Multivariate non-stationarity for floods in Canada, testing and modeling

Código

I-EBHE0067

Tema

WG 1.10: Hydrologic Design - Solutions & Communication

Resumo

Floods are the most common natural hazard in Canada, posing significant risks to both human lives and the environment. However, with global climate change and human activities, stationary approaches fail to capture the non-stationary characteristics of flood events. Moreover, these events are multivariate, described mainly through their peak and volume as dependent variables. This study aims to characterize the preferred non-stationary models for flood peak and volume series across Canada for the period 1970-2019, using data from a network of 169 watersheds. Various model candidates (univariate and copula) are considered, with parameters potentially dependent on covariates in linear or quadratic forms. The selection procedure for the "best" models for marginal distributions and copulas relies, as a decision process, on an extensive set of steps and statistical tools (graphics, criteria, univariate and multivariate trend tests). These steps and tools do not always provide results in agreement with each other, which make the decision more complex. The large number of the considered stations allows to be confronted to a variety of real-world situations.