Are 100 ensemble members enough to capture the remote atmospheric response to 12°C arctic sea ice loss?

• Main Authors: Peings, Y, Labe, ZM, Magnusdottir, G
• Format: Article in Journal/Newspaper
• Language: unknown
• Published: eScholarship, University of California 2021
• Subjects: Arctic; Sea ice; Atmosphere-ocean interaction; Teleconnections; Numerical analysis; modeling; Climate variability; Atmospheric Sciences; Oceanography; Geomatic Engineering; Meteorology & Atmospheric Sciences; Pacific; Global warming; North Atlantic
• Online Access: https://escholarship.org/uc/item/5r49771s

This study presents results from the Polar Amplification Multimodel Intercomparison Project (PAMIP) single-year time-slice experiments that aim to isolate the atmospheric response to Arctic sea ice loss at global warming levels of 128C. Using two general circulation models (GCMs), the ensemble size is increased up to 300 ensemble members, beyond the recommended 100 members. After partitioning the response in groups of 100 ensemble members, the reproducibility of the results is evaluated, with a focus on the response of the midlatitude jet streams in the North Atlantic and North Pacific. Both atmosphere-only and coupled ocean–atmosphere PAMIP experiments are analyzed. Substantial differences in the midlatitude response are found among the different experiment subsets, suggesting that 100-member ensembles are still significantly influenced by internal variability, which can mislead conclusions. Despite an overall stronger response, the coupled ocean–atmosphere runs exhibit greater spread due to additional ENSO-related internal variability when the ocean is interactive. The lack of consistency in the response is true for anomalies that are statistically significant according to Student’s t and false discovery rate tests. This is problematic for the multimodel assessment of the response, as some of the spread may be attributed to different model sensitivities whereas it is due to internal variability. We propose a method to overcome this consistency issue that allows for more robust conclusions when only 100 ensemble members are used.