Modeling Climate Change Impacts on Historical and Projected Water Resources Vulnerability using Machine Learning and SWAT Model

Sustainable and adaptive water management strategies require holistic approaches to understand complex systems, mitigate risks from shifting weather patterns, and manage disruptions to hydrological cycles. The main objective of this study was modeling the impact of climate change on water resources vulnerability using machine learning (ML) and the SWAT model, leveraging CMIP6 Global Climate Models (GCMs) under Shared Socioeconomic Pathways (SSPs) in Upper Gilgel Gibe Watershed, Ethiopia. Six ML models were tested for predicting hydroclimatic events, with Extremely Randomized Trees (ERT) and Categorical Boosting (CatBoost) outperforming others in simulating ensemble climate interactions. The ensemble SWAT model performance demonstrated strong agreement between simulated and observed values, supported by indicators Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and Percent Bias (PBIAS) values of 0.93, 0.91, and -1.08 for calibration and 0.94, 0.93, and -2.32 for validation periods respectively, confirming reduced input uncertainties using bias-corrected datasets. A novel Hydrological Vulnerability Index (HVI) framework was developed based on water balances to quantify watershed vulnerability across baseline and future scenarios. The HVI ranges from low to extreme, with lower values indicating resilience to hydrological stress and higher values reflecting severe vulnerability. Baseline assessments revealed 54.03% of areas with low HVI, indicating strong resilience, whereas SSP245 showed a significant decline in low HVI (26.48%) and an increase in extreme HVI (43.45%), driven by higher evapotranspiration and extreme drought conditions. SSP370 showed improved hydrological balances, with low HVI covering 49.21% and extreme HVI decreasing to 10.98%. Conversely, SSP585 displayed a slight increase in low HVI (49.51%) but persistent vulnerabilities, with high HVI (14.01%) and extreme HVI (18.02%) concentrated in key regions. The findings highlight substantial spatial variability in hydrological stress, emphasizing the need of scenario-specific water management strategies. Moderate HVI reflects intermediate vulnerability, while extreme HVI denotes sensitive risks of water scarcity, drought, and flooding, with severe implications for ecosystems and communities. Extreme rainfall events under SSP585 pose additional challenges, such as soil erosion, land degradation, and increased water treatment costs. Effective water conservation measures and adaptive infrastructure are essential to mitigate these risks. Furthermore, increased atmospheric water demand under SSP370 and SSP585 raises the potential for drought, threatening agricultural productivity and ecological health. Precipitation patterns under SSP245 suggest manageable water stress, while SSP370 and SSP585 reveal greater challenges from higher emissions, including extreme rainfall and associated flood risks. The HVI framework integrates climate projections with actionable insights, offering a comprehensive approach to sustainable water management, adaptive infrastructure, and targeted interventions. Hence, innovative policies are critical to address extreme HVIs ensuring resilience against water scarcity and ecosystem degradation. This study underscores the importance of coupling data-driven hydrological analysis with climate responsiveness for effective watershed management and environmental sustainability.

Funding: This Research was carried out under 2025 KICT Research Program (Development of IWRM-Korea Technical Convergence Platform Based on Digital New Deal) funded by the Ministry of Science and ICT.