Developing a Multimodel Ensemble Framework for Improved Streamflow Simulation in the Amudarya River Basin

Abstract There is a broad consensus that an ensemble of global climate models (GCMs) generally outperforms individual models and is, therefore, preferred in water resources’ impact assessment studies for better planning and management. However, the reliability of future projections depends on the historical performance of these models. This study first aimed to evaluate the performance of different GCMs’ ensembles by simulating streamflow and, second, to develop a new ensemble framework for improved hydrologic simulation. We used daily precipitation and temperature data (1931–2012) of all available GCMs from the Coupled Model Intercomparison Project phase 6. Six datasets were created from a best-performing GCM, five top-ranked GCMs, and all available GCMs, both with and without bias correction (BC). These datasets were evaluated by simulating streamflow with a hydrological model at different hydrometric stations in the Amudarya River basin, which shares the borders of Afghanistan, Tajikistan, Kyrgyzstan, Turkmenistan, and Uzbekistan. Initial simulations using these conventional ensembles—both raw and bias-corrected—showed poor performance, with Nash–Sutcliffe efficiency E values often below 0.50 and underestimation of peak flows. To address this, we proposed a multimodel ensemble using a ranking-based weighted mean of bias-corrected GCMs, further disaggregating into a daily scale using a weather generator. This approach demonstrated substantial improvements in streamflow simulation, increasing E and correlations by up to 0.72 and 0.85, respectively, on average, and a more accurate reproduction of extreme flows. So the proposed framework can be used with confidence in simulating hydrological processes, particularly for extreme events.