U.S. Dept. of Commerce / NOAA / OAR / PMEL / Publications

A sea ice free summer Arctic within 30 years?

M. Wang1 and J. E. Overland2

1Joint Institute for the Study of the Atmosphere and Ocean (JISAO), University of Washington, Seattle, Washington USA

2Pacific Marine Environmental Laboratory, National Oceanographic and Atmospheric Administration, Seattle, Washington, USA

Geophys. Res. Lett., 36, L07502, doi:10.1029/2009GL037820
Copyright 2009 by the American Geophysical Union.
Further electronic distribution is not allowed.

2. Climate Model Selection and Projections

Many projections from CMIP3 models show an increased rate of September sea ice reduction when the sea ice extent is near the present 4.6 M km2 mark, compared to the rate of sea ice reduction found before 2000 [Holland et al., 2006; Stroeve et al., 2007]; also see the future sea ice losses projected by the CCSM3, CNRM and MIROC(med) models in Figure 1. This change in rate of sea ice reduction supports our further investigation of the sea ice projections from CMIP3 models to estimate the potential timing of a summer sea ice free Arctic, given the conditional state that an observed September sea ice extent of 4.6 M km2 has already been reached.

figure 1

Fig. 1. September sea ice extent as projected by the six models that simulated the mean minimum and seasonality with less than 20% error of the observations. The colored thin line represents each ensemble member from the same model under A1B (blue solid) and A2 (magenta dashed) emission scenarios, and the thick red line is based on HadISST analysis. Grey lines in each panel indicate the time series from the control runs (without anthropogenic forcing) of the same model in any given 150 year period. The horizontal black line shows the ice extent at 4.6 M km2 value, which is the minimum sea ice extent reached in September 2007 according to HadISST analysis. All six models show rapid decline in the ice extent and reach ice-free summer (<1.0 M km2) before the end of 21st century.

Confidence that climate models provide credible quantitative projections of future climate is build upon their demonstrated ability to reproduce observed features of recent climate [IPCC, 2007; Gerdes and Köberle, 2007]. It is therefore important to apply an observational constraint on the CMIP3 models, and eliminate "outlier" models from further consideration. Inspired by Knutti et al. [2006], we require that models simulate the seasonal cycle and the mean of September sea ice extent to within ±20% of HadISST analysis for the period of 1980–1999 (Figure S1 of the auxiliary material).1 Reproducing the correct magnitude of the seasonal cycle of sea ice extent is one way of demonstrating the models sensitivity to changes in external forcing, e.g., solar insolation. The September mean sea ice extent is an efficient constraint to eliminate models with systematic biases. Our constraints are based on comparisons to the HadISST sea ice concentration analysis, which was made more homogeneous by compensating satellite microwavebased sea ice concentrations for the impact of surface melt effects on retrievals in the Arctic [Rayner et al., 2003]. The combination of the seasonal cycle and mean conditions is an improved constraint relative to previous studies [Stroeve et al., 2007; Overland and Wang, 2007]. The selection process not only reduces the range of uncertainty in the future projections by these models, but also shows that models with reasonable seasonal cycle relative to observations project a faster future decline of September sea ice extent (Figure S2).

Applying the observational constraints results in the retention of 6 of 23 CMIP3 models; their projected September sea ice extents under the IPCC A1B and A2 emission scenarios (colored thin lines in each plot) are shown in Figure 1. Using both scenarios provides at least two ensemble members per model to account for the influence of natural variability. The justification for combining the two emissions scenarios is that the difference in CO2 concentrations before 2050 between the A1B and A2 emissions scenarios is small. The A1B scenario actually has a slightly faster CO2 emission growth rate during this early period [IPCC, 2001, Figure 5]. Although an evaluation of why some models perform better than others is difficult [Gleckler et al., 2008], we do note that among the six selected models, three (CCSM3, CNRM-CM3 and UKMO-HadGEM1) include a multiple sea ice thickness distribution as part of a sophisticated sea ice physics and dynamics package, a feature only present in five of the current generation of the CMIP3 models [Zhang and Walsh, 2006].

All ensemble members from the six models show sea ice extent reaching current value (4.6 M km2) sometime in the 21st century with rapid declines afterward, similar to the observed time series (thick red line). Contrasting these sea ice extent projections with those from control runs (i.e., without anthropogenic forcing, grey lines) clearly shows that a necessary factor for the future decline of summer sea ice extent is the presence of external anthropogenic forcing. The different timing for reduction of sea ice extent in different model projections is a consequence of natural variability in the simulated climate and the differences and limitations of current sea ice models. There is little difference in the projected trajectories of sea ice extent between the A1B (blue solid line) and A2 (magenta dashed line) emission scenarios, especially before 2050.

In Figure 2 we show the time interval for sea ice extent to be reduced from 4.6 to 1.0 M km2 for all ensemble members of the six models under both A1B and A2 emission scenarios. The 1.0 M km2 limit is chosen because models suggest that the regions north of Greenland/Canada will retain some sea ice in the future even though the Arctic can be considered as "nearly sea ice free" at the end of summer. The median duration interval for the sea ice to reduce from 4.6 to 1.0 M km2 is 30 years with quartiles at 21 and 41 years. The overall mean interval is 32 years. This provides an expected value (based on the median) for a September nearly sea ice free Arctic in the year 2037. The first quartile of the distribution for the timing of a September sea ice loss will be reached in 2028. The uncertainty in future timing for a September sea ice free Arctic is strongly influenced by the chaotic nature of natural variability. This uncertainty in timing is reflected in the wide range of time intervals for sea ice loss from the set of ten CCSM3 model ensemble members, which span from 15 to 42 years.

figure 2

Fig. 2. Estimated number of years for sea ice extent to drop from the current value (4.6 M km2) to less than 1.0 M km2 (summer ice free Arctic) based on six models under IPCC emission scenarios A1B (grey) and A2 (white). Each bar represents one ensemble member from each model. The model name is listed on the far-left bar if multiple ensemble members are provided. The far right bars show the model mean (black) and median (hatched) based on all ensemble members combined from both emissions scenarios.

Our expected time frame of ∼30 years to reach a September sea ice free Arctic is based on current conditions in the Arctic and information from the currently available set of fully coupled CMIP3 atmosphere-ocean-ice General Circulation Models (GCMs). Confidence in the reduced set of six models comes from their basis in established physical laws and their ability to simulate 20th century seasonality in Arctic sea ice extent. While our approach is a less than an ideal assumption compared to re-running the CMIP3 models with new initial conditions, we can justify transposing the model sea ice declines to earlier years as follows: the present day September Arctic sea ice cover has already decreased to 4.6 M km2, and all the six models show rapid sea ice declines after this sea ice extent threshold is reached. The basic physical processes of September sea ice reduction would also apply to re-running GCM models with new initial conditions: once there is considerable open water in the summer central Arctic, albedo feedback and ocean-atmosphere heatflux feedback are major and rapid contributors to continued sea ice reductions.

In addition to the reduction in sea ice extent, sea ice thickness will also decrease as more areas are replaced by first year ice. Figure 3 displays the spatial ice thickness fields in the Arctic averaged over the six models. In the year when the six models have their September sea ice extent reach 4.6 M km2, much of the central Arctic is covered by sea ice less than 2.5 m in March (Figure 3a). By September much of the remaining sea ice is less than 1.2 meters thick in the central Arctic (Figure 3b). At the time of a nearly sea ice-free Arctic (1.0 M km2 in September) about 30 years later, March sea ice is thinner, with much of the area being covered by sea ice less than 2.0 m (Figure 3c). The distribution of remaining September sea ice for our "nearly sea ice free" definition of 1.0 M km2 is shown in Figure 3d; the region north of the Canadian Archipelago and Greenland remains a sea ice refuge.

figure 3

Fig. 3. Mean sea ice thickness for (left) March and (right) September based on ensemble members from six models under A1B emissions scenario. (a and b) Year when the September ice extent reached 4.6 M km2 by these models and (c and d) year when the Arctic reached nearly sea ice conditions (less than 1.0 M km2) in September, i.e., ice free summer.

1 - Auxiliary materials are available in the HTML. doi:10.1029/2009GL037820.

Return to Previous section

Go to Next section

PMEL Outstanding Papers

PMEL Publications Search

PMEL Homepage