Abstract. Accurate prediction of land-use/land-cover (LULC) change is essential for sustainable urban and environmental planning. This study analyzes and models multi-decadal LULC dynamics in Shahrekord, Iran, using Landsat Surface Reflectance images for 1990, 2000, 2009, 2020, and 2024. Four classes were identified: Built-up/Barren (UB), Agriculture (AG), Water (WT), and Vegetation/Orchards (VG). LULC maps were produced using the Maximum Likelihood Classification (MLC) method in ArcGIS, based on pre-processed data from Google Earth Engine (GEE).Two predictive approaches were compared in TerrSet: Cellular Automata–Markov (CA-Markov) and Land Change Modeler (LCM) with a multilayer perceptron (MLP). Model performance was evaluated using hindcasts for 2020 and 2024, applying Overall Accuracy (OA), Kappa, and Pontius metrics. The CA-Markov model achieved higher accuracy and was therefore selected to predict LULC for 2030.Between 1990 and 2024, the UB class remained dominant, while AG increased in certain periods; WT and VG showed minor fluctuations. The findings confirm that neighborhood-based transitions drive most changes, enabling reliable short-term projections. The main limitations are the merged UB class and irregular time intervals. Recommendations for class refinement and temporal standardization are provided to improve future modeling and reproducibility.
Sarhadi et al. (Fri,) studied this question.