Spatial–temporal analysis of land surface temperature in Ahmedabad city using Landsat-derived indices and machine learning–based regression assessment (1995–2025)

Abstract

Land surface temperature (LST) have significantly increased as a result of rapid urbanization, worsening environmental stress, and raising threats to public health. This study investigates the spatial-temporal assessment of LST dynamics in Ahmedabad city over the last decades (1995–2025) using Landsat-derived multispectral indices, namely LST, NDBI, NDMI, NDWI, NDVI, and SAL for summer and winter seasons. Seasonal and decadal analyses reveal a steady increase in LST, with summer mean temperatures rising from 36.47 °C (1995) to 39.87 °C (2025) and winter temperatures increasing from 23.14 °C to 29.53 °C. Correlation analysis was conducted for the selected indices to assess their interrelationships and effect on LST. The strong non-linear correlation between surface moisture, built-up intensity, and LST. The same has been further confirmed by a 3D scatter-based interaction analysis, emphasizing the need for multi-variable heat-resilient planning. The results highlight NDBI and NDMI as dominant factors of LST, while vegetation and water bodies provide localized cooling effects. A comparative evaluation of machine learning regression models for LST prediction was assessed using artificial neural networks (ANN), random forest (RF), and linear regression. The results confirm that the RF model (R² = 0.90) provides the most accurate match between selected spectral indices and LST. The study recommends some essential actionable policy and planning insights for climate-responsive urban development, emphasizing the implementation of blue-green infrastructure to reduce heating effects in semi-arid cities.