Preprints
https://doi.org/10.5194/hess-2021-229
https://doi.org/10.5194/hess-2021-229

  25 May 2021

25 May 2021

Review status: this preprint is currently under review for the journal HESS.

Applying Non-Parametric Bayesian Network to estimate monthly maximum river discharge: potential and challenges

Elisa Ragno, Markus Hrachowitz, and Oswaldo Morales-Nápoles Elisa Ragno et al.
  • Delft University of Technology, Faculty of Civil Engineering and Geosciences, 2628 CN, Delft, Netherlands

Abstract. Non-Parametric Bayesian Networks (NPBNs) are graphical tools for statistical inference widely used for reliability analysis and risk assessment. However, few hydrological applications can be found in the literature. We therefore explore here the potential of NPBNs to reproduce catchment-scale hydrological dynamics by investigating 240 catchments with contrasting climate across the United States from the CAMELS dataset. First, two networks, one unsaturated (UN-1) and one saturated network (SN-1) based on hydro-meteorological variables are used to generate monthly maximum river discharge considering the catchment as a single element. Then, the saturated network SN-C, based on SN-1 but additionally including physical catchments attributes, is used to model a group of catchments and infer monthly maximum river discharge in ungauged basins based on the attributes similarity. The results indicate that the UN-1 model is suitable for catchments with a positive dependence between precipitation and river discharge, while the SN-1 model can reproduce discharge also in catchments with negative dependence. Furthermore, in ~40 % of the catchments analysed the SN-1 model can reproduce statistical characteristics of discharge, tested via the Kolmogorov-Smirnov (KS) statistic, and Nash-Sutcliffe Efficiencies (NSE) ≥ 0.5. Such catchments receive precipitation mainly in winter and are located in energy-limited regions at low to moderate elevation. Further, the SN-C model, in which the inference process benefits from catchment similarity, can reproduce river discharge statistics in ~10 % of the catchments analysed. However, in these catchments a common dominant physical attribute was not identified. In this study, we show that, once a NPBNs is defined, it is straightforward to infer discharge, when the remaining variables are known. We also show that it is possible to extend the network itself with additional variables, i.e. going from SN-1 to SN-C. Despite these advantages, the results also suggest that there are considerable challenges in defining a suitable NPBN, in particular for predictions in ungauged basins. These are mainly due to the discrepancies in the time scale of the different physical processes generating discharge, the presence of a “memory” in the system, and the Gaussian-copula assumption used by NPBNs for modelling multivariate dependence.

Elisa Ragno et al.

Status: open (until 20 Jul 2021)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on hess-2021-229', Anonymous Referee #1, 08 Jun 2021 reply
  • RC2: 'Comment on hess-2021-229', Anonymous Referee #2, 10 Jun 2021 reply

Elisa Ragno et al.

Elisa Ragno et al.

Viewed

Total article views: 236 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
187 44 5 236 13 1 2
  • HTML: 187
  • PDF: 44
  • XML: 5
  • Total: 236
  • Supplement: 13
  • BibTeX: 1
  • EndNote: 2
Views and downloads (calculated since 25 May 2021)
Cumulative views and downloads (calculated since 25 May 2021)

Viewed (geographical distribution)

Total article views: 233 (including HTML, PDF, and XML) Thereof 233 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 23 Jun 2021
Download
Short summary
We explore the ability of Non-Parametric Bayesian Networks to reproduce monthly maximum river discharge events in a catchment when the remaining hydro-meteorological and catchment attributes are known. We show that a saturated network evaluated in an individual catchment can reproduce statistical characteristics of discharge in about ~40 % of the cases, while challenges remain when a saturated network evaluated considering all the catchment together is considered.