Thursday 12 May 2016: Statistical Science Seminar: Regional flood frequency analysis for Norway
Thirds Thorarinsdottir -
In Norway, hydropower comprises 95% of the total electricity generation
with over 2300 classified dams connected to the power
production. Extreme flooding and dam safety are thus important societal
issues. The aim of flood frequency analysis is to estimate streamflow
quantiles, or design floods, for given exceedance probabilities or
return periods, e.g. the 100-year flood. Design flood estimation forms
the basis for hazard management related to flood risk and is a legal
obligation when building infrastructure such as dams, bridges and roads
close to water bodies. The streamflow quantile estimates required for
these applications might range from an estimate of the index flood, or
the mean annual flood, to quantiles higher than the 1000-year flood.
Furthermore, applications commonly require design flood estimates at
locations for which only a short series of discharge data is available.
To this end, we propose a regional flood estimation approach based on
the general extreme value (GEV) distribution where the parameters of the
distribution depend on meteorological, hydrological and geographical
information for the catchment. Estimation in the model is performed via
Bayesian model averaging resulting in a mixture of models where the
different mixture components vary in the particular covariates used for
each model parameter. Additionally, we discuss approaches for verifying
frequency analysis models and pose some questions in this regard.