Arctic sea ice: use of observational data and model hindcasts to refine future projections of ice extent

Abstract

This manuscript presents an evaluation of global climate models to guide future projections of Arctic sea ice extent (SIE). Thirty-five model simulations from Coupled Model Intercomparison Project, Phase 5 were examined to select model subsets using comparison to observational data (1979–2013). The study extends previous work by highlighting the seasonality of sea ice trends, utilizing a multi-step selection process to demonstrate how the timing of an ice-free Arctic varies with the hindcast performance of the models, and extending the analysis to include sudden ice loss events (SILE). Although the models' trends for the historical period are generally smaller than observed, the models' projected trends show a similar seasonality, largest in September and smallest in March to April. A multi-step evaluation process is applied to obtain progressively smaller subsets of the best-performing models. As the number of models retained becomes smaller, the simulated historical trend becomes larger and the median date of a projected ice-free Arctic becomes earlier. An examination of SILE through the historical period and model projections from 2014 through 2099 shows that SILE can account for between half and all of the future net loss of SIE. We created an application for exploring sea ice data: http://spark.rstudio.com/uafsnap/sea_ice_coverage/.

Acknowledgments

We thank NSIDC for providing Arctic sea ice data. This work was supported by the NOAA Climate Program Office through Grant NA110AR4310172, by NSF through award number 1023131, and by the Alaska Climate Science Center through Cooperative Agreement Number G10AC00588 from the United States Geological Survey. The contents are solely the responsibility of the authors and do not necessarily represent the official views of NOAA or the USGS. All statistical analyses were performed using the R language and environment for statistical computing and graphics. For more information, see http://www.r-project.org/.