Comparability of NARX model to SWAT model in simulating future water resources scenarios using CMIP6 climate model outputs over UASB, Ethiopia

Abstract Quantifying water resource potential is crucial for making well-informed decisions in planning, managing, and developing water resources within a given study area. This study utilizes an ensemble of climate variables derived from five CMIP6 climate models (EC-EARTH3, GFDL-ESM4, MPI-ESM1-2-HR, MRI-ESM2, and INM-CM5-0) to simulate future monthly streamflow conditions over the Upper-Awash Sub-Basin (UASB) for three Shared Socioeconomic Pathway (SSP) scenarios (SSP1.26, SSP2.45, and SSP5.85) until the end of the twenty-first century. Streamflow simulations are conducted using both a non-linear data-driven model, NARX, and a physically based model, SWAT. These models are trained and validated using observed streamflow data from a gauging station at the outlet of the sub-basin. During training, the NARX model exhibits a Nash–Sutcliffe Efficiency (NSE) of 94%, while the SWAT model achieves 88%. In validation, NARX maintains a high NSE of 92%, compared to SWAT's 82%. Overall, the NARX model demonstrated superior performance and applicability for quick streamflow simulation with fewer input variables. However, it struggles with peak flow simulations due to its sensitivity to outliers in the training phase. Despite differences in their capabilities, both models project an increase in future monthly streamflow across all scenarios and time periods. Seasonal projections indicate a rise in streamflow during the rainy seasons of spring and summer, while dry periods (Dec–Jan–Feb) experience a decrease. The anticipated increase in streamflow during rainy seasons may exacerbate flood incidences, especially when combined with escalating industrialization and population growth within the sub-basin.