Attribution of extreme temperature changes during 1951–2010

• Published in: Climate Dynamics
• Main Authors: Kim, YH, Min, SK, Zhang, X, Zwiers, F, Alexander, LV, Donat, MG, Tung, YS
• Format: Article in Journal/Newspaper
• Language: unknown
• Published: Springer Nature 2016
• Subjects: 13 Climate Action; anzsrc-for: 0401 Atmospheric Sciences; anzsrc-for: 0405 Oceanography; anzsrc-for: 0406 Physical Geography and Environmental Geoscience; Arctic; Pacific
• Online Access: http://hdl.handle.net/1959.4/unsworks_42966; https://doi.org/10.1007/s00382-015-2674-2

An attribution analysis of extreme temperature changes is conducted using updated observations (HadEX2) and multi-model climate simulation (CMIP5) datasets for an extended period of 1951–2010. Compared to previous HadEX/CMIP3-based results, which identified human contributions to the observed warming of extreme temperatures on global and regional scales, the current results provide better agreement with observations, particularly for the intensification of warm extremes. Removing the influence of two major modes of natural internal variability (the Arctic Oscillation and Pacific Decadal Oscillation) from observations further improves attribution results, reducing the model-observation discrepancy in cold extremes. An optimal fingerprinting technique is used to compare observed changes in annual extreme temperature indices of coldest night and day (TNn, TXn) and warmest night and day (TNx, TXx) with multi-model simulated changes that were simulated under natural-plus-anthropogenic and natural-only (NAT) forcings. Extreme indices are standardized for better intercomparisons between datasets and locations prior to analysis and averaged over spatial domains from global to continental regions following a previous study. Results confirm previous HadEX/CMIP3-based results in which anthropogenic (ANT) signals are robustly detected in the increase in global mean and northern continental regional means of the four indices of extreme temperatures. The detected ANT signals are also clearly separable from the response to NAT forcing, and results are generally insensitive to the use of different model samples as well as different data availability.