Does the performance enhancement through multi-model averaging at the catchment outlet gets translated to the interior ungauged points?

Abstract

Accurate prediction of streamflow is essential at both the outlet and interior points for effective water resource management. This study aimed to investigate whether improved simulation results at the catchment outlet through Multi-Model Averaging (MMA) can be translated to interior ungauged points in a catchment. To achieve this, the study considered ensemble combinations of 81 conceptual hydrologic models and calibrated each model using four objective functions for two study areas (Jagdalpur and Wardha subbasins of Godavari basin). Six MMA methods were applied to evaluate their performance at interior points in the catchment. The findings revealed that, when compared to the best performing hydrologic model, the best MMA method showed an improvement in terms of Nash-Sutcliffe efficiency, ranging from 2.2% to 3.8% at the interior point of Jagdalpur and 2% to 12% at the interior points of Wardha. MMA methods with KGE-calibrated models as input exhibited better performance in terms of NSE, while those using LogarithmicNSE-calibrated and parent models as input ensembles showed better performance in terms of Percentage bias and Skill Score. Bayesian model averaging and Genetic Algorithm based averaging methods were identified as the most appropriate MMA methods, and the study findings suggest that the performance of these techniques varies at different interior points based on their spatial and temporal variation of flow characteristics. Overall, this study would be beneficial for better flow prediction at internal ungauged points, which is crucial for effective water resource management in areas where there is limited data or where new developments have occurred.