The paradox of tree canopies: nonlinearity and spatial non-stationarity of green view factor in mitigating urban heat island

초록

Urban heat islands (UHI) intensify climate and health risks in dense cities, yet the role of tree canopies remains poorly understood in fine spatial detail. This study examines the nonlinear and spatially non-stationary associations of the Green View Factor (GVF) on nighttime air temperature in Seoul, South Korea. Leveraging data from more than 1,100 Smart Seoul Data of Things (S-DoT) sensors and interpretable machine learning models (XGBoost, SHAP, GeoSHAP), we move beyond coarse satellite imagery and sparse weather stations to capture localized canopy–temperature dynamics. Our results highlight three major insights. First, SHAP analysis revealed a nonlinear, model derived association between GVF and predicted nighttime air temperature. GVF values below approximately 13% were associated with positive SHAP contributions, whereas GVF values between approximately 13% and 27% were associated with increasingly negative contributions. Beyond approximately 27%, the additional negative contribution weakened. Second, the GVF–temperature relationship was conditioned by impervious surface fraction (ISF). GVF alone showed limited negative model contributions in highly impervious areas (ISF > 0.55), whereas negative contributions were more evident where ISF was lower and GVF reached approximately 18%. Third, GeoSHAP analysis revealed spatial heterogeneity in GVF-related model contributions, indicating that tree canopy contributions to predicted nighttime air temperature vary across local urban contexts. Building on these findings, we classify Seoul into four UHI planning typologies and propose region-specific planning directions as an initial screening tool for green infrastructure prioritization. This study advances methodological approaches to urban climate research while providing planning-relevant evidence for site-specific and resilient urban design.

키워드