Preprints
https://doi.org/10.5194/hess-2022-295
https://doi.org/10.5194/hess-2022-295
 
16 Aug 2022
16 Aug 2022
Status: this preprint is currently under review for the journal HESS.

Continuous streamflow prediction in ungauged basins: Long Short-Term Memory Neural Networks clearly outperform hydrological models

Richard Arsenault1, Jean-Luc Martel1, Frédéric Brunet1, François Brissette1, and Juliane Mai2 Richard Arsenault et al.
  • 1Hydrology, Climate and Climate Change Laboratory, École de technologie supérieure, 1100 Notre-Dame West, Montréal, Québec, H3C 1K3, Canada
  • 2Department of Civil and Environmental Engineering, University of Waterloo, 200 University Ave W, Waterloo, Ontario, N2L 3G1, Canada

Abstract. This study investigates the ability of Long Short-Term Memory (LSTM) neural networks to perform streamflow prediction at ungauged basins. A series of state-of-the-art, hydrological model-dependent regionalization methods is applied to 148 catchments in Northeast North America and compared to a LSTM model that uses the exact same available data as the hydrological models. While conceptual model-based methods attempt to derive parameterizations at ungauged sites from other similar or nearby catchments, the LSTM model uses all available data in the region to maximize the information content and increase its robustness. Furthermore, by design, the LSTM does not require explicit definition of hydrological processes and derives its own structure from the provided data. The LSTM networks were able to clearly outperform the hydrological models in a leave-one-out cross-validation regionalization setting on most catchments in the study area, with the LSTM model outperforming the hydrological models in 93 to 97 % of catchments depending on the hydrological model. Furthermore, for up to 78 % of the catchments, the LSTM model was able to predict streamflow more accurately on pseudo-ungauged catchments than hydrological models calibrated on the target data, showing that the LSTM model’s structure was better suited to convert the meteorological data and geophysical descriptors into streamflow than the hydrological models even calibrated to those sites in these cases. Furthermore, the LSTM model robustness was tested by varying its hyperparameters, and still outperformed hydrological models in regionalization in almost all cases. Overall, LSTM networks have the potential to change the regionalization research landscape by providing clear improvement pathways over traditional methods in the field of streamflow prediction in ungauged catchments.

Richard Arsenault et al.

Status: open (until 26 Oct 2022)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • CC1: 'Comment on hess-2022-295', Jonathan Frame, 25 Aug 2022 reply
    • AC1: 'Reply on CC1', Richard Arsenault, 25 Aug 2022 reply
  • CC2: 'Comment on hess-2022-295', John Ding, 31 Aug 2022 reply

Richard Arsenault et al.

Data sets

LSTM regionalization datasets and codes Richard Arsenault, Jean-Luc Martel, Frédéric Brunet, François Brissette, Juliane Mai https://osf.io/3s2pq/

Model code and software

LSTM regionalization datasets and codes Richard Arsenault, Jean-Luc Martel, Frédéric Brunet, François Brissette, Juliane Mai https://osf.io/3s2pq/

Richard Arsenault et al.

Viewed

Total article views: 976 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
647 315 14 976 34 4 4
  • HTML: 647
  • PDF: 315
  • XML: 14
  • Total: 976
  • Supplement: 34
  • BibTeX: 4
  • EndNote: 4
Views and downloads (calculated since 16 Aug 2022)
Cumulative views and downloads (calculated since 16 Aug 2022)

Viewed (geographical distribution)

Total article views: 832 (including HTML, PDF, and XML) Thereof 832 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 28 Sep 2022
Download
Short summary
Predicting flow in rivers where no observation records are available is a daunting task. For decades, hydrological models were setup on these gauges and their parameters were estimated based on the hydrological response of similar or nearby catchments where records exist. New developments in machine learning have now made it possible to estimate flows at ungauged locations more precisely than with hydrological models. This study confirms the performance superiority of machine learning models.