How does PIOMAS function without melt ponds?

[Nightvid Cole]

It's quite puzzling to me how PIOMAS does so well despite having no melt ponds whatsoever in the model  (!!!) despite the fact that multiple other simulations have shown marked improvement when melt ponds are added (September Arctic sea-ice minimum predicted by spring melt-pond fraction:

 http://www.nature.com/nclimate/journal/v4/n5/full/nclimate2203.html?WT.ec_id=NCLIMATE-201405 )

Is there an explanation for this? Is there some effect PIOMAS overestimates that almost perfectly cancels out its lack of melt ponds due to pure happenstance?

[viddaloo]

It's quite puzzling to me how PIOMAS does so well despite having no melt ponds whatsoever in the model  (!!!) despite the fact that multiple other simulations have shown marked improvement when melt ponds are added (September Arctic sea-ice minimum predicted by spring melt-pond fraction:

Being the subject of so much uncertainty during the last year, what is your base for saying "PIOMAS does so well"? How do you know it's doing well?

[Nightvid Cole]

It's quite puzzling to me how PIOMAS does so well despite having no melt ponds whatsoever in the model  (!!!) despite the fact that multiple other simulations have shown marked improvement when melt ponds are added (September Arctic sea-ice minimum predicted by spring melt-pond fraction:

Being the subject of so much uncertainty during the last year, what is your base for saying "PIOMAS does so well"? How do you know it's doing well?

Well, it's amazing (IMO) how well PIOMAS gets the shape and size of the ice pack in September even with no assimilation (see figure 1b in http://psc.apl.washington.edu/zhang/Pubs/Zhang_etal2008GL033244.pdf ,for example). The models used by the IPCC are not nearly this good - and predict no nearly-ice-free summers until the second half of the century!

[viddaloo]

That figure 1 says it did fairly well (compared to a test–run with sea ice conc data) back in 2007. But how do you know it's doing "so well" in 2015?

[Nightvid Cole]

Their seasonal outlook did quite well - even the shape of the ice pack was spot on:

http://psc.apl.washington.edu/zhang/IDAO/seasonal_outlook.html

Again, I ask, how do they perform so well without melt ponds?

[viddaloo]

That page is pretty confusing, Nightvid. Can you walk me through it and show me where the "Correct Answer" is, the correct answer that is then used to determine whether PIOMAS is doing well in 2015?

[DavidR]

The forecast on the PIOMAS website is produced on Aug 1st 2014, for a September  minimum that was only  six weeks away.  It  doesn't really  say  much about the long term accuracy of their predictions.  It  also makes the forecast long after the period when meltponds are considered a valuable predictor, which is about May.

[seaicesailor]

Melt ponds may appear in May - June but their influence (or lack thereof) is felt during the max insolation period May-August. To take them into account, PIOMAS cannot just assimilate their May state as a "valuable predictor" but somehow model their effect throughout the season.
Melt-pond evolution depends heavily on June weather too. They may refreeze or halt their spread if cold weather

[jdallen]

Melt ponds may appear in May - June but their influence (or lack thereof) is felt during the max insolation period May-August. To take them into account, PIOMAS cannot just assimilate their May state as a "valuable predictor" but somehow model their effect throughout the season.
Melt-pond evolution depends heavily on June weather too. They may refreeze or halt their spread if cold weather

I think the big problem - one most difficult to solve - is understanding how the heat gets distributed (or ejected, via albedo).  With so few sensors, I don't think it's going to be something we can do skillfully enough to get accurate predictions past a couple weeks or so.

We honestly don't have much of a clue except for after the fact - via changes in SST and ice coverage, so we can tell *when* the heat arrived (past tense), but not how much we will get (future tense).

[Michael Hauber]

It's quite puzzling to me how PIOMAS does so well despite having no melt ponds whatsoever in the model  (!!!) despite the fact that multiple other simulations have shown marked improvement when melt ponds are added (September Arctic sea-ice minimum predicted by spring melt-pond fraction:

 http://www.nature.com/nclimate/journal/v4/n5/full/nclimate2203.html?WT.ec_id=NCLIMATE-201405 )

Is there an explanation for this? Is there some effect PIOMAS overestimates that almost perfectly cancels out its lack of melt ponds due to pure happenstance?

I don't think melt pond is nearly as good a predictor as the paper claims.  I believe there is a serious issue of 'overfitting' the model.  Melt ponds in May are mostly in fringe areas and cover less than 2% of the entire Arctic.  There are many different fringe areas, and in many years there is no ice in those fringe areas.  Wherever there is a grid square with a melt pond in one year, and no ice in many other years there is an opportunity for the regression model to choose coefficients to more accurately predict the small number of years with ice in that grid square and having no effect on any year without ice.  Regression coefficients are capped to limit this effect to some extent (no single grid square can be given too much weighting), but this effect is guaranteed to exist and it is only a question of whether it is too small to matter, or big enough to account for nearly all of the correlation.

Another possibility (which is mentioned in the paper) is that melt ponds do not as much 'cause' the final extent through feedback processes, but predict the final extent because they reflect the state of the ice, which PIOMAS may have made reasonable estimates of through other means.

[Wipneus]

Another possibility (which is mentioned in the paper) is that melt ponds do not as much 'cause' the final extent through feedback processes, but predict the final extent because they reflect the state of the ice, which PIOMAS may have made reasonable estimates of through other means.

Indeed, correlation is not the same as causation. We have physical reasons that melting (ponds) will help melting by decreasing albedo but that does not rule out other major factors that force both.

If I remember correctly crandles found a remarkable correspondence between May Piomas volume drop and drop in the rest of the melting season. It is somewhere in the archives of this forum discussed in the spring of 2013. It was the first indication for me that 2013 would not see the expected volume drop but a remarkable "recovery" instead.

[Nightvid Cole]

The forecast on the PIOMAS website is produced on Aug 1st 2014, for a September  minimum that was only  six weeks away.  It  doesn't really  say  much about the long term accuracy of their predictions.  It  also makes the forecast long after the period when meltponds are considered a valuable predictor, which is about May.

Irrelevant. If PIOMAS does not incorporate melt ponds at all, at any time, then the only way (one would think) to get a good result from August 1 would be if the ponds' effect somehow was built into the assimilated observations prior to August 1, or equivalently, that the ponds don't significantly further impact melt after August 1. Why should we take the latter to be the case, given that ponds increase bottom melt as well as top melt, and bottom melt in August is quite significant?

[Michael Hauber]

And while PIOMAS may not assimilate melt pond data, it does assimilate temperature and cloud data.  Surely it would generate a reasonably accurate representation of surface melt within its models which must include melt ponds, and the effect on albedo?  How else could it hope to accurately model reduction in ice thickness if it doesn't model surface melting in some way?

[epiphyte]

I think that the time to measure the accuracy of PIOMAS predictions against reality is in mid-September. Even then we might be taken by surprise if we rely on it. We can only pinpoint inaccuracies when the model predicts >0 ice depth, and there is observed to be no ice.

There have been many references in this forum and others to experienced researchers on the ground observing that the ice over much of the arctic is qualitatively very different to what it has been in the past. Viz: This TED talk -  , in which David Barber relates a cruise through 2M thick ice at 13 knots in a ship which can manage only 13.5 knots in open water.

The calculations underlying the PIOMAS model predate whatever changes to the mechanisms governing the growth and melting of arctic ice have caused these observable phenomena. The same is true of most of the submarine and other observations which were used to validate it.

The latest possible direct comparison between the PIOMAS model and reality using the link at the top of this thread is at a point in time when most of the surface has a model-predicted thickness > 1M.

So if the model is overestimating the average thickness by 99CM, or if it has the thickness right but the composition of the last remaining ice is such that the model cannot predict what will happen to it, or when, we wouldn't know.

My guess is that for the last 1M, at least, PIOMAS is as clueless as the rest of us.

[LRC1962]

As a followup to what has said, the question then becomes if there had been sufficient melting for melts ponds to be made or not. If there had been enough melting for melt ponds to be made but they did not appear then it seems to me that the water had a way of getting through the ice to water below. At that point the volume still would be the same, the difference is in the interpretation on what PIOMAS is saying. If you use a formula for melting based on what PIOMAS states and treat it as solid ice then you would be grossly exaggerating how much energy is needed to melt the ice. The problem is that there is actually no way of knowing how much energy is need to melt the current amount of Arctic Ice. That is what would fall into the category of mass. Mass and  volume are two different numbers PIOMAS is interested in volume and does a fairly commendable job considering the extremely challenging job it is to come up with that number. The mass on the other hand is impossible to get even if you know the exact volume.

[Nightvid Cole]

And while PIOMAS may not assimilate melt pond data, it does assimilate temperature and cloud data.  Surely it would generate a reasonably accurate representation of surface melt within its models which must include melt ponds, and the effect on albedo?  How else could it hope to accurately model reduction in ice thickness if it doesn't model surface melting in some way?

Have you read any of the papers describing what went into the model? Zhang and Rothrock 2001 and Zhang and Rothrock 2003 describe it in fair detail, IIRC.

There are no melt ponds simulated in PIOMAS, believe it or not!

[Michael Hauber]

I had tried to track such papers down before and failed, however with a further attempted at googling 'Zhang and Rothrock 2003' I've found it

Zhang And Rothrock reference Winton 2000 as the source of their ice thermodynamic model.  The ice is modeled in 3 layers - one for snow, and the other two for ice.  It models the surface status as either melting and non melting, and sets the surface albedo as 0.8 for frozen snow, 0.75 for melting snow, and 0.65 for ice with no snow.  For the most part I would expect ice with no snow on it to be melting ice, with some exceptions when temperatures drop under freezing, and before snow has had a chance to fall.

So yes PIOMAS includes a representation of surface melting, and has a mechanism to reduce albedo due to the surface melting.

[Nightvid Cole]

65% albedo is significantly higher than that of heavily ponded ice...

[epiphyte]

65% albedo is significantly higher than that of heavily ponded ice...

Given melting conditions, is ponding more likely on thin ice or thick ice? I'd hazard maybe the latter - going on the theory that thin ice is more permeable to liquid water than thick ice?

If that's the case, then one might argue that any hypothetical systemic error in PIOMAS thickness arising from underestimating the albedo of melt ponds might not be detected until very late on in the process, since the underestimate would occur mostly on relatively thick ice, which presumably would be the last to go...

[Peter Ellis]

65% albedo is significantly higher than that of heavily ponded ice...

Yes, but I think it's also significantly lower than that of unponded ice - which means that this parameter is effectively a measure of the average proportion of melt ponding.

Ponding tends to occur more on thinner ice, because it's flatter and thus forms more extensive, shallower pools.

Remember also that PIOMAS integrates concentration constantly.  If melt ponding is more extensive than usual in a given year, then the heavily-ponded bits of ice may melt out before PIOMAS expects them to.  But as soon as it's gone, PIOMAS is able to assimilate the fact that there's no longer any ice in that pixel.

[Nightvid Cole]

65% albedo is significantly higher than that of heavily ponded ice...

Yes, but I think it's also significantly lower than that of unponded ice - which means that this parameter is effectively a measure of the average proportion of melt ponding.

Ponding tends to occur more on thinner ice, because it's flatter and thus forms more extensive, shallower pools.

Remember also that PIOMAS integrates concentration constantly.  If melt ponding is more extensive than usual in a given year, then the heavily-ponded bits of ice may melt out before PIOMAS expects them to.  But as soon as it's gone, PIOMAS is able to assimilate the fact that there's no longer any ice in that pixel.

Actually, PIOMAS assimilation of concentration is mainly aimed at the ice edge, not the interior of the ice pack:

R. W. Lindsay and J. Zhang, 2006: Assimilation of Ice Concentration in an Ice–Ocean Model. J. Atmos. Oceanic Technol., 23, 742–749.
doi: http://dx.doi.org/10.1175/JTECH1871.1

That said, perhaps the SST assimilation, rather than the concentration, does the trick...