Work

Modelling with uncertain and potentially disinformative discharge data: Part II. A GLUE limits-of-acceptability approach to ensemble calibration

Year: 2026
Type: preprint
Abstract: The information content of uncertain hydrological data imposes limits on the extent to which a modeller can identify the model structures that most adequately represent an assemblage of real-world states and fluxes. This study demonstrates an ensemble modelling strategy for investigating catchment functioning. It builds on the first paper in this two-part series, where a Generalised Likelihood Uncertainty Estimation approach to specifying limits of acceptability was developed based on propagating the joint effect of uncertainties in precipitation and discharge data on a timestep-by-timestep basis through rainfall–runoff ratio (RRR)-based benchmarks. Here, these a priori defined rejection criteria were used to discriminate among 3.5 million parameter sets within 70 model structures configured using the Framework for Understanding Structural Errors (FUSE) for a data-poor, baseflow-dominated Philippine catchment. Analysis of predictive reliability–precision trade-off and other diagnostics was performed to compare the effectiveness of the conventional limits-of-acceptability definition based on rating curve (RC) uncertainty with that of the proposed RRR-based definition, which also applies an informal likelihood measure designed to downweight the influence of residuals associated with potentially disinformative observations on model conditioning. No model realisations survived the RC-based limits, suggesting that discharge data uncertainty alone may underestimate the effective observational error relevant to calibration. Although both definitions converged on the same dominant mechanisms, the RRR-based definition improved discrimination among tested process representations. Differences in behavioural ensemble composition further illustrate how epistemic uncertainties in data, if unaccounted for, can bias hypothesis-testing inference, raising the risk of Type I or Type II errors.
Language: English
Open Access status: green
OpenAlex ID: W7235938413
Cites:
Cited by:
Related to: 10