Multivariate Extreme Value theory in Flood Mapping
Caroline Keef, David Kearney, Philip Emonson, Duncan S. Faulkner
JBA Consulting

Mapping the extent of large floods is undertaken for a number of purposes these include; deciding where to locate new houses; deciding where to locate new flood defences, and how large these should be; meeting the EU floods directive; and calculating expected insured losses. The standard tools used in all these situations are hydraulic modelling engines. Although the different engines vary in detail and complexity all take a grid of elevations and extreme river flows as inputs. The extent covered by a single hydraulic model can vary from a few hundred metres along a single stretch, to many kilometres covering a whole river catchment including tributaries.

In models that include tributaries the flows on each of them must be included separately as multiple inflows. To do this accurately it is necessary to assess which extreme river flows are likely to occur at the same time. Methods for estimating extreme river flows at a single location are well established. However statistical methods for simultaneous extreme values are less common. In this paper we present a method that uses extreme value theory to answer the practical question of how to set multiple inflows.

Keywords: Multivariate extremes; Flood risk

Biography: Caroline is an applied statistician working in the area of flood risk. Following on from her PhD in spatial dependence of river floods and extreme rainfall, she has developed and applied methods to help solve various applied problems. In addition to the topic she will present here these include estimating the probability of widespread flood events, and methods to accurately assess the probability of flooding above a certain depth.