Essay: Predicting a date for an ice free-Arctic Ocean

The Working Group’s Fifth Assessment Report (AR5) concluded that, for the last two decades, Arctic sea ice has continued to fall in both volume and extent. Annual mean Arctic sea ice extent very likely decreased by 3.5%-4.1% per decade from 1979 to 2012, and there is medium confidence this is unprecedented over the last 1,450 years. Figure SPM.3 (b) shows a fall in Arctic summer sea ice extent from approximately 10.5 million km2 in 1900 to 6 million km2 by 2010. More models than at the time of AR4 reproduce the observed downward trend in Arctic ice extent, if with a large inter-model spread. Anthropogenic influences have very likely contributed to sea ice loss since 1979. Year round reductions of up to 94% for RCP8.5 are projected by the end of the 21st century with a medium confidence – leading to a likely (medium confidence) nearly ice-free Arctic Ocean in September before 2050 for RCP8.5.
Projections on Arctic ice extent “are known to be sensitive to model selection,” (Laliberté, Howell and Kushner, 2016, 256), particularly in areas where models have not accurately mapped Arctic ice loss, such as the Northwest Passage sector. Given the challenges in adopting a ‘pan-Arctic’ model for ice loss, Laliberté et al. therefore attempt a model selection criteria building on CMIP5 to include only models which sufficiently map out each region, evaluated each month from June to October. Through adopting an ensemble model, the authors conclude a September sea ice-free Arctic will likely become a reality between 2045 and 2070, with a median of 2050 (Laliberté, Howell and Kushner 2016, 262).
Jahn et al. (2016) lend further credence to this result. They conclude that, using the Community Earth System Model (CESM) to asses the impact of annual fluctuations in Arctic ice extent, September sea ice extent first crosses the 106 km2 threshold for 5 consecutive years between 2032-2053 or 2040-2056. The difference in range is down to whether a large (40 member) or medium (15 member) ensemble is used. However, the authors also focus their argument on the difficulties of accurately modelling precisely when the Arctic will be ice free on a year-round basis. In their suite of CIMP5 models, of which the CESM is just one component, they find a large spread in sea ice extent. They conclude that “we cannot predict the timing of a summer ice-free Arctic with an uncertainty of less than 21 years.” Additionally, both Laliberté et al. — “this large spread …makes projection of the year when a given region will become ice-free uncertain, even with an ensemble of state-of-the-art climate models” — and Tilling et al. — “it has been difficult to quantify trends in sea ice volume because…past observations have been spatially incomplete and temporally sporadic” — argue it is near impossible to predict when we will see a year-round ice free-Arctic ocean.
Despite this, the scientific community appears to consider modelling as being more accurate than it was at the time of AR5 owing to improvements in several areas. Perhaps most importantly, since 2013, scientists have been able to collect data from the underwater Lomonosov Ridge because of collapses which left new sediment exposed. This helped scientists “reconstruct the…climate history and the tectonic evolution of the central Arctic Ocean” (Stein et al. 2015). This is particularly useful because the data goes beyond instrumental records of past climate conditions and represents vastly different times in the Earth’s history, and therefore can be used to test climate models’ predictions for those times —thus as an indicator of their overall accuracy.
Paleoclimate data is therefore demonstrably useful in evaluating the accuracy of today’s models. Stein et al. identified for the first time a proxy biomarker for sea ice in sediments as old as the late Miocene, which ranges between 5 and 10 million years ago. They conclude there would have been spring sea-ice concentrations of 20-70%, and a “relatively warm and ice-free summer” from this new data. The AOGCM model which they employ is found to stand up well to this finding, as it predicts a spring sea ice concentration of 20-60% for 450 p.p.m. CO2 — and a summer season of ice-free conditions, too. Researchers in other papers discussed here tend to ignore the implications of paleoclimate data — to their own detriment. Stein et al. can make an objective judgement, to the extent that such a judgement is possible, of the accuracy of the model they have chosen. Other authors make no effort to judge models against historical records, which could enhance the accuracy of the models they so lament. As Stein et al. highlight, “this type of records giving detailed information about past Arctic sea-ice conditions are still very rare.” Further research in this field, and on further proxies which can be used to understand ice cover, will likely see greater accuracy in modelling — as they suggest.
The pertinent point to take away from Stein et al.’s findings, though, is that models are ultimately becoming more accurate. For example, the areas of thickest sea ice around Greenland and Ellesmere Island can now be modelled with a far greater degree of accuracy than before, which is vitally important for modelling sea ice volume. This was a ‘blind spot’ in accurate scientific modelling, as satellite altimetry previously didn’t previously extend to these regions, as well as to the central Arctic [Tilling et al., 2015; Laliberté, Howell and Kushner 2016, 256]. Perhaps most encouraging, though, is an acknowledgement that pan-temporal, pan-Arctic models will not yield the most accurate forecasts — a topic raised mainly by Laliberté, Howell and Kushner (2016). By employing an ensemble modelling method, the authors can more accurately account for ice cover of all regions at different times from June to October. Importantly, they also verify the extent to which particular model configurations can affect the prediction it makes, and conclude that “the median year [2050] of ice-free conditions is not really affected by the selection” as different model configurations still point to ice free years “within a decade of each other for all regions” (Laliberté, Howell and Kushner 2016, 261).
Despite these improvements, Jahn et al. nevertheless argue a prediction with an uncertainty of fewer than 21 years due to is impossible when considering the impact of internal variability alone. This result holds for both the large ensemble (40) and medium ensemble (15) CESM model the authors employ. Furthermore, accounting for other factors such as “scenario uncertainty” adds another five years to the 21-year window. Hence, they conclude, “accurate predictions of the exact year or even the exact decade we could first reach a summer ice-free Arctic decades in advance are not possible.” This result seems unsatisfactory, but bearing in mind their starting point, the result points to a prodigious improvement in accuracy. The difference in CIMP5 outcomes, from which they started their research, ranged over 100 years. To this, it is clear modelling has markedly improved since AR5.
If accurately predicting a date for an ice free-Arctic Ocean is difficult, scientific literature confidently points to the long-term trend of both Arctic sea ice extent, and volume, continuing to fall. This is addressed by Tilling et al., who highlight both intra- and inter-annual fluctuations in sea ice extent, and highlight this as a cautionary tale against reading too much into short-term trends. Between 2010 and 2012, total autumn sea ice decreased in volume by 14%, which was followed by a 41% increase in 2013, and then just a 6% decrease in 2014, using CryoSat-2 measurements. They further highlight that only one of the 16 regions measured as part of their study, shows a trend in sea ice volume over the period of the survey. This highlights the importance of keeping in mind the long-term trends associated with ice loss over many decades, accurate data for which is still missing. It is important to take this into account regarding the models Laliberté et al. and Jahn et al. employ to predict future ice conditions. Although they acknowledge the implications of internal variability for the accuracy of their models, a discussion of the explanations for inter-annual variability, and how that could be built into their models, is lacking – unlike in Tilling et al.’s paper. Tilling et al. first rule out snow-load and “wind-driven ice convergence.” They then turn their attention to thermodynamic forcing – specifically, how the number of ‘melting’ days correlates to Arctic sea ice volume. The correlation is found to be strong, with r2=0.73. They, therefore, conclude “thermodynamics play an important role” in determining Arctic sea ice volume. It is then no coincidence that the significant 2013 increase in Arctic sea ice volume corresponded to a sharp drop in the number of melting degree days (Appendix 1), and therefore ultimately was considered a “relatively cool year, with temperatures comparable to those of the late 1990s.” This discussion is important as it points to further ways in which modelling can be improved to account for anomalous results like that of 2013’s autumn Arctic ice volume— however, such a discussion is lacking on the parts of Laliiberté et al. (2016) and Jahn et al. (2016).
These studies on improved modelling, and discussions on when we will first see an ice-free Arctic, and on Arctic sea ice extent itself, all speak to the data presented in AR5, on several counts. Firstly, there is clearly further “robust evidence” (IPCC, 2013) that models are continuing to reproduce the downward trend in Arctic summer sea ice extent. Scientists now appear to be concerning themselves less with the matter of whether ice in the Arctic Ocean is melting or not, but rather with trying to predict our first summer ice-less Arctic Ocean. They are also making these predictions with seemingly little inter-model spread (Laliberté, Howell and Kushner 2016, 261). It is therefore very likely that Arctic sea ice cover will continue to shrink and thin. This is in line with the IPCC report of 2013. Additionally, a medium confidence should still be attached to the predicted ice reductions for the end of the 21st century, as this is too far away a time to predict anything with more than a medium level of confidence, as suggested mainly by Jahn et al. (2016)
IPCC scientists can predict a likely nearly ice-free Arctic Ocean in September before mid-century for RCP8.5 with high confidence. This is one confidence interval above that reflected in AR5, owing to several factors. A model incorporating a 42-model ensemble, using the respective individual models only to model the regions and times where they are known to be accurate, has predicted a nearly ice free-Arctic Ocean in 2045-2070, with a median of 2050. While this alone may not be enough to update the confidence reflected in AR5, the fact even changing the selection criteria for models does not alter the median ice-free year by more than a decade points to a strong level of certainty. Although Jahn et al. (2016) caution against trying to predict a summer ice-free Arctic Ocean accurately. If given a 25 year period, their model suggests the first ice-free Arctic Ocean in September by 2043-2058, and ice-free Arctic Ocean summers as the norm by the 2060s. Given the improvements in modelling delineated by Tilling et al. and Stein et al. there is reason to update the confidence associated with when we will reach this landmark occurrence in human history. In any case, it can be considered very likely with high confidence that we will see ice-less summers in the Arctic Ocean by 2065, as all models point to this eventuality. This should be added to the IPCC report to reflect current scientific consensus.