Multivariate Models to Estimate Hydrometeorological Extremes Risk

Extreme hydro-meteorological events are due to a combination of several interrelated occurrences. Their impacts generally have effects on a regional scale. Classes of random processes have been developed under certain assumptions of stationarity and spatial continuity to improve the representation of extremes. Conventional multivariate models allow, in addition to probabilistic representation, to incorporate the effects of covariates that may further explain the characteristics of the conditional distribution. In the present study, a spatial quantile regression model is proposed to estimate the quantile curve for a given probability of non-exceedance, as function of locations and covariates. Canonical vines copulas are considered to represent the spatial dependence structure. A case study illustrates the introduction of regional climate variability to estimate flood risk in Eastern Canada.