A stacked generalization-based multi-model ensemble framework for precipitation downscaling and drought characterization in a semi-arid region

Drought, as a recurrent climatic phenomenon driven by prolonged precipitation deficits, has intensified in recent decades, causing widespread impacts on agriculture, water resources, and socio-environmental sustainability. This trend is particularly evident in arid and semi-arid regions such as Iran, highlighting the critical need for continuous monitoring and assessment. This study proposes a novel downscaling framework using Stacked Generalization, which differs from traditional ensemble methods by using base model predictions as inputs to a secondary model. This two-stage approach captures complex dependencies, leading to improved downscaling performance. The refined outputs were then used to analyze drought characteristics based on the best-performing Global Climate Models (GCMs). Then, the next objective is to better determine future drought characteristics by utilizing the more accurate results obtained from the proposed downscaling approach. It further aims to improve the identification of future drought characteristics using the more accurate results obtained from the proposed downscaling approach. The results indicated that the Stacked method consistently outperformed the individual base ones (MLP, SVR, and RF), achieving the highest Nash-Sutcliffe Efficiency (NSE) across all stations and climate models, and exhibiting the lowest Mean Squared Error (MSE) compared to the other methods. Additionally, the Standardized Precipitation Index (SPI) was calculated using a parametric method at 3-, 6-, and 12-month timescales. The findings indicated that, while short-term drought characteristics remain stable, long-term droughts, as represented by SPI-12, are projected to become longer and more severe, particularly in certain regions.