Understanding Land Use/Land Cover Dynamics and Land Surface Temperature Variations Through Machine Learning and Geospatial Approach: A Case Study of Kullu District, Himachal Pradesh, India

Abstract

Machine learning (ML)-based classification provides an effective approach for extracting land use/land cover (LULC) information from multi-temporal satellite observations. This study evaluates Maximum Likelihood Classification (MLC), Support Vector Machine (SVM), and Random Forest (RF) for mapping LULC changes in Kullu district, Himachal Pradesh, India, using Landsat data from 1991 to 2022. SVM achieved the highest classification accuracy and was therefore used for subsequent LULC mapping. Six LULC classes were identified and their temporal changes were examined in relation to Landsat-derived land surface temperature (LST). The built-up area increased from 4.37% in 1991 to 20.40% in 2022, whereas vegetation declined from 34.41% to 16.23% and snow cover from 15.20% to 3.02%. The area-weighted mean LST was 4.10 °C in 1991, 10.03 °C in 2001, 8.43 °C in 2011, and 4.18 °C in 2022, representing a marginal net increase of 0.08 °C over the study period. Class-wise analysis showed that barren and agricultural lands generally recorded higher mean LST, while vegetation and waterbodies exhibited comparatively lower values. The relatively high snow-cover LST in 2022 may reflect reduced and fragmented seasonal snow cover and snowmelt conditions during image acquisition. Correlation analysis revealed an inverse relationship between LST and vegetation and snow-cover indicators, with R2 values reaching 0.66 for NDVI and generally exceeding 0.62 for NDSI. Overall, the findings demonstrate substantial LULC transformation and pronounced spatial differences in surface temperature, emphasizing the role of vegetation and snow cover in regulating thermal conditions and providing useful information for sustainable land management and spatial planning in the ecologically sensitive Himalayan environment.