Spatial-temporal modeling of heat hazards for a high-density subtropical city: A case study of Hong Kong

Heatwaves are becoming more frequent, intense, and prolonged over the last two decades. Air temperature-based indicators more accurately capture heat hazards than commonly used land surface temperature measures, yet few studies have explored their relationship with urban characteristics or undertaken long-term spatial-temporal modeling. To fill these gaps, this study explores the applicability of land use regression, machine learning methods, and stacking ensemble learning to map Very Hot Day Hours and Hot Night Hours from 2000 to 2023 in Hong Kong, using meteorological data from 40 ground-level stations. Feature importance analyses are used to clarify how urban characteristics influence heat hazards. With the yearly heat hazard maps produced, the long-term trend of heat hazards in different areas is also examined. It is found that the stacking ensemble learning method further reduced the mean absolute error from 38.32 to 35.19 for Very Hot Day Hours, and from 96.55 to 88.73 for Hot Night Hours. Wind and elevation are found to be critical in mitigating daytime heat hazards, while vegetation is found to be more important in mitigating nighttime heat hazards. Practical implications for increasing community heat resilience are also provided.