Integrating multi-dimensional urban green space configuration in land surface temperature modelling: A seasonal and methodological evaluation for southern Delhi

Abstract

Increasing urban growth and global warming necessitate effective temperature mitigation strategies to support sustainable development goals (SDGs) 11 (Sustainable Cities and Communities) and 13 (Climate Action), especially in developing megacities such as Delhi. This study (i) investigates seasonal relationship between vertical (3D) and horizontal (2D) spatial configuration of urban green spaces (UGS) and land surface temperature (LST) and (ii) evaluates the effectiveness of non-spatial, spatial and non-parametric regression methods in explaining LST variability using GEDI LiDAR, Sentinel 2 A and Landsat 8 data for New Delhi and South Delhi (NSD). GEDI-derived vertical metrics reliably captured the 3D structure of UGS when validated against laser rangefinder measurements (avg. seasonal R² = 0.76), with vertical height (Vert_HT), canopy complexity (fhd), and fractional gap (pgap_th) emerging as strong predictors. Among 2D metrics—interspersion-juxtaposition (IJI) exhibited strong negative associations with LST during summer, indicating enhanced cooling over interspersed landscape in summertime. In contrast, shape-complexity metrics such as perimeter-area fractal dimension (PAFRAC) and edge density showed a seasonal cooling in post-monsoon conditions, while proportional adjacency (PLADJ) demonstrated cooling across transitional season. Integrating 3D metrics with 2D improved LST prediction, reducing RMSE by ∼33% (from 0.6 °C to 0.4 °C) during post-monsoon season. Comparative modelling revealed superior performance of spatial regression models, with geographically weighted regression emerging as the most robust approach in the NSD region (R² ∼ 0.73 avg. across UGS classes and seasons), indicating strong spatial dependency in landscape-LST relationships. Although non-parametric machine learning models demonstrated moderate LST explanatory power, their performance varied significantly across UGS classes and seasons, influenced by vegetation density and activity. The findings provide actionable insights for climate-resilient landscape planning and management in rapidly urbanising environments.