Performance evaluation of various hydrological models with respect to hydrological responses under climate change scenario: a review

Studies reviewed in this paper show anomaly for temperature pertaining to streamflow and rainfall showing different trends, especially in Ethiopia to support the research findings and interpretation. There are many hydrological models, including 54 physically distributed, lumped, and conceptual hydrological models, of which 28 have been used in Ethiopian river basins. The models include the most adaptable and commonly used SWAT model applicable from small areas up to large basins. It is indeed a challenge to use a single hydrological model as the data rely on consistency, limitation-free, and exactly fitted output. The overall performance of individual physically-based, conceptual, and machine learning (ML) models varied at different watersheds. Reasonably, ML performs very well, up to 0.99 for R2 and NSE and up to 0.001 for PBIAS. Inopportunely, using a single hydrological model has its limitations; ensemble multi-individual models, coupling or hybridization of physical or conceptual models with machine learning, combining evolutionary optimization algorithms with ML, and also comparisons of multi-single hydrological models, and selecting the best one are recommended options. No single hydrological model is indispensable and can be termed as better than the other for any watershed. Somewhat, ML outperforms SWAT but cannot be considered an absolute substitute. The size of the watershed, the number of data used, and the ratio between calibrations year to validation year do not have a clear correlation with the performance, particularly for the SWAT model accounted for in this review. Optimization algorithms explore multiple options and choosing the right one is a tedious task before a final decision is taken.