Nonstationary Compound Weather Extremes in Canada based on Large Ensemble Climate Simulations

Abstract

<p>Compound weather extremes including warm-wet and warm-dry events can lead to catastrophes such as wildfires, droughts and flooding. We use three large ensembles (3 × 50 members) of climate simulations to study the non-stationarity of compound events based on an ensemble pooling approach: the Canadian Regional Climate Model Large Ensemble (CanRCM4-LE), and the Canadian Large Ensembles Adjusted Datasets (CanLEAD1&2). The CanLEAD products include daily precipitation, maximum and minimum temperature from CanRCM4-LE that are bias-corrected using a novel statistical approach, which preserves the multivariate structure of the climate variables and corrects for univariate biases. Each ensemble member is validated against the NRCANmet observed data over Canada for 1951-2000 using a hierarchical Bayesian framework. Additionally, the performance of the models to mimic the dependence structure of the observation is tested using copulas. Extreme climate indices are estimated for a baseline period and changes in extremes are explored across four future warming scenarios corresponding to +1.5°C, +2.0°C, +3.0°C and +4.0°C warming above the pre-industrial period of 1850-1900. The ensemble pooling approach allows for the quantification of changes in the dependence structure and its subsequent effects on compound extremes in the future. </p><p>Results show that the CanLEAD products can reduce warm and wet biases in CanRCM4-LE over the majority of Canadian regions in all seasons except for winter. The ensembles unanimously project significant warming and wetting trends over most of southern Canada excluding the Canadian Prairies in summer, which show a drying trend towards the end of the 21<sup>st</sup> century. The overall trend shows an increase in hot extremes in central and southeastern Canada and a significant increase in wet extremes in western coastal regions. Results from compound extreme analysis show that there is significant under-estimation of extremes when the dependence between temperature and precipitation is ignored. For example, a 100-year hot and dry event under the assumption of independence becomes a ~60-year event when the dependence is characterized using copulas.</p>