As the annual September sea ice minimum in the Arctic approaches, the usual questions arise about whether this year will set a new record for the extent or volume of ice left at the end of the summer. Although there was a new winter record low in 2017 it is looking unlikely that the summer will also set a record for extent, but there is still a month to go.
We understand that sea ice melt is erratic – we should not expect new records every year, but the overall trend is towards less and less ice. But, what about looking further ahead? And, can we understand sea ice variations during the historical period?
Mahlstein & Knutti (2012) suggested that there was an approximate linear relationship between global temperature change and Arctic sea ice area decline, and inferred that the Arctic would be effectively ‘ice free’ (<1Msqkm) at around 2C above pre-industrial levels, with some uncertainty. However, the observed relationship looked suggested faster melt than in the simulations from CMIP3. [Note comment below by James Screen on this paper.]
Now we have longer estimates for both global temperatures and Arctic sea ice and newer simulations. Notz & Stroeve (2016) recently suggested that there was a linear relationship between cumulative carbon emissions and Arctic sea ice. They found that each metric tonne of CO2 emitted caused around 3m2 of sea ice area to melt.
Can we go back even further to see whether these simple linear relationships hold? Recently, Walsh et al. (2017) published an Arctic sea ice dataset from 1850-2013. Note that the amount of data going into this dataset reduces significantly further back in time so there will be large uncertainties in the extent values.
However, this dataset can be used to examine how Arctic sea ice and global temperatures have evolved together over the past 164 years. As global temperatures have increased (upside down in the figure below), the extent of Arctic sea ice in September has declined.
What is striking when comparing these indices, which are constructed entirely independently, is their agreement on the long-term trends and multi-decadal fluctuations, especially from 1910 onwards. However, this relationship appears to break down from 2007, which was a record low in the observations at the time.
There are several possible explanations for this divergence after 2007. (1) The sea ice observations further back in time are too uncertain. (2) There is a real acceleration in sea ice loss per degree of global warming. (3) A large internal variability fluctuation has caused the Arctic sea ice to melt more than expected from 2007 onwards.
Which is right? I suspect a combination of factors. Note that Arctic Amplification does not necessarily explain this feature, unless the amplification factor has also changed over time.
To demonstrate that (3) is at least a possibility, the figure below from Jahn et al. (2016) shows four simulations with the CESM GCM with the same radiative forcings. The only difference between the simulations is the internal variability.
Realisation #13 follows the observations quite closely but then shows a large increase in sea ice, before rapidly declining again. Realisations #9 & #14 are towards the upper end of the range before declining more rapidly. These examples highlight that climate simulations do show significant natural fluctuations in sea ice extent, on top of the long-term decline.
However, GCMs show a large diversity in the simulated amplitude of sea ice variability, as shown in Swart et al. (2015). And, it is not necessarily the most variable in global temperature which show the largest sea ice variations.
Decadal variations in sea ice have been explored in detail in one model (GFDL CM2.1) by Zhang (2015) which discusses the important role for oceanic and atmospheric heat transports, but more research in this area is needed.
Overall, Arctic sea ice trends are strongly related to changes in global mean temperature. There are hints that the ice may be becoming more sensitive to temperature in the observations, but this is not clear in the GCMs. Our simulations show large internal variations in sea ice which we need to better understand and examine whether the same mechanisms may be working in the real world.
Imagine if the real world behaved like realisation #13 from the CESM ensemble?