CFE 2019: Start Registration
View Submission - CMStatistics
Title: Approximate Bayesian inference for spatial flood frequency analysis Authors:  Birgir Hrafnkelsson - University of Iceland (Iceland) [presenting]
Raphael Huser - King Abdullah University of Science and Technology (Saudi Arabia)
Arni Johannesson - University of Iceland (Iceland)
Haakon Bakka - King Abdullah University of Science and Technology (Saudi Arabia)
Stefan Siegert - University of Exeter (United Kingdom)
Abstract: Extreme floods cause casualties and damage to vital civil infrastructure. Predictions of extreme floods within gauged and ungauged catchments are crucial to mitigate these disasters. A latent Gaussian model is proposed for predicting extreme floods using the generalized extreme-value (GEV) distribution and a novel multivariate link function for its location, scale and shape parameters. This link function is designed to separate the interpretation of the parameters at the latent level and to avoid unreasonable estimates of the shape parameter. Structured additive regression models are proposed for the three parameters at the latent level. Each of these regression models contains fixed linear effects for catchment descriptors. Spatial model components are added to the two first latent regression models, to model the residual spatial structure unexplained by the catchment descriptors. To achieve computational efficiency for large datasets with these richly parametrized models, we exploit a Gaussian-based approximation to the posterior density. This approximation relies on site-wise estimates, but, contrary to typical plug-in approaches, the uncertainty in these initial estimates is properly propagated through to the final posterior computations. We applied the proposed modeling and inference framework to annual peak river flow data from 554 catchments across the United Kingdom. The framework performed well in terms of flood predictions for ungauged catchments.