A regional Bayesian POT model for flood frequency analysis

TL;DR

This paper introduces a regional Bayesian Peak Over Threshold (POT) model for flood frequency analysis, improving estimates for short data series by leveraging regional information and Bayesian methods.

Contribution

It proposes a less restrictive regional Bayesian model that enhances flood quantile estimation for short records, outperforming traditional index flood and local models.

Findings

Bayesian model outperforms traditional methods in accuracy.

Larger, more homogeneous regions yield better estimates.

Model performs well across different degrees of regional homogeneity.

Abstract

Flood frequency analysis is usually based on the fitting of an extreme value distribution to the local streamflow series. However, when the local data series is short, frequency analysis results become unreliable. Regional frequency analysis is a convenient way to reduce the estimation uncertainty. In this work, we propose a regional Bayesian model for short record length sites. This model is less restrictive than the index flood model while preserving the formalism of "homogeneous regions". The performance of the proposed model is assessed on a set of gauging stations in France. The accuracy of quantile estimates as a function of the degree of homogeneity of the pooling group is also analysed. The results indicate that the regional Bayesian model outperforms the index flood model and local estimators. Furthermore, it seems that working with relatively large and homogeneous regions may lead to more accurate results than working with smaller and highly homogeneous regions.