Guest blog: Arctic Prediction Models

Here's another guest blog, sent to me by Lewis Clark, who is also a commenter on the Arctic Sea Ice Forum. Lewis describes this year's melting season and how several forecasts have played out in relation to it.

I like the term 'thin ice age'. Thin ice makes it more difficult to rely on conventional wisdom, although this year has proven that a lack of melting momentum during May and June followed by weather conditions during July and August that do not favour melt/export/compaction can still prevent a record, even if volume was at a record low for much of the year.

Thanks, Lewis!

-----

Which Arctic Prediction Models Are Relevant in the Thin Ice Age?
by Lewis Clark

2017 provides a stress-test for Arctic sea ice models. Following an unusually warm winter, the melting season began with record lows (post-1979, the satellite era) for both maximum ice extent and volume. After a slowdown in ice loss, however, measures are back to a near-normal range. This development seems to partly validate some models that predicted a rebound year, but the jury is still out. With a few weeks go before the minimum, large areas of thin ice might yet hold surprises.

This just in, courtesy of Wipneus from the Arctic Sea Ice Forum (ASIF) sea ice area and extent data thread: an update of the Basin-only graphs. On NSIDC's numbers, both area and extent are tracking close to several low but non-record years. (Note that this excludes Barents and Greenland Seas and the Canadian Arctic Archipelago, which collectively aren't likely to alter the trend much in the next month).

Despite the big swing this year in overall ice state, the lack of extreme conditions during the Arctic summer offers a good opportunity to evaluate the performance of prediction models. Weather is always a wild card that can throw off any or all seasonal models in any given year. The head-turning decline in sea ice in 2007, for example, was caused mainly by sunny skies and a prolonged dipole: high pressures over the Beaufort Sea, combined with lows near Siberia, pulled in considerable warmth from the Pacific side and increased ice export to the Atlantic. In 2012 an intense and long-lasting cyclone caused an extra decline in ice cover during August en route to a record low minimum.

This summer, in contrast, the weather has been less extreme. Conditions have tended toward cloudy and cool, with low-pressure systems meandering about the Arctic but no major cyclones. Wave action has been strong from winds over open water, though only intermittently. Ice export through the Fram Strait was somewhat high in May, but since dropped to near zero. Surface albedo and ice concentration, two key measures of melting potential, are currently near the middle of the range for recent years.

Because the sea ice models use different variables and weightings, they can offer insights into which features most affect a specific melting or freezing season. This post is not meant to be a full exposition of the models or variables, but will give a brief overview of categories and a closer look at some models that have garnered discussion in the ASIF. The focus here is on seasonal prediction models, rather than long-term models as at the Polar Prediction Workshop 2017. (Some of those models do cover seasonal cycles along with longer-term conditions: see e.g. section 3.4 of neXtSIM: a new Lagrangian sea ice model.)

A good starting point is the Sea Ice Prediction Network (SIPN), which uses four categories of models in its September Sea Ice Probability (SIP) forecasts: Heuristic, Statistical, Mixed, and Dynamical. As Figure 2 from their July Sea Ice Outlook shows, the statistical models have a higher prediction than other methods on average for September ice extent. Three of the four methods have higher average predictions as of July than the June predictions had. The month-to-month increase reflects more favorable conditions for ice retention as low pressure patterns came to dominate the central Arctic (see Figure 9 of the same report). Weather patterns have remained generally similar since the latest SIPN report, so comparing current ice trends to those projections might offer insights into the predictive value of specific variables.

One example of the statistical method, from citizen scientist Rob Dekker, is based on seasonal heat absorption in the northern hemisphere. It uses three variables: land snow cover, ice concentration, and ice area. Though the first one seems like a loose proxy for snow cover on sea ice, it does affect the total heat budget. Rob's correlation work demonstrates that "the June land snow cover signal is clearly present in the September ice extent numbers." The graph shows the best fit of the three variables in a hindcast (using current data). Based on June 2017 data, it predicted a September average extent of about 5.4 million km2.

Rob noted a few weeks ago that ice extent measures were tracking below the model's forecast. This indicates the influence of variables that are not in the model. Perhaps some of the ice is thin or mobile enough to continue melting even under relatively cloudy and cool conditions.

Though Rob uses this model to forecast September NSIDC extent - the basis of the SIPN comparisons - his calculations show it has actually had greater accuracy in predicting NSIDC area. This presents an interesting test for 2017. Much of the ice that is currently part of the area measure is relatively thin, susceptible to bottom melt and wave action even while surface temperatures decline. If it disappears in the next month, the ratio of area to volume would show a steeper drop than usual. So if the Dekker model produces an atypical miss in its area prediction, it would indicate that thickness distribution is particularly important to the seasonal minimum. We'll circle back to this point at the end of the post.

Also in the statistical category is the model developed by the late Drew Slater, now maintained by Andrew Barret's team at NSIDC.  They run the forecast with a 50-day lead time, though it has predictive skill out to 90 days.  It uses only sea ice concentration as an input, computing the probability of future concentrations by location.  Because forecasts for all points are independent in space and time, the model can adapt to changing conditions.  Observed values were running below the forecast in recent weeks, but the gap has narrowed with a flattening of the extent graph.  

The track record of this probabilistic model underscores the findings of a recent paper by Kashiwase et al. Their research examines the effects of more mobile ice cover since 2000. Even though it causes only a small reduction in concentration - the input to the Slater model - "this trigger accelerates ice melt through the enhanced solar heat input over the open water fraction." This form of ice-ocean albedo feedback proves to be a major influence on both seasonal and interannual variations in sea ice.

A different take on albedo is at the core of the CPOM (a.k.a. University of Reading) model. Based on research by Dr David Schröder and others, it models the amount and location of melt ponds in May and June. This feature is a key factor in "melting momentum" because it is both a cause and effect of ice loss. Pond area is likely to be greater with early-season heat and moisture advection, under clear skies near the solstice, and over first-year ice (which tends to be flatter than multi-year ice, so meltwater can spread out more). The ponds also have much lower albedo than bare ice or snow cover. Their formation typically leads to thinner and/or lower-concentration ice later in the season.

The CPOM contribution to SIPN forecasts a September average extent of 5.1 million km2. Partly because of low ponding and melting in the eastern Arctic, the accompanying text specifically states that "...we predict the September ice extent to be quite large in spite of the lowest Arctic ice volume in recent months."

In more recent research, Schröder and others find that forecasts based on sea ice concentration and melt onset have similar to higher skill values compared to the melt pond model. They also show that some forecast models are better than others at different times of year. Such a comparison will be useful to revisit after melt season 2017, which began with extreme anomalies yet appears to be headed toward a fairly normal conclusion

This year has seen a lively debate on ASIF about whether factors that favor ice retention - such as snow cover, compactness, melt ponding and cool/cloudy weather - are sufficient to offset low volume and large expanses of first-year ice. While the question is not yet settled, Wipneus and others have taken a stab at it based on the PIOMAS model. Ice that is thin at midsummer is more likely to melt by the September minimum. Solid lines in the graph show modeled thickness distribution. The dotted line does not compare directly to the SIPN numbers: it is for a specific date, not the annual minimum, and is for area rather than a monthly average extent.

Wipneus' take is:

After 2007 the ice below 0.26 m is almost always gone at the September minimum. For most years that is it, but in 2008, 2012 and 2016 thicker ice (up to 0.71 m) melts as well.

This interpretation of the area graph as a step-change, rather than a continuous decline, makes sense in relation to ice age. In 2007, enormous quantities of multi-year ice disappeared, and since then, first-year ice has constituted a larger share of most PIOMAS thickness bands. It is more prone to melting than multi-year ice, partly because of lower brine content. So with increasing heat in the Arctic over time - or with weather in any given year that's more conducive to melting - the area measure is more likely to overtake the next-thicker PIOMAS band in the graph.

Though the PIOMAS model indicates that overall volume anomaly has bounced back from its extreme in the last few months, thickness distribution might still matter for the season's minimum area and extent. Its impact may be even more telling for the lack of big late-season storms, which would stir the waters and muddy the picture. In another month we will have more information to revisit the fundamental question that Rob Dekker posed: "does the Arctic summer melt 'extent' or does it melt 'volume' ?" A single year will not be enough to resolve the thin-ice debate, but 2017 has already advanced our understanding considerably.

Comments

[Jim Hunt]

A very interesting winter, followed by a very interesting summer! And the show's not over until the fat lady sings.

I'm not about to make such a prediction, but I wonder if Lewis and/or Neven would care to predict the value (and date?) of the summer 2017 minimum for one or more Arctic sea ice metrics?

Here's some more suggestions from the ASIF:

http://forum.arctic-sea-ice.net/index.php/topic,2126.0.html

[iceman]

At this late date and under near-neutral conditions, it shouldn't be too hard to pick a number. But I'm still baffled by the factors in opposition: concentration vs. thickness distribution, among others. Thinking on a V-shaped extent minimum, which would require compaction continuing beyond next week and a further push from bottom melt. So I'll guess 4.1 million km2 for the JAXA daily min you linked to.

Interestingly, on the other poll's measure of NCSID September average extent, the SIPN August Report's median of 4.54 million km2 is up slightly from July's, but still low enough for a 3rd-place finish.
https://www.arcus.org/sipn/sea-ice-outlook/2017/august

[Artful Dodger]

Hi Lewis,

Good summary of the year-to-date. Thank-you for all your effort!

Slight errata: you wrote "first-year ice ... is more prone to melting than multi-year ice, partly because of lower brine content."

Of course, that should read "higher brine content" for 1st-yr sea ice.

Brine rejection continues for about 3 yrs by when MYI seaice is mostly brine-free and is highly resilient to melting. Most MYI is lost through transport out of the Arctic via the transpolar drift, rather than melting out in-situ.

Or at least it used to work that way before the southern Beaufort sea turned into the place MYI ice goes to die. Oh, the changes we have seen in our life times.

Cheers,
Lodger

[Hans Gunnstaddar]

http://www.bbc.com/news/science-environment-41037071

'First tanker crosses northern sea route without ice breaker'

The specially-built ship completed the crossing in just six-and-a-half days setting a new record, according to the tanker's Russian owners.

"The 300-metre-long Sovcomflot ship, the Christophe de Margerie, was carrying gas from Norway to South Korea.

Rising Arctic temperatures are boosting commercial shipping across this route."

[Jim Hunt]

Hans - There's much more on that story on the ASIF:

http://forum.arctic-sea-ice.net/index.php/topic,854.msg123499.html#msg123499

[Hans Gunnstaddar]

Interesting thread, thanks Jim.

[Hans Gunnstaddar]

https://seaice.uni-bremen.de/data/amsr2/today/Arctic_AMSR2_nic.png

This is about the arctic so hopefully on topic enough, although not directly related to the guest post.

Indulge me for a moment for an interesting observation. If you look at the above link you'll notice a round wide open expanse of water along the Gakkel Ridge, in an area of ice concentration that is otherwise at much higher concentrations.

In a YouTube Video I saw a while back there was mention and a map showing a spot along the Gakkel Ridge that has in recent years supposedly been emitting mantle methane, and I think it coincides with that open water location. My question is this; Once the seasonal melt reaches a certain point in which the ice has been weakened, could rising methane bubbles push away the ice to form a round shaped open area? I ask in part because that same round area has been there for a few weeks now. Below is a map showing the Gakkel Ridge.

http://4.bp.blogspot.com/-jywSP0Wq3yc/UmYWKs8k1aI/AAAAAAAALsA/IR6NvaBd_zc/s1600/Arctic+Ocean.jpg

[Rob Dekker]

Thank you Lewis, for this nice overview of prediction models.

Early in the melting season, we saw this large negative anomaly of volume, which was caused by an exceptionally warm winter.

Luckily, the summer registered a rather cool melting season, which I think is partially caused by that large positive land snow anomaly that Rutgers Snow land recorded :
http://climate.rutgers.edu/snowcover/chart_anom.php?ui_set=1&ui_region=nhland&ui_month=6
Some 4 million km^2 of land snow more in June than last year.
That has got to leave a significant dent on absorbed heat in the Northern Hemisphere.

So I think that the negative volume anomaly in winter and the positive land snow anomaly in summer balanced each other out and we end up with a September SIE that is neither exceptionally low nor exceptionally high.

Other years we may not be so lucky.

[Wade Smith]

Not exactly the right place to talk about this, but it's semi-relevant.

[No, it's not the right place to talk about this and so I'm deleting the other comments as well; you can discuss Harvey over on the ASIF; N.]

[Wade Smith]

Wait, sorry, but what is that? I don't know how to get to that site...

[It's here; N.]

[iceman]

Good catch Lodger, thanks for the correction. I wonder whether there is some way to derive data from the Tschudi ice age model to quantify the change in MYI melting within the Arctic vs. being exported. The animation on Climate.gov certainly supports your point.
https://www.climate.gov/news-features/videos/old-ice-arctic-vanishingly-rare

Rob, generally agree with your assessment, though it omits the unknown (and possibly greater) influence of snow cover over sea ice. Would that we had better observational data.

[Rob Dekker]

Lewis, thanks.
One point of clarification : I include land snow cover in my model since it clearly affects albedo and thus the heat budget over the Northern Hemisphere.
Snow on ice affects the albedo much less and thus in my opinion is not that important for the heat budget.

[Rob Dekker]

As Wipneus reports on the ASIF :

According to NSIDC NT sea ice concentration, extent in 2017 is currently the 8th lowest, area is 7th.

Looks like we are not going to see any records broken this year.

[Hans Gunnstaddar]

https://seaice.uni-bremen.de/data/amsr2/today/Arctic_AMSR2_visual.png

Ok, everybody ignored my mantle methane, Gekkal Ridge theory about a round open water area. Well, it's still there. Look at the above link and then enlarge the view (control +) and you'll see it's about as round as could be expected considering that area is surrounded by high concentration ice. If it isn't methane bubbling up opening that part of the ice or clathrate methane, remaining in that exact same location week after week, then what is causing it?

[Jim Hunt]

Hans - You may be interested in this?

http://forum.arctic-sea-ice.net/index.php/topic,1834.msg125795.html#msg125795

There's little support for the polynya occurring every year, which restricts the space of possible explanations.

[Hans Gunnstaddar]

"A case could be made that 2012 and perhaps 2013 had a persistent polynya in a fairly similar place to 2017's. However there's little support for the polynya occurring every year, which restricts the space of possible explanations."

How about seismic activity along the Ridge is not constant, but instead random.

https://seaice.uni-bremen.de/data/amsr2/today/Arctic_AMSR2_visual.png

https://volcanohotspot.wordpress.com/2017/06/16/1999-2001-eruption-sequence-in-the-gakkel-ridge/

Coincidentally the polynya just happens to be located directly above the Gekkal Ridge? I think this is one to keep an eye on.

There is a woman with a very distinctive voice (but I can't recall her name) who did a YouTube Vid on GW, and I can't find it now, but she suggested mantle methane was rising along the Gekkal Ridge and had a map showing the location and it is the same location as the polynya. I'm going to keep trying to find it.

[Rob Dekker]

Hans, About this polynya, I presented a different theory, that it may be caused by the Coriolis forces driving the Atlantic warm water upward while making a U-turn at that very location against the Gekkal Ridge :
http://forum.arctic-sea-ice.net/index.php/topic,1834.msg125728.html#msg125728

[Hans Gunnstaddar]

Hi Rob, yes, saw that first in the link Jim provided and also in your link. That certainly is possible.

Would be nice to go to that location and do some on site research to try and confirm, but haven't got access to a research vessel - lol.

By the way, congrats on your more accurate prediction of this years melt vis a vis above average snowfall. CAB ice held on Pacific side much better than I expected with how fast it was declining earlier. Wonder if the repeating 5 year cycle would have hit a new low if there had been average snowfall.

This is all good though because it dodges a bullet or as you put it a possible cannonball with how low volume had gotten during winter. More time to deploy more renewables.

[AnotherJourneybyTrain]

Congratulations to Rob,

"You stole the show .."

I still think there will be some sort of knock-on-effect from the low volumes experienced earlier in the year but I suppose this year simply becomes a time for re-engineering of theory.

(..back to the drawing board...)

[Hans Gunnstaddar]

https://www.youtube.com/watch?v=a9PshoYtoxo

Ok, I found that YouTube Video by Jennifer Hynes (distinctive voice) and part of it is on Mantle methane. Take a look first at 1:08:40 which shows a methane emissions chart with high levels of emissions from the Laptev Sea (from earthquakes occurring along the Gekkal Ridge, resulting from increased seismic activity from isostatic rebound of Greenland from less mass holding it down). Watch it from there if you like or you can jump to 1:11:35 and see a red dot she's pointing to on a map where there have been high levels of methane emissions from the Gekkal Ridge. That spot appears to line up well with the polynya in question (from earlier posts).