Evaluation and joint projection of temperature and precipitation extremes across Canada based on hierarchical Bayesian modelling and large ensembles of regional climate simulations

Abstract

Individual and joint variations of extreme temperature and precipitation are assessed across Canada using the large ensemble of Canadian Regional Climate Model simulations (CanRCM4-LE) and two corresponding multivariate bias-corrected datasets (Canadian Large Ensembles Adjusted Datasets, CanLEAD-E & S). The overall performance of the three 50-member ensembles is evaluated against the NRCANmet gridded observation for 1951–2000. A hierarchical Bayesian framework is then applied to analyze the biases of each ensemble member, characterize the ensemble uncertainties associated with internal climate variability, and assess the trends and the joint distributions of temperature and precipitation. Further, projected changes of extreme climate indices are assessed across the three ensembles in the historical period and four future scenarios corresponding to +1.5 °C to +4.0 °C warming above the pre-industrial level of 1850–1900. Results show that the CanLEAD products have lower warm and wet biases compared to CanRCM4-LE over most regions and in all seasons except in winter. CanLEAD-S significantly reduces precipitation biases and represents the behaviour of extreme climate indices better than the other two ensembles. All ensembles consistently project strong warming and wetting trends over most parts of southern Canada excluding the Canadian Prairies in summer, which shows a drying trend towards the end of the 21st century. The ensembles show increases in hot extremes in central and southeastern Canada and increases in wet extremes in western coastal regions. The results of the study suggest that CanLEAD-E&S can be used as reliable products for climate change impact assessments at regional scales particularly for the analysis of nonstationary compound events.