Work

Assessing the role of gridded evapotranspiration products in improving streamflow simulation and reducing hydrological modeling uncertainty

Year: 2026

Type: article

Abstract: Hydrological models are typically calibrated using streamflow measurements, which poses challenges in ungauged or poorly gauged basins. In addition, calibrating solely with streamflow cannot guarantee accurate estimates of other hydrological variables, leading to large uncertainty in hydrological predictions. The inclusion of other variables, such as evapotranspiration (ET) data, offers a potential approach to overcome these limitations. This study systematically evaluated how to incorporate gridded ET products to constrain hydrological models and reduce the uncertainty in hydrological modeling. To this end, three calibration experiments, established based on streamflow observations (benchmark), gridded ET products, and the combination of both data, were applied to constrain a process-based hydrological model in the Baihe River basin, located in the upper Han River basin. Results indicated that: (1) Single-variable calibration with ET products provided acceptable monthly streamflow estimates, with the average Nash-Sutcliffe Efficiency (NSE) of 0.75 and 0.67 during the calibration and validation periods, respectively. This provides valuable support for direct model calibration in ungauged catchments. For multi-variable calibration, using additional ET data achieved similar monthly streamflow simulation compared to the benchmark, with the average NSE being 0.84 and 0.87, respectively. (2) Model calibration using a single variable (only streamflow or ET data) could result in large uncertainty in parameters and simulated results. In contrast, multi-variable calibration using their combination improved the reliability of model predictions and reduced the overall parameter uncertainty, providing better constraints for representing hydrologic processes. (3) Compared to the original ET data, the bias-corrected ET data performed better in reducing prediction uncertainty and improved the estimation of median and low flows, with average NSE and Kling-Gupta efficiency (KGE) improved by 0.08 and 0.37, respectively. These findings provide valuable insights for applying gridded ET products in hydrological model calibration and prediction.

Source: Ecological Informatics

DOI: 10.1016/j.ecoinf.2026.103698