Reassessing English flood frequency relationships in light of widespread new instrumental stage maxima

Abstract

Flood frequency estimation typically involves applying well-established statistical methods to instrumental peak discharge data. However, even notwithstanding any potential climate change trends, such analyses may have a tendency towards underestimation. The shortness of the available records is the proposed cause, whilst the prominence of flood frequency results in practical applications heightens the concern. Following the passage of the extra-tropical storms Desmond and Eva in December 2015, previous river stage maxima were widely surpassed across northern England. Indeed, preliminary analyses indicated that peak flows on the Rivers Eden, Tyne and Lune were higher than any previously recorded in England and Wales (CEH, 2016). Herein, the effects of including these exceptional observations on flood frequency estimates produced using established methods are investigated. Annual maxima series were extended at 155 stations, and models fitted on a single-site basis with and without the additional data were compared. Predictably, return period flow estimates generally increased with the additional data; the mean 1-in-100-year change (i.e. across all sites) was +7%. Spatial patterns of change correspond closely with the event footprint, whilst associations between change and both record length and catchment area were found to be only weak. The ‘enhanced single-site’ method was then applied at a subset of stations (without inclusion of the latest data). Interestingly, these estimates were not substantially higher that those produced from the same samples using the single-site method, implying a certain dependence between the pooling group stations. In a final set of analyses, the estimates were found, in many cases, to demonstrate less sensitivity to the choice of statistical distribution than to the sample or method used. Overall, these findings lend some support to the notion that standard methods may underestimate flood frequency. However, the difficulty of disproving probabilistic predictions reduces the confidence with which this assertion can be made. Going forward, reinvigorating efforts to incorporate longer-term hydrological data more routinely into hazard assessments may prove fruitful. It could also be appropriate for flood frequency estimates to be updated more frequently as new instrumental data become available.