Reliability-weighted approach for streamflow prediction at ungauged catchments

Abstract

Understanding the hydrological processes and the determination of the frequencies and magnitudes of streamflow are crucial for better oversight of available water resources. A significant obstacle to accurately assessing the water resources of a basin is the accessibility of data for hydrologic modeling. For example, nearly 70% of Ethiopia's Omo-Gibe River basin is ungauged, despite the fact that it is currently undergoing major socioeconomic activity, including the development of cascade hydropower generation. As a result, regionalization remains a critical difficulty in creating solid projections in these ungauged catchments. Thus, this study aimed to provide a dependable technique for accurately determining surface water resources in ungauged catchments. To accomplish this, a new approach called Reliability-weighted (RB) has been introduced to predict runoff in regions with unreliable data. This method combines the advantages of three commonly used parameter transfer methods: global mean, physical similarity, and spatial proximity through weighted averaging. The weights are computed using the reliability-weighted values of each parameter transfer strategy, which are also computed using the donor catchment's hydrological model's reliability value during the calibration and validation periods. The suitability of the proposed scheme was checked in the Omo Gibe and Upper Blue Nile River basins, in Ethiopia. After analyzing parameter transfer techniques using the jackknife algorithm at several target-gauging stations, it was found that the reliability-weighted method outperforms all three regionalization approaches by about 30% for the test catchments. In addition, approximately 85% of the time, the proposed strategy's regionalization performance has a metric Nash-Sutcliffe efficiency greater than 0.50. Alternatively, according to the median Efficiency of the Nash-Sutcliffe model, the newly proposed strategy reduces the spatial and temporal losses by around 0.01 and 0.02, respectively, when compared to the second-best performing option. Furthermore, the performance of the presented approach's runoff prediction at the test gauged catchments was very comparable to the performance of the on-site calibrated model. The newly proposed approach therefore proved to be a useful tool for assessing the potential of available water resources in ungauged catchments. Finally, the uncertainty in runoff prediction at ungauged basins was reduced by combining the strengths of parameters transfer strategies relying on their reliability at test gauging stations.