1. Mechanisms causing reduced sea ice loss in a coupled climate model
Alex West, Ann Keen, Helene Hewitt

2. Overview
Much work has gone into investigating causes of rapid ice loss – but what about reduced ice loss?
(Well it’s not something we’re seeing at the moment, but hey-ho…)
One of the Met Office coupled models, HadGEM1, gives a surprisingly strong signal for a temporary slowing of ice loss from (even a cessation in some ensemble members).
This model produced a relatively good simulation of historical sea ice, at least by CMIP3 standards; therefore it was felt that this behaviour at least warranted investigation.
We use an energy budget analysis of the Arctic to examine the causes of this slowdown, and look at temporal variation of the fluxes passing between the components of the climate system.
Several mechanisms are identified, including briefly reduced ocean heat transport and ice export.
We assess the likelihood of this slowdown being observed in the real world (possibly at a later date??)

3. CMIP3, HadGEM1 and observations
It’s well known that the CMIP3 multi-model ensemble tended to underestimate the sea ice decline even before the recent record minima (Stroeve et al. 2007).
But some models underestimate the decline less than others (Wang and Overland, 2009). The model in question, HadGEM1, was named as being among the better simulations.
x 106 km2
Mean value
Annual linear trend
Detrend-ed stddev
ALL
6.44
-0.072
0.48
ANT
6.69
-0.058
0.45
HadISST
6.30
-0.088
0.58
Stats of Sept ice extent,
Strikingly, while substantial ice loss is simulated in the historical period, very little loss is simulated from

4. Examining rates of change more closely
We can examine the slowdown more closely by computing 15 year running gradients of the rate of ice loss.
We can also attempt to rule out the possibility of the slowdown being due to random variations in Arctic summer weather by computing the standard errors for each gradient, and constructing a gradient interval for each period of [gradient – 2xstderr,gradient+2xstderr].
We say a run exhibits a significant slowdown if there exist successive 15-year periods such that the gradient intervals do not overlap.
-> 4 out of 7 runs exhibit a significant slowing in ice loss between 2000 and 2020.
When we consider volume, the number rises to 5 out of 7. Therefore in these runs, more fundamental mechanisms are likely to be responsible for the slowdown.

5. Examining the Arctic energy budget
The Arctic Ocean region (bottom right) was divided into the three components of atmosphere, ice and ocean.
Energy transfers between these regions were computed for the period of interest ( ), as well as energy entering from outside the Arctic, and uptake within the ice and ocean components.
In the zero-layer thermodynamic model used in HadGEM1 (based on the appendix to Semtner et al., 1976), the ice has no heat capacity; ice heat uptake is therefore proportional to ice volume loss, and ice heat transport out of the Arctic proportional to ice export from the Ocean.

6. Examining the Arctic energy budget: results
Ice energy uptake shows a clear minimum in the 2010s decade, corresponding to the slowdown.
In the all-forcings ensemble, this is immediately associated with a drop in ocean-to-ice heat flux – itself associated with a temporary drop in ocean heat transport.
In the anthropogenic ensemble it’s more directly linked to a drop in ice heat transport, or ice export. However, a similar drop occurs in the all-forcings ensemble, a decade earlier.

7. Changes in ice export
Ice export was recalculated as ∫hicevice.dS (i.e. integrated around the Arctic boundary).
The h and v fields were replaced in turn by their time means. Only when the v field was so replaced did the timeseries remain qualitatively the same – i.e. the decrease is caused by thinning of exiting ice.
The h field was then replaced by its time mean plus ice thickness anomalies in various parts of the Arctic, including the Western Arctic and Eastern Arctic below.
Only for the WA anomalies was the shape maintained (centre).
Decreasing ice export is a feature of virtually all runs (top centre).
This isn’t in itself surprising – thinning of the ice pack will tend to reduce the rate of ice export unless accompanied by a proportional increase in outflow velocity.
But decreasing ice export by this mechanism alone would be unlikely cause a slowdown in ice volume loss.
Moreover, there’s marked variation within the timeseries; the all-forcings runs display sharp reductions in the 1990s, the anthropogenic forcing runs in the 2000s.

8. Changes in ice export (continued)
Conclusion: the decreasing ice export is caused by changes in ice thickness in the region north of Greenland and Ellesmere Island, with changes in ice thickness in other Fram Strait ‘feeder regions’, and in ice velocity, not making a significant contribution
Observe the timeseries of ‘Western Arctic’ ice thickness; there are sharp decreases in around 1998 and 2008 for the two ensembles respectively, corresponding to the time of fastest decrease in ice export.
The heat budget analysis was repeated for the WA region only, to assess the causes of these sudden changes.
A single cause was hard to determine; for many runs the sharp decrease was preceded by a ‘flushing event’ (year of exceptionally high export), for others by periods of strong oceanic heat convergence.

9. Changes in ocean heat transport
Ocean heat transport is generally increasing throughout the period of interest – but most runs exhibit periods where the rise briefly slows or reverses, generally corresponding to the time of reduced ice loss.
Ocean heat transport can be calculated as ∫Tocnvocn.dS (integrating around the Arctic Ocean boundary). Again, each field was held constant in turn and the resulting OHT fields compared.
Holding the velocity field constant produced a very similar picture; holding the temperature field constant, there was no long term trend, but the changes in gradient were similar.
Therefore correlated changes in temperature and velocity are acting in concert to produce periods of reduced / increased OHT rise.
Examining the transport across various straits, most of the variability was found to occur at the Barents Sea boundary, on the Atlantic side of the Arctic.

10. Changes in ocean heat transport (continued)
Two important aspects of the North Atlantic ocean circulation were examined: the Meridional Overturning Circulation (MOC) and Sub-Polar Gyre (SPG).
They were defined as the maximum overturning streamfunction, and maximum gyre streamfunction respectively, from 66˚ - 80˚ N, so as to concentrate on the effects at the Arctic-Atlantic boundary.
As in many coupled models, the MOC shows a systematic decrease – but the ALL ensemble shows a particularly sharp jump in around 2008, corresponding to the drop in OHT in this ensemble.
No systematic decline in the SPG – but ALL-4, the run with the most ‘severe’ slowdown, exhibits a very low index from

11. Summary of findings
A majority of the HadGEM1 runs show a clearly slowed rate of ice loss between
This is probably due to three factors, two closely related:
Decreased ocean heat entering the Arctic, probably as a result of small but sudden changes in the MOC (and the SPG in the most extreme case)
A negative feedback on ice export; thinning of the ice cover implies lessened export
Sudden reductions in ice thickness in a small ‘feeder region’, north of Greenland and Ellesmere Island.
Notice the lack of atmosphere-related mechanisms posited. No clear reductions were seen in the atmospheric heat transport or atmosphere-to-ice heat flux around the time in the slowdown. Nevertheless, known modes of variability were plotted over the time period (first 3 EOFs of the Arctic SLP pattern) and no obvious trend was visible.

12. What relevance does this have to the real world?
When we first noticed this, in 2010, the observations and all-forcings ensemble matched extremely well…
With three more data points, they are diverging dramatically, as ice loss has temporarily stopped in HadGEM1.
Nevertheless, it’s not impossible that a temporary slowing of ice loss, albeit possibly for a shorter period, will be observed in the future, before a summer ice-free Arctic becomes a reality. It would be important to attribute such an ‘event’ reasonably quickly and decisively. ! ?

13. Real world continued…
This study confirms the simulation of a sustained period of reduced ice loss within a long-term decline, in an experiment forced with steadily increasing GHGs
It suggests three mechanisms to investigate were such a situation observed: the negative feedback of ice export; a preceding sudden thinning of ice north of Greenland and Ellesmere; and decreased oceanic heat entering the Arctic as a result of Atlantic ocean circulation changes.
However, other factors could easily be involved in a slowdown of ice loss in the real world – for example, a cessation of the persistent Arctic Dipole that has characterised Arctic summer weather since 2007 (Overland et al., 2012), and which was not captured by our model.
For more detail, read West et al., 2012, currently under interactive discussion at The Cryosphere.

14. Questions and answers