SC080030 - Risk-based Probabilistic Fluvial Flood Forecasting for Integrated Catchment Models

Theme:Incident Management and Community
Project status:Completed
Start date:20/08/2008
End date:24/08/2011
Keywords:
  • Flood Defence
Contractor:
  • WS Atkins,
  • CEH - Wallingford
Contact:

PROJECT SUMMARY

This Environment Agency R&D project investigated how to use probabilistic river flood forecasting to help understand and reduce the uncertainty in flood forecasts made as part of flood risk assessments.

Reliable flood forecasts are vital to providing a flood warning service to people and businesses at risk from flooding. For fluvial (river) flood forecasting, rainfall-runoff, river flow routing and hydraulic computer models are often combined into computer model cascades and are run automatically in the Environment Agency’s National Flood Forecasting System (NFFS).

However, the accuracy of flood forecasts can be influenced by a number of factors such as the accuracy of the data fed into the model, model structure, parameters and state (initial conditions of the scenario modelled). Having a good understanding of these modelling uncertainties is vital to maintaining and improving the flood forecasting service provided by the Environment Agency.

This R&D project developed internal practical guidance for conducting risk-based probabilistic fluvial flood forecasting for integrated catchment models. The main aim of the project was to develop and test practical probabilistic methods to quantify and, where possible, reduce uncertainties in fluvial flood forecasts from sources other than predicted rainfall.

The project provided advice on how to use probabilistic fluvial flood forecasting techniques, including selecting the most suitable method depending on the situation being forecast. The target audience for the information are Environment Agency flood forecasting technical specialists and others involved in maintaining and improving forecasting models.

The advice is based on experience gained and methods developed during this R&D project, which evaluated probabilistic techniques for the Upper Calder, Lower Eden, Ravensbourne and Upper Severn catchments in North East, North West, Thames and Midlands regions of the UK. The locations for the case studies were chosen after consulting regional flood forecasting and warning staff early on during the project.

Following an introduction to concepts in probabilistic forecasting, and its potential uses, this project outlined a number of possible methods, including forward uncertainty propagation, data assimilation and forecast calibration techniques. A concise version of an uncertainty framework was developed as a guide for selecting appropriate techniques.

Forward uncertainty propagation techniques include the following methods for propagating individual sources of uncertainty through integrated catchment models, such as:
• Rainfall inputs derived from raingauge weighting schemes.
• Model parameter uncertainty from MCRM, TCM and PDM rainfall-runoff models.
• Rating curve uncertainty.
These key sources of uncertainty were identified as priorities in consultations during project. Two data assimilation techniques are also described (adaptive gain and the data-based mechanistic approach) which, unlike the deterministic approaches currently used in the Environment Agency, estimate uncertainty as well as improving the forecast.

The work confirmed that forecast calibration (or conditioning) has a key role to play in calibrating the probabilistic content of forecasts, based on long runs of historical data (hindcasts). The methods which are considered are quantile regression (also known as historic flood forecast performance tool) and autoregressive moving average (ARMA) error prediction.

Suggestions were also made for possible probabilistic performance measures which might be used, whilst noting that this topic is considered in other projects. Run-time issues were investigated where they affect the methods considered, and possible options for reducing run times were considered, such as computational improvements, and reconfiguration and emulators for hydrodynamic models.