Risk-based Probabilistic Fluvial Flood Forecasting for Integrated Catchment Models: Phase 2 - final. Science Report – SR SC080030

Sene, K.; Weerts, A.; Beven, K.; Moore, R.J.; Whitlow, C.; Beckers, P.; Minett, A.; Winsemius, H.; Verkade, J.; Young, P.; Leedal, D.; Smith, P.; Cole, S.; Robson, A.; Howard, P.; Huband, M.; Breton, N.. 2009 Risk-based Probabilistic Fluvial Flood Forecasting for Integrated Catchment Models: Phase 2 - final. Science Report – SR SC080030. Bristol UK, Environment Agency, 152pp. (CEH Project Number: C03755, CEH Project Number: C04217) (Unpublished)

Before downloading, please read NORA policies.

Text SC080030_-_Phase_2_Report_-_final.pdf - Submitted Version Restricted to NORA staff only Download (5MB) | Request a copy

Abstract/Summary

Robust forecasts are vital in providing a comprehensive Flood Warning Service to people and businesses at risk from flooding. For fluvial flood forecasting, rainfall-runoff, flow routing and hydraulic models are often combined into model cascades and are run automatically in the Environment Agency’s National Flood Forecasting System (NFFS). However, it is widely known that the accuracy of flood forecasts can be influenced by a number of factors, such as the accuracy of input data, and the model structure, parameters and state (initial conditions). Having a sound understanding of these modelling uncertainties is vital to assess and improve the flood forecasting service that the Environment Agency provides. This report describes the findings from Phase 2 of the project “Risk-based Probabilistic Fluvial Flood Forecasting for Integrated Catchment Models”, whose main aim is to develop and test practical probabilistic methods to quantify and, where possible, reduce uncertainties around fluvial flood forecasts from sources other than predicted rainfall. The project started in November 2008 and will complete in 2010. The aims of Phase 2 were as follows: • Task 2.1 - Uncertainty Framework - To further develop the high-level unified uncertainty framework for quantifying uncertainty for the major sources of uncertainty (identified in Task 1.2) in integrated catchment models and suggest suitable validation measures • Task 2.2 - Model Run Times - To recommend and investigate alternative ways of reducing run-times for real-time probabilistic models without significantly increasing uncertainty. • Task 2.3 - Case Studies – To demonstrate and validate the probabilistic treatment of uncertainties for selected integrated catchment models through case studies and test configurations in NFFS. • Task 2.4 - Flood Mapping - To briefly test how uncertainties in flood forecasting may affect flood extent and depths and to make recommendations for future research on how to carry forward the outputs of probabilistic flood forecasting to the generation of probabilistic flood warning maps (extent, depths) The locations for the case studies were selected following consultations during Phase 1 of the project and consisted of the following four catchments: • Rapid Response Catchment – Upper Calder in North East Region • Flow Routing Reach – Lower Eden in North West Region • Integrated Catchment Model 1 – Ravensbourne in Thames Region • Integrated Catchment Model 2 – Upper Severn in Midlands Region The uncertainty framework consists of a combination of guidance notes, flowcharts, worksheets and tables. For a given integrated catchment model, the main aim is to assist with selecting an appropriate uncertainty estimation technique taking account of the level of risk, lead-time requirements and a number of other factors (e.g. operational requirements, run-times, performance measures). A simple catchment classification scheme is introduced to assist with identifying the main sources of model and forecast uncertainty (e.g. rainfall forecasts, catchment average rainfall estimates). The framework was in part developed through trials on the four case study catchments, and for a case study for the Netherlands, and will form the basis for the guidelines to be developed during Phase 3 of the project. The investigation of run-times considered the potential improvements from computational improvements, model reconfiguration, statistical approaches, and model emulators. A distinction was made between gains from reducing the number of model runs required, faster model run-times, and reductions in pre- and post-processor times. Sample run-times were also provided from some of the case study catchments. Some potential areas for improvements include developments to NFFS workflow configurations, reconfiguration of hydrodynamic models, and parallel processing. A module adapter for emulation was also developed and its use demonstrated successfully on the Lower Eden and Upper Severn case studies. During the case studies, the uncertainty estimation techniques which were developed and evaluated included the following approaches: • Forward Uncertainty Propagation methods – for rainfall inputs, rainfall-runoff model parameters, rating curves • Data Assimilation – adaptive gain, ensemble Kalman Filter, Data Based Mechanistic (DBM) methodology • Conditioning (or forecast calibration) – ARMA error prediction, quantile regression, Bayesian Model Averaging The selection of techniques was agreed during Phase 1 of the project, and these tasks were completed during the period May to December 2009, providing 7-8 months for implementation. The focus for development was therefore on adaptation of existing methods, development of NFFS configurations, and assessing those techniques with the most potential for operational application. Some new techniques were also developed and evaluated where this was possible within the time available. NFFS model adapters were developed for the adaptive gain, ensemble Kalman Filter and Bayesian Model Averaging methods, with an additional adapter developed for quantile regression, beyond the planned scope for Phase 2 of the project. More than 20 test configurations were developed, with associated factsheets describing the methods and the main strengths and limitations of each approach (including run-time considerations). A series of technique summary sheets also summarise issues such as ease of use, data requirements, probabilistic interpretation, and forecast run-time issues. The rainfall inputs which were considered included raingauge observations, radar- rainfall estimates, deterministic rainfall forecasts, and ensemble rainfall forecasts (STEPS). The combination and conditioning of these and other sources was demonstrated in some cases. For the data assimilation techniques, in contrast to existing Environment Agency techniques, the methods chosen also provide an estimate of uncertainty as part of the assimilation process. The final task, Task 2.4 - Flood Mapping, was a relatively minor component of the project to explore the main issues with using the existing mapping module in NFFS with probabilistic inputs. The factors considered included run-times, visualisation of outputs and the ease of use in configuring and using the module in this way. As part of this work, an existing mapping module for one of the case study catchments (the Lower Eden) was upgraded to match the current operational forecasting model, and was used to test and demonstrate some of the techniques. The report concludes with recommendations for Phase 3 of the project, in which guidelines and an implementation plan will be developed. This includes a discussion (and samples) of how the guideline document might be presented, and initial proposals for the key items to appear in the implementation plan. In addition to technical considerations, these also include recommendations on further software development, further development of techniques, and operational implementation.

Item Type:	Publication - Report
Programmes:	CEH Topics & Objectives 2009 - 2012 > Water > WA Topic 3 - Science for Water Management > WA - 3.1 - Develop next generation methods for river flow frequency estimation and forecasting
UKCEH and CEH Sections/Science Areas:	Boorman (to September 2014)
Additional Pages:	+ appendices
Funders/Sponsors:	Environment Agency, CEH Wallingford
Additional Information. Not used in RCUK Gateway to Research.:	Dissemination Status: for internal Environment Agency use only Citation: Environment Agency (2010) Risk-based Probabilistic Fluvial Flood Forecasting for Integrated Catchment Models:Phase 2 Report. Science Report – SR SC080030. Authors: K. Sene, A. Weerts, K. Beven, R.J. Moore, C. Whitlow, P. Beckers, A. Minett, H. Winsemius, J. Verkade, P. Young, D. Leedal, P. Smith, S. Cole, A. Robson, P. Howard, M. Huband, N. Breton. Research Contractor: Atkins; Collaborators: Deltares, Lancaster University, CEH Wallingford, Edenvale Young. Environment Agency, Bristol, UK, 152pp plus Appendices.
Additional Keywords:	probabilistic, flood, forecasting, ensemble, catchment, rainfall-Runoff, hydrodynamic
NORA Subject Terms:	Meteorology and Climatology Hydrology Atmospheric Sciences
Date made live:	27 Sep 2012 14:32 +0 (UTC)
URI:	https://nora.nerc.ac.uk/id/eprint/13831

Actions (login required)

View Item

Document Downloads

More statistics for this item...