Objective Under the combined influence of climate warming and high-density urbanization, extreme heat and heat-wave events have become more frequent, lasted longer, and produced stronger impacts. Central urban areas, where population, buildings, impervious surfaces, and socioeconomic activities are highly concentrated, have become among the most prominent areas of heat health risk. Existing studies have mainly focused on the cooling effect of green space or on single vegetation indices, while the combined role of two-dimensional landscape pattern and three-dimensional vegetation structure remains insufficiently understood. Therefore, this research constructs a heat health risk assessment system, extracts two-dimensional and three-dimensional green space characteristics, identifies nonlinear response patterns and threshold effects, and compares green space regulation mechanisms across LCZ types.
Methods The study area covers the central urban area of Beijing. First, based on the IPCC risk framework, a heat health risk index was constructed from three dimensions: heat hazard, population exposure, and social vulnerability. Heat hazard was represented by land surface temperature during heat-wave periods, retrieved and integrated from Landsat 8/9 OLI/TIRS and ECOSTRESS imagery. Population exposure was represented by population density. After range standardization, the entropy method was used to determine the weights of vulnerability indicators, and the spatial distribution of heat health risk was then obtained. Second, according to land-cover classification, forest land, grassland, and water areas were defined as urban green space. Green space patches were extracted based on ESA WorldCover data, and a two- and three-dimensional green space feature set was constructed. The two-dimensional features included normalized difference vegetation index, fractional vegetation cover, patch area, percentage of landscape, number of patches, largest patch index, landscape shape index, fractal dimension index, patch cohesion index, landscape division index, perimeter-area ratio, and aggregation index. The three-dimensional features included mean green space height, height standard deviation, green space volume, and leaf area index. Third, LCZ types were classified on a 300 m × 300 m grid using key morphological parameters. Finally, with the heat health risk index as the dependent variable and green space characteristics as independent variables, an XGBoost−SHAP interpretable machine-learning framework was constructed. SHAP values were used to identify the contribution, direction of effect, and nonlinear response relationship of each feature factor.
Results Heat health risk in the central urban area of Beijing showed clear spatial differentiation, with higher risk in the central urban area and lower risk in the peripheral areas. From the perspective of LCZs, the heat health risk of built-type LCZs was generally higher than that of natural-type LCZs. The XGBoost−SHAP results indicated significant differences in the contribution and direction of effect among different green space characteristics. Grassland patch area, fractional vegetation cover, normalized difference vegetation index, and leaf area index had relatively high contributions and all showed negative contributions, suggesting that increasing effective green quantity, improving vegetation coverage, and enhancing vegetation growth conditions can help reduce heat health risk. In contrast, height standard deviation and indicators of patch fragmentation and boundary complexity showed positive contributions, indicating possible reductions in ventilation and heat dissipation. The risk-mitigation effect of grassland patch area increased rapidly at the initial stage, but tended to stabilize when the area reached approximately 1 hm2. Fractional vegetation cover and normalized difference vegetation index showed S-shaped responses, with relatively high marginal benefits in the ranges of 0.6−0.7 and 0.34−0.37, respectively. Leaf area index showed strong risk-mitigation capacity in the range of 2−3, while its marginal benefits gradually weakened as it continued to increase. After the height standard deviation exceeded approximately 3, its contribution to risk increased rapidly. LCZ3 was more sensitive to the vertical structure and configuration of green space. Leaf area index and forest aggregation helped reduce risk, while excessive mean height and height differences may hinder near-surface ventilation. In LCZ4, normalized difference vegetation index and fractional vegetation cover were more important, indicating that risk mitigation in this type of area mainly depends on green quantity, vegetation quality, and green space continuity. In LCZ5, leaf area index had the highest contribution, suggesting that in open mid-rise areas, improving community leaf area and shading−transpiration capacity through multilayer vegetation configuration is more efficient than simply expanding green space area.
Conclusion The study shows that the role of green space in mitigating heat health risk in the central urban area of Beijing is not determined by a single indicator, but is jointly affected by green quantity, vegetation quality, and spatial structure. However, green space optimization does not mean unlimited increases in area, coverage, or structural complexity. Different green space characteristics have effective ranges and marginal effects, and a balance should be maintained among increasing green quantity, improving vegetation quality, and optimizing structural configuration. Excessive fragmentation, boundary complexity, and vertical structural differences may reduce air movement and heat diffusion efficiency, thereby weakening the overall regulation capacity of green space. The LCZ-based analysis further shows that differences in urban form determine differences in the priorities of green space regulation for heat health risk. LCZ3 is more sensitive to the vertical structure and configuration of green space, LCZ4 depends more on green quantity, vegetation quality, green space regularity, and continuity, while LCZ5 is more suitable for enhancing risk mitigation through multilayer vegetation structure. Overall, this study reveals the nonlinear response mechanism of multidimensional green space characteristics to heat health risk and demonstrates the necessity of differentiated green space regulation based on LCZ types.
DownLoad: