Latest PIOMAS Volume update

[oren]

Making a probable summation, Beaufort + ESS + CAA should be around 250 km³ by the minimum (compared to 50km³ for 2012), and Greenland Sea another 250 km³. Adding the CAB should get the total to less than 4400 km³, with the lower limit not well defined.

Overall, I expect 2019 to pass 2016 (4400 km³) and 2011 (4300 km³) and reach 2nd place in volume. I'd be quite surprised if it manages to break 2012's 3670 km³ record.

In reality, Beaufort + ESS Laptev + CAA was a low 180 km³, and with Greenland Sea the minimum volume outside the CAB was a meager 290 km³ (similar to 2012 on minimum date and a bit lower than 2011's 320 km³). This helped to clinch the 2nd lowest title, which totaled a very respectable 4050 km³, but as expected did not break 2012's record.

All in all, the predictions were not far off the mark (but had I known in advance that August melting would be so poor I would have predicted higher numbers).

[Wipneus]

PIOMAS gridded thickness data has updated. Volume calculated from thickness was 4.57[1000km³] on 30th September, second lowest for the day.
The 2019 minimum was around the 10th, 4.06[1000km³]. As these number are not exactly the same as the official PIOMAS volume numbers, wait for the release of those.

Here is the animation.

[Wipneus]

The volume and volume-anomaly graphs.

[Wipneus]

Some people have use for the updated regional data files:

monthly:
https://sites.google.com/site/arctischepinguin/home/piomas/data/PIOMAS-regional-monthly.txt.gz
daily:
https://sites.google.com/site/arctischepinguin/home/piomas/data/PIOMAS-regional.txt.gz

[Wipneus]

More graphs to prepare (the ones with annual minimum), but that will be much later. Paint job to do.

[blumenkraft]

Thanks a million, Wipneus!

[Shared Humanity]

Your delivery of data in a consistent manner like a finely tuned Swiss watch is what makes this site so valuable to the rest of us. I cannot thank you enough. I now anticipate the other members here who use the data to provide other insights.

I am a spectator here and value all of you who do the work you do. It is not lost on me how much time you take to do this.

[Jim Hunt]

PIOMAS gridded thickness data has updated.

Thanks very much Wipneus.

Since I’m currently involved is a heated “debate” about Arctic sea ice trends over at ex Prof. Judy’s, here’s an added bonus. The September monthly volume trend:

http://GreatWhiteCon.info/2019/10/facts-about-the-arctic-in-october-2019/#Oct-05

[slow wing]

Thanks Wipneus!

The September monthly volume trend:

Thanks for this plot. There is clearly some year-to-year correlation but, that aside, it seems to me that the Arctic sea ice volume data seems reasonably well described by a linear decline.

Eye-balling, it looks like it reaches zero volume in about 10 years time (and may fluctuate down to that sooner). I wouldn't rule that out!

[Stephan]

Thank you Jim for the link. [http://GreatWhiteCon.info/2019/10/facts-about-the-arctic-in-october-2019/#Oct-05]. From the two graphs it can be easily ruled out that sea ice thickness (expressed as volume divided by extent) Sept 2019 has been lower than in Sept 2012.

[El Cid]

The September monthly volume trend:

The trendline on this chart is obviously faulty.

A trendline is right if the datapoints are more or less randomly distributed above and below the trendline - this is simple statistics. If the datapoints are not randomly distributed above and below the trendline then it is
a) not a linear trend
b) the dataset is not right

In this case I suspect a), since from 1985-2005 all datapoints except 2 are above the trendline and then 6 consecutive ones are below. This is clear signal that this linear trend is not OK.
Also, volume is not going anywhere these past 10 years. 

b): Everyone uses the data from 1979 and creates linear/polinomial etc.  trends starting there just because that is where the dataset starts. That is a pretty terrible reason to start the trendlines from that point. What if you had the data since 1950? What if only from 1990? Would you get the same result? of course not. First, you need to have some proof as to when (natural processes leading to) the volume reduction  actually started and start the linear trend from that point (eg from 1988 or what have you).

Summarizing: It is obviously statistically NOT a linear trend and even if it were the dataset used for them is probably not the right one.

[Jim Hunt]

The trendline on this chart is obviously faulty.

Did I mention that "I’m currently involved is a heated “debate” about Arctic sea ice trends over at ex Prof. Judy’s"?

The displayed linear trend was created using the Excel "TREND" function. Surely you don't want to start a "debate" with Microsoft too?

[binntho]

The September monthly volume trend:

The trendline on this chart is obviously faulty.

A trendline is right if the datapoints are more or less randomly distributed above and below the trendline - this is simple statistics. If the datapoints are not randomly distributed above and below the trendline then it is
a) not a linear trend
b) the dataset is not right

I would caution against calling any trendline "right" - it's only a model, an attempt to decipher underlying tendencies and predict future values.

In this case, 18 datapoints are under the line and 22 above. Pretty near random. Long stretches are above or below, which can be explained by inter-annual memory in the system (i.e. a low year tends to be followed by a low year and vice versa - not necessarily true at all times but perhaps more likely to be true than not).

The linear trend is clearly not perfect, but I'd suggest that it is the best we can get given the paucity of data points. Any eyeballing of other behaviour in the underlying data may seem compelling, but cannot be shown to be statistically valid.

The claim that "volume is not going anywhere" in the last 10 years is not correct since it's been swinging pretty much over that period, going both up and down. An outlier in 2012 + few datapoints + system memory may well create what looks like a stall, but isn't.

[Stephan]

I kindly ask you not to overestimate the effects of the nature of the trendlines. I experimented with linear, logarithmic and polynomial trendlines. In the end (e.g. "When will a certain dataset reach a certain value?") the differences are small, within a few percent or even less of error margin.

[Jim Hunt]

I would caution against calling any trendline "right" - it's only a model, an attempt to decipher underlying tendencies and predict future values.

The claim that "volume is not going anywhere" in the last 10 years is not correct since it's been swinging pretty much over that period, going both up and down. An outlier in 2012 + few datapoints + system memory may well create what looks like a stall, but isn't.

Axel Schweiger is one of the authors of this paper, referenced over at Judy's place:

https://doi.org/10.1038/s41561-018-0256-8

The relative contribution and physical drivers of internal variability in recent Arctic sea ice loss remain open questions, leaving up for debate whether global climate models used for climate projection lack sufficient sensitivity in the Arctic to climate forcing.

Here, through analysis of large ensembles of fully coupled climate model simulations with historical radiative forcing, we present an important internal mechanism arising from low-frequency Arctic atmospheric variability in models that can cause substantial summer sea ice melting in addition to that due to anthropogenic forcing. This simulated internal variability shows a strong similarity to the observed Arctic atmospheric change in the past 37 years.

[Wipneus]

Continuing with the graphs...

Here is the updated Fram volume export graph. September export was below normal.

[grixm]

Did I mention that "I’m currently involved is a heated “debate” about Arctic sea ice trends over at ex Prof. Judy’s"?

Can you post the raw link? Archive.is doesn't work with 1.1.1.1 DNS servers.

[Wipneus]

The graphs updated with the annual minima.
Here is my collection of possible math fits to the timeline.

This graph was originally as a reply posted on the ASI blog. In those days the minima closely followed an exponential decline.
Not so in recent years. Every year the extrapolated day of zero ice is postponed about one year. Similar for most other mathematical functions. More about that in the next posts. 

A click will give you a slightly better image.

[Wipneus]

Here is the exponential fit. The 2019 minimum is above the trend, shifting the extrapolated zero ice after  2025 where with only data up to 2018 it was 2014.

[Wipneus]

Continuing discussion this, here is an attempt to show how the estimates of zero ice have been shifting. It is the same graph as in the previous post with extrapolations form former years added.
Here the dark blue line is the extrapolation using data up to 2018; The next greenish line data up to 2017
And so fort.
Many years the zero ice prediction was close to 2015 but that has been postponed, on average 1 year every year for the last 6 years or so.
 

[Wipneus]

Same as the previous but now a Gompertz regression, preferred by some. The Gompertz function approaches but never reaches zero, but still the same postponement to later and later dates is clear.

[Wipneus]

Finally a linear trend. The 2019 data point lies near perfect on the trend line. That means that the zero ice extrapolation has not shifted either and stays at 2032. That, and the fact that is has not changed much for many years gives some confidence to this date.

So for comparison I have added the same graph that I posted in 2012: also a 2032 zero ice extrapolation. The 'prediction' for 2019 is spot-on.

[Wipneus]

Finally the thickness map comparisons. Here are thickness map, compared with previous years and their diff's on the 10th of September (approximate day of annual minimum).

[Pavel]

Thanks for the update. It looks the Fram export was low in volume because the survived ice is very thin

[Jim Hunt]

Can you post the raw link?

I'm afraid not. Neven would have my guts for garters if I did that!

Perhaps you could employ the services of the Great God Google instead?

https://lmgtfy.com/?q=ex+prof+judy+sea+ice

[Jim Hunt]

Finally a linear trend. The 2019 data point lies near perfect on the trend line.

The 'prediction' for 2019 is spot-on.

Thanks for all the assorted "extrapolations" Wipneus.

Please bear in mind that in conformance with what appears to be standard practice at ex Prof. Judy's my graph was for the PIOMAS monthly volume numbers for September, not the September daily minimum volume.

[binntho]

Thanks as always for your postings Wipneus, and specifically for the extrapolation posts - it does seem that the humble linear prediction has the best track record so far, in spite of the apparently non-linear behaviour at times.

[Stephan]

Continuing discussion this, here is an attempt to show how the estimates of zero ice have been shifting. It is the same graph as in the previous post with extrapolations form former years added.
Here the dark blue line is the extrapolation using data up to 2018; The next greenish line data up to 2017
And so fort.
Many years the zero ice prediction was close to 2015 but that has been postponed, on average 1 year every year for the last 6 years or so.
 

I visualised this shift in "time to zero dates" derived from this graph

[El Cid]

The trendline on this chart is obviously faulty.

The displayed linear trend was created using the Excel "TREND" function. Surely you don't want to start a "debate" with Microsoft too?

I will not debate excel functions. I debate their use (or even usefulness). I did take quite a few statistics classes and if 15 datapoints in a row are above your trendline than it is obvious that you should not use that trendline (whatever the type of the trend), because it is not that sort of trend. That is all I wanted to say. There are various methods to determine randomness in your residuals but you don't need any (other than your eyeballs) if a long string of datapoints are above or below the trendline. Any trendline.

With all that said, I also want to thank Wipneus for his great work. May we never see his trendlines hitting that zero point!

 

[binntho]

The trendline on this chart is obviously faulty.

The displayed linear trend was created using the Excel "TREND" function. Surely you don't want to start a "debate" with Microsoft too?

... and if 15 datapoints in a row are above your trendline than it is obvious that you should not use that trendline (whatever the type of the trend), because it is not that sort of trend.

It's not at all obvious.

Wipneus has showed that the linear trendline is the only one that has any validity - so are you saying that there is no trendline that can be used on this data?

System memory may well cause 15 datapoints to fall to one side and 15 to another, even if the underlying trend over 30 points is linear.

[dnem]

The trendline on this chart is obviously faulty.

The displayed linear trend was created using the Excel "TREND" function. Surely you don't want to start a "debate" with Microsoft too?

I will not debate excel functions. I debate their use (or even usefulness). I did take quite a few statistics classes and if 15 datapoints in a row are above your trendline than it is obvious that you should not use that trendline (whatever the type of the trend), because it is not that sort of trend.
With all that said, I also want to thank Wipneus for his great work. May we never see his trendlines hitting that zero point!

Did they teach you that linear regression minimizes the sum of the SQUARE of the residuals? The raw count of residuals above/below the line is not a factor in fit.

Very useful stuff Wipneus and thank you for all your work.

[El Cid]

Wipneus has showed that the linear trendline is the only one that has any validity - so are you saying that there is no trendline that can be used on this data?

That is EXACTLY what I am saying. The linear trend is still the best of all the bad solutions (but that is just a coincidence because luckily we have data from 1979, which was almost the time when we started to lose ice seiously!!!) Had we started gathering data in 1950 or 60 or 70 it would all look very different...

I suspect it is so, because there are multiple trends. eg there was probably basically nontrending behaviour from 1950-1985 (CO2 growth counterbalanced by various aerosols), then a linear trend (loss of periphereal seas) until cca 2006-10, and now we have something else (I don't know what, but the Central Arctic seems different) until it changes again.


/last one on linear trends: the residuals should be a white noise, and there should be no autocorrelation, in our case it is clearly NOT a white noise/

 

[blumenkraft]

Wipneus has showed that the linear trendline is the only one that has any validity - so are you saying that there is no trendline that can be used on this data?

That is EXACTLY what I am saying.

+1

[uniquorn]

Thanks Wipneus. Here's an animation of the piomas thickness difference collage, feb-sep, for 2019

[dnem]

Last one on this. I totally agree with your last statement El Cid, including about autocorrelation. (But still, you don't just count positive and negative residuals!).

I have frequently railed here against curve fitting in such a rapidly changing environment and I will continue to do so!

[philopek]

The trendline on this chart is obviously faulty.
Summarizing: It is obviously statistically NOT a linear trend and even if it were the dataset used for them is probably not the right one.

The following is not meant personal but general and I only use the quoted post as an arbitrary trigger post.

My take on such discussions is that the trend is obviously downward and that's most probably the only thing that matters and is of interest.

Generally experts from various fields of expertise should stop to overshadow the relevant information with discussions about technical detail.

Ice cover is shrinking, ice thickness is shrinking, ice-volume is shrinking and that's all we have to focus on with the REAL goal to find ways to stop it or to slow it down significantly.

The later goal is difficult enough to achieve and even more difficult to take sufficient fellow humans with us onto that path.

All discussions about who has more insight and how exactly such a trend looks like is not target leading IMO or at least only as a sideshow.

[Juan C. García]

According to NSIDC, 2007 is still the second lowest on record. But if we look at JAXA, 2007 is the fourth lowest on record. So, if we still have AMSR2 data on the next decade, JAXA will put as the default graph, the years 2012, 2016 and 2019, but 2007 will be out.

More important, if we analyze the PIOMAS volume, 2007 is now the tenth lowest on record. September 2019 end up being less than 2/3 of the September 2007 volume (64.2%).

There has been critics that as a Forum, we have the tendency to forecast low values. It has been true in cases when there is a forecast of a blue ocean event, but from my point of view, not completely true if you look at volume.

It is true that the forecast of less than 1,000 km3 has been moving forward (from what we expected after September 2012) but we will still have the chance to see this figure (less than 1,000 km3) on the next decade. From my point of view, less than 2,000 km3 will be very bad and there is a greater chance that we have this figure on 2020-29.

Thank you, Wipneus! Great graphs!

[oren]

Thanks Wipneus. Here's an animation of the piomas thickness difference collage, feb-sep, for 2019

Wonderful animation.
Comparing with 2016 and 2012, this year had huge success in killing off initially positive anomalies in the Kara and Barents, and southern Laptev. Another big win was the Lincoln Sea, going from zero anomaly to terribly negative. OTOH, 2019 failed to kill off the last of the anomaly along the FJL-SVALBARD line, an area that is normally not hard to kill off, and lost its early big advantage on the Chukchi-Beaufort side. This was the main effect of the calm August.
Very interesting year.
Note: all win lose gain etc. should be in "". Not meant as good/bad.

[gerontocrat]

PIOMAS Volume as at 30 September 2019  4.568  km3 '000

The standard graphs and tables as I use for the JAXA extent data are attached.

Volume loss in the first half of September mostly below average, gain in the 2nd half mostly above average.

But 2019 volume is still 2nd lowest in the satellite record, by 601 km3 (15%) ABOVE 2012, and 221 k (5%) below 2011 the 3rd lowest, and less than 2018 by 549 km3 (11%).


_______________________________________________________________
The last table is a look at projections to the next maximum. Far too early to take it seriously, although it is obvious that 2012-13 disappears from the minimum maximum possibilities, replaced by 2016-17.

2012 shown again to be an outlier
Graph vol-4 shows the impressive decadal loss trend

[gerontocrat]

Just for interest here is the plume of projections for October. Will the current slow refreeze of area and extent be reflected in volume data?

Also shown is how 2012 did produce a new record 365 day average volume low but not so impressive compared with 2016-17. 365 day data is currently progressing to a new record low - perhaps by the end of 2021 .

An just a reminder of how the 2019 minimum was at a record low by the normal amount if 2012 is excluded or averaged with the bounce back the following year.

[crandles]

I have to admit the linear trend is surprisingly stable over last several years. I was surprised by this. 4 parameter gompertz fits can only manage such consistency since 2014.

To me a question remains of whether the trend in the data from 2012 to 2019 being pretty much an extension of the prior linear trend is just coincidence having wandered in different directions as indicated by the residuals being too similar.

Alternately perhaps autocorrelation explains the persistent residuals and the linear trend is more reliable than using too many parameters.

Either view seems quite reasonable as far as I can tell. I tend to prefer to give the shape models suggest more weight. Tamino finding statistically significant slow down in extent decline also improved my confidence in some form of sigmoid shape. But Wipneus has made me stop and think maybe this isn't yet enough to consider it done and dusted.

[Archimid]

I like using both the maximum and the minimum to glimpse into the future of the ice. Attached is an animation using the intersection of the trendlines of maximum volume and volume loss from 2007 to 2019.  Date of intersection estimated by sight, but it shouldn't be off by more than 1 year.

2032 has indeed been remarkably stable.

[binntho]

I don't think there are any true linear rates of change in nature. Many systems will in isolation exhibit polynomial rates of change, but linear? I can't think of a single instance.

Expecting complex and compound systems to exhibit any behaviour that can be expressed in a single mathematical formula (linear, polynomial or other) is probably pure wishful thinking. Which means that every and all attempts at extracting trends from data are "wrong" in that they will never be able to express what the system is really doing.

For any and all real-world data, finding the "least bad" option is the best we can do. And I think we have seen enough examples over the last couple of days to lay this to rest once and for all: The least-bad modelling option of past satellite-era behaviour of arctic sea ice is by far the linear one. And at the same time, it may well be the worst option for predicting future behaviour - we'll just have to wait and see!

[Jim Hunt]

Expecting complex and compound systems to exhibit any behaviour that can be expressed in a single mathematical formula (linear, polynomial or other) is probably pure wishful thinking.

I've been extolling the virtues of "The Slow Transition" concept over at Judy's, where my words of wisdom haven't fallen on entirely deaf ears!

I also was a reader (silent) in the ASIF but in the melting saison 2018 I stopped it because some members wrote to much boring stuff.

[Shared Humanity]

I've been extolling the virtues of "The Slow Transition" concept over at Judy's, where my words of wisdom haven't fallen on entirely deaf ears!

While this is based solely on my reading the work of others here, I also think the slow transition concept has merit.

[kaixo]

I like using both the maximum and the minimum to glimpse into the future of the ice. Attached is an animation using the intersection of the trendlines of maximum volume and volume loss from 2007 to 2019.  Date of intersection estimated by sight, but it shouldn't be off by more than 1 year.

2032 has indeed been remarkably stable.

This way of estimating the trendline of the september volume minimum at least includes the two basic processes determining it, freezing and melting.

To get into this a little bit more, it could be interesting to estimate the trendlines in decadal montly volume averages. To me Jim Pettits Arctic death spiral graph http://iwantsomeproof.com/extimg/siv_annual_polar_graph.png
suggests that over the last 40 years the trendlines in decadal average monthly volumes are different in winter months to those in summer/autumn.

I am aware that with four decades of measurements you only have four data points for each month, which makes any result from this exercise untrustworthy. Still, the arctic death spiral graph consists of quite a few individual months and the shape suggests a rather smooth change over the months of the year and over the years. And the 'dent' in the graph in recent years from august-october i find especially intruiging. To me it suggests some sort of change in dynamics.

Maybe one could use 5 yearly averages and thus create 8 data points for each month. Anyway, maybe like this one could estimate the decadal (or 5-yearly) monthly averages all year around for the near future.

Does anyone know where to find the monthly volume data since 1979? Maybe i will give it a try in a few months time when we also have 2019 numbers, although i haven't had anything to do with statistics for 30 years, so it will be quite a challenge    If you have any suggestions or think this is a waste of time because of a lack of data or other reasons, let me know.

[Jim Hunt]

Does anyone know where to find the monthly volume data since 1979?

You could try:

http://psc.apl.washington.edu/wordpress/wp-content/uploads/schweiger/ice_volume/PIOMAS.2sst.monthly.Current.v2.1.txt

[blumenkraft]

WOW, kudos to Apple. I could copy&paste this into a blank Numbers sheet and it converted it to .CSV just like that.

This is 1979 vs. 2019

[Juan C. García]

The always great video of Arctic Sea Ice Minimum Volumes 1979-2019, by Andy Lee Robinson:

[Neven]

Latest update on the ASIB: PIOMAS October 2019