Unravelling and improving the potential of global discharge reanalysis dataset in streamflow estimation in ungauged basins

Abstract

Mastery in forecasting the streamflow is of great importance in environmental and sustainability research. Although many global-scale reanalysis products provide a new way to overcome the lack of streamflow records in ungauged basins, streamflow estimation through hydrological models remains a great challenge mostly due to inevitable biases. In this study, we developed a novel bias-correction system equipped with the proposed Piecewise Random Forest (P-RF) model to improve the potential of GloFAS-ERA5 (GloFAS), a global-scale river discharge reanalysis product, as a calibration benchmark for building hydrological models in ungauged basins. Considering three ungauged scenarios, several cases of temporal, spatial, and spatiotemporal bias-corrections were implemented with a total of 13 river gauges located in the Min River Basin in China, and the Fuji River Basin and the Shinano River Basin in Japan. Then, the well-improved GloFAS discharge was applied for the calibration of the Block-wise use of the TOPMODEL (BTOP) model to evaluate its performance in substituting the discharge observations. The results show that: (1) the bias-correction system performs better on the temporal scale, which applies to ungauged basins lacking long-term continuous observations; (2) the integrity and adequacy of the samples used for training the P-RF model have a significant impact on the spatial and spatiotemporal bias-corrections, and they can be reliably estimated by the proposed metric, Ratio of the Valid samples' Proportion; and (3) the statistical metric differences between the simulated discharges obtained by the calibrated BTOP model using observations and GloFAS discharge, are reduced by 25%–50% through the bias-correction.