Land Surface Temperature as a Proxy for Outdoor Thermal Comfort?

The complexity of urban environments requires advanced methods to assess human outdoor thermal comfort (OTC). Remote sensing of Land Surface Temperature (LST) has been extensively used to map urban thermal patterns, yet its applicability in assessing OTC remains controversial due to inherent limitations. This study aims to bridge the gap between LST observations and high-resolution OTC assessments using the Universal Thermal Climate Index (UTCI).

We compare Landsat LST data with high-resolution UTCI maps from the deep learning-based OTC model HTC-NN (Briegel et al., 2024) to evaluate LST as a proxy for OTC across various urban and rural areas. Cluster analysis examines spatial variations in the LST-UTCI relationship, while random forest modeling assesses the predictive ability of LST, meteorological data (ERA5 Land), and spatial variables for UTCI values and heat stress classes, using the ERA5 HEAT dataset (Di Napoli et al., 2020) as a baseline.

Our findings show a linear relationship between LST and UTCI under non-heat stress conditions, which becomes non-linear during heat stress events. Cluster analysis identifies distinct spatial patterns in the LST-UTCI relationship, influenced by land cover and urban form. In densely built-up areas, LST and UTCI show less agreement, with an average difference of ~10K, compared to -0.8K in vegetated areas, which highlights the limitations of LST in capturing pedestrian-level thermal stress in dense urban environments. Random forest models using LST alone show low predictive power for UTCI and heat stress classes, but performance improves when combined with ERA5 Land data. Models incorporating both achieve 82% accuracy for UTCI stress classes, surpassing the 77% accuracy of the ERA5 HEAT dataset. In urban clusters, the random forest model demonstrates significantly lower error (RMSE 2.6K) compared to ERA5 HEAT (RMSE 4.8K). Overall, this study underscores the potential limitations of LST as a standalone metric for OTC evaluation.