −1 K to +9 K highlights the significant impact of urban morphology on the LST-UTCI relationship, leading to substantial intra-urban variability. This variability appears to persist even within similar urban typologies (e.g. LCZs/clusters), with only limited reduction in spatial variability. Random forest models confirm these findings: those based solely on LST or global-scale predictors struggle to capture intra-urban UTC variability, while models incorporating local urban morphology and land cover data outperform them (even without LST input). This suggests that the contribution of LST to neighborhood-scale UTC modeling is limited under certain conditions and environments.