| Abstract | Non-stationary flood frequency analysis (FFA) has gained much momentum in recent years. However, most of the applications have considered time as the sole covariate, which may not be fully adept given the varied evolution of flood predictors. Incorporating physical covariates representing dominant flood generating mechanisms can enable more precise flood risk assessments. This study applies non-stationary FFA with flood-relevant physical covariates to assess dam vulnerability across Canada, using transient climate change simulations from a regional climate model, one-way coupled to a routing model, for the current 1991–2020 and future 2070–2099 periods. From a set of 10 non-stationary models, considering maximum snow water equivalent, average annual temperature, spring-summer rainfall, fall-winter rainfall, and their combinations as plausible physical covariates, the preferred model for each grid cell is selected through a corrected form of the Akaike Information Criterion. The spatial patterns of physical covariates of the preferred models are, in general, found consistent with the flood generating mechanisms. Little support is found for the traditional stationary FFA approach compared to the non-stationary FFA with physical covariates. Application of the preferred model for the RCP8.5 scenario suggest that 67.5 % of dams, from a pre-identified set of medium and large dams, are located in vulnerable regions, far exceeding the stationary model estimates. The study provides useful insights, both for the development of non-stationary FFA models considering physical covariates and their application in the vulnerability assessment of critical water infrastructure in Canada, thereby contributing to the development of climate change adaptation strategies. |
|---|
