Mechanisms of non-stationary influence of urban form on the diurnal thermal environment based on machine learning and MGWR analysis

The non-stationary impact of urban form at block scale on the diurnal thermal environment has not been extensively studied. To fill the gap, a comprehensive multi-scale research framework was constructed, integrating machine learning algorithms and multiscale geographically weighted regression (MGWR) analysis. The urban form types were extracted by machine learning algorithms from 2282 blocks of Dalian, a coastal city, and the contribution to the diurnal land surface temperature (LST) was evaluated. Based on MGWR, the non-stationary effects of urban form, human activity and spatial location on the diurnal LST were quantified, which indicated the contributing factors to the diurnal thermal differences among the form types. The result was as follows: 1) Blocks characterized by low vegetation and mid/low-rise buildings had the highest warming contribution for diurnal LST. 2) The impact of sky view factor (SVF) on diurnal temperature amplitude exhibited no significant spatial-temporal heterogeneity. 3) Building density had a prominent effect on diurnal temperature amplitude. 4) High-vegetation and open (SVF > 0.5) mid/mid-high/low-rise (15–50 m) buildings were recommended. This study provides a more precise basis for policymakers to develop climate adaptation strategies throughout day and night, particularly for coastal cities.