Bayesian probabilistic projections of future climate over Canada based on the RCM ensemble

Abstract

In this research, a series of approaches are proposed to address the challenges in generating robust probabilistic projections of climatic variables (e.g., temperature, precipitation, and Intensity-Duration-Frequency curves) and analyzing the associated uncertainties based on the Regional Climate Model (RCM) ensemble. The proposed approaches have been applied to Canada for demonstrating their effectiveness. Specifically, a new discriminant-Bayesian Model Averaging (BMA) ensemble climate modeling (DBEC) approach is developed to help address the neglected spatiotemporal variations of model biases. Through the proposed method, the BMA weights are used as prior distributions to drive the Bayesian discriminant analysis in order to generate refined weights for individual ensemble models according to their spatiallyand temporally-clustered performance. The results suggest that the DBEC approach can improve both the accuracy and reliability of ensemble projections to some extent, especially in winter and Arctic regions. The probabilistic projections of temperature for three future periods under two emission scenarios are then obtained through the proposed DBEC model. The results indicate that comparatively larger temperature increases can be observed in Arctic regions. In addition, the magnitude of uncertainties is found to be negatively correlated to the elevation. Then a new multi-dimensional discriminant-BMA ensemble approach (MDBE) is developed to quantitively characterize the relationships between the modeling performances and climatic conditions. Through the comparative assessments of the proposed approach against three other ensemble methods, its effectiveness in generating the probabilistic projections of annual and seasonal precipitation over Canada has been illustrated. In detail, the R2 and percentage coverage will increase up to 0.15 (from 0.52 to 0.67) and 20% (from 60% ~ 80%), respectively. The generated projections suggest that significant precipitation increases are observed in future periods, especially in the Arctic regions. The warming climate could be the primary reason for such increases. Moreover, the intensified atmospheric radiative cooling is also a possible explanation for the winter precipitation increase. Finally, a new CDF-distance-based method is proposed to generate ensemble projections of IDF curves over Canada. Compared with the traditional ensemble methods, the proposed CDF-distance-based ensemble approach depends less on the simulated accuracy of annual maximum precipitation time series. Consequently, it can improve both the accuracy and reliability of the probabilistic projections in IDF curves. The proposed method has been applied to Canada for assessing the future changes of the IDF curves. The results suggest that the upward shifts of the IDF curves under all return periods are observed under changing climate conditions. Moreover, the percentage changes of precipitation intensities increase with return periods.