Thursday 18 June 2015

Are we facing a crash in 2015?

This blog post was based on data from the US Navy's ACNFS model (known as HYCOM in this post). It has turned out that the massive melt shown in the model was due to a technical glitch. This post is kept as I don't delete posts, but it is no longer considered valid.

Correction statement with links to details here.


It's bad form to blog on consecutive nights, but I'm now going to say something that has been on my mind for over a week.

Last year, and the year before, I recollect a lot of excitable discussion at the forum in June about how we were going to see a crash in sea ice. A lot of talk about how the weather would shift and massive losses would start, of course as we now know, that just didn't happen.

This year I'm seeing the same sort of discussions over at the forum, but I am more persuaded by hard numbers than qualitative arguments. I am by nature conservative so I am not going to change my opinion now, I'm going to wait until I can look back at June in its entirety. But there is one strand of evidence that has me suspecting we might be in for a crash of the same order as 2012.

In my previous post I showed how compactness has finally started dropping. But the same thing happened this time in 2013, and look where that summer took us. Drops in compactness tell us nothing about the coming melt season at this point in June, because absent other information they can go either way, drops can be followed by periods where compactness ceases to drop as much. Extent anomaly remains rather level, while area anomaly falls, such a situation is not exceptional for June.

It is only the US Navy's HYCOM-CICE model that is getting me quietly excited about the possibility of this year seeing a crash like 2012 and I don't just mean in terms of area or extent, I mean in terms of the pivotal impact of a storm.

I posted a few days ago, closing with this image.

The region indicated, from Laptev to Chukchi being my 'watch this space' region for the 2015 season. Now here is what the model is showing for the recent past and up to this time next week.

On around the 10 June a massive storm came from Barents over the pack towards Chukchi, discussed as a future point of interest in Neven's blog on 4 June 2015. This is shown in GFS, here shown on 10 June 2015 from Wetterzentrale.

That plot shows geopotential height in colours and sea level pressure in isobar contour lines. In the HYCOM model the impact of this was, as seen above, massive. Indeed it was far more massive than I had expected at the time.

Is there any reasonable expectation that what HYCOM predicts could be real? Below I show HYCOM for 18 June 2015 as predicted on 10 June 2015 and the actual for 18 June 2015.

Considering that the upper image is for predicted weather and the lower image is for the model run using actual weather, the similarities are close. But this is just for one day (and the other days I have checked are for a few more days), furthermore this doesn't rule out weather over the coming week changing from what is forecast. However consider the ice state shown for 25 June 2015, this is the worst state for the middle of June that I can recall seeing for any index.

For the sake of argument, let's accept this as a true picture of ice state next week. This state is under peak seasonal insolation, and suggests a lot of dispersed ice in the upper left quarter of that plot. It is not unreasonable to consider that almost everything under 2m thick will be gone by this September as ice/ocean albedo feedback works on the ice. That could leave a similar extent of ice as in 2012, possibly less.

It might be supposed that this would support claims of snow cover driving the ice loss in summer. One problem with that is that in 4 June 2014 ice thickness was similar to 4 June 2015.

The other problem is that the swath of thin ice started with the storm around 10 June and clearly evolved from the immediate impact of that storm. The proponents of the idea that snow cover will drive this season will have to await another year to test their theory. That's not a problem, if the theory is right, it will be successful again under better test conditions in another year.

HYCOM may not be correct. If that is the case we will know by the end of this month, the massive thinning and dispersion predicted for 25 June will be clear by the 30 June as highly abnormal concentration in the sea ice maps, and will result in strong decline in compactness across Chukchi, East Siberia Sea, and the Laptev Sea, it will likely be clearly abnormal in MODIS. At present, in my opinion the case for such corroborative evidence is not made.


An Anonymous commenter has raised the issue of a disagreement between the two HYCOM versions, HYCOM-Arctic & HYCOM-Global. In the above I used HYCOM-Global, here is an update with HYCOM Arctic.

If HYCOM is correct then the ice state this year is far worse than 2012 across the entire Siberian Coast, and as 2012 was the last major crash it remains reasonable to ask if 2015 will see similarly low sea ice by September. However HYCOM Arctic is significantly thicker than HYCOM Global. But as it is only 22 June it is too early to tell from the sea ice concentration data if this is a real state, we should know by the end of this month.

Anyone thinking about accusing me of bias in selecting the thinnest model should read my posts on the Slow Transition. ;)


Peter said...

Exciting times indeed! I would however take issue when you say the following:

s there any reasonable expectation that what HYCOM predicts could be real? Below I show HYCOM for 18 June 2015 as predicted on 10 June 2015 and the actual for 18 June 2015.

The second sentence doesn't follow from the first - there is as yet no comparison to "real" for the ice thickness data. Yes, you can conclude that the weather forecast was accurate, and therefore the forecast and the nowcast were similar. But, there is no independent check on the model predictions.

The only buoys in the area are 2015B, 2014F (Obuoy10) and 2014G (Obuoy12), and none of them show a dramatic recent thinning, either by the thermistor strings, sonar or webcams.

Over the next couple of weeks it would be well worth looking at the forecast and nowcast ice concentration maps as well as the thickness maps, since the concentration model prediction can be compared directly with microwave observations.

Neven said...

Yes, the Chukchi and ESS are going to get punched hard in days to come, with clear skies and relatively high temps. The same goes for Baffin and Hudson Bay (and Greenland). But other than that there is not much movement. If 2015 wants to get serious, it will have to start showing SLP and SAT patterns like we saw in the top years, not just on the fringes.

Either way, my focus is on that dispersed multi-year ice on the Pacific side of the Arctic, and the value it represents volume-wise. The rebound that lingered on after the freezing season, might vanish again, leaving us in the same position of the past couple of years, where one freak year could really obliterate the sea ice close to ice-free levels.

Chris Reynolds said...


That's why I said "HYCOM may not be correct."

I know it's not a popular position, but I have serious doubts about the practical utility of buoys and whether they tell us anything more than conditions regarding the particular lump of ice they are mounted in.


In a previous post I pointed out the dissimilarity between 2015 and two of the 2007 to 2012 years, as you imply it stands for the other 2007 to 2012 years, though not all showed the pattern in early June. However what you show as a dipole between the Canadian and Eurasian Arctic on the forum may be the beginning of a more typical 2007 to 2012 set up. If Blaine is correct as commented on your blog, loss of sea ice from Chukchi to Laptev may drive the formation of the 2007 to 2012 set up.

Anyway, there is nothing to do now but wait and see if what HYCOM suggests will happen will indeed happen.

I don't know if I could manage it, but I am tempted to do other things until the end of June and see what has happened when I update then.

Chris Reynolds said...

BTW - Peter, good idea regards the concentration maps from HYCOM.

Neven said...

I don't know if I could manage it, but I am tempted to do other things until the end of June and see what has happened when I update then.

Forget about it, you're not going to do that. And neither will I. ;-) :-B

Peter said...

Hmm, the buoys are by and large in the spine of older ice that doesn't show so much change in HYCOM, but I'd have expected at least some melt ponding and snow loss even if we discount the thickness measurements.

Nightvid said...
This comment has been removed by the author.
Nightvid said...

"One problem with that is that in 4 June 2014 ice thickness was similar to 4 June 2015."

How is this a problem? None of us are saying that snow loss leads to a thickness of ice reduction with an onset that early in the season. If you still had the same thickness on July 4, then you might have a point. But June 4? No.

"The other problem is that the swath of thin ice started with the storm around 10 June and clearly evolved from the immediate impact of that storm. The proponents of the idea that snow cover will drive this season will have to await another year to test their theory."

How does your argument not apply equally well to other years like 2012 and 2013, which still managed to end up as suggested by June snow cover despite the confounding variable?

If one model lines up with the data but not with the combination of another model and another assumption (i.e. that thin ice will melt while thicker ice won't), then that suggests not that the first model is wrong, but rather that either the second model is wrong or that the auxiliary assumption is wrong.

sedziobs said...

Thanks Chris, I enjoyed the analysis.

Having watched HYCOM models and read through the documentation over the years, I think the new GLB product is an excellent model as far as daily output and forecasts go. Large scale fluxes and weather/concentration assimilation seem to be adequately implemented, with improvement efforts now focused on micro-scale processes such as local fracturing and surface melt water flows.

One local process that I think still needs attention is Nares Strait. HYCOM doesn't seem to allow for an arch there, and ice is seen flowing through the strait at all times, likely causing an underestimation of thickness near Lincoln Sea. Obviously that will not affect any of your analysis in this piece, but I just wanted to point it out.

As en engineer, I am quantitatively minded like you, so I really appreciate all of your contributions.


sedziobs said...

Nightvid, I can't speak for Chris, but I believe he is simply saying that the cyclone has created a large enough impact that 2014 can no longer be considered an adequate control. Directly comparing 2015 extent with snow cover losses would suffer from omitted variable bias by ignoring cyclone impacts. Now, I think any two given measures directly compared in one season would suffer from the same bias (for instance, you could feasibly say that "due to anomalously low snow cover in 2015, the proponents of the idea that HYCOM thickness will drive this season will have to await another year to test their theory "), but I think that's what Chris is getting at.

Neven said...

Two years ago in June the ACNFS model also forecast a spectacular thinning event due to a cyclone (though less spectacular than this one). I wrote about it in this blog post: If this is real...

A bit of convergence did happen, but it didn't quite pan out as forecast. I can't help but be wary of the ACNFS and TOPAZ models. But perhaps this year is different.

Neven said...

You also wrote about it here, Chris, BTW.

Chris Reynolds said...


This year if HYCOM is right, what may lead to a record low is a storm, not snow cover. My point regards 4 June is merely that it was only after that, with the storm that the deterioration started, exactly where the storm hit. Your point regards 2012 confirms the suspicion I have, because a major factor in the final ice edge in 2012 was the late August storm.

Sedziobs states "I can't speak for Chris, but I believe he is simply saying that the cyclone has created a large enough impact that 2014 can no longer be considered an adequate control" S/he gets it, that is exactly what I am saying.

You have two or three years of data.

I have two and a bit cycles of the August Extent Oscillation, I conclude this may be worth following. With the 'may' in italics to stress it.

I have two recent instances, and some past instances supporting my proposed early warning for sea ice crashes.
I start off by saying "Time will tell whether it is of any use in the future."

I waited for four years (2011, 12, 13, 14) before being persuaded that the April volume 'stalling' at a volume to be expected for 2m thickness was significant.

So yes, I think you are arguing from a weak case, but are being too strident given the small body of evidence you have.

Furthermore, I have seen something very similar to what you and Vergent are seeing.
A poster came up with a similar prediction technique. When you have as many as four years of data you might want to try the second test used there.

But there is more. In "Abrupt Climate Change: An Alternative View" Carl Wunsch demonstrates how in timeseries with similar power spectra it is easy to find false correlations by virtue of the similar power spectra. Hold that idea for a moment, now consider how it references to an enclosed ocean, with a set geometry, consistent processes causing winter cold and ice thickness to be similar from year to year, consistent processes meaning that the Atlantic ice edge recedes less than the ice edge off the Alaskan and Siberian coasts.

Expanding Wunsch's insight shows us that apparent visual correlations (such as you use over short sets of data) within the Arctic Ocean are more likely to be illusory than not. And that to reach significance the correlation must be stronger than might be assumed because the degrees of freedom are more limited than might be assumed.

Yes I know how you and Vergent would interpret a 2012 style low this year, but I think you would be wrong to interpret it in such a manner. That is why I addressed the issue as early as I suggested that we might see a 2012 style low.

However, as I have noted elsewhere, if one accepts the evidence of previous years, the TOPAZ snow distributions that Vergent has been posting clearly imply that this year will see a crash well below 2012.

Chris Reynolds said...

Sedziobs (Brian),

Thanks for making clear what I failed to do when I wrote the post last night - yes my concern is that this year is already a poor control for the snow idea.

HYCOM CICE is good, better than the old version I used to look at, but along with Nares the ice still seems to be a bit 'sticky' as shown when there are breakups from ice moving westwards from Banks Island. Satellites show parallel fracturing where HYCOM shows open water forming. But overall it is far better than it was.


I'd forgotton I'd mentioned that, but it was 2 years ago! In defence of this post I wouldn't have written this post if what HYCOM is showing was so small. ;)

Leave it until June? You're probably right, but I have been enjoying my spectrum analyser, and have bought myself a USB mixing console. This time last year it was totally work and sea ice. Not so this year.

Chris Reynolds said...


I should be clear after all the above. When Barnes showed that the Francis & Vavrus paper on 'wavier jetstreams' was not showing what it claimed to, my position was that I thought it was happening, but it was not proven by Francis & Vavrus 2012.

Likewise I think you have a point, that initial snow cover plays a role, but I don't think you have demonstrated it robustly, and I think you might be over estimating how strong a role it plays (although with such limited data it is hard to be sure how important it is).

Anonymous said...

It looks like HYCOM now has a completely different look/concentration than even their "actual weather" model run on 6/18:

Even today's (6/21) looks to be in better shape than either of the 6/18 ones Chris posted:

I'm sort of new to this, so, am I missing something? How can the disparity be so great?

george_formerly_of _the jungle said...

Yes, the Navy has 2 different models. GLB is the Global HYCOM model. I clicked on the link and it asked for a password. The CICE model which was developed by Los Alamos Lab is available without a password and account. It shows more ice thickness than the HYCOM model. The CICE model does not agree with other thickness models such as PIOMAS. CICE model thickness is greater than other models.

CICE is good for comparing different years since 2012, but don't trust the absolute value.

Anonymous said...

Thanks for the thorough explanation, george! Has cleared things up for me quite a bit.

Chris Reynolds said...

Anon, George,

As a result of following the two models and discussion at the forum I had switched to using GLB (global) as opposed to ARC (Arctic). Now I find that GLB is password protected...

It doesn't change the basic conclusion of the post though, and using ARC I can show a comparison between 2012 and 2014 for the current date. GLB only gave me back to 2013 which is why I used 2014.

I'll update the post with a PS.

Anonymous said...

Hey, thanks for the update, Chris! Just as striking a contrast in that comparison/juxtaposition.

Chris Reynolds said...

Hi Anon,

It is striking, but I'm not yet convinced it's real. If we take the HYCOM ARC images you originally linked to it suggests around 1m thick. But that won't be 1m all over, and it should mean that area within the grid boxes 'marked' as extent will fall - that is a drop in compactness. My guess is that by the end of this month it should be apparent in the satellite data, from which compactness is calculated. So I'm not going to get into a running commentary on this blog - changes over a few days can often indicate either way.

At present I don't see compelling information either way to say whether a very low September is going to happen or not, so I'm not getting excited right now. But I think HYCOM raises the question in a way none of the other data does.

Nightvid said...
This comment has been removed by the author.
Nightvid said...


My suspicion (just a WAG at this point) is that the thinning of ice in the GLB simulations in response to cyclones is heavily modulated by the local SST's at the time of cyclone passage. I suspect (again just a WAG) that the reason this year's cyclones cause so much thinning in the model is that the SST's are abnormally high for this time of year, as a result of the early snow retreat and subsequent albedo loss of the ice.

But unfortunately, I can't really test this idea without being able to re-run their simulation under different conditions, so it remains just a guess and nothing more.

It would be very nice if the models' source code were available, but oh well.

If the transition to ice free Arctic summers is not too fast, we should still be able to test these ideas with more data.

Chris Reynolds said...

Hi Nightvid,

This page has a link to the source code for PIOMAS.

But it is behind a password. I've never asked Dr Zhang if I could have the password because a) I suspect it would take a lot of work to get it running b) I know there would be a good chance of me messing up somehow as a victim of Dunning Kruger.

In my May post on four dipoles I suggested that influx of Pacific Water might have interesting effects, I too suspect that SSTs (and deeper through vertical mixing) might be a large player.

A fast transition would allow us to get more data to really get to grips with various factors in the post 2007 New Arctic Ice Pack. But a fast transition would be more exciting. ;)

Nightvid said...


I am commenting again here because the nature of one of your comments is such that "waiting until September" won't resolve the issue.

You argue that I have not robustly demonstrated the snow hypothesis. Fair enough - but then, neither have you, by your standards, demonstrated the cyclone hypothesis - because you have not included previous years with cyclones in your analysis, apart from a brief mention of 2012.

By the standards you are attempting to hold my analysis to, you would have to look at the statistics of cyclones and September ice over the last 5-10 years before you have a good analysis.

This is very different from simply showing the storm breaking up the ice in 1 year, or even in 2 years (2012 and 2015).

Chris Reynolds said...

What I have done is show a substantial change in the behaviour of anomalies after the storm hit, but yes, only in a model can it be seen how things may play out without the storm. I accept the PIOMAS message - that drops were to come anyway. I just suspect it underplays.

None of which is like your snow hypothesis, you're not assessing a one time event, you're proposing a persisting mechanism. When you have some more years data (or can get earlier data for a long hindcast performance analysis), I'll consider it again. There is nothing wrong with being patient.