Refinement of future Arctic sea-ice projections


 Arctic sea ice has been retreating at unprecedented pace over the past decades. Model projections show that the Arctic Ocean could be almost ice free in summer by the middle of this century. However, the uncertainties related to these projections are relatively large. Here we use 33 global climate models from the Coupled Model Intercomparison Project 6 (CMIP6) in order to reduce these uncertainties. We select the models that best capture the observed Arctic sea-ice area and volume and northward ocean heat transport to refine model projections of Arctic sea ice. This model selection leads to smaller Arctic sea-ice area and volume relative to the multi-model mean without model selection and summer ice-free conditions could occur as early as around 2035. These results highlight a potential underestimation of the future Arctic sea-ice loss when including all models.


17
The retreat of Arctic sea ice is one of the most striking consequences of global warming and has 18 strong implications for local and remote climate, biosphere and society 1 . The total area of the 19 Arctic Ocean covered by sea ice, the Arctic sea-ice area, has decreased by about 2 million km 2 20 (yearly average) in the past 40 years of satellite observations, with more pronounced loss in the 21 summer 1-3 . As sea ice has also thinned by 1.5 -2 m in the Central Arctic since 1980 4, 5 , the 22 total Arctic sea-ice volume has substantially decreased at a rate of 3800 km 3 per decade between 23 1979 and 2010 6 . The current Arctic sea-ice losses are strongly connected to the rising global 24 temperatures 7-9 , and thus to cumulative greenhouse gas emissions into the atmosphere 3, 10 . Thus, 25 the observed sensitivity of sea-ice changes to cumulative greenhouse gas emissions has been used 26 to provide an estimate of the future Arctic sea-ice area 10 . 27 However, this simple linear extrapolation strongly neglects non-linearities in the climate sys-28 tem and ocean-ice-atmosphere interactions and feedbacks 11,12 , resulting in strong short-and long-29 term deviations from the ongoing negative trend in sea-ice area and volume 13 . In order to include 30 these non-linearities and interactions, climate models can be used to provide more reliable projec-31 tions of the fate of Arctic sea ice 14,15 . In particular, global climate models coupling the atmosphere, 32 ocean and sea ice are well suited to make such projections [16][17][18] . The inclusion of these models in the 33 different Coupled Model Intercomparison Project (CMIP) phases 19-21 allows to provide estimates 34 of Arctic sea-ice area and volume projections in the next decades to centuries. The latest CMIP6 35 modelling effort 21 will feed into the next Intergovernmental Panel on Climate Change (IPCC) As-36 2 sessment Report 6 and includes climate model projections that follow different greenhouse gas 37 emission scenarios using the Shared Socioeconomic Pathways (SSPs) 22 . 38 In our study, we use CMIP6 model outputs with the aim to reduce uncertainties in the future 39 projections of Arctic sea ice. We select the models that best represent the present Arctic sea ice 40 and northward ocean heat transport, as the latter is a major driver of the recent sea-ice loss, and we 41 compare this model selection to the case without selection. We find that the sea-ice loss over this 42 century is stronger using different model selection criteria compared to the average over all models 43 without model selection. In particular, we find that summer ice-free Arctic conditions could occur 44 as early as 2035 in the selection case, compared to 2061 in the no-selection case. We also find that 45 some individual models strongly diverge from the multi-model mean and are associated with an 46 outdated sea-ice model.

47
Results and discussion 48 Projections without model selection. In our study, we focus on both the high-emission SSP5-49 8.5 and low-emission SSP1-2.6 scenarios, which correspond to a global warming of around 4 • C 50 and 1 • C, respectively, over this century (2081-2100 relative to 1995-2014) 23  March Arctic sea-ice area and volume are reduced by 45 % and 78 %, respectively, in 2096-2100, 53 compared to 2015-2019, in the high-emission scenario ( Fig. 1a and Supplementary Fig. 3a). In 54 September, the Arctic sea-ice area and volume are decreased by 90 % and 98 %, respectively, at 55 3 the end of the century ( Fig. 1b and Supplementary Fig. 3b). The Arctic Ocean becomes almost 56 ice free (sea-ice area lower than 1 million km 2 12 ) in September in 2061 for the multi-model mean 57 (Fig. 1b). These Arctic sea-ice area and volume changes are considerably slowed down in the 58 low-emission scenario: the multi-model mean March sea-ice area and volume are reduced by only 59 8 % and 28 %, respectively, at the end of the century, while the September sea-ice area is decreased 60 by 49 % and thus never reaches almost ice-free conditions during this century, and the September  sumes that all models are equally plausible and that the range of their projections is representative 72 of the uncertainty 25 . As some models better represent a specific aspect of the observed climate, e.g.

73
Arctic sea ice in our case, we can argue that these models will provide more accurate projections 74 of this specific aspect. A good agreement with observations does not constitute a final evidence 75 that the models are correct, but a bad agreement with such observations clearly indicates some 76 problems of the models 25 . Different approaches have been taken to try to reduce the model spread 77 in projections of Arctic sea-ice area for a given emission scenario. One such approach consists in 78 giving a weight to each model based on its performance relative to observations during the his-79 torical period: models that strongly agree with observations receive more weight than models that 80 poorly agree 23, 25 . Another approach is to select models based on their historical performance and 81 exclude models that do not satisfy the selection criteria 16, 18, 26 .

82
In our study, we adopt the latter approach, i.e. model selection, as it allows to exclude model  When applying our selection criteria, we find that the Arctic sea-ice area and volume gen- . This is mainly due to stronger 95 5 reductions in sea-ice area and volume over the twenty-first century in the selected models, and 96 also to a smaller initial present-day Arctic sea-ice area to a lesser extent. The stronger reduc-97 tions in sea-ice area and volume over the twenty-first century probably stem from the fact that the 98 selected models have a larger sensitivity to anthropogenic global warming than the non-selected 99 models 18 . Also, the smaller present-day sea-ice area in the selected models is due to the fact that 100 the multi-model mean without selection overestimates the observed sea-ice area (Fig. 3a,b); thus, 101 the selection of models closer to observations allows to reduce this overestimation, explaining the 102 smaller present-day sea-ice area.

103
The loss in sea-ice area and volume over this century is most pronounced when selecting the 104 models that best represent the historical Atlantic and Pacific ocean heat transports, in combination ice area and volume reach less than 7 million km 2 and less than 5,000 km 3 , respectively, by the 108 end of the twenty-first century, and the September sea ice totally disappears (Fig. 3). Selecting 109 the models that best represent the observed mean sea-ice area and volume and trend in sea-ice 110 area also provides a stronger reduction in the future Arctic sea-ice area and volume compared to 111 no selection, especially in September, but the sea-ice loss is less strong than with the ocean heat ice area and volume variability, this is partly linked to the fact that these quantities are directly 117 related to atmospheric variability 9 . In turn, the latter does not highly depend on the total amount 118 of ice. Thus, even a model with too much (or not enough) sea ice can have a realistic atmospheric 119 variability, leading to a realistic sea-ice variability. 120 An additional model selection criterion that we include in our analysis is the minimum num-121 ber of ensemble members. We select all models that have at least five members, as this allows 122 to both keep the models that take into account the uncertainty linked to internal variability and to 123 have about a third of the total number of models. We find that the multi-model mean averaged over 124 these models also leads to a stronger sea-ice loss relative to no selection, with no remaining sea ice 125 in September by the end of the century and reductions of 60% and 87% in March sea-ice area and 126 volume, respectively, in the high-emission scenario (Figs. 2-3). This strengthens our main finding 127 that the reduction in sea ice is stronger with model selection.

128
Our model selection based on the historical performance allows to exclude outliers that have 129 either too much or not enough Arctic sea ice. For the winter months, outliers are mainly located on 130 the high end as most models overestimate the observed sea-ice area (Fig. 3a), while for the summer 131 months, outliers are located on either end (Fig. 3b) 18 . Thus, our model selection allows to narrow in September for at least some years before the end of this century, but with a sea-ice area staying 140 close to the 1 million km 2 threshold until the end of the century (Supplementary Fig. 2b).  As the NEMO and MOM components are both shared by more than five different CMIP6 150 models, we further investigate the individual models using these two ocean components. This 151 reveals that the multi-model mean sea-ice area associated with these two ocean components is 152 clearly driven by specific outliers. The below-average March sea-ice area from NEMO in the 153 high-emission scenario is driven by two CMIP6 models that have a very low sea-ice area over the 154 whole twenty-first century (Fig. 4b). It is worth noting that these two models use version 2 of the 155 8 Louvain-la-Neuve sea Ice Model (LIM2), which is a former version of the LIM3 sea-ice model 32 .

156
In particular, LIM2 only includes one sea-ice thickness category, while LIM3 has five thickness 157 categories, making it more reliable in terms of sea-ice area variability compared to observations 33 .

158
Four other models using NEMO have a strong reduction in March sea-ice area in the high-emission 159 scenario at the end of the twenty-first century (Fig. 4b). 160 The above-average March sea-ice area from MOM in the high-emission scenario is driven 161 by one specific CMIP6 sea-ice model that has a sea-ice area about 4 million km 2 larger than the 162 MOM multi-model mean (Fig. 4b); this is also the case in the low-emission scenario (Supple-