Understanding how specific urban morphology features influence the Urban Heat Island (UHI) effect requires robust characterisation of Local Climate Zones (LCZs) and appropriate definition of thermal source areas. This study developed separate daytime and nighttime Random Forest (RF) models to determine the spatial scale that best represents the thermal source area based on surface cover fractions. The response variable, UHI intensity (UHII), was derived from air temperature data collected from Citizen Weather Stations (CWS) and professionally operated weather stations (PRWS) under stable atmospheric conditions (low cloud cover, low wind speed, low relative humidity, and no precipitation within the preceding 24 hours).
Predictor variables included percentages of impervious surfaces, trees, shrubs, wet and dry grass, water, and bare soil. LCZs were generated using the LCZ Generator. Four hexagon diameters (250 m, 500 m, 750 m, and 1000 m) were tested in 12 directional sectors (0°–330° at 30° intervals), aligned with prevailing wind directions to account for turbulent heat transport toward sensors. Results show that daytime UHII is generally lower than nighttime UHII. Nighttime patterns are similar for LCZ 3 (compact low-rise) and LCZ 8 (large low-rise), whereas LCZ 6 (open low-rise) exhibits rapid heating and cooling. The highest UHII values occurred during consecutive hot days, indicating heat accumulation. Model performance suggests a 500 m diameter as optimal for LCZs 6 and 8 (R² = 0.814; 0.859), while LCZ 3 performs best at 250 m (R² = 0.617), reflecting differences in morphology and heat storage capacity.