28 Nov 2023
 | 28 Nov 2023
Status: this preprint is currently under review for the journal HESS.

A National Scale Hybrid Model for Enhanced Streamflow Estimation – Consolidating a Physically Based Hydrological Model with Long Short-term Memory Networks

Jun Liu, Julian Koch, Simon Stisen, Lars Troldborg, and Raphael J. M. Schneider

Abstract. Accurate streamflow estimation is essential for effective water resources management and adapting to extreme events in the face of changing climate conditions. Hydrological models have been the conventional approach for streamflow inter/extrapolation in time and space for the past decades. However, their large-scale applications have encountered challenges, including issues related to efficiency, complex parameterization, and constrained performance. Deep learning methods, such as Long Short-Term Memory networks (LSTM), have emerged as a promising and efficient approach for large-scale streamflow estimation. In this study, we conducted a series of experiments to identify optimal hybrid modelling schemes to consolidate physically based models with LSTM aimed at enhancing streamflow estimation in Denmark.

The results showed that the hybrid modelling schemes outperformed the Danish Water Resources Model (DKM) in both gauged and ungauged basins. While the standalone LSTM rainfall-runoff model outperformed DKM in many basins, it faced challenges when predicting streamflow in groundwater-dependent catchments. A serial hybrid modelling scheme (LSTM-q), which used DKM outputs and climate forcings as dynamic inputs for LSTM training, demonstrated higher performance. LSTM-q improved the median Nash-Sutcliffe Efficiency (NSE) by 0.18 in gauged basins and 0.11 in ungauged basins compared to DKM. Similar accuracy improvements were achieved with alternative hybrid schemes, i.e., by predicting the residuals between DKM-simulated streamflow and observations using a LSTM. Moreover, the developed hybrid models enhanced the accuracy of extreme events, which encourages the integration of hybrid models within an operational forecasting framework. This study highlights the advantages of synergizing existing physically based hydrological models with LSTM models, and the proposed hybrid schemes hold the potential to achieve high-quality, large-scale streamflow estimations.

Jun Liu, Julian Koch, Simon Stisen, Lars Troldborg, and Raphael J. M. Schneider

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on hess-2023-235', Anonymous Referee #1, 26 Dec 2023
  • RC2: 'Comment on hess-2023-235', Anonymous Referee #2, 27 Dec 2023
  • RC3: 'Comment on hess-2023-235', Anonymous Referee #3, 03 Jan 2024
Jun Liu, Julian Koch, Simon Stisen, Lars Troldborg, and Raphael J. M. Schneider
Jun Liu, Julian Koch, Simon Stisen, Lars Troldborg, and Raphael J. M. Schneider


Total article views: 361 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
247 96 18 361 4 3
  • HTML: 247
  • PDF: 96
  • XML: 18
  • Total: 361
  • BibTeX: 4
  • EndNote: 3
Views and downloads (calculated since 28 Nov 2023)
Cumulative views and downloads (calculated since 28 Nov 2023)

Viewed (geographical distribution)

Total article views: 350 (including HTML, PDF, and XML) Thereof 350 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 25 Feb 2024
Short summary
We developed hybrid schemes to enhance national scale streamflow predictions, combining LSTM with a physically based hydrological model (PBM). A comprehensive evaluation of the hybrid setups across all of Denmark indicates that LSTM forced only by climate data and catchment attributes perform well in many regions but face challenges in groundwater-dependent basins. The hybrid schemes supported by PBM possess higher performance in reproducing long-term streamflow behavior and extreme events.