Abstract:Coarse resolution global climate models (GCMs) have inherent difficulty simulating a reliable climate regime in coastal areas, as in northern Canada, where sea ice and snow cover are highly sensitive to fine-scale climate forcings. As a result, strong biases are present in GCM temperature regimes in this region, and the direct use of raw-GCM climate change signals at the local scale is problematic. However, fine resolution climate change information for use in impact studies can be obtained via statistical downscaling (SD) methods. This study investigates the regression-based SDSM model with respect to its potential to simulate reliable and plausible changes in mean values as well as probabilities of extreme temperatures, in some specific locations in northern Canada. Four sets of independent climate predictors, from the outputs of two GCMs (i.e. CGCM2 and HadCM3) and using two SRES emission scenarios (i.e. A2 and B2), are used by the SDSM model to construct climate scenario information for this region over the period 2070-2099. The results demonstrate that the SD model is able to capture the major part of the temperature change signal, with a plausible climatic regime for higher warming in winter than in summer and in A2 than in B2 runs. The combination of relevant atmospheric predictors in the SD process is able to take into account most key factors of the temperature change signal, with strong convergence in the magnitude and the timing of the changes in all results. The downscaling signals are more consensual and physically-plausible in comparison with the raw GCM anomalies, with relatively better skill using HadCM3 predictors than those from CGCM2. The study also confirms that scrupulous analysis of the climate change regime and its temporal and spatial distribution at the scale of interest is essential for it to be useful in impact studies.