A non-parametric data-based approach for probabilistic flood forecasting in support of uncertainty communication
Introduction
At Flanders Hydraulics Research, a division of the Flemish governmental authorities in Belgium, hydrodynamic models are used to forecast discharges and water levels at a number of locations along the navigable rivers in the Flanders region of Belgium. The forecasts are issued several times a day and the system produces water level forecasts up to 48 hours in the future. The obtained forecasts are subject to uncertainties originating from the lack of accuracy in the input data (e.g. rainfall forecasts), the quality of the hydrological and hydrodynamic sub-model structures, the accuracy of parameter calibrations, etc. The models provide deterministic results of forecasted water levels and discharges, but perform less well than hoped during the majority of extreme flood events. Similar observations have been made in other countries (Spencer et al., 2006). The disappointing level of performance is not surprising, considering that each large flood event has unique characteristics and that it is very difficult to calibrate a model for these rare events. Inevitably, any model attempting to forecast complex hydrologic and/or hydraulic processes will do so with some (often significant) degree of error (Leedal et al., 2010, Pappenberger et al., 2005). Fig. 1 shows an example of observed versus forecasted water levels along the river Demer at the gauging station of Zichem. Differences between the observation and the forecast up to 1.5 m can be noticed. Given that the bottom level of the river Demer at Zichem is at 14.5 mTAW, this gives a relative error up to 60%, whereas maximum relative errors of 10% are typically considered as acceptable by the responsible authority (Van Looveren et al., 2000).
It consequently can be advised to provide to end users of the forecasts, the water and flood crisis managers and other decision makers, probabilistic information based on estimation of the forecast uncertainty. Such information should take indeed a central role in the presentation of model output to water managers and decision makers. The need for a range of forecast stages with associated probabilities was already addressed by Ingram (1997).
Leedal et al. (2010) state two main obstacles in achieving the goal of estimation and presentation of uncertainty in real-time flood predictions: (1) because calculation time and stability are essential requirements for real time flood forecasting applications, mathematically complex and time consuming methods for uncertainty estimation may not be appropriate, and (2) probabilistic information on model output has to be presented in a useful form, avoiding confusion or misinterpretation. This paper addresses these two obstacles and presents a simple, but robust approach for estimating the uncertainty of real-time flood forecasts.
A review of the existing approaches is given in Section 2 followed by Section 3 which discusses aspects of communication of uncertainty. The methodology selected in this study for uncertainty estimation on river flood forecasting results is explained in Section 4. In that section also the considered flood forecasting models and study cases are presented. In Section 5 results on the applied approach are shown and discussion on the use of these results in communication of the uncertainty to water managers and the larger public. Section 6 summarizes the different conclusions.
Section snippets
Existing approaches for uncertainty estimation
The interest in assessing “uncertainty” in flood forecasting models has grown considerably within the scientific communities of meteorologists and hydrologists. This paper focuses on the predictive uncertainty, i.e. the uncertainty of a predictand given on the available information and, in particular on one or more forecastst (Todini, 2009). Emulative uncertainty or validation uncertainty is not considered. Various approaches for assessing the predictive uncertainty in hydrological and
Uncertainty communication
Communication of the uncertainty is an as even important task as the calculation of the uncertainty or might even be more important (Kloprogge et al., 2007). The uncertainty information has to be presented is such a way that it avoids confusion and that it can form a base on which sound decisions can be taken, potentially with the aid of decision support systems (Honghai and Altinakar, 2011, de Kort and Booij, 2007). The communication of uncertainty can be done in various ways: linguistic,
The flood forecasting models
The flood forecasting models considered in this study are the ones, which are currently in operational use by Flanders Hydraulics Research. They consist of a combination of catchment hydrological and river hydrodynamic models and a data-assimilation (DA) method for real-time model updating. The hydrological models are implemented in the NAM module of the Mike11 software of DHI Water & Environment (DHI, 2007a, DHI, 2007b). NAM is a lumped conceptual rainfall-runoff model (Madsen, 2000). The
Confidence intervals and exceedance probability
The non-parametric method was implemented in software, that is connected to the database of the forecasting system. Hereby it is possible to automatically update the error matrix, based on new simulations and observations. By making use of the calculated confidence intervals, also the exceedance probability of alert and alarm levels could be calculated and visualised. Fig. 12a shows the forecasted water levels before and after bias correction and the confidence intervals for one of the gauging
Conclusion
A method has been presented to produce probabilistic water level forecasts. The method was implemented and tested for three river basins in the Flanders region of Belgium and allows calculation of confidence intervals as well as exceedance probabilities of given alert and alarm levels in function of the time horizon. The method tackles the problem of heteroscedasticity of the forecast residuals and does not apply any predefined probability distribution. After comparison with a Bayesian
Acknowledgements
The results presented in this paper were obtained by a research project on flood forecasting for Flanders Hydraulics Research of the Flemish Government of Belgium.
References (41)
- et al.
Fuzzy neural networks for water level and discharge forecasting with uncertainty
Environmental Modelling & Software
(2011) - et al.
Real-time probabilistic forecasting of flood stages
Journal of Hydrology
(2007) - et al.
Decision making under uncertainty in a decision support system for the Red River
Environmental Modelling & Software
(2007) - et al.
Evaluation of statistical models for forecast errors from the HBV model
Journal of Hydrology
(2010) Automatic calibration of a conceptual rainfall-runoff model using multiple objectives
Journal of Hydrology
(2000)- et al.
Reply to comment by Keith Beven, Paul Smith and Jim Freer on “Hydrological forecasting uncertainty assessment: incoherence of the GLUE methodology”
Journal of Hydrology
(2007) Order tests for two-sample problem and their power II
Indagationes Mathematicae
(1953)Order tests for two-sample problem and their power III
Indagationes Mathematicae
(1953)- et al.
Uncertainty communication in environmental assessments: views from the Dutch science-policy interface
Environmental Science & Policy
(2008) A time series tool to support the multi-criteria performance evaluation of rainfall-runoff models
Environmental Modelling & Software
(2009)
The future of distributed models: model calibration and uncertainty prediction
Hydrological Processes
An analysis of transformations
Journal of the Royal Statistical Society
Recent developments in predictive uncertainty assessment based on the model conditional processor approach
Hydrology and Earth System Sciences Discussions
Mike11: A Modelling System for Rivers and Channels. Reference Manual
Mike11: A Modelling System for Rivers and Channels. User Guide
A GIS-based decision support system for integrated flood management under uncertainty with two dimensional numerical simulations
Environmental Modelling & Software
Advanced hydrologic forecasting for flood and drought mitigation
Cited by (49)
Gaussian process emulation of spatio-temporal outputs of a 2D inland flood model
2022, Water ResearchCitation Excerpt :At high temporal and spatial resolutions, although parallel CPU and GPU implementations can significantly reduce the computation time, a single simulation of such models can take many hours to complete (Abolfathi et al., 2018; Goodarzi et al., 2020; Shaw et al., 2021). Despite the computational limitations, in many cases, numerical modelling is the most reliable tool for simulation and prediction of flooding and inundation especially when no alternative data-driven approaches (Van Steenbergen, et al., 2012; Abolfathi et al., 2016) can be utilised due to sparse empirical observations. The adoption of machine learning (ML) approaches (Mosavi, et al., 2018; Yang & Chang, 2020; Lin, et al., 2020) in flood modelling could alleviate the high computational costs associated with numerical modelling, allowing a wider range of scenarios to be considered in a reasonable amount of time and with more accessible computational resources.
A GIS-based spatial multi-index model for flood risk assessment in the Yangtze River Basin, China
2020, Environmental Impact Assessment ReviewSystematic analysis of uncertainty in 2D flood inundation models
2019, Environmental Modelling and SoftwareCommunicating model uncertainty for natural hazards: A qualitative systematic thematic review
2019, International Journal of Disaster Risk ReductionError correction-based forecasting of reservoir water levels: Improving accuracy over multiple lead times
2018, Environmental Modelling and SoftwareBayesian flood forecasting methods: A review
2017, Journal of Hydrology