Geospatial and Regression-Based Modelling of Land Surface Temperature in Aligarh: A Comparative Study of OLS and GWR

Understanding the impacts of urbanization and environmental transitions in rapidly developing regions such as Aligarh District, Uttar Pradesh, requires a comprehensive assessment of land use/land cover (LULC) and land surface temperature (LST). For this study, a semi-automated hybrid classification approach, integrating maximum likelihood classification with object-based image analysis, was applied to Landsat-8 OLI imagery from 30 May 2022 to map LULC. LST was derived from thermal band 10 using a four-step procedure that converted the satellite-recorded digital numbers (DNs) into accurate land surface temperature values. Accuracy assessment using 250 reference sites yielded an overall accuracy of 94.4% and a Kappa coefficient of 0.93, confirming high reliability. LST analysis revealed considerable spatial and thermal variability, with summer temperatures ranging from 26.48°C to 46.40°C (mean: 36.32°C). Pearson’s correlation results indicated consistent relationships between LST and key remote sensing indices. NDVI and SAVI showed moderately negative correlations with LST, demonstrating the cooling influence of vegetation, while NDBI exhibited a strong positive correlation, highlighting the urban heat island effect. NDWI showed a negative relationship with LST, and NDBaI displayed a weaker positive correlation, underscoring the moderating effect of water bodies on surface temperature. The Ordinary Least Squares (OLS) regression model explained 70.06% of LST variance, with an Akaike Information Criterion (AICc) value of 3792.15. Coefficient patterns indicated that NDBI contributed to LST intensification, whereas NDVI, NDWI, and SAVI significantly reduced surface temperatures. The Geographically Weighted Regression (GWR) model substantially improved explanatory power, achieving an R² of 0.9405 and reducing residual spatial autocorrelation, as reflected by the decline in Moran’s I from 0.30 (OLS) to 0.02 (GWR). Overall, the findings demonstrate that LULC dynamics drive surface temperature fluctuations in the Aligarh district, and GWR's ability to capture geographical variations makes it highly effective for environmental modelling.

Keywords: Land surface temperature (LST), Geographically weighted regression (GWR), Spatial regression, Ordinary least squares (OLS), Geospatial analysis.