FEH Local: Improving flood estimates using historical data

Ilaria Prosdocimi,Lisa Stewart,Duncan Faulkner,Chrissy Mitchell

doi:10.1051/e3sconf/20160701006

Abstract

The traditional approach to design flood estimation (for example, to derive the 100-year flood) is to apply a statistical model to time series of peak river flow measured by gauging stations. Such records are typically not very long, for example in the UK only about 10% of the stations have records that are more than 50 years in length. Along-explored way to augment the data available from a gauging station is to derive information about historical flood events and paleo-floods, which can be obtained from careful exploration of archives, old newspapers, flood marks or other signs of past flooding that are still discernible in the catchment, and the history of settlements. The inclusion of historical data in flood frequency estimation has been shown to substantially reduce the uncertainty around the estimated design events and is likely to provide insight into the rarest events which might have pre-dated the relatively short systematic records. Among other things, the FEH Local project funded by the Environment Agency aims to develop methods to easily incorporate historical information into the standard method of statistical flood frequency estimation in the UK. Different statistical estimation procedures are explored, namely maximum likelihood and partial probability weighted moments, and the strengths and weaknesses of each method are investigated. The project assesses the usefulness of historical data and aims to provide practitioners with useful guidelines to indicate in what circumstances the inclusion of historical data is likely to be beneficial in terms of reducing both the bias and the variability of the estimated flood frequency curves. The guidelines are based on the results of a large Monte Carlo simulation study, in which different estimation procedures and different data availability scenarios are studied. The study provides some indication of the situations under which different estimation procedures might give a better performance.

Highlights

(gauged or ungauged) the L-moments are estimated as the average values of the L-moments of a group of stations which are deemed to be similar to the site of interest
Results show that some improvement can be obtained for some return periods when the Partial Probability Weighted Moments (PPWM) approach is used in the presence of historical information
The data randomly generated from such a distribution can often have negative values, and seems to be not very representative of the peak flow values observed in the British catchments

Summary

Introduction

(gauged or ungauged) the L-moments are estimated as the average values of the L-moments of a group of stations which are deemed to be similar to the site of interest (the pooling group). Events for which this type of evidence is available rather than the evidence given by a human artefact are generally referred to as paleo-floods (see for example [4] for a review on the topic) This type of information can be used to inform the estimation of rare events, it is often the case that information on paleo-floods is much more uncertain than the typical information obtained from historical sources. Some practical guidelines on the identification and evaluation of historical events are given in [5], where different methods to use available historical information for design event estimation are presented. There is not at present an easy way to formally include historical data in an analysis which uses the standard FEH statistical method, but [5] give some advice on how to evaluate the validity of an estimated flood frequency curve in light of the presence of historical floods in the area under study. The s=40 years long systematic record of gauged peak flows is shown (black bars)

Objectives

Methods

Results

Conclusion