Liu, Y. and Konzen, E. and Tozer, N.P. (2026) Statistical methods for risk-based coastal flood risk analysis. In: EVAN 2026 (7th International Conference on Advances in Extreme Value Analysis and Application to Natural Hazards), 21-23 July 2026, Delft University of Technology. (Submitted)
Full text not available from this repository.Abstract
During the winter of 2013/2014, the coast of England experienced an exceptional sequence of severe storms that caused widespread damage to coastal defences. In response, the Environment Agency commissioned the State of the Nation project in 2015 to update the National Flood Risk Assessment (NaFRA). This programme led to the development of a risk based statistical framework for national scale coastal flood assessment founded on Multivariate Extreme Value Analysis (MEVA). The original State of the Nation approach combined MEVA of key storm drivers, including wind, offshore waves and water levels, with a linked modelling chain incorporating numerical wave modelling and empirical overtopping relationships. This paper reframes and extends that approach through the development of more advanced statistical methods within a MEVA setting, with the aim of incorporating previously unaccounted for temporal dependencies. In particular, the framework represents short term dependence through full storm profiles and intermediate term dependence through storm clustering, enabling a more realistic representation of compound and successive storm effects in coastal flood risk estimation. Storm clustering is examined through a dedicated statistical analysis. Exploratory analysis of long term meteorological and oceanographic records from the east coast of the UK reveals clear temporal dependence in storm occurrence. Storm arrivals are modelled using a flexible non homogeneous Poisson process with time varying intensity, coupled with MEVA based storm characteristics. Model behaviour is evaluated using an extremal dependence diagnostic based on a temporal application of the Heffernan–Tawn model. To enable large scale inference, Gaussian Process Emulators (GPEs) are developed for both SWAN2D wave transformation and wave overtopping processes. These emulators act as efficient statistical surrogates for computationally intensive numerical models, while allowing uncertainty to be quantified and propagated through the risk assessment. Overall, the proposed framework integrates multivariate, non stationary and cluster aware extreme value methods with statistical emulation to support risk based coastal flood analysis. The approach provides an improved representation of temporal dependence, uncertainty and design relevant extremes, and supports more robust coastal flood risk assessment and adaptation planning.
| Item Type: | Conference or Workshop Item (Paper) |
|---|---|
| Subjects: | Coasts > General |
| Divisions: | Coastal |
| Depositing User: | Helen Stevenson |
| Date Deposited: | 11 May 2026 10:44 |
| Last Modified: | 11 May 2026 10:44 |
| URI: | http://eprints.hrwallingford.com/id/eprint/1733 |
Actions (for site administrators only - login required)
 |
View Item |
