the Air Vent

A guest post by Jonathan Drake who runs a well named site Questioning Climate

Jonathan, has done an enormous amount of investigation into ice level measurement and has found some interesting details. Jonathan sent me this well referenced post which I’m sure I will use down the road for it’s links to data sources. He has said he will be available for questions as much as he can. I think you’ll find his open style similar to my own, but this is his post so please direct questions to him.

The copying process was very difficult requiring me to screen grab the graphics and add back the bolds. I did my best to keep the graphics reasonable sized and recopy the post. Let me know if there are problems.

—-

Day after day we are bombarded with images of polar bears on ice bergs and collapses of ice sheet into the ocean with apocalyptic headlines that the Arctic ice will vanish within a few years. It is of particular interest that, according to Dr. Vicky Pope, head of Climate Change at the UK Met Office, “Recent headlines have proclaimed that Arctic summer sea ice has decreased so much in the past few years that it has reached a tipping point and will disappear very quickly. The truth is that there is little evidence to support this. Indeed, the record-breaking losses in the past couple of years could easily be due to natural fluctuations in the weather, with summer sea ice increasing again over the next few years.”

http://www.metoffice.gov.uk/corporate/pressoffice/2009/pr20090211.html

So, on that basis, what is going on? Could there be some kind of misinterpretation of the data? Might someone have made a wild extrapolation? With these questions in mind the Arctic sea ice records are examined to see what, if anything can be learnt. The two commonly cited measures for sea ice are extent and area. Many people are unaware that there are two different measures, let alone a difference or what it is and so a simplified definition would seem useful. Both are presently derived from satellite and calculated from images. Each pixel of the image is assigned an area.

- Sea ice Extent is the area over which sea ice is found at a concentration of at least 15% by area. It is calculated by summing the number of pixels showing at least 15% sea ice multiplied by the pixel area. Thus one pixel area will be recorded if 15% or more of it is covered by sea ice.

- Sea ice Area is in reality a better measure of the amount of ice. It is essentially the extent multiplied by a factor representing the actual area that the ice covers, that factor being between 0.15 and 1.00. If for example the pixel is covered by 50% sea ice, the area of that pixel will be multiplied by 0.5.

It is important to note that there is a region in the measurement named the Pole Hole that cannot be measured. For the sea ice extent it is assumed to always have at least 15% concentration, whereas for the sea ice area pixels in this sector are not used. For further information see:

http://nsidc.org/data/smmr_ssmi_ancillary/area_extent.html

This article will solely concentrate on Arctic sea ice extent. One thing to bear in mind is that the satellite data starts in October 1978. Quite obviously this is a rather short record on which to examine trends in climate, but nonetheless climatologists and politicians maintain that it is long enough to ascertain climatic trends. So let’s go with the flow.

Were there no observations made prior to this? Yes, there were and at least two reconstructions can be found. Below is a graphic from Walt Meier and Julienne Stroeve of NSIDC:

http://nsidc.org/sotc/sea_ice.html

They state: [Bold emphasis added]

“Mean sea ice anomalies, 1953-2008: Passive microwave-derived (SMMR / SSM/I) sea ice extent departures from monthly means for the Northern Hemisphere, January 1953 to September 2008. Image by Walt Meier and Julienne Stroeve, National Snow and Ice Data Center, University of Colorado, Boulder.”

“Passive microwave satellite data reveal that, since 1979, winter Arctic ice extent has decreased about 4.2 percent per decade (Meier et al. 2006). Antarctic ice extent is increasing (Cavalieri et al. 2003), but the trend is small.”

“Satellite data from the SMMR and SSM/I instruments have been combined with earlier observations from ice charts
and other sources to yield a time series of Arctic ice extent from the early 1900s onward. While the pre-satellite
records are not as reliable, their trends are in good general agreement with the satellite record and indicate that
Arctic sea ice extent has been declining since at least the early 1950s.”

Why was the graph showing the 1900’s onwards not used on the NSIDC website instead of the above? Maybe it will become more apparent later? In the text, one of the statements is particularly questionable:

“Arctic sea ice extent has been declining since at least the early 1950s.”

This is somewhat misleading since the 12 month running mean of ice extent is within about +/- ¾ of one standard deviation (SD) of the 1 SD mark from 1953 until mid 1979 when it drops for the first time below the 0 SD line, indicated by the black dotted line. It must surely be a coincidence that the decline began at the start of the satellite record?
The following graphic illustrates the point. It is the same illustration, but with some trends added by eye:

At the start of the satellite data, marked with a light blue dotted line, there was a rapid and significant decline in the sea ice extent. The interpretation of this feature is left to the reader.

Moving along, Chapman and Walsh from the University of Illinois have also created an Arctic sea extent reconstruction covering the era 1870 to 2008 based upon the following eight basic data sources:

1. Danish Meteorlogical Institute
2. Japan Meteorological Agency
3. Naval Oceanographic Office (NAVOCEANO)
4. Kelly ice extent grids (based upon Danish Ice Charts)
5. Walsh and Johnson/Navy-NOAA Joint Ice Center
6. Navy-NOAA Joint Ice Center Climatology
7. Temporal extension of Kelly data (see note below)
8. Nimbus-7 SMMR Arctic Sea Ice Concentrations or DMSP SSM/I Sea Ice Concentrations using the NASA Team Algorithm

They state that the more historical data should be treated cautiously and split the dataset into three sections, identifying fundamental changes in the sources:

1870-1952 Mostly climatologically determined but with increasing amounts of observations.
1953-1971 Mainly hemispheric observational data of varying reliability, but described as generally accurate.
1972-2008 Hemispheric satellite data with ’state-of-the-art’ accuracy.

The University of Illinois northern hemisphere seasonal sea ice extent data can be found here:

http://arctic.atmos.uiuc.edu/SEAICE/ more specifically, the data file is: http://arctic.atmos.uiuc.edu/

SEAICE/timeseries.1870-2008

Notes:
1. Most of the direct observations (1870-1971) are from ships.
2. It would appear that Chapman and Walsh have extended the satellite data backwards in time, but for the purposes of this analysis let us stick to their definitions.

Plotting the mean annual values from the time series gives:

It should be noted that at the time the the dataset was downloaded it dd not include a point for 2008.

Since this is the longer of the two datasets, and more easily accessible, it will be examined in preference. But for now, it is fair to say that the two pretty much agree in the overlapping period, at least in terms of general character.

The Chapman & Walsh data does not seem to show anything particularly extraordinary, apart from a couple of obvious points; the first around 1952/3 and second at 2007. It merely shows relatively flat character at first, with maybe a slight downward slope that becomes much more rapid after about 1970.

Now if we look at the data split into the sections specified by Chapman & Walsh some interesting features appear. Least squares linear fits have been added to each section independently to highlight the trends.

As can be seen, the two early eras have almost flat trends and there is a noticeable step in the data between them. That said, the step between trends appears to be within the variability. Now turning to the satellite era, it is clear that it joins the 1953-1971 data quite well. However, there is a very distinct difference in the trends with the satellite showing a rapid decrease in sea ice extent.

The calculated linear trends are:

1870-1952 0.00114 x 106 km2/year
1953-1971 -0.00080 x 106 km2/year
1972-2008 -0.04172 x 106 km2/year
1870-1971 (mean) 0.00017 x 106 km2/year

So when the measurement method swapped to satellite, the trend altered abruptly, from close to zero, to significantly declining. In fact, it is a factor of 52 greater than that of the 1953-1971 period. Is this just coincidental? It is very difficult to say, but it is interesting that Chapman and
Walsh consider the 1953-1971 data to be ‘generally accurate’ and data for the 1972-2008 era as having ‘state-of-the-art accuracy’. They also state that the pre-1953 reconstruction should be treated with particular caution.

If the first part of the set is rejected, it leaves two supposedly good adjacent datasets showing significantly different trends. Which is correct, or more correct? This looks like a classic two clock problem.

News Flash: http://nsidc.org/arcticseaicenews/

[The bold emphasis added]

“February 18, 2009: Satellite Sensor Errors cause Data Outage

On February 16, 2009, as emails came in from puzzled readers, it became clear that there was a significant problem—sea-ice-covered regions were showing up as open ocean. The problem stemmed from a failure of the sea ice algorithm caused by degradation of one of the DMSP F15 sensor channels. Upon further investigation, we found that data quality had begun to degrade over the month preceding the catastrophic failure. As a result, our processes underestimated total sea ice extent for the affected period. Based on comparisons with sea ice extent derived from the NASA Earth Observing System Advanced Microwave
Scanning Radiometer (EOS AMSR-E) sensor, this underestimation grew from a negligible amount in early January to about 500,000 square kilometres (193,000 square miles) by mid-February (Figure 2). While dramatic, the underestimated values were not outside of expected variability until Monday, February 16. Although we believe that data prior to early January are reliable, we will conduct a full quality check in the coming days.

Sensor drift is a perfect but unfortunate example of the problems encountered in near-real-time analysis. We stress, however, that this error in no way changes the scientific conclusions about the long-term decline of Arctic sea ice, which is based on the consistent, quality-controlled data archive discussed above.

Some people might ask why we don’t simply switch to the EOS AMSR-E sensor. AMSR-E is a newer and more accurate passive microwave sensor. However, we do not use AMSR-E data in our analysis because it is not consistent with our historical data. Thus, while AMSR-E gives us greater accuracy and more confidence on current sea ice conditions, it actually provides less accuracy on the longterm changes over the past thirty years. There is a balance between being as accurate as possible at any given moment and being as consistent as possible through long time periods. Our main scientific focus is on the long-term changes in Arctic sea ice. With that in mind, we have chosen to continue using the SSM/I sensor, which provides the longest record of Arctic sea ice extent.”

- End of News Flash.

Surely this doesn’t imply that they have decided to use a specific type of sensor because the new one would show a different trend? The mixing of data sources has not been a problem for Chapman and Walsh, or Meier and Stroeve before, as can be seen above. Does this new, more accurate
detector have greater long term instability than the old one? It seems doubtful.

Another, issue worth a quick mention, but nothing more, is that the sensor drift was to lower sea ice coverage. Would this have been noticed had it been a smaller and more gradual error?

Now would seem a good time to move onto a different dataset, so let us look at the JAXA AMSRE. The data is updated daily and available here:

http://www.ijis.iarc.uaf.edu/en/home/seaice_extent.htm

This is, of course, the new sensor which has high accuracy. The actual dataset used was dated 31/1/2009 and looks like this:

It is difficult to visualise the trend in such a graph and so further processing was done to reveal the trend. Firstly, interpolation has been used to achieve a regular sampling interval of one day and to fill in the missing data, and then a moving average filter of 365 points was applied.

Rather than applying padding or other techniques to extend the filtered result, the series was terminated.

The apparent melt in 2007 stands out as a very rapid decrease immediately following a slight upturn. This period also stands out as anomalous. However, just as importantly the subsequent increase is significantly faster than the general decline in the first part of the record, with the late
2008 extent heading for levels similar to those of 2004.

This scaling makes the trend look more dramatic than it really is. Putting it in context with the seasonal swing yields:

It can now be seen that the trend is a tiny compared to the annual swing in sea ice extent.

It is irritating not to be able to see the trend over more of the data. So, in order to see the changes nearer to today, the simplest method is to examine the rate of change. Annual rates of change have been calculated from the raw data and plotted as a time series. Three fits have been applied in order to help visualise the underlying trend. In all three cases the rate of change is towards more positive values, and increasing positive values. This suggests that not only is the sea ice extent increasing, but it’s expanse is accelerating.

Now let us look at the longer NASA team record. The data is available here:
ftp://sidads.colorado.edu/pub/DATASETS/seaice/polar-stereo/trends-climatologies/iceextent/nasateam/

Specifically the following dataset:

gsfc.nasateam.daily.extent.1978-2007.n

Again, interpolation was used to achieve regular sampling, and then a 365 point moving average filter was applied. The result is shown below.
An important point is that the melt in 2007, the extreme of which is just captured, is not unusual when compared to other parts of the time series, notably 1995. Thus it can be reasonably stated that the 2007 media hysteria regarding the melt was unfounded, despite it apparently being a record low, and particularly in light of the recent gains which can be seen from JAXA.

As can now be seen, the change in the trend is a tiny proportion of the seasonal swing.

To put further context to this, the uncertainties in the measurements have been estimated from:

“Confidence Level/Accuracy Judgment:

Estimates of the accuracy of the NASA Team algorithm vary depending on sea ice conditions, methods, and locations used in individual studies. Cavalieri et al. (1992) summarizes several of these studies. In general, accuracy of total sea ice concentration is within +/- 5% of the actual sea
ice concentration in winter, and +/- 15% in the Arctic during summer when melt ponds are present on the sea ice. Accuracy tends to be best within the consolidated ice pack when the sea ice is relatively thick (greater than 20 cm) and ice concentration is high. Accuracy decreases as the
proportion of thin ice increases. See Cavalieri et al. (1992), Steffen et al. (1992), and other listed references for an overview of the algorithm performance.”

“The goal of this data set is to provide a long term, consistent sea ice concentration product in which sea ice extent and area differences between the sensors are reduced and could serve as a baseline for future measurements.”

http://nsidc.org/data/docs/daac/nsidc0051_gsfc_seaice.gd.html Revision: April 2008

It is understood from this that the +/-5% and +/-15% apply to both measures. It was assumed that these vary sinusoidally throughout the season such that an uncertainty of +/-5% applies in peak of winter (greatest extent) and +/-15% in summer (lowest extent). Calculating the uncertainties and plotting them with the seasonal variation yields:

Now calculating the trend and uncertainty bands provides a clearer picture, thus:

The apparent decline in Arctic sea ice extent is about the same magnitude as the estimated measurement uncertainty.

We will now compare the SSM/I (NASA Team) against AMSR-E (IRAC JAXA). For this the same datasets are used, and the temporally overlapping data are compared. In both cases the data has been interpolated and re-sampled to contiguous daily values.

Interestingly there are notable differences between them. They agree reasonably well in the summer, but show large discrepancy during winter, peaking at about 7%. This discrepancy is quite possibly the result of the differing algorithms. The annual rates of change over the period 21/6/2003 to 31/12/2007 were calculated each day and plotted:

And the NASA Team annual rate of change chart follows:

In both cases linear fits have been applied in order to obtain a measure of the relative trends. The equations of the fits are:

AMSR-E (IRAC JAXA) Y = 3695647 -100.688X (-36800 km2/year/year)
SSM/I (NASA Team) Y = 5253773 -141.285X (-51600 km2/year/year)

Thus it can be seen that, over this period, the SSM/I has a slope about 40% greater than AMSR-E.

It is therefore little wonder that NSIDC do not want to switch to AMSR-E since it would definitely show a marked change in the long term trend.
Having noted this, how well do the datasets match over this era? Since the period is relatively short and AMSR-E is claimed more accurate in the short term it is used as the reference and thus the SSM/I is plotted against it:

Two linear fits have been applied, one forced through the origin and the other unforced. Both have a reasonable R2, and are described by the equations:

Y = 1.0561X R2 = 0.9967

Y = -562213 + 1.1076X R2 = 0.9991

Clearly there is a significant mismatch between the two measurements, and this would definitely need to be taken into account if ever the two were used in the same time series.

Another way of considering this is to look at the rate of change between the measurements as is shown in the following:

The offset is close to zero and therefore of little consequence, but there is still an almost 5% difference in slope.

Yet another way of analysing the difference is the calculate the ratio of one to the other and plot it as a time series. In the next image, this has been done and a 365 point moving average has been added to reveal the underlying trend. A linear fit was then applied to the moving average resulting in the following:

The difference between the two satellites varies seasonally, however, the trend is relatively smooth, but declining, as shown by the red line. From the linear fit to the trend, a decline of about 0.0017 per year relative to AMSR-E (IRAC JAXA) is evident. Obviously, this is a rather short time frame from which to draw any major conclusions. However, assuming this trend is representative of the entire satellite era and that AMSR-E provides more stable measurements, then the observed differences might account for a substantial proportion of the reported Arctic sea ice decline. In simple terms, this may be indicative of instrumentation or other drift.

In summary, the following major points are highlighted:

1. Reconstructions of sea ice extent show no obvious decline until the advent of the satellite.
2. The claim that Arctic sea ice extent has been declining since 1953 is unjustified since there was no significant decline until 1979 when it exceeded 1 standard deviation below the 1953 to 1978 mean.
3. The satellite sensors drift, and as has been aptly displayed by SSM/I recently, there is at least one failure mode that results in underestimation of the sea ice extent.
4. The proclaimed unprecedented Arctic sea ice melt of 2007 does not appear particularly unusual and a similar but much more severe episode occurred in 1995, although at that time the measurements showed a higher starting level.
5. Presently, based upon the reportedly “most accurate” satellite data from AMSR-E (IRAC JAXA), Arctic sea ice extent is increasing at an accelerating rate.
6. Significant differences in winter ice extent are observed between NASA team SSM/I and AMSR-E IRAC JAXA during the overlapping period that might be explained by differing algorithms.
7. NASA team SSM/I shows a decline in extent relative to AMSR-E IRAC JAXA that might account for a substantial proportion of the apparent Arctic sea ice loss since the start of the satellite surveys.

Jonathan Drake 16/3/2009
Questioning Climate