Abstract
A semiparametric copula joint framework was proposed to model wind gust speed (WGS) and maximum temperature (MT) in Canada, using Gaussian kernel density estimation (GKDE) with parametric copulas. Their joint probability estimates allow for a better understanding of the risk of power blackouts and the demand for air conditioning in the community. The bivariate framework used two extreme sample groups to define extreme pairs at different time lags, i.e., 0 to ± 3 days, annual maximum WGS (AMWGS) and corresponding MT and annual highest MT (AHMT) and corresponding WGS. A thorough model performance comparison indicated that GKDE outperformed the parametric models in defining the marginal distribution of selected univariate series. Significant positive correlations were observed among extreme pairs, except for Calgary and Halifax stations, with inconsistent correlation variations based on selected cities and lag time. Various parametric 2-D copulas were selected to model the dependence structure of bivariate pairs at different time lags for selected stations. AMWGS or AHMT events, when considered independently, would be stressful for all stations due to high estimated quantiles with low univariate RPs. The bivariate events exhibited lower AND-joint RPs with moderate to high design quantiles, indicating a higher risk of power blackouts and heightened air-conditioning demands, which varied inconsistently with time lags across the station. The bivariate AMWGS and corresponding MT events would be stressful in Regina, Quebec City, Ottawa, and Edmonton, while AHMT and corresponding WGS events in Toronto, Regina, and Montreal. Conversely, Vancouver poses a lower risk of joint action of pairs AHMT and corresponding WGS events. These hazard statistics can help in better planning for community well-being during extreme weather.