University of Birjand-Research Group of Drought and Climate Change Journal of Drought and Climate change Research 3092-6076 Articles in Press 2026 05 25 Predicting the Impacts of Climate Change on Vegetation Dynamics (NDVI) in Arid Ecosystems Using Machine Learning Models: A Case Study of Hormozgan Province Predicting the Impacts of Climate Change on Vegetation Dynamics (NDVI) in Arid Ecosystems Using Machine Learning Models: A Case Study of Hormozgan Province 3991 10.22077/jdcr.2026.10754.1200 FA Massomeh Forozanfard Phd Student, Department of Natural Resources Engineering, Faculty of Agricultural and Natural Resources Engineering, University of Hormozgan, Bandar Abbas. 0009-0008-0608-3149 Yahya Esmaeilpour Ascociat Professor, Department of Natural Resources Engineering, Faculty of Agricultural and Natural Resources Engineering, University of Hormozgan, Bandar Abbas. 0000-0002-9816-899X Hossein Zamani Department of Statistics and mathematics, Faculty of Science, University of Hormozgan, Bandar Abbas, Iran. 0000-0003-1126-6288 Mohammadi Mojtaba Assistant Professor ,Department of Desert Management and Control, Faculty of Environmental Sciences, Planning and Sustainable Development, University of Saravan, Saravan, Iran. 0000-0002-5614-8376 Journal Article 2025 12 23 Climate change threatens the fragile ecosystems of arid regions such as Hormozgan province. This study aims to predict future vegetation dynamics (NDVI) up to the year 2100 by comparing advanced machine learning models under the pessimistic climate change scenario (SSP3-7.0). To this end, MODIS NDVI time series (2000-2018) and ERA5 climate data were extracted for two representative points within the province. The performance of four machine learning algorithms (GPR, GAM, RF, and XGBoost) was evaluated using rigorous statistical metrics, and the optimal model was employed to project future NDVI using data from the GFDL-ESM4 model under the SSP3-7.0 scenario. Evaluation results indicated that the Gaussian Process Regression (GPR) for the first point and eXtreme Gradient Boosting (XGBoost) for the second point achieved the highest performance in the testing phase, with Kling-Gupta Efficiency (KGE) values exceeding 0.88. Projections towards the 2100 horizon revealed two divergent ecological responses: the first point is projected to experience a significant 42% increase in NDVI (a “greening” phenomenon), whereas the second point is expected to show modest growth before stabilizing (a 10.9% increase). This spatial heterogeneity indicates that the region’s ecosystems exhibit varying responses. We conclude that drought management strategies must be location-specific and tailored to the unique growth potential and resilience of each area. Climate change threatens the fragile ecosystems of arid regions such as Hormozgan province. This study aims to predict future vegetation dynamics (NDVI) up to the year 2100 by comparing advanced machine learning models under the pessimistic climate change scenario (SSP3-7.0). To this end, MODIS NDVI time series (2000-2018) and ERA5 climate data were extracted for two representative points within the province. The performance of four machine learning algorithms (GPR, GAM, RF, and XGBoost) was evaluated using rigorous statistical metrics, and the optimal model was employed to project future NDVI using data from the GFDL-ESM4 model under the SSP3-7.0 scenario. Evaluation results indicated that the Gaussian Process Regression (GPR) for the first point and eXtreme Gradient Boosting (XGBoost) for the second point achieved the highest performance in the testing phase, with Kling-Gupta Efficiency (KGE) values exceeding 0.88. Projections towards the 2100 horizon revealed two divergent ecological responses: the first point is projected to experience a significant 42% increase in NDVI (a “greening” phenomenon), whereas the second point is expected to show modest growth before stabilizing (a 10.9% increase). This spatial heterogeneity indicates that the region’s ecosystems exhibit varying responses. We conclude that drought management strategies must be location-specific and tailored to the unique growth potential and resilience of each area. https://jdcr.birjand.ac.ir/article_3991_dfbd282c18300fa0eccceea6c5fac41f.pdf