Clarifying the factors that influence land surface temperature (LST) is crucial for proposing specific LST mitigation strategies. This article focuses on the Beijing–Tianjin–Hebei (BTH) region and investigates the LST influencing factors of various local climate zone (LCZ) built types from perspectives of urban morphology, land cover, and human activity. The results suggest that the areas of urban LCZ built types vary across cities within the BTH region, attributed to differences in city size and gross domestic product (GDP). The area of urban LCZ built types in Beijing and Tianjin, cities with significantly high city sizes and GDP, exceeds 2000 km ² . In contrast, in Qinhuangdao, Zhangjiakou, and Chengde, which have relatively low GDP, this area is less than 500 km ² . However, the main LST influencing factors for the same LCZ built type across cities are highly consistent. Building coverage ratio (BCR), average building height (ABH), and pervious surface fraction (PSF) are the three most important factors. The correlation between BCR and LST is mainly concentrated on compact high-rise and open high-rise types, with Pearson correlation coefficient ( r ) ranging from 0.2 to 0.44; the correlation between ABH and LST is mainly concentrated on compact high-rise, compact mid-rise, open high-rise, and open mid-rise types, with r ranging from −0.2 to −0.52; the correlation between PSF and LST is concentrated on almost all LCZ built types in cities within the BTH region, with r ranging from −0.2 to −0.56. By integrating these findings with the features of each LCZ built type within the BTH region, the specific LST mitigation strategies were further proposed. This article can help to develop specific LST mitigation strategies within the context of coordinated development of the BTH region, thereby promoting healthy and sustainable development in the region.

In recent years, the phenomenon of urban warming has become increasingly serious, and with the number of urban residents increasing, the risk of heatstroke in extreme weather has become higher than ever. In order to mitigate urban warming and adapt to it, many researchers have been paying increasing attention to outdoor thermal comfort. The mean radiant temperature (MRT) is one of the most important variables affecting human thermal comfort in outdoor urban spaces. The purpose of this paper is to predict the distribution of MRT around buildings based on a commonly used multilayer neural network (MLNN) that is optimized by genetic algorithms (GA) and backpropagation (BP) algorithms. Weather data from 2014 to 2018 together with the related indexes of the grid were selected as the input parameters for neural network training, and the distribution of the MRT around buildings in 2019 was predicted. This study obtained very high prediction accuracy, which can be combined with sensitivity analysis methods to analyze the important input parameters affecting the MRT on hot summer days (the days with the highest air temperature over 30 °C). This has significant implications for the optimization strategies for future building and urban designers to improve the thermal conditions around buildings.

Accurate assessment of microclimatic characteristics in urban blocks is a prerequisite for climate-adaptive urban design. However, traditional numerical simulations face high computational costs. This study proposes a novel data-driven prediction framework that combines semantic computing of multi-source Street View Imagery (SVI) with stacked ensemble learning to demonstrate the spatial heterogeneity and diurnal asymmetry of the urban microclimate. Focusing on five typical urban functional zones, three mainstream computer vision datasets were comparatively tested to extract morphological parameters. A two-layer stacked prediction model was constructed to interpret the nonlinear spatiotemporal driving mechanisms of six microclimatic variables. The results indicate that: (1) The Stacking model achieves high accuracy, with R2 values for air temperature and relative humidity exceeding 0.97. ADE20K outperforms other models in predicting Mean Radiant Temperature (Tmrt ) and Predicted Mean Vote (PMV) due to its fine-grained semantic granularity. (2) Microclimatic driving mechanisms exhibit prominent spatial heterogeneity: high-density central business districts are dominated by geometric morphology (SVF/Enclosure), the influence of vegetation (GVI) is more significant in low-density areas, and old communities are affected by the thermal capacity effects of Impervious Surface Ratio (ISR). (3) Significant diurnal asymmetry exists: long-wave radiation dominates at night, and the explanatory power of static visual features is higher than during the day. This study also proposes the concept of the ”Steady-state Variance Trap” to explain the physical reasons for the decline in model performance during stable atmospheric periods. These findings provide a reference for morphology-driven microclimatic interventions in urban systems.