A Probabilistic Framework for Predicting Spatiotemporal Intensity and Variability of Outdoor Thermal Comfort

Abstract

Thermal conditions in the urban canopy exhibit stochastic variability driven by varied radiative fluxes and turbulent wind fields, requiring probabilistic rather than deterministic prediction methods. This study presents a probabilistic framework for predicting the spatial and temporal intensity and variability of outdoor thermal comfort in tropical urban environments. The framework integrates ground-measured meteorological data and remote sensing urban morphological data to calculate Physiological Equivalent Temperature (PET), and applies K-means, XGBoost, and Monte Carlo simulations on PET training and inference. The prediction model achieved strong performance, with R2, RMSE, and SMAPE values of 0.93, 0.81 degC, and 1.34% for PET_mean, and 0.85, 0.38 degC, and 10.44% for PET_std, respectively. A case study showed clear spatial heterogeneity of outdoor thermal comfort. Locations with dense tree canopies and vegetated surfaces displayed a normalized percentage of acceptable thermal comfort (NATC) up to 65%, whereas built-up zones dominated by impervious surfaces, such as industrial estates and high-density residential areas, recorded NATC below 30%. Greenery was found to mitigate both the intensity of heat stress and its variability, producing a stable and comfortable microclimate. Daytime PET_std ranged from 4.0-4.5 degC in built-up areas to 1.5-2.0 degC in greenery-covered zones, while nighttime PET_std decreased to 2.2-2.4 degC and 1.2-1.4 degC, respectively. These findings emphasize the critical role of greenery in mitigating thermal variability and enhancing outdoor thermal comfort, while revealing the stochastic nature of thermal comfort across different urban morphologies.