A Decomposition of Feedback Contributions to the Arctic Surface Temperature Biases in the CMIP5 Climate Models

This study examines the systematic surface temperature biases over the Arctic Ocean in the eight Coupled Model Intercomparison Project 5 (CMIP5) models (BCC-CSM1-1, CESM1-CAM5-1-FV2, GFDL-CM3, GISS-E2-H, IPSL-CM5A-LR, MIROC-ESM-CHEM, MIP-EMS-LR, and NorESM1-M) and investigates the quantitative attribution of feedback processes that are responsible. In general, the cold biases are predominant over the entire Arctic Ocean in the CMIP5 models. These systematic temperature biases are decomposed into the biases by physical and dynamical processes, based upon the Climate Feedback-Responses Analysis Method (CFRAM). The multi-model ensemble mean shows that dynamical processes are more important in the Pacific sector of the Arctic, while physical ones are more important in the Atlantic sector, although this feature is not consistently found in all models. Dynamical processes involve surface heat fluxes, and frictional, turbulent, convective, horizontal heat transports, etc., which cannot be decomposed further by the method used. In contrast, the physical processes can be further decomposed into cloud, water vapor, and albedo feedbacks. According to our results, most of the physical processes are attributable to the albedo feedback in all models. Cold (warm) biases due to the albedo feedback almost correspond to the positive (negative) biases of sea ice concentration. Hence, our results suggest that for better simulation of the Arctic surface temperature, it tells us which part of the model processes we should efficiently focus on improving.