What is the future of Arctic Sea Ice? Part 1.

Recently I blogged about a paper showing the cause of the post 1995 volume decline in PIOMAS, this being the period of greatest and most persistent volume loss. The cause turned out to be self-acceleration of volume loss caused by the ice albedo feedback, as this is such a fundamental process and as the later version of the PIOMAS has proven to be so successful, I am in little doubt that Lindsay & Zhang's hypothesis is correct and that in the 1990s atmospheric impacts on the ice ushered in a period of decline largely driven by the ice albedo feedback.

There is really only one issue I can move on to address now, and that is the future of the sea ice.


Stroeve and Barrett have produced a set of slides for a presentation on CMIP5 (Coupled Model Intercomparison Project 5) models reproduction of Arctic sea ice metrics (see refs at end). CMIP5 being the latest generation of GCMs that shows improvements in terms of sea ice compared to CMIP3. In the following graphics SRESA1B is an emssions scenario used with CMIP3, RCP4.5 is a emissions scenario used with CMIP5.

Firstly it's worth looking at the seasonal cycle.

Whilst the seasonal cycle in CMIP5 (RCP 4.5) covers the observations, there appears to be a high bias from winter to summer, with observations occupying the lower end of the CMIP5 spread (red). This is important because the seasonal cycle is a critical measure of a model's response to an externally forced perturbation of the sea ice. Human driven climate change and the increase in July 65degN insolation during the Holocene Climatic Optimum are two examples of an externally forced perturbation of sea ice that has lead to severe sea ice recession in the Arctic. However the seasonal cycle due to changes in surface insolation is also an externally imposed perturbation, albeit one with an annual cycle, as opposed to acting over many years.

Wang and Overland updated an earlier paper and used the fidelity of seasonal cycle of modelled extent to observations to select a subset of the models that best reproduce the seasonal cycle, on the basis of response to external forcings as described above. Using this subset of GCMs, they find that a September extent of less than 1M km^2 is likely only by the 2030s.

The following graphic will be familiar to those following the Arctic, showing observed sea ice extent for September (black) for the period 1979 to 2011, compared with CMIP3 (blue) and CMIP5 (red).

Here CMIP5 is shown to be a large improvement on CMIP3, and in terms of extent one could construe the observations as being similar to a conceivably aggressive ice loss episode in CMIP5. However one has to consider what is driving the observed decline of extent, and regular readers of this blog will know that I am convinced that volume is the driver. As volume declines so does thickness and the seasonal thinning from April to September is then able to expose open water more easily, resulting in a decline of sea ice extent.

The intimate relationship between extent and volume for the September average is shown by the above scatter plot, with a simple power function seeming to me to be the simplest way to make both reach zero at the same time.

In the case of CMIP5 models, once again one could charitably interpret the PIOMAS decline in volume as being  an instance of aggressive volume decline within the range of CMIP models. Considering the top plot, MIROC I think, there is a similarly aggressive decline in the early part of the 21st century, albeit from a much higher baseline volume (or thickness).

The above graphic bears out a comment I've read in a paper (can't remember which paper) to the effect that volume decline in CMIP models is related to initial thickness, the thicker the ice is initially the greater it thins and the greater the resultant volume loss. Bearing this in mind it seems hard to take the charitable interpretation of PIOMAS as an aggressive volume melt in line with the CMIP5 volume losses, because PIOMAS does not start from a very high initial volume, even taking into account the Lindsay and Zhang paper discussed previously where PIOMAS volume in the 1980s is about the same as in the 1950s.

Restating the findings in the final page of the Stroeve/Barrett presentation:

• In general, CMIP5 models better capture the observed decline in Arctic sea ice than CMIP3.
• In general, the models exhibit a stronger seasonal cycle in both sea ice extent and ice volume.
• Large inter-model scatter remains in both ice extent and ice volume, particularly during summer.
• The spatial variability of thickness, a large-scale slowly varying climatic feature of the ice cover remains not well produced by the majority of the models.

Notably they find that thickness is not well reproduced, hence the massive spread in volume. At this point it is worth bearing in mind that PIOMAS volume is close to Cryosat 2, in the following graphic PIOMAS is line and circle, Cryosat 2 is the triangles.

There remains some doubt about Cryosat 2 and PIOMAS, both are proxies for volume and actual volume is not known with great certainty. However comparing the lowest data points (October) with the spread in summer minimum volume for the CMIP5 models shows how well PIOMAS is doing when compared to the closest thing we have to actual volume measurements, Cryosat 2.

So which is correct, the GCMs projecting survival of a perennial sea ice pack into the 2040 (and beyond) or PIOMAS? A clue here is that the other model producing aggressive volume decline is the US Navy Postgraduate Naval School's model, NAME.

In a 2009 conference presentation, shown in this document from FreshNor, Maslowski drew linear trends from NAME October/November ice volume to hit zero around 2012 to 2016, although in reality, as drawn on the graphic there is a supposed remnant of ice surviving summer melt. In the more recent Maslowski et al 2012 NAME modelled volume reaches about 1.4m thick ice in the summer by 2004, unfortunately I can't find further results beyond that date. However it's worth noting that 1.4m is in the same ball park as PIOMAS which has a late summer thickness of about 1.2m in 2004, link.

It seems to me that the link between PIOMAS and NAME that is key to the aggressive volume decline is that both assimilate atmospheric data and sea ice concentration data, so they are constrained by observations of actual changes as they proceed to model sea ice and model volume. Whereas GCMs are hindered by problems such as that found by Boe et al: That the models tend to create too strong a boundary layer inversion which means a far cooler atmosphere and stronger ice formation. That study relates to CMIP3 models, CMIP5 seems to have the same problem, but I've not got the time to properly delve into that - this isn't my day job, sorry.

Another significant issue with GCM's as opposed to assimilating models (PIOMAS and NAME) is that of albedo. Koenig et al find that CMIP5 GCMs have a massive spread in albedo (e.g. 0.3 to 0.7 for July) and that the models fail to reproduce the seasonal evolution and distribution of albedo changes. Whereas assimilating models will pick up impacts of albedo changes due to both temperature assimilation (2m temperature) and sea ice concentration assimilation. This issue is particularly important because whilst April volumes do not suggest a rapid transition to ice free in the summer, the greater loss of volume in spring/summer (notably post 2010 in PIOMAS) does have the potential to play a role in severely reducing summer volumes in the coming years. Previous related posts, The PIOMAS Spring Volume Loss, Sea Ice Albedo Changes and the Seasonal Cycle.

PIOMAS has been used to project sea ice volume into the future. In Zhang et al 2010, PIOMAS was fed simulated future data using available NCEP/NCAR weather for past years which was applied randomly to a regional warming scenario derived from emissions scenarios applied to a GCM. The volume projections are shown below. Dotted lines are for September mean ice volume, solid lines for annual, after 2010 the IPCC emissions scenarios are as follows: green is A1, red is A2, blue is B1, black is B2.

What is notable is the discontinuity and change of trend from the dashed line, hindcast before, projection after. This is even more marked in the future projections of temperature where all GCM produced temperature trends show a step drop as the projection starts together with a reduction in the rate of temperature increase. So, as noted in the paper, the projection method may be neglecting atmospheric changes impacting sea ice.

Those issues aside, Zhang et al find that the reason for the long persistence of sea ice volume is that thinner ice has a faster growth rate over autumn/winter, so as volume loss through the spring and summer increases so does volume gain over autumn and winter, this is known as the thickness/growth feedback.

The following scatter plot shows volume gain in the melt season as a function of NSIDC Extent at the end of the preceding melt season.

There is a strong relationship (R2 = 0.65) between the two variables, in simple terms; the lower extent gets at the end of the melt season, the greater the volume increase in the following melt season. It should be noted that as the recent years have shown greater losses, stretching data points towards the top left, so the growth in the following melt season has been all the more vigorous.

However despite this strong negative feedback, volume at the end of the melt season has continued to plummet. 2013 was an anomaly due to weather, not part of the long term trend.

I've been waiting for this negative feedback to stem losses, and have in the past few years had concluded it won't. But in my next post I will show why I may have called it too soon, and why although GCM projections of summer perennial ice well into this century are not supported by the evidence, the first year with a virtually sea ice free state may be around 2030.


References.

Boe et al, 2009, Current GCMs’ Unrealistic Negative Feedback in the Arctic, PDF.

Koenig et al, 2014, Summer Arctic sea ice albedo in CMIP5 models, PDF.

Maslowki et al, 2012, The Future of Arctic Sea Ice, PDF.

Stroeve & Barrett presentation, Assessment of Arctic Sea Ice in the CMIP5 Climate Models, PDF.

Wang & Overland, 2012, A sea ice free summer Arctic within 30 years-an update from CMIP5 models. Abstract.

Zhang et al, 2010, Arctic sea ice response to atmospheric forcings with varying levels of anthropogenic warming and climate variability. Abstract.