Sunday 23 March 2014

What caused the volume loss in PIOMAS?

Behind all of my blogging on sea ice is the volume loss, which is driving the changes. The volume loss is shown by other data (e.g. Submarine upward looking sonar, ICESat, IceBridge, Cryosat 2), but the most detailed picture is that provided by PIOMAS. What is driving the volume loss in PIOMAS and is this process at work in reality?

In my previous post I used a graph showing the cumulative sum of interannual differences of September average NSIDC sea ice Extent. In the following copy of that graph blue (deltas) is the interannual difference, red is the running sum of those numbers.

The sum of the interannual differences isn't really a physical number in the sense that in no way do the differences between years accumulate like that. This is unlike volume where such differences do accumulate to produce the overall volume loss, indeed the plot of volume can be plotted on top of the running sum of interannual extent differences and the two track fairly closely.

The point of the running sum of inter-annual differences is merely to show that up to 1997 for each loss, on average there was a compensating gain in extent, so although there was a downward trend it was slight, with the series staying close to the zero axis. It was after 1997 that the losses began to dominate over gains such that the running sum sharply veered off the zero extent line and started to decline. Something happened around this time that lead to a change in behaviour from near balance (with only a slight decline in extent) to a marked decline trend post 1997.

Using this observation as a basis for splitting NSIDC Extent CT Area data, the following graphic shows the periods before and after in terms of NSIDC Extent CT Area, with their trends. EDIT - graph incorrectly used CT Area data.

The slope of the loss trend after 1997 is four times greater than before. Using the lower left equation to project extent forward, extent reaches zero in 2038, and drops below 1M km^2 (virtually sea ice free) in 2030. Omitting 2013 as an outlier and using only the years 1997 to 2012 gives a zero extent in 2033, and virtually sea ice free (less than 1M km^2) in 2026.

I've plotted the relationship between NSIDC Extent and PIOMAS volume for September. A power function has been used to relate the two because of the available simple functions this one converges to zero volume at the same time as zero extent.

The clustering found underlines that there is a clear relationship between extent and volume in the month of September, the month of minimum volume.Using a linear fit produces a similar R2 (0.8778), but is unphysical, giving 3.23M km^2 extent at zero volume. What this suggests to me is that as volume declines below about 2.5k km^3 we should see increasingly aggressive open water formation with attendant ice albedo feedback (low September extent and volumes are a result of low extent and volumes earlier in the summer). However it provides no information about the speed of progression along the fit curve, or even if the fit will continue to hold. That noted, due to considerations of thickness and resultant efficiency of open water formation, it is clear that the volume loss is driving the declines in extent that are seen in the NSIDC September extent, as argued in my previous post. The thinner ice is at the start of the melt season, the more open water will be formed by the end of the melt season, and as more open water forms so ice/ocean albedo feedback becomes stronger (white ice reflects more sunlight than darker ocean, as as the ice retreats the ocean gains more energy which can increase melt).

The reason I wanted to show that, and turn thoughts towards open water formation efficiency and the ice/ocean albedo feedback is that there is research using an earlier version of PIOMAS that shows what caused the volume loss in that version. The process shown is so fundamental, it also applies to the volume drop we have been seeing in PIOMAS, and as I will argue, this is probably the process behind the virtually simultaneous increase in loss rate shown in the NSIDC data set after 1997.

Lindsay & Zhang 2004 examine whether a tipping point was passed in the 1990s, but in studying that also account for what has caused the volume decline in the 'feedback' period, i.e. the strongest period of volume decline in the PIOMAS model data.

They identify three distinct phases as shown in the above graph, although the study runs the model from 1948 to 2003. The phases are as follows:
  1. Preconditioning. During this period, 1948 to approximately 1989, warming winter air temperatures reduced the thickness of undeformed ice while the thickness of ridged ice increased. Undeformed ice is that which has thickened thermodynamically, as heat is lost through the ice to the atmosphere new ice grows on the underside of ice floes. Critical to this ice formation is the heat flux through the ice which is set by the temperature difference between the ice/ocean boundary at the underside of the ice (freezing point) and the temperature of the atmosphere at the ice surface, changes in atmospheric temperature dictate thermodynamic freezing. Ridged (or deformed) ice thickens as mechanical compression of ice causes the ice to buckle under compression and form ridges. So as the air warmed over this early period it reduced the thickening of ice over winter by reducing heat flux through the ice and reducing growth of new ice. 
  2. Trigger. (Approximately 1990 to 1995) During the 1990s the Arctic Oscillation and Pacific Decadal Oscillation lead to a shift in the strength and centre of action of the Beaufort Gyre, from 1990 this lead to larger amounts of open water in summer. Due to export of ice through the Fram Strait (as noted by other authors, e.g. Rigor et al) there is a large drop in deformed ice thickness as multi-year ice North of Greenland and the Canadian Arctic Archipelago (CAA) was reduced by large volumes exporting out of the Fram Strait. This period can be seen in the first graphic of this post as a succession of large interannual variations in NSIDC sea ice extent.
  3. Feedback. (Approximately 1996 onwards) Following the thinning of the pack in the Preconditioning and Trigger phases, the stage was set for ice albedo feedback to take effect. Ice was thin enough such that in the marginal oceans summer thinning was able to open up more open water and lower concentrations of ice, allowing heat gain within the open water (and the pack behind the ice edge - not stated in the paper). Lindsay & Zhang find that from 1988 to 2003 the loss of thickness due to ocean heat flux is about 2m, this is countered by infra-red emission of heat from freeze season open water and thin ice which accounts for a gain of about 2m thickness of ice. However the change in absorbed solar flux is equivalent to about 3m of ice loss, and as the model doesn't simulate yearly changes in insolation this must be due to albedo reducing as the sea ice recedes.
The ice albedo feedback is such a basic feature of the physics of the model, and shows such a strong role in that model, that it must apply to the later model - what we now know as PIOMAS. So ice albedo feedback is the largest component driving the massive decline in volume seen in recent years in PIOMAS. But what of the real world?

That the model shows ice albedo feedback is driving most of the volume decline should not be a surprise, nor should it be a surprise if the model results can be transferred to the real world, for some time it has been known as a basic factor in Arctic sea ice, e.g. Curry et al. Furthermore as found by other researchers ice/ocean albedo feedback is causing ocean warming, e.g. Perovitch et al, and ice albedo is falling, as discussed previously. Referring back to the first graphic and previous post, I think that the change seen in NSIDC Extent is evidence that the process found by Lindsay & Zhang in their model has been happening in reality. This is evidently a case of self-acceleration. That the incidence of loss years has increased in NSIDC Extent, such that the cumulative sum went negative at almost precisely the same time as the model showed ice albedo feedback driven self acceleration kicked in, seems too much of a coincidence to me.

Furthermore PIOMAS is validated against available observational data, Schweiger et al, Laxon et al (i.e. figure 3), and is able to reproduce large scale patterns of multi-year ice, see here. In my opinion, taking into account that corroboration and long term detailed data, PIOMAS is the best available proxy for sea ice volume. The above noted behaviour of NSIDC Extent and the coincident timing with the start of the 'feedback' period from Lindsay & Zhang 2004 adds to that opinion.

One counter intuitive issue is that the volume loss has mainly been coming from the thicker ice within the Central Arctic. The following graphic is for September PIOMAS volume based on gridded data.

Bitz & Roe show that because first year ice can grow back in a season, but multi-year ice takes years to grow back, in a declining ice pack the first year ice will appear to be largely stable, while the thickest oldest ice effectively has a long memory of losses. However ice/ocean albedo feedback hadn't penetrated into the middle of the pack in the early years of the Feedback period. A particular mechanism relevant here will have been the failure of the Beaufort Gyre Flywheel. The following graphic is amended from Cryosphere Today.

In past years ice moving in the Beaufort Gyre survived the summer, so was able to age (4 years typically) and return to the central Arctic via the transpolar drift (from Siberia to the CAA/Northern Greenland). With the lower summer extents in recent years, ice exported out from the Central Arctic region into Beaufort and Chuckchi has tended to melt out in summer, especially in a year like 2012.

Lindsay and Zhang find that in the recent feedback period internal thermodynamic process (ice albedo feedback) dominate the loss of volume, not external forcing (i.e. anthropogenic global warming, AGW). However the atmospheric data they use to drive the model is from NCEP/NCAR, this includes the signal of AGW because it is reanalysis data derived from observational data. So does Lindsay & Zhang really discount a role for AGW in Arctic sea ice loss?

Notz & Marotzke found no evidence of self acceleration in sea ice loss. However they looked at sea ice extent and concluded that any year of increased sea ice could have broken the trend in loss, therefore self-acceleration was not at work. They did not look at volume, where the decline has been staggering and it is clear why even a single year like 2013 is insufficient to break the trend of losses, the Arctic would need a succession of cold years to increase volume and start to rebuild the ice, bucking the trend and starting a recovery. But that is not going to happen.

Notz & Marotzke plotted scatter plots of sea ice extent and various factors that could be reasonably assumed to be affecting sea ice.

The only factor that shows the sort of grouping one might expect of a causal factor is CO2 concentration, the others including the AMO, don't cut it. What is going on is volume loss caused by self acceleration that is triggered, and continues to be enabled, by anthropogenic global warming (AGW).

The alternative view would require that the sort of loss of sea ice we are seeing happens from time to time. I might suggest that the 1930s warming didn't lead to the sort of precipitous crash in sea ice seen at present, that could countered by citing the following cooling that occurred.

But taking the long view, that what happening now is occurring merely as a coincidence with AGW, and that AGW has no causal role, such a stance becomes hard to maintain given sea ice throughout much of the history of civilisation (fig 3a of Kinnard et al).

Going back further there is no evidence of such low sea ice levels as at present since the Holocene Climatic Optimum, 6 to 8 thousand years ago. In the following graphic low sea ice concentration from various lines of evidence is shown in dark turquoise. The black line trace in the lower panel is a proxy for temperature from a Greenland ice sheet ice core. The blue trace in the lower panel is July insolation at 65degN (strength of sunlight).

So the last time that sea ice was in a worse state than at present was under summer sunlight that was about 10% stronger than at present.

Then there are the model results. Climate models only show sea ice decline with anthropogenic factors included in the model runs. In the following from Wang & Overland 2012 (fig2) coloured plots are for the sea ice simulations of seven different models with natural and anthropogenic forcings, the grey shaded plots are for simulations with natural forcings only.

Despite the different behaviour of the models and the differences between model designs the common behaviour is clear: Without anthropogenic forcing the ice does not decline.

So while Lindsay & Zhang state that internal thermodynamic processes dominate the volume loss from the mid 1990s to 2003, not external forcings. To take that as discounting external forcing would be wrong. The wider evidence supports the view that the self acceleration is happening against an enabling background of external anthropogenic forcing.

The Arctic sea ice volume loss is driven by anthropogenic forcing, but with the background of that forcing the sea ice is now in a state of self accelerated decline due to the ice albedo feedback. With ten more years of data and published science since Lindsay & Zhang was published in 2004, the position has become more clear, within a few decades at most we will almost certainly see the Arctic ocean in a state it hasn't been in for about 7000 years, virtually sea ice free at the end of the summer melt.


Bitz & Roe, A Mechanism for the high rate of thinning in the Arctic Ocean.

Curry et al, 1999, Sea ice albedo climate feedback mechanism.

Jakobsson et al, 2010, New insights on Arctic Quaternary climate variability from palaeo-records
and numerical modelling.

Kinnard et al, 2011, Reconstructed changes in Arctic sea ice over the past 1,450 years

Laxon et al, CryoSat-2 estimates of Arctic sea ice thickness and volume.

Lindsay & Zhang, 2004, The Thinning of Arctic Sea Ice, 1988–2003: Have We Passed a Tipping Point? Abstract - PDF should be available for free via that link.

Notz & Marotzke, 2012, Observations reveal external driver for Arctic sea-ice retreat.

Perovitch et al, 2007, Increasing solar heating of the Arctic Ocean and adjacent seas, 1979–2005: Attribution and role in the ice-albedo feedback.

Rigor et al, 2002, On the Response of Sea Ice to the Arctic Oscillation.

Schweiger et al, 2011, Uncertainty in Modeled Arctic Sea Ice Volume.

Wang & Overland, 2012, A sea ice free summer Arctic within 30 years? - CMIP5 Update.


crandles said...

I agree that the Beaufort gyre plays a role in the central Arctic thinning despite open water not reaching there. In the past there was thick ice that the gyre could move back into central Arctic region. Now that ice melts out in Beaufort/Chuchi/ESS/Laptev there isn't ice to move back into central region.

However, I would suggest a possible additional/compounding effect is that there is now less mass of ice. In the past winds and currents acting on large mass would cause more and higher ridges.

Chris Reynolds said...


Zhang et al uses PIOMAS in "Recent changes in the dynamic properties of declining Arctic sea ice: A model study" They find that a decline in mechanical strength due to thinning increases deformation rates by 17%. So I don't think that greater deformation happened in the past. Although throughout a longer 'lifetime' of a parcel of MYI - total integrated ridging would likely be more than for the short lifetimes of MYI at present.

In Maslanik's work it seems reasonable to imagine the age distribution for ice below 5years to be continued into ice over 5 years, especially in the past.
Despite current greater amounts of thinner ice, a time period of over 5 years may be enough to overcome reduced deformation due to greater mechanical strength.

However a paper I forgot to include in this post is Kwok & Cunningham "Contribution of melt in the Beaufort Sea to the decline in Arctic multiyear sea ice coverage: 1993–2009" They find that 2005 to 2008 accounts for 490k km^2 of the area loss of MYI for the period 1993 to 2009 the loss of MYI in Beaufort is 947k km^2.

That supports the idea that recent open water in Beaufort has caused a substantial destruction of MYI.

However for the 2005 to 2008 period Fram export accounts for 20% of total loss of MYI area, while Beaufort accounts for about 25%. This leaves 55% loss of MYI area within the central Arctic. The authors suggest that a significant part of this may be due to - "Convergence, especially when the ice is pushed against the northern Greenland coast and Ellesmere Island, could cause significant reduction in area that could be misconstrued as loss due to melt or export."

Jai Mitchell said...

Trigger? :


(note the reference at the bottom of slide 12)

what we are seeing is an artefact of ocean heat content increases due to TOA imbalance.

I believe that the next El Nino will be a monster that will increase globally averaged temperatures by .4C in the space of 2 years and that we will see a combination of the two effects above and an increase in freshwater volume that will push September 2016 Piomas volume below 2,000 km^3.

Chris Reynolds said...


Ocean heat is fashionable as a suggested cause of the volume loss in PIOMAS. Yet, to quote from my post, Lindsay & Zhang shows:

"from 1988 to 2003 the loss of thickness due to ocean heat flux is about 2m, this is countered by infra-red emission of heat from freeze season open water and thin ice which accounts for a gain of about 2m thickness of ice. However the change in absorbed solar flux is equivalent to about 3m of ice loss, and as the model doesn't simulate yearly changes in insolation this must be due to albedo reducing as the sea ice recedes."

The model uses the following inputs atmospheric SLP and 2m temperature from NCEP/NCAR, specific humidity and longwave and
shortwave radiative fluxes are calculated from SLP and 2m temperature. The model includes precipitation and river run off.

There is no prescribed input of wider ocean warming due to AGW.

The paper references a 1998 paper, available here:

Once again in that source there is no mention of wider ocean warming being used to drive the model, only atmospheric drivers.

Furthermore, referring to figure 8 of Lindsay & Zhang, the annual flux of thickness is about -3m due to solar, and less than -0.5m due to ocean heat flux, both go slightly more negative towards the end of the series. This can be explained by the ice abedo feedback warming the ocean, it doesn't require recourse to wider global ocean warming.

The title of the post specifically refers to the volume loss within the PIOMAS model, and within the the early version of the PIOMAS model the largest cause of loss is change in albedo, which is due to ice albedo feedback. This is so fundamental and so strong a factor that it seems a reasonable conclusion that it applies to the volume loss in the latest PIOMAS version.

In the real world ocean warming plays a role. But as PIOMAS validates well against observational data, ocean warming does not appear to be a necessary factor in the observed volume loss.

Chris Reynolds said...


I forgot to add - I don't get what you're saying about snow data and the feedback issue.

Kevin O'Neill said...

Chris, one of your best posts.

Chris Reynolds said...

Thanks Kevin.

jai mitchell said...


the Maislowski presentation is putting together the work that he will submit to publish later this year. it shows that during the winter months, the accumulated ocean heat content below 30 meters is being contained in a thermocline.

he has also correlated regions in the bearing and berents seas that reveal increased thinning with this OHC accumulation in the middle depth.

Deeper temperature studies have shown decadal increases as well. I am not sure but I think that the real feedback is due to this thermocline heat storage during the winter months leading to a feedback mechanism. Note he also mentions increased heat content from surface melt (rivers).

The idea I had is that the early feedback period around 1988 onward was due to an albedo feedback and convection melt. Remember, the majority of the melt at that time was in old ice accumulations, this set the stage for later melt.

Just a thought.

jai mitchell said...


Summer and Fall snow anomalies

to your "trigger" years

also can be correlated to summer melt temperatures here:

Chris Reynolds said...


I've deleted the comment you withdrew to protect your identity.

Maslowski has a strong point with regards certain regions, Barents and Bering are two classic examples. In the case of Barents the decline of winter sea ice since the early 1900s is astounding and the most reasonable cause is warmer Atlantic Water. There's a directory of historic sea ice charts from DMI here.

With regards snow, I think the reteat of snow in summer is being caused by the retreat of sea ice. I've done similar analysis in winter, but not looked at summer continental warming's association with sea ice loss.

jai mitchell said...


Thanks! For fear of getting into a land-snow/sea-ice (chicken/egg) scenario. I am sure you are right. The land-based regional snowmelt will increase with arctic ice loss-albedo feedback.

However, the vast majority of the regional temperature changes, and snow loss during the trigger period are associated with significant warming in areas that are well below the arctic circle. Note: The anomalies that I posted were for the month of April-peak ice extent period.

Therefore, the early regional loss of snow and ice, once melted prevented the capture of additional heat in the latent heat of melt and deposited it into the land, then to the air via convective currents. Leading to the observed surface volume loss of MYI at this time.

Chris Reynolds said...


The warming associated with snow retreat is not necessarily the sole cause of the snow retreat. Again we're at 'chicken and egg' because as the snow line retreats absorption of insolation jumps and the surface is warmed - there's also the issue that once the snow has melted the insolation that would have gone into melting snow (latent heat) manifests itself as sensible heat (heat which can be sensed).

The resultant atmospheric warming plays in tandem with the ocean warming to create a background of forcing that acts against ice survival. The distant loss of snow may, I suspect, have a role in forming the summer pattern associated with high pressure and clear skies over the Arctic Ocean and Greenland - but that's really little more than a suspicion.

Note that the loss of snow, both distant and close to the Arctic (the latter being later in the spring) will cause a warming of air in the NCEP/NCAR data, which is based on observations. However Lindsay & Zhange state:

"Notably the change in ice thickness due to changes in the turbulent sensible and latent heat terms is relatively small. These fluxes are largely determined by the air temperature (relative to the ocean temperature), so the recent changes in the mean ice thickness are not primarily due to recent changes in the surface air temperature."

Rob Dekker said...

Thank you Chris. Great post !

Your hypothesis of "pre-conditioning", "trigger" and "feedback" makes sense, especially when one considers the physics of ice growth and melt under an ever increasing small CO2 forcing.

Two remarks :
1) Your figure 2 (NSIDC Extent decline) appears to be incorrectly scaled on the Y-axis. Currently it suggests that 2012 was 2.3 M km^2 and that at no time since 1979 extent was larger than 6 M km^2.
That can't be right.

2) Regarding snow, I think Jai has a point. In spring and early summer, snow extent has declined much faster than sea ice over the past decade :

Common sense would tell that snow cover is as much part of the albedo feedback as is ice cover.

And theory predicts that especially in late spring/early summer snow will have a very significant influence on the amount of heat absorbed by the Northern areas (adjacent to the Arctic) of our planet.

So I do not quite understand your remark that "I think the retreat of snow in summer is being caused by the retreat of sea ice."

The causal effect (reduced snow cover in June causes reduced ice extent in Sept) even shows up rather strongly in the data, and there does not seem to be any (theoretical nor empirical) evidence that snow cover retreat in summer is caused by the retreat in sea ice.

Could you please comment on what you meant with that assertion ?

Chris Reynolds said...


Thanks for catching that error, I'd accidentally used CT Area data not NSIDC Extent.

Careful - this isn't my hypothesis, it's the hypothesis of Lindsay & Zhang.

Snow/ice/warming. See for example Lawrence et al 2008:

They find that during rapid ice loss events warming extends up to 1500km inland - the pattern being similar to the current warming and snow loss regions.

They use RILEs identified in Holland et al 2006,
see para's 9 and 10. The ice loss events are driven thermodynamically by ice albedo feedback on the ocean. Also para 13 - ocean heat pulses often precede RILEs in the GCMs used -no mention of snow.

See also this post:

Notably figure 7 - in years with low snow there is a warming that starts low level over the ocean/pack and reaches inland in the boundary layer.

Rob Dekker said...


I'm not quite sure if we are talking about the same "snow" effect.

Lawrence et al 2008 is talking about the warming extends up to 1500km inland in autumn (thus AFTER the albedo feedback did its work). So that warming may very well be a RESULT of spring/early summer snow cover decline.

Holland et al 2006 in paragraph 9 and 10 talks mostly about the non-linear effect of open water during the melting season. Specifically :

"As such, ‘‘the efficiency of open water production’’ (defined as the percent open water formation per cm of ice melt over the melt season) (Figure 2b) increases nonlinearly as the ice thins."

This non-linearity in the albedo feedback during summer is very important, since that is what makes it so potent, especially for early summer albedo effect.

Which is why it is surprising that the paper does not address the albedo feedback of spring/early summer snow cover at all.

I did not see your "eurasian-snow-cover-and-atmospheric" before yet, and I find it very interesting.
Especially your finding that land temperature increased (in May) in line with the snow cover decline is encouraging evidence of the albedo effect in the far North during the melting season, and that it's early (May) effect lies in snow cover retreat.

But I'm not sure if that info sustains your argument that "I think the retreat of snow in summer is being caused by the retreat of sea ice."

In fact, I think that you would have a point about snow cover in autumn and winter being affected by summer ice retreat, but I think you have snow and ice effect in summer backward.

Basic physics and also correlation data suggest that the cause-and-effect is the other way around :

Spring and early summer snow retreat facilitates Arctic ice melt. Not the other way around.

Where did Lindsay & Zhang state otherwise ?

Chris Reynolds said...


See figure 2b of Lawrence et al, during RILEs there is significant summer and spring warming as compared to periods not in RILEs.

Lindsay & Zhang find that sensible warming is not driving the volume loss in PIOMAS after 1995, they state: " the recent changes in the mean ice thickness are not primarily due to recent changes in the surface air temperature" i.e. the thinnning (volume loss) was not due to warming from loss of adjacent snow cover. They find that the losses are from sea ice albedo feedback.

Likewise Holland et al find that the increases of open water formation efficiency with thinning is the major player during RILEs. They don't mention snow because as Lindsay and Zhang find it it not a necessary component.

It's worth noting that the sustained low summer extents following 2007 have been following a major thinnning and volume loss in 2007. If you're finding high correlation between snow and sea ice I would suggest you are not detrending. So the declining trends in ice and snow are what the correlation has found.

As I show in my post on Eurasian snow cover and warming, anomalies of June and July snow cover show a linear decline after 2005.
While April thickness shows step drops after 2007 and after 2010 which correspond with step increases in April to September Volume loss.

And as Holland point out - thinning leads to greater open water formation.

So I fail to see a link whereby snow loss controls sea ice loss by direct themodynamic impacts.

Where I think there may be a link is in snow loss acting as a precursor to the formation of an Arctic dipole - which may have a limited effect on sea ice loss. I think the evidence suggests that ice state is the main mover, not atmosphere: