Decadal Dynamics of Rangeland Cover Using Remote Sensing and Machine Learning Approach

Rangeland degradation in arid and semi-arid regions poses significant environmental and socioeconomic challenges globally. This study aims to assess the Spatio-temporal dynamics of rangeland changes in Khushab district, Pakistan, between 20 0 0 and 2020 by developing an integrated approach combining remote sensing, vegetation indices, and machine learning techniques. The specific objectives were to: (1) quantify rangeland extent changes using multi-temporal Landsat imagery, (2) evaluate rangeland health through multiple vegetation indices, and (3) analyze the primary drivers of rangeland transformation. The methodology integrated Landsat-derived land use land cover (LULC) classification using Random Forest and SMILE CART algorithms, analysis of six vegetation indices (NDVI, GNDVI, SAVI, EVI, ARVI), and land surface temperature (LST) assessment. The classification accuracy exceeded 90% for Random Forest and 87% for SMILE CART across all time periods. Results revealed significant rangeland degradation, with area declining from 9% to 6% of total land between 20 0 0 and 2020. Cropland expansion was the primary driver, increasing from 16% to 29% and converting 218 sq km of rangeland. Vegetation indices showed stable NDVI but declining GNDVI maximums from 0.37 to 0.36, indicating deteriorating plant health. Rising minimum LST from 27.82 degrees C to 31.81 degrees C suggested increasing heat stress on vegetation. This research demonstrates the effectiveness of integrating multiple remote sensing approaches with machine learning for comprehensive rangeland monitoring. The findings provide crucial baseline data for evidence- based policy making and sustainable rangeland management in Pakistan's semi-arid regions. Future work should incorporate ground validation and socioeconomic surveys to better understand degradation drivers and develop targeted conservation strategies. (c) 2025 The Society for Range Management. Published by Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.