The wetlands of the province of Taza (Morocco) are highly sensitive to climate variability due to pronounced spatial heterogeneity, with arid plains and wetter high-altitude areas. This study investigates the spatio-temporal dynamics of vegetation cover in these wetlands and predicts future changes under climate change scenarios using an integrated approach that combines remote sensing (NDVI) and machine learning algorithms (MaxEnt, Random Forest, and XGBoost). A set of bioclimatic, topographic, and hydrological variables was used to model vegetation distribution, and model performance was evaluated using AUC metrics. Results indicate that XGBoost and Random Forest outperform MaxEnt, providing highly accurate predictions of vegetation dynamics. Current NDVI patterns show that low-density vegetation dominates approximately 42% of the study area, while medium and high NDVI classes cover 33% and 26%, respectively. Projections to 2060 and 2100 under SSP 245 and SSP 585 scenarios suggest a slight decline in low NDVI areas, relative stability of medium NDVI classes, and moderate expansion of high NDVI areas, indicating resilience in some ecosystems. Spatial analysis further identifies specific stations (Oued El Bared, Lac Tamda, Oued M’soun, and Oued Chaouya) as highly vulnerable to combined climate and anthropogenic pressures, whereas others (Bab Louta and Ras El Ma) show greater resilience. This study demonstrates that integrating NDVI with machine learning enables robust prediction of vegetation responses to climate change, providing a scientific basis for adaptive management and conservation strategies in vulnerable wetland ecosystems.