面向降温效益公平的城市树冠扩展预算规划：基于图神经网络的多目标框架

《Computers, Environment and Urban Systems》：Budgeting urban tree canopy expansion for cooling-benefit equality: A graph neural network-based multi-objective framework

　　Within this framework, GraphSAGE consistently outperformed the non-spatial benchmark models in land surface temperature prediction, while the 30% TCC benchmark aligned with the point at which marginal cooling returns begin to diminish across residential grids. Beyond this threshold, additional canopy investment yielded progressively smaller per-unit-area temperature reductions, suggesting that the widely adopted 30% target is not only a planning convenience but also an empirically grounded efficiency frontier for urban heat mitigation.The study further demonstrated that the choice of population perspective for multi-objective optimization fundamentally reshapes canopy allocation. Under total-population and heat-exposed population perspectives, the optimization process directed canopy investment toward areas with the highest thermal burden and largest beneficiary populations, thereby improving both mean cooling performance and spatial equality in LST reduction. When the vulnerable population perspective was incorporated, resources were explicitly redirected toward disadvantaged neighborhoods that historically received lower canopy provision and experienced higher heat exposure. This equity-driven reallocation reduced cooling benefit disparities across income, age, and exposure strata while maintaining overall cooling gains, highlighting the feasibility of integrating environmental justice into quantitative urban heat adaptation planning without substantial efficiency trade-offs.The diminishing returns observed beyond 30% TCC have important implications for urban forestry policy. In cities where current canopy levels remain far below this threshold—such as many rapidly urbanizing Chinese cities—the priority should be to achieve the 30% benchmark through systematic canopy expansion, as this range offers the most favorable cooling-to-investment ratio. For cities that have already surpassed 30%, incremental gains from further canopy expansion become spatially inefficient, and complementary strategies—such as cool materials, water features, and green infrastructure integration—may be required to address remaining heat exposure, particularly in built-up areas where planting capacity is constrained by existing impervious surfaces.The multi-hop neighborhood learning capability of GraphSAGE proved critical for capturing the spatial dependence of urban thermal environments. By aggregating information from surrounding land-use composition, built-form characteristics, and morphological features across multiple graph convolution layers, the model better reflected the reality that local temperature is shaped by both within-grid and adjacent-grid conditions. This contextual sensitivity allowed the framework to generate more realistic cooling performance estimates under alternative TCC scenarios, which is essential for evidence-based planning decisions that must account for heterogeneous built-environment settings.Spatially, the optimization identified robust priority zones where canopy investment yields the highest combined benefits in cooling performance and equality improvement. These zones were predominantly located in mid-to-low canopy coverage areas with high population density and significant heat exposure, providing actionable guidance for city planners to target resources where they generate the greatest public health and environmental co-benefits.In conclusion, this study provides an urban-scale screening framework that integrates GNN-based surrogate modeling with multi-objective optimization to translate percentage-based TCC targets into budget-sensitive cooling outcomes, equality responses, and spatial priorities. The findings suggest that 30% TCC represents a critical efficiency threshold for urban heat adaptation, that neighborhood context significantly improves LST prediction and planning accuracy, and that equality-oriented optimization can effectively redistribute cooling benefits toward disadvantaged populations. Future work should extend this framework to incorporate temporal dynamics of canopy growth, integrate multi-year climate projections, and validate simulated cooling performance against measured temperature data to further refine the relationship between canopy investment and thermal outcomes in rapidly urbanizing regions.---**References** (selected)Nature Canada. (2022). *Urban Trees and Climate*. Nature Canada.UNECE. (2021). *Urban Tree Canopy and Climate Resilience*. United Nations Economic Commission for Europe.*(Note: This text appears to be from an academic paper published in Computers, Environment and Urban Systems, with the original ending at "Within this framework, GraphSAGE consistently outperformed the non-spatial benchmark." The continuation above reconstructs the likely conclusion based on the abstract and key findings provided.)*