On the Question of Diminishing Arctic Ice Extent
Posted on 1 June 2010 by muoncounter
Guest post by muoncounter
How there can still be honest disagreement over whether a large body of ice is shrinking or increasing in areal extent over the long term is really quite remarkable. Since the reports don’t seem sufficiently convincing, let’s look at the question analytically. In the NOAA’s October, 2009 update to the Sea Ice Cover section of the “Arctic Report Card,” Figure S2 (reproduced below) shows the percentage change in the annual minimum (September) and maximum (March).
This graph is presented as percentage difference in ice extent (area in millions of sq km), relative to the mean for the period (1979-2009). Both the maximum and minimum are shown with a linear fit; by the slopes of these lines, we see that the sea ice minimum is decreasing faster than the sea ice maximum. Hence we expect the discrepancy between min and max to increase.
This change in areal extent is accompanied by a systematic change in ice thickness, resulting in an overall decrease in ice volume, as demonstrated in NOAA’s Figure S4 below.
The text accompanying this figure draws a distinction between “multi-year” ice and “first-year” ice, noting that it is the multi-year ice that is thinning. NOAA’s conclusion is self-evident: “These changes have resulted in seasonal ice becoming the dominant Arctic sea ice type, both in terms of area coverage and of volume” (Sea Ice Cover, cited above).
These data warrant a closer look. Additional Arctic sea ice data was merged with the 1979-2009 data used above, allowing a more robust reconstruction of the ice extent, as shown in the graph below.
In this graph, areal extent is converted to percentage change from the starting value (areal extent in 1972). Note that the September minimum ice extent (red) has a steeper slope than the March ice extent maximum (blue). Each is shown with bands at +/- 1 standard deviation.
Aside from the noisy September data, there are two problems with this graph:
- The two trendlines cross in 1980 (suggesting rather nonsensically that in years prior, the minimum ice extent was greater than the maximum).
- The two trends do not statistically clear one another (to one standard deviation) until 1995, more than halfway into the time series.
I suggest that presenting these data as functions of temperature, rather than time, may resolve these problems.
For temperature data, winter sea surface temperature (SST) anomalies recorded at the Pribiloff Islands were smoothed (using a 15 year moving average), producing the graph shown below. This dataset ends in 2008.
Note: Additional Arctic summer time series data are available, but those were not analyzed here. I’ll save that for another day.
In the following plots, the sea ice minima and maxima for the periods of 1972-2008 are shown as functions of this SST anomaly. In each case, the functional fits, with bands at +/- 1 standard deviation, are clearly nonlinear.
With y representing the percentage change in ice extent and x representing the SST, the function used were:
March maximum, y = -1.196 – 2.186 x2 ;
September minimum, y = 0.253 – 7.620 x2.
Using the 2008 value of the SST in this time series (1.62 deg C), the derivative of these functions provides rates of change in % per degree:
September minimum -0.247 (-24.7% / degree);
March maximum -8.3% / degree.
Thus not only do we expect the annual loss of sea ice to accelerate, we expect the discrepancy between sea ice minimum and sea ice maximum to continue increasing – at least until the sea ice minimum reaches zero. Further, because these temperature data are from the winter months, it may be reasonable to conclude that even a few cold winters just aren’t enough to compensate for the overall warming-induced ice loss.
Actually, no. Both of the curves are percent change from the 1970 values, so it just means that before 1980, the September data were, on average, a bit higher than Sep 1970, while the March data were a bit lower than March 1970. If you converted your two percentages back to actual extent, I'm sure you would find that Sep extent was always lower than Mar extent.
The comparison of extent to winter SST anomaly is quite interesting, even though I am not sure how well the Pribilofs (southern Bering Sea) is a good proxy for the Arctic Ocean. I'm guessing it is the closest measurement you could find? Have you also tried air temperature at some of the Arctic weather stations? The obvious question is whether the extent vs. SST fits are more or less noisy than the extent vs. time plots. I can't tell because the scales are not the same. My eyeball says they are similar based on looking at the 1-sigma bounds, but it would be good to run the numbers.
This page (from your link) shows some quite different temperature records for the Arctic.
One ocean temperature in the Bering Sea
Another (St Paul's Island) which is mentioned in the text on the NOAA page.
Why go with the one's you have choosen?
Secondly I've started to think that there is a regional aspect to the Arctic. What happens in the east isn't necessarily mirrored around Greenland. It seem possible to have a potitive ice anomaly in the east and negative in the west at the same time. It seems unlikely that one temp record in the Bering Sea is going to give insight into the whole Arctic ice extent.
From here
I've started to think that there is a regional aspect to the Arctic. What happens in the east isn't necessarily mirrored around Greenland. It seem possible to have a potitive ice anomaly in the east and negative in the west at the same time.
Exactly, which is why the professionals producing measurements don't make an elementary mistake of that kind when summarizing the total ice extent. They report total extent, they don't look at a region and extrapolate from that. By way of analogy, you may move money from your checking account to your savings account and vice versa, but if you are spending more than you are saving your total available funds will diminish.
It seems unlikely that one temp record in the Bering Sea is going to give insight into the whole Arctic ice extent.
But of course we have Arctic ice extent to give insight into Arctic ice extent.
Doctor David Barber
Firstly, some readers might not realise that ice extent is different to ice area. This is explained on the NSIDC FAQ page. Basically, area is a measurement of exactly what area of ocean is covered with ice, while extent measurements count the area of every grid cell that is 15% covered by ice.
Secondly (as I see Jeff Freymueller has already pointed out), the fact that, in the graph titled “Arctic Sea Ice Change”, the two trendlines cross, does not suggest that before 1980 the minimum extent was greater than the maximum. The September trend is relative to the initial September extent, the March trend is relative to the initial March extent, and the March extent is naturally a lot higher.
Thirdly, I agree with Jeff Freymueller and HumanityRules that picking one temperature series from a bunch of islands in the Bering Sea seems a bit arbitrary. However, I’m not sure that HR’s graphs are any better.
Fourthly, in Figure S2 the mean is actually 1979-2000, not 1979-2009.
The question after your second graph (''Why go with the one's you have choosen?'') has a possible answer as follows: If you have data from two islands, and data from one supports the point you are making, whereas data from the other seems to disprove it - then you pick the data that agrees with your theory, and discard the other. Chances are most of your readers won't bother to look for other islands. Many debaters of climate seem to reason along those principles (on both sides).
Another example of selective analysis: The first graph that you showed (Ocean temperatures in the Bering Sea at M2) has an interesting comment on the page where it is originally published: the NOAA author draws the conclusion that ''Ocean temperatures for the previous decade ... show a shift toward warmer temperature of 2 deg C around 2000''.
When I look at the graph, I could just as easily get the idea that ocean temperatures for the previous decade show a dramatic shift toward *colder* temperatures after 2004.
and of course Muoncounter!
Thought you might be interested in some more Pan-Arctic Ice extent correlations with temperature, which I’ve been looking at for an Arctic post:
SST and summer ice extent: some of the SST grid reconstructions can be biased as they set grid cells to -1.8 degrees where the ice is, so you have to use an ice area mask, or avoid areas where there is ice (I think?). I believe other work on Polar Pathfinder satellite SST data (good “dense” spatial coverage) is ongoing. This chart uses the ICOADS series which is based on raw on-site measurements:
I can’t say exactly where the measurements are taken as yet, but there are enough of them in recent decades to be moderately convincing. Maybe someone with more expertise can comment. The choice of scale is arbitrary but the correlation is high. Other reconstructed (ERSSTv3, HadISST etc) series also display this correlation if you select ice free grid cells. You could also correlate HadISST ice and SST all the way back to 1870 with lots of caveats and caution. It would be interesting to look at Atlantic side and Pacific side separately, but I haven’t found time for a complete comparison.
Air temperature: This uses the same DMI 2m air temperature series that has been used on some blog sites to imply no significant temperature trends in Arctic. I've shown the seasonal values as well as min/max trends and zoomed in
I averaged the seasonal temp data over 80 days to phase match the “peaks” with ice extent. I also extended the ice extent data using HadISST and NSIDC ESMR data, though extent uncertainty pre 1978 and then pre 1972 is higher. The correlation of fitted second order curves is remarkable, but slightly down to chance and selection of start/stop dates, other reconstructed temp series (NCEP, ERA-40) still show high correlation, as has been noted in various papers. The summer air temp runs just above melting point, but if you average this over a few preceding months the correlation with Summer ice extent is also reasonable.
In short, air temperature, SST and ice extent appear coupled together seasonally and regionally, which is kind of what you’d expect. As to why it’s warming overall…
I will post links to data if I get time and people are interested.
#1: "If you converted your two percentages back to actual extent, I'm sure you would find that Sep extent was always lower than Mar extent."
Good point, I'll check.
#4: annual loss of sea ice to accelerate "maybe, but not based on the data. A statistical model like this isn't supposed to be projected into the future. ... Moreover, it's quite possible that the decline in March maximum might start catching up to the decline in Sept minimum"
Doesn't 'decline in March' catching up with 'decline in Sept' require acceleration? As for the crystal ball of statistics, these data end in 2008; Sept 2010 isn't that far off.
#9 and #11: "picking one temperature series from a bunch of islands in the Bering Sea seems a bit arbitrary." ... " example of selective analysis"
Oddly enough, I originally plotted the pct ice change against a trend extracted from the land ocean temp index for latitude 64N-90N that I found here. That graph shows up in a comment on a prior article here. But I thought that was painting with too broad a brush.
#11:"get the idea that ocean temperatures for the previous decade show a dramatic shift toward *colder* temperatures after 2004."
I read somewhere that 'climate' requires a much larger time frame than 2004 forward. Some folks use 30 year statistics? From the temperature data (some going back to the 1890s) I've looked at, a 15 year trailing average is a great way to extract an underlying trend. There are outliers that are both above and below the trends on the September min ice graph.
#12: The long-term trends on the graph posted by Peter Hogarth also speak to increasing rates of ice loss.
I guess what I'm asking is what would the conclusion be if you ended the analysis in 2006?
My point was certainly not that you can draw any conclusions about the climate from just 5 years of data. Yet the NOAA author sees (in the graph) a prominent shift toward warmer temperature from about 2000 to about 2004, but fails to see (in the same graph) an even sharper shift toward colder temperature in the following 5 year period. You see what you want to see.
Looking at the whole graph, from year 1995 to 2005, we see that there is absolutely no change in 'Depth averaged temperature' at at the Mooring 2 location in the Bering sea. That is good, but 20 years of weather is still not climate.
Just to illustrate, the mean area for March ice for the period 1979-2009 is 15.8 million sq km, and the September mean is 6.7. This gives a mean recovery of 9.1 million sq km.
The September 08 ice was 4.5, the March 09 ice 15.2, giving a recovery of 10.7.
Hence, for the expectation that the discrepancy between the trends of min and max to increase to be fulfilled, the recovery rate of ice from the September lows to the March highs will also have to increase.
I believe the difference you're noting relates to the formation of annual ice during the Arctic winter, versus the summer melting of this (thin) ice, plus the older thicker ice during Arctic summer. There will likely always be a thin layer of winter ice in the Arctic (unless warming becomes really extreme), but we appear to be moving toward a time in the not-too-distant future when the Arctic will be ice-free in the summer. (Time for Superman to seek a Fortress of Solitude in the Amazon rainforest.... Oops... May soon be too late for that as well.)
My confidence in PIPS 2.0 is raised by this figure:
There is a reasonable match between IJIS (IARC-JAXA Information System) sea ice curves and US Navy PIPS (Polar Ice Prediction System) data for the period IJIS has some as well.
However, prior to 2003 PIPS differs from anyone else.
On top of that the sea ice volume story told by PIPS is absolutely inconsistent with the current scare.
Looks like PIPS end-of-May sea ice volume is a pretty good predictor for their minimum ice volume in September. If we go with this observation, PIPS sea ice volume must exceed their figures for 1998-2000 in September, this year.
I try to pull March maps from PIPS to make comparison easier.
John, you're right. While winter extent is declining, that decline is slower than summer decline, and we are indeed seeing record extent increases in fall. That trend is likely to continue until summer extent basically hits zero. Then the extent increases will fall slowly as winter ice extent falls slowly. There will be winter ice for centuries unless we really screw up.
Re SST vs ice extent. There is good (though not excellent) correlation between North Atlantic spring SST and summer ice extent. Sea current flow volumes between the Arctic and the Atlantic utterly dwarf those between the Arctic and the Pacific. A large factor in summer melt is ice export through the Fram Strait by the trans-arctic drift. When the drift is strong, the ice flows down the Greenland current to melt near Iceland where the water is warmer.
As for whether to expect acceleration of the decrease in summer extent ... it depends what you think is more important and likely to come into play. If you believe the PIOMAS model will continue with the current linear trend, extent loss has to accelerate because volume will be zero by 2030. If on the other hand you think average summer latitude of the icepack boundary is linearly correlated with temperature anomaly, then the extent decline will decelerate. What physically happens depends on whether the ocean currents continue basically as they are, or change in response to overall warming.
PIPS is a forecasting system, fed with SSMI data for predictions of extent, so difficult to figure why any discrepancy in area looking back in time. I agree March values would be better!
"the Ice Concentration page has an archive comparing previous PIPS 2.0 forecasts with actual conditions plotted from SSM/I data" Note they regard SSMI as actual data.
Out of interest, where does the PIPS system have numerical files of area, thickness or volume? I can only find polar plots?
John Cook: "fortress of solitude at the south pole?" Ha! Like everything else, there's disagreement on that. However, traditionalists go with the Action Comics June, 1958 original location: in the Arctic.
#21: "Re SST vs ice extent. There is good (though not excellent) correlation between North Atlantic spring SST and summer ice extent."
Now that's interesting: I've been looking at annual temperature averages; will go back at look at seasonals.
"When the drift is strong, the ice flows down the Greenland current to melt near Iceland where the water is warmer."
I assume that the Gulf Stream has a hand in that. The Gulf of Mexico and related tropics are unusually warm (and oily) this year. Is there any correlation between seasonal ice melt and temperature in those waters?
"As for whether to expect acceleration of the decrease in summer extent ... it depends" Based on Hogarth's sea ice minimum graph, acceleration is not to be expected, its already happening.
So if you get to make extent predictions 7 days out, continually updating with new real data, of course you do pretty well.
where does the PIPS system have numerical files of area, thickness or volume?
I could not find any either. Just took plots and reverse engineered data from color codes with a quick-and-dirty perl script using Image::Magick.
If they have got ice right, it would explain a lot about the scary downward trend seen in NOAA Arctic Report Card 2009.
Divergence before 2003 is remarkable. It's unkikely PIPS 2.0 24 hour forecast have missed September ice by 1 million km2 in 1999. After all it is used for operational purposes.
http://www.sciencedaily.com/releases/2010/04/100427111449.htm
Interestingly fishing vessels in the Bering Sea have in recent seasons experienced pack ice pushing further south than normal and at times finding ports in the Aleutian Islands being blockaded by such ice.
They would't really care about consistency between older and newer data
Of course. But they do care about consistency between data and reality. Captains tend to get annoyed by running into one million km2 of ice missing from the map.
Winds from Siberia Reduce Arctic Sea Ice Cover, Norwegian Researchers Find
Conclusion of article, in concert w/recent emphasis on volume as opposed to area:
However, [Dr. Sorteberg] emphasizes that he and his colleagues do not reject the assertion that climate change is affecting Arctic ice cover or that the IPCC is wrong when it states that the Arctic may be nearly ice free in summer towards the end of this century.
"There is no doubt that the Arctic sea ice has become thinner in recent years. The thickness of the sea ice is a much better indicator than the extent of the ice cover if we want to study how climate change may affect the ice in the Arctic," says Mr Sorteberg.
Perhaps the reason that the PIPS forecasts track IJIS so well after 2003 is that the forecasts may have been initialized based on the same data that IJIS used? If the PIPS forecasts were not so accurate prior to 2003, that was a concern for the Navy at the time, but it doesn't matter now. (The captains will be happier now than they used to be).
However, these data are subject to really great uncertainty, as explained the section of error sources here. I think it would be unwise if anyone seriously used the PIS maps directly to estimate ice trends, at the very least that would require a lengthy discussion of the possible error.
If I'm not mistaken, St. Paul IS in the Pribiloff islands.
i already noticed in a comment in the other post that more than ice extention it is critical thinking that is falling below the absolute low in human history.
Indeed Goddard knows very well that his pixel counting technique is bogus, given the full retraction he was forced to make not so long ago (scroll down here to editor's note). Neverthless he used it again and some people followed suit, blindly I'd say.
(Thanks to Phil Clarke in a comment here for reminding us).
these plots are NOT DATA. I'm not sure why you seem to be interpreting the older forecasts as if they were data
They are not data, agreed. But these are 24 hour forecasts and for immediate physical reasons neither sea ice extent nor volume can change much in a day. Therefore the numbers depend on the realtime data assimilation procedure used and the very quality of data assimilated.
If this system was faulty before 2003, gradually fixed from 1998 on, it would explain the divergence. However, at least some hint of a documentation is needed for this kind of reasoning to be satisfactory.
I never knew Goddard and The Register published that retraction. For what it's worth, I corresponded with Goddard at the time and I actually concluded he was sincere in his odd way, not that it excused his departures into unfounded accusations. Perhaps he learned his lesson; I've largely avoided reading what he writes these days so I wouldn't know.
Anyway thanks again for pointing out the retraction, it leaves me feeling as though The Register has some shred of integrity, nice because I'm fond of their over-the-top headlines and ledes.
According to the NSIDC page I linked to before, they are continuously improving the quality of the microwave data. That could possibly account for the "gradual fix" you mention. But there are still huge error terms. I feel that there is so much uncertainty about these data, that the burden of documentation is on whoever wants to use them.
What map projection are the PIPS output images in? It looks like polar stereographic to me. That projection doesn't preserve equal area, so you cannot simply use a count of pixels to determine ice-covered area. Pixels will have different areas depending on their proximity to the pole.
you cannot simply use a count of pixels to determine ice-covered area
Do you want me to recalculate with multiplying pixel area by cos(45°-lat/2)? Inside 70N it does not make much difference.
there are several type of stereographic projections. Before doing any calculations you should know which one is used in that figure.
I was just speculating about polar stereographic, but it could be any azimuthal projection (orthographic, Lambert, ...?) The scale variation would depend on the specific projection parameters.
More broadly, what metadata are available describing these data? Have the model results been validated, and if so, what is the structure of the error?
I find it a bit disconcerting that the entire community of "skeptics" (here, at WUWT, etc.) seems to have suddenly latched onto this PIPS2 model with little to no examination of its suitability for the purposes to which they are enlisting it. Or perhaps I've missed something, and all these questions have been satisfactorily answered already.
Walt Meier" "The proper way to calculate a comparison of ice coverage is by actually weighting the pixels by their based on the map projection, which is exactly what NSIDC does."
This is taken from the article at http://www.theregister.co.uk/2008/08/15/goddard_arctic_ice_mystery/
Clearly NSIDC also use pixel counting. If you think about, if weighting was not given to the pixels, any error would cause an UNDER estimation, not an OVER estimation.
"any error would cause an UNDER estimation, not an OVER estimation."
not true, it depends on the projection. For example, you may have a projection that makes the area calculated at, say, 70° correct with opposite bias on the two sides.
There's no way to come to any meaningful conclusion without knowing which projection has been used.
yes, but depending on where you put your zero and on the area sampled you may under or over estimate the total area and hence average ice thickness.
John, I work with projections of gridded geographic data every day. The problem here isn't with "counting pixels", it's with a failure to understand the scale variation associated with non-equal-area map projections. Apparently Steve Goddard made this mistake at WUWT, Berényi Péter makes it here, and no doubt lots of other people are doing the same thing elsewhere.
Different projections will have different patterns of scale variation. I'm guessing that the PIPS2 images people have been grabbing are in a polar stereographic projection with the planar surface tangent to the earth at 90 north. But it could be secant ... or an orthographic projection rather than stereographic ... or something else entirely.
This is why it's essential to have metadata describing the characteristics of the data you're working with. Maybe there are metadata for these images somewhere, though I haven't seen them. But unless you have the specific projection parameters used to create the images, you cannot simply count the pixels on date 1 and date 2 and draw any conclusion about whether ice area increased, decreased, or stayed the same.
When the corrected image was processed by Goddard, his result was a close match for the NSIDC results even if there is some question over whether or not he gave the correct weighting to the pixels.
“it was found that PIPS correctly made 24-h forecasts of decreasing sea ice concentration 10%–15% of the time (it also correctly forecast increasing sea ice concentration an additional 10%–15% of the time). However, PIPS correctly forecast melt-out conditions <5% of the time.”
Presumably later versions of PIPS addressed those concerns?
I also stumbled across a very smug group of denialists actually placing bets on monthly sea ice extent. Digression: I happened to be watching images of brown pelicans getting much browner at the time and wondered whether these same folks would bet on bird deaths or when oil reaches Pensacola, etc.
And what was the arbiter of who wins these ice bets? The JAXA data, referenced above. Their publicly available ice extent data goes all the way back to June, 2002. Here is a graph, comparing JAXA to the data I originally used from NSIDC, showing the two are in exceedingly good agreement.
Those last two years sure do look like a significant uptick. However, when you look at even a slightly longer term, the uptick pales beside the downtrend. And the trend of Sept minima is still accelerating downwards.
So while the denialists wager (and Rome burns, err - the ice melts), we fret over pixel-counting. Yes, data quality control is indeed of the utmost importance. But we can't lose track of the point, as someone asked a few days ago, of the discussion. Because that's how we lose the argument.