Urban growth and climate change intensify urban heat islands, yet operational frameworks that integrate high-resolution field-based ecological data with remote sensing to guide urban climate adaptation remain scarce, particularly in Global South megacities. This gap limits the ability of public managers to identify which ecological attributes of urban green spaces effectively drive thermal regulation and reduce socio-environmental inequalities. To address this gap, this study analyzed 13 public squares in São Paulo by integrating field-based floristic data (composition, structure, and biomass) with remote sensing variables, land surface temperature (LST) and vegetation cover fraction (FRACVEG), derived from Landsat 8/9 imagery. A machine-learning pipeline combining K-means clustering, PCA, and XGBoost was applied to identify functional typologies and quantify the relative importance of ecological predictors. Results demonstrate that vegetation structure, rather than species diversity, is the primary driver of thermal regulation, providing empirical support for functional, rather than taxonomic, approaches to urban climate adaptation. FRACVEG emerged as the dominant predictor (R 2 = 0.757), with a threshold of approximately 40% vegetation cover associated with a temperature reduction of 1.5–2.0 °C. Three ecological profiles were identified, aligned with São Paulo's socio-environmental inequalities. The coolest cluster (29.93 °C) was dominated by exotic species (67.18%), indicating a trade-off between short-term cooling and long-term ecological stability, reinforcing the need to prioritize native functional equivalents. Warmer and structurally simplified areas were concentrated in the East Zone. These findings support revising urban greening strategies toward increasing vegetation cover and prioritizing ecological functionality to reduce heat exposure and environmental inequalities. • Vegetation structure is more important than species diversity for thermal comfort. • Exotic species dominated the coolest cluster, valuing functional attributes over origin. • The modeling validates the need for management directed by functional typology, not square size. • Increasing Biomass above 30,084 tC reduces temperature by more than 1 °C.
França et al. (Thu,) studied this question.