Budgeting urban tree canopy expansion for cooling-benefit equality: A graph neural network-based multi-objective framework

Urban tree canopy coverage (TCC) targets are increasingly used for heat mitigation. However, translating fixed percentage targets into context-specific cooling benefits, population coverage across cooling thresholds, and spatial priorities is essential for making them actionable in planning practice. This study develops a spatially explicit framework that couples a Graph Neural Network (GNN)-based land surface temperature (LST) surrogate model with multi-objective optimization to evaluate TCC increment scenarios in residential areas. Using Shanghai, China, as a case study, we first show that incorporating neighbourhood context improves LST prediction, with the GraphSAGE model consistently outperforming non-spatial machine-learning models. We then compare a uniform 30% TCC baseline with three equality-oriented scenarios using population-weighted, heat-exposure-weighted and vulnerability-weighted perspectives. With mean summer LST as the baseline, the equality-oriented knee solutions required 18.0–20.2% more canopy investment than the uniform baseline while increasing mean LST reduction by 45.3–48.8%. The three equality-oriented scenarios produced similar city-scale cooling patterns but led to different local canopy-allocation priorities. Marginal cooling gains diminished as added TCC increased, with the most rapid gains occurring within the first 20 percentage points of TCC increment, bringing average TCC close to the 30% benchmark in Shanghai. At a 20-percentage-point TCC increment, about 65% of the population achieved at least 1 °C LST reduction, whereas stronger cooling thresholds remained difficult to reach even under high-budget scenarios. Overall, this study translates fixed-percentage canopy targets into budget-sensitive cooling outcomes, equality responses and spatial priorities, providing an urban-scale screening framework for TCC planning.