Open Access

Qi Sun, Joël Arnault, Patrick Laux, Luca Glawion, Harald Kunstmann

• Hydrological models are essential tools for water resource management and for mitigating extreme hydrological events risks. Although they are crucial for flood forecasting, these models often exhibit substantial uncertainties, including input data uncertainties (e.g., precipitation) and structural uncertainties of the models themselves. This study aims to explore the implications of different precipitation datasets and hydrological model structures on streamflow simulation, by evaluating the effects of multiple precipitation products and employing an enhanced model version to reduce structural uncertainty. This study evaluated the hydrological applicability of three representative precipitation products—reanalysis-based (the land component of the fifth-generation European Reanalysis, ERA5-Land), satellite-based (Integrated Multi-satellite Retrievals for GPM, IMERG), and machine learning-based (the first deep learning based spatio-temporal downscaling of precipitation data on a globalHydrological models are essential tools for water resource management and for mitigating extreme hydrological events risks. Although they are crucial for flood forecasting, these models often exhibit substantial uncertainties, including input data uncertainties (e.g., precipitation) and structural uncertainties of the models themselves. This study aims to explore the implications of different precipitation datasets and hydrological model structures on streamflow simulation, by evaluating the effects of multiple precipitation products and employing an enhanced model version to reduce structural uncertainty. This study evaluated the hydrological applicability of three representative precipitation products—reanalysis-based (the land component of the fifth-generation European Reanalysis, ERA5-Land), satellite-based (Integrated Multi-satellite Retrievals for GPM, IMERG), and machine learning-based (the first deep learning based spatio-temporal downscaling of precipitation data on a global scale, spateGAN-ERA5), using the offline version of WRF-Hydro, a distributed hydrological model. Additionally, this study evaluated the performance of an enhanced version of WRF-Hydro, incorporating an overbank flow module for reducing the model structural uncertainty in a large, flood-prone tropical river basin, Irrawaddy River Basin in Myanmar. The findings indicate that: (1) Simulations driven by IMERG precipitation outperformed those driven by ERA5-Land and spateGAN-ERA5 in terms of accuracy in streamflow, with average NSE values of 0.77, compared to 0.19 and 0.09, respectively; (2) The modified model with enabled overbank flow showed consistent improvements over the default model. The average NSE improved from 0.09–0.77 (default) to 0.31–0.78 (modified); (3) The water balance analysis reveals that incorporating the overbank flow module reduces surface runoff, accompanied by an increase in soil moisture storage, and slightly enhancing underground runoff and evapotranspiration (ET) during the rainy period. After the end of the rainy period, the increase soil moisture storage gradually contributes to an increase in surface runoff. These results highlight the significant impact of accurate precipitation data and the overbank flow module on hydrological processes, particularly in flood-prone areas, and suggest that the modified model and high quality precipitation data may enhance hydrological forecasting capabilities.…