This talk has been automatically captioned. You can remove these by pressing CC on the video toolbar.

Name: Silius M. Vandeskog

Talk Title: Modelling extreme sub-daily precipitation with the blended generalised extreme value distribution

Abstract: Short-term extreme precipitation can cause flash floods, large economic losses and immense destruction of infrastructure. In order to prepare for future extreme precipitation, a Bayesian hierarchical model is applied for estimating return levels for the yearly maxima of sub-daily precipitation in Norway. 

A modified version of the generalised extreme value (GEV) distribution, called the blended GEV (bGEV) distribution, is used as the model for yearly maxima of sub-daily precipitation. Inference with the GEV distribution is known to be difficult, partly because its support depends on its parameters.

The bGEV distribution has the right tail of a Fréchet distribution and the left tail of a Gumbel distribution, resulting in a constant support that allows for more stable inference. Fast inference is performed using integrated nested Laplace approximations (INLA).

We propose a new model for block maxima that borrows strength from the peaks over threshold methodology by linking the scale parameter of a block maximum to the standard deviation of observations larger than some threshold. The new model is found to outperform the standardblock maxima model when fitted to the yearly maxima of sub-daily precipitation in Norway. Evaluation is performed with the threshold weighted continuous ranked probability score (twCRPS), where the weight function only focuses on large quantiles.

This talk is a contributed talk at EVA 2021.