Anticipating Semi-Arid Landscape Change: Machine Learning and Sentinel-2 for Land Use and Land Cover Prediction

Understanding land use and land cover changes (LULCC) is crucial for sustainable urban planning and environmental management, particularly in rapidly developing regions. This study predicts future LULC changes in the Mitidja plain, a semi-arid region in northern Algeria, using Sentinel-2 satellite imagery and advanced machine learning techniques. By systematically analyzing past landscape transformations, we developed a robust predictive model to assess patterns of urban expansion and evolving land use trends over time. The methodology involved creating classified LULC maps for 2019 and 2023, which were then integrated to generate a comprehensive transition map revealing spatial changes across the landscape. A Random Forest (RF) machine learning model was trained using these historical LULC maps, topographic elevation data, and temporal variables to capture the complex drivers of land change. The trained model successfully predicted LULC distribution for 2030, 2075 and 2100 revealing significant urban growth occurring predominantly at the expense of agricultural croplands and natural forests, with minor fluctuations observed in water bodies and barren lands. These projected trends underscore the region's rapid urbanization trajectory and its associated environmental impacts, including habitat fragmentation, biodiversity loss, increased flood risks, and agricultural land degradation. The study demonstrates the effectiveness and reliability of Random Forest modeling for spatially explicit LULC prediction, offering valuable insights and evidence-based recommendations for policymakers, urban planners, and environmental managers seeking to balance development with ecological sustainability.