eprintid: 723
rev_number: 8
eprint_status: archive
userid: 17
dir: disk0/00/00/07/23
datestamp: 2018-05-27 17:33:56
lastmod: 2018-05-27 17:33:56
status_changed: 2018-05-27 17:33:56
type: report
metadata_visibility: show
creators_name: Munday, Paul
creators_name: Pearson, Natalie
creators_name: Lang, Georgina
creators_name: Warneford, Emma
creators_name: Siganporia, Zubin
corp_creators: Mike Vaughan
title: Probabilistic flood forecasting
ispublished: pub
subjects: environment
subjects: utilities
studygroups: esgi85_v2
companyname: Environment Agency
full_text_status: public
abstract: The Environment Agency provides a forecasting and warning service to people at risk from flooding. However, flood forecasts are inherently un- certain. Efforts to quantify the uncertainty based on quantile regression have failed to capture the full extent of the uncertainty associated with significant flooding events.
An investigation into factors that may be correlated with the uncertainty lead to the observation that there are structural biases in the model. It is possible to remove these, and thereby reduce the mean square error of the predictions, but the benefit of this is apparent in the prediction of ’normal’ conditions, rather than in flood predictions.
Additionally, a tweak to the linear fit in the quantile regression is sug- gested which is better suited to the data.
problem_statement: (1.1.1) The Environment Agency provides a forecasting and warning service to people at risk from flooding. However, flood forecasts are inherently un- certain. The differences between forecast time series of river level and subsequent observations can be relatively large, so understanding uncer- tainties is useful when interpreting forecasts for decision support.
(1.1.2) Historic flood forecast performance data has been analysed to give an estimate of unceratinty of current flood forecasts in real time. This ap- proach assumes that previous error relationships continue to hold. The paper (Weerts et al, 2009) describes a technique for doing this based on quantile regression.This is used to determine non-parametric relationships between the quantiles of the error distribution of the flood forecasts and the forecast magnitude and the lead-time of the forecast.
(1.1.3) An evaluation of the method has found that a greater-than-expected pro- portion of observed flood peaks fall above the upper uncertainty bounds. For instance, restricting observations to significant events it was found for some locations that typically more than 50% of peaks exceed the 5% level. However, overall the uncertainty bounds were found to contain the correct proportions of observations. The problem is that it is the peaks that are of most interest to forecasters and the public.
(1.1.4) It is thought that this may be due to non-stationarity of the errors. In particular, rivers tend to rise more quickly than they fall, and the errors in prediction are much smaller for falling water levels. Another possible factor is the use of forecasts (which may be inaccurate) for rainfall. How- ever, even when the actual observed rainfall is used as a model input, the problem persists.
(1.1.5) Amongst the quantities that are of most interest are the magnitude of the flood peak, and the time at which a particular threshold is crossed. Currently, the uncertainties in the predictions of these are not quantified.
date: 2012
citation:   Munday, Paul and Pearson, Natalie and Lang, Georgina and Warneford, Emma and Siganporia, Zubin  (2012) Probabilistic flood forecasting.  [Study Group Report]     
document_url: http://miis.maths.ox.ac.uk/miis/723/1/ESGI85-EA_CaseStudy.pdf