Fast and direct nonparametric procedures in the L-moment homogeneity test

TL;DR

This paper introduces fast, nonparametric methods for the L-moment homogeneity test in regional frequency analysis, improving power and simplicity over traditional parametric approaches, with validation through simulations and a real case study.

Contribution

It develops nonparametric procedures integrated into the L-moment homogeneity test, addressing limitations of the Hosking-Wallis test regarding distribution assumptions and rejection thresholds.

Findings

Permutation and bootstrap methods outperform the HW test in power.

Nonparametric tests are faster and simpler to implement.

Real case study confirms nonparametric tests' effectiveness.

Abstract

Regional frequency analysis is an important tool to properly estimate hydrological characteristics at ungauged or partially gauged sites in order to prevent hydrological disasters. The delineation of homogeneous groups of sites is an important first step in order to transfer information and obtain accurate quantile estimates at the target site. The Hosking-Wallis homogeneity test is usually used to test the homogeneity of the selected sites. Despite its usefulness and good power, it presents some drawbacks including the subjective choice of a parametric distribution for the data and a poorly justified rejection threshold. The present paper addresses these drawbacks by integrating nonparametric procedures in the L-moment homogeneity test. To assess the rejection threshold, three resampling methods (permutation, bootstrap and P\'olya resampling) are considered. Results indicate that permutation and bootstrap methods perform better than the parametric Hosking-Wallis test in terms of power as well as in time and procedure simplicity. A real-world case study shows that the nonparametric tests agree with the HW test concerning the homogeneity of the volume and the bivariate case while they disagree for the peak case, but that the assumptions of the HW test are not well respected.