Preprints
https://doi.org/10.5194/hess-2021-113
https://doi.org/10.5194/hess-2021-113

  11 Mar 2021

11 Mar 2021

Review status: a revised version of this preprint is currently under review for the journal HESS.

Evaluation and interpretation of convolutional-recurrent networks for regional hydrological modelling

Sam Anderson and Valentina Radic Sam Anderson and Valentina Radic
  • Department of Earth, Ocean, and Atmospheric Sciences, University of British Columbia, Vancouver, V6T 1Z4, Canada

Abstract. Deep learning has emerged as a useful tool across geoscience disciplines; however, there remain outstanding questions regarding the suitability of unexplored model architectures and how to interpret model learning for regional scale hydrological modelling. Here we use a convolutional-recurrent network, a deep learning approach for learning both spatial and temporal patterns, to predict streamflow at 226 stream gauges across the region of southwestern Canada. The model is forced by gridded climate reanalysis data and trained to predict observed daily streamflow between 1979 and 2015. To interpret the model learning of both spatial and temporal patterns, we introduce two experiments with evaluation metrics to track the model's response to perturbations in the input data. The model performs well in simulating the daily streamflow over the testing period, with a median Nash-Sutcliffe Efficiency (NSE) of 0.68 and 35 % of stations having NSE > 0.8. When predicting streamflow, the model is most sensitive to perturbations in the input data prescribed near and within the basins being predicted, demonstrating that the model is automatically learning to focus on physically realistic areas. When uniformly perturbing input temperature timeseries to obtain relatively warmer and colder input data, the modelled freshet timing and intensity changes in accordance with the transition timing from below- to above-freezing temperatures. The results demonstrate the suitability of a convolutional-recurrent network architecture for spatiotemporal hydrological modelling, making progress towards interpretable deep learning hydrological models.

Sam Anderson and Valentina Radic

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-2021-113', Anonymous Referee #1, 20 Apr 2021
    • AC1: 'Reply on RC1', Sam Anderson, 30 Jun 2021
  • RC2: 'Comment on hess-2021-113', Anonymous Referee #2, 03 Jun 2021
    • AC2: 'Reply on RC2', Sam Anderson, 30 Jun 2021

Sam Anderson and Valentina Radic

Sam Anderson and Valentina Radic

Viewed

Total article views: 522 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
364 145 13 522 3 3
  • HTML: 364
  • PDF: 145
  • XML: 13
  • Total: 522
  • BibTeX: 3
  • EndNote: 3
Views and downloads (calculated since 11 Mar 2021)
Cumulative views and downloads (calculated since 11 Mar 2021)

Viewed (geographical distribution)

Total article views: 458 (including HTML, PDF, and XML) Thereof 458 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 24 Jul 2021
Download
Short summary
We develop and interpret a spatio-temporal deep learning model for regional streamflow prediction at more than 200 stream gauge stations in Western Canada. We find the novel modelling style to work very well for daily streamflow prediction. Importantly, we interpret model learning to show that it has learned to focus on physically interpretable and physically relevant information, which is a highly desirable quality of machine-learning based hydrological models.