Evaluating the influence of biophysical factors in explaining spatial heterogeneity of LST: Insights from Brahmani-Dwarka interfluve leveraging Geodetector, GWR, and MGWR models

Abstract

The present study aims to investigate variations in land use and land cover (LULC) while examining how land surface temperature (LST) and biophysical parameters in the Brahmani-Dwarka Interfluve zone change over time. It also tries to discover LST clusters and outliers and investigate how biophysical variables influence the spatially stratified heterogeneity of LST. This study introduces an innovative perspective by concentrating on the Brahmani-Dwarka Interfluve, an area characterized by stone crushing and mining operations. It employs Geodetector, Geographically Weighted Regression (GWR), and Multiscale Geographically Weighted Regression (MGWR) to thoroughly evaluate the LST heterogeneity, thereby enhancing its uniqueness. Our analysis revealed LST saw a substantial rise of 7.05 °C, equivalent to an average yearly increase of around 0.235 °C between 1991-2021. The regions primarily occupied by stone crushing and mining activities showed elevated LST values compared to other LULC categories in the years 2001, 2011, and 2021. The factor detector findings suggested that the Normalized Difference Latent Heat Index (NDLI) had the most influence in 2001 (q-value 0.39) and 2021 (q-value 0.50), whereas the Normalized Difference Moisture Index (NDMI) had the highest influence in 1991 (q-value 0.39) and 2011 (q-value 0.29). Soil-adjusted Vegetation Index (SAVI) had the most negligible influence (q-value between 0.09-0.15). The Interaction detector findings identified NDLI with NDMI (q-value 0.42) and Modified Normalized Difference Water Index (MNDWI) (q-value 0.42) in 1991, NDLI with Normalized Difference Bareness Index (NDBaI) (q-value 0.41) and MNDWI (q-value 0.41) in 2001, and NDLI with Normalized Difference Vegetation Index (NDVI) (q-value 0.56) and SAVI (q-value 0.56) in 2021 as the most influential interacting variable. In 2011, significant interactions were identified between MNDWI and NDVI (q-value 0.31), NDMI, and SAVI (q-value 0.31), among others. A comparative assessment between regression models using AICc, R², and adjusted R² revealed that all MGWR models performed (R² between 0.92-0.94 & adj. R² between 0.89-0.92) much better than GWR and Ordinary Least Square (OLS) models during the research period. The results will undoubtedly assist policymakers, environmentalists, and stakeholders by providing valuable insights for enhancing environmental management and policies.