The Atlas cedar forests in the Moroccan Middle Atlas, particularly the Sidi M'Guild region, are undergoing rapid degradation due to increasing climatic stress and anthropogenic pressure. This study introduces a hybrid modelling approach integrating random forest (RF), cellular automata (CA) and Markov chains to simulate forest cover dynamics from 1990 to 2032. The model integrates remote sensing data from Landsat 4, 8 and Sentinel-2, bioclimatic variables (temperature, seasonality, rainfall of the driest quarter) and indicators of human influence (density of occupancy, proximity to forest edges). The results project a 91% decline in Cedrus atlantica and a 74% decline in juniper, contrasted with a 1,290% expansion of holm oak, indicating a major ecological shift to drought-tolerant hardwoods. The RF–AdaBoost classifier achieved 98% accuracy, and the RF–CA–Markov framework demonstrated strong predictive power (Kappa = 0.72). These results offer a solid tool to anticipate forest transitions and guide adaptive forest management strategies, aligned with Morocco's national reforestation efforts.