Explaining Urban Thermal Hotspots in Ankara Using Remote Sensing and Spatial Statistics

Ankara has experienced rapid urban growth in recent years, making the spatial distribution of surface heating an increasingly important planning issue. This study examines warm-season land surface temperature (LST) patterns in Ankara for May-October 2025 using Landsat 8/9-derived LST and a 500 m grid-based spatial framework. The analysis combines remote sensing indicators of vegetation and built-up structure with topographic and settlement-related variables, including NDVI, NDBI, albedo, elevation, slope, built-up fraction, and population density. Spatial clustering of LST was first evaluated using the Getis-Ord Gi∗ statistic for both the full domain and the urban core. The relationships between LST and its potential drivers were then assessed by ordinary least squares (OLS), while residual spatial dependence was examined with Moran's I. To address spatial non-stationarity within the urban core, geographically weighted regression (GWR) was applied. The results show that warm-season LST in Ankara is strongly clustered in space, with coldspots dominating the full domain and hotspot patterns becoming more pronounced in the urban core. OLS suggested global relationships between LST and several explanatory variables in both domains; however, residual autocorrelation remained substantial, especially in the full area (Moran's I ~ 0.821), indicating that the global model did not fully capture the spatial structure. In the urban subset, GWR improved model performance (global R² ~ 0.874) and reduced residual spatial dependence markedly (Moran's I ~ 0.043). These findings indicate that Ankara's urban thermal structure is spatially heterogeneous and that local modelling provides a more realistic basis for interpreting thermal hotspots.