Evaluating a conceptual hydrological model at gauged and ungauged basins using machine learning-based limits-of-acceptability and hydrological signatures

This is a Preprint and has not been peer reviewed. This is version 2 of this Preprint.

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.

Downloads

Download Preprint

Authors

Abhinav Gupta, Mohamed Hantush, Rao Govindaraju

Abstract

Hydrological models are evaluated by comparisons with observed hydrological quantities such as streamflow. A model evaluation procedure should account for dominantly epistemic errors in measured hydrological data such as observed precipitation and streamflow and avoid type-2 errors (rejecting a good model). This study uses quantile random forest (QRF) to develop limits-of-acceptability (LoA) over streamflow that accounts for the measurement uncertainties. A significant advantage of this method is that it can be used to evaluate models even at ungauged basins. In this study, this method was used to evaluate a hydrological model – namely the Sacramento Soil Moisture Accounting (SAC-SMA) in St. Joseph River Watershed (SJRW) – in gauged and hypothetical ungauged scenarios. Using LoA alone to account for uncertainty in data yielded a large number of models as behavioral, suggesting the need for additional measures to develop a more discriminating inference procedure. Five streamflow-based signatures (i.e., autocorrelation function, Hurst exponent, baseflow index, flow duration curve, and long-term runoff coefficient) were used to further eliminate physically unrealistic models which were considered behavioral by LoAs. The combination of LoAs over streamflow and streamflow-based signatures helped constrain the set of behavioral models in both gauged and ungauged scenarios. Among the signatures used in this study, Hurst exponent and baseflow index were the most useful ones. The NSEs of behavioral models ranged from 0 to 0.65. Very wide predictive uncertainty bounds were obtained in the ungauged scenario. Many of the behavioral models resulted in underestimation (overestimation) of observed high (low) flow. Overall, the methodology used in this study showed promise as a model inference strategy.

DOI

https://doi.org/10.31223/X5PD8S

Subjects

Engineering

Keywords

Model (in)validation, Limits-of-Acceptability, machine learning, Prediction at ungauged basins, streamflow, Limits-of-acceptability, machine learning, Prediction at ungauged basins

Dates

Published: 2024-01-30 17:18

Last Updated: 2024-02-01 08:04

Older Versions
License

CC-By Attribution-NonCommercial-NoDerivatives 4.0 International

Additional Metadata

Conflict of interest statement:
None

Data Availability (Reason not available):
Data are publicly available and the links an be found in manuscript.