LULC Change Detection and Prediction Using Random Forest Classification and CA–ANN Modeling: A Geospatial Study of Sarkaghat Tehsil, Himachal Pradesh, India

Land-use/land-cover (LULC) change affects the balance between settlements, agricultural land, and natural ecosystems, making its assessment important for land-use planning in mountainous regions. This study integrates Random Forest (RF) classification and Cellular Automata–Artificial Neural Network (CA-ANN) modeling to assess historical LULC change in Sarkaghat Tehsil, Himachal Pradesh, and generate baseline projections for 2031 and 2046. Landsat imagery from 2016, 2021, and 2026 was classified into six classes: Forest Land (FL), Agricultural Land (AL), Built-up Land (BUL), Barren Land (BL), Water Body (WB), and Grass Land (GL). Overall classification accuracy was 95%, 96%, and 93%, respectively, with Kappa coefficients of 0.91, 0.94, and 0.89. The 2026 map had the lowest accuracy; the lower agreement was particularly evident for Grass Land, which had a user’s accuracy of 0.83. Historical comparisons from 2016 to 2026 showed increases in BUL (0.46%), BL (0.04%), and GL (0.16%), and decreases in AL (0.45%) and FL (0.21%); WB changed negligibly. The CA–ANN simulation of 2026, evaluated against the independently classified 2026 map, achieved an overall accuracy of 97.898% and a Kappa coefficient of 0.965. Relative to 2016, the 2031 baseline projection indicates changes of FL 0.00%, AL −0.37%, BUL +0.27%, BL −0.03%, WB 0.00%, and GL +0.13%. For 2046, the corresponding changes are FL −0.01%, AL −0.37%, BUL +0.27%, BL −0.03%, WB 0.00%, and GL +0.14%. The projections are conditional on stable historical transition relationships and should be interpreted as baseline scenarios rather than certain forecasts. The integrated workflow provides a locally evaluated basis for assessing land transitions and informing land-use planning in Sarkaghat Tehsil.