Considering historical flood events in flood frequency analysis: Is it worth the effort?

Abstract

Information about historical floods can be useful in reducing uncertainties in flood frequency estimation. Since the start of the historical record is often defined by the first known flood, the length of the true historical period M remains unknown. We have expanded a previously published method of estimating M to the case of several known floods within the historical period. We performed a systematic evaluation of the usefulness of including historical flood events into flood frequency analysis for a wide range of return periods and studied bias as well as relative root mean square error (RRMSE). Since we used the generalized extreme value distribution (GEV) as parent distribution, we were able to investigate the impact of varying the skewness on RRMSE. We confirmed the usefulness of historical flood data regarding the reduction of RRMSE, however we found that this reduction is less pronounced the more positively skewed the parent distribution was. Including historical flood information had an ambiguous effect on bias: depending on length and number of known floods of the historical period, bias was reduced for large return periods, but increased for smaller ones. Finally, we customized the test inversion bootstrap for estimating confidence intervals to the case that historical flood events are taken into account into flood frequency analysis.