Antarctic Agreements and Disagreements

Guest Post by Willis Eschenbach

Steven Mosher has pointed out a Science Magazine article (subscription only) about Antarctica. It is a discussion of the temperature changes in the West Antarctic Peninsula (WAP). And where might that be?

Figure 1. Location of the West Antarctic Peninsula. Yellow push-pin markers show the location of temperature stations. Yellow outline shows the enclosing area used for temperature calculations. This is the smallest area using 5°x5° gridcells that contains all of the WAP temperature stations.

The Science Magazine article contains the following statement about the WAP, which set my bad number alarm bells ringing:


Physical Changes in the WAP

Changes in the WAP are profound. Mid-winter surface atmospheric temperatures have increased by 6°C (more than five times the global average) in the past 50 years (14, 15).

I had never looked closely at the WAP temperatures. However, that seemed way high for the changes in the WAP air temperature, no matter what month we are talking about. The references for that statement are:

14. P. Skvarca, W. Rack, H. Rott, T. I. Donángelo, Polar Res. 18, 151 (1999).

15. D. G. Vaughan et al., Clim. Change 60, 243 (2003).

I couldn't find a copy of either of those on the web … so I did what I always do. I went to get the data, to see what is happening.

Initially, the situation looks good. There are thirty stations on the peninsula. Figure 2 shows the location of some of these stations:

Figure 2. Location of the temperature stations in the WAP

So, what's the problem? As you might imagine, many of these stations are only occupied part of the year. Others have been occupied intermittently, or have closed entirely. As a result, we don't have anywhere near the coverage that thirty stations would imply. Here's a graphic showing the dates of coverage for each of the thirty stations:

Figure 3. Dates of coverage for each of the thirty stations in the WAP. Only a half dozen or so show coverage over most of the last fifty years

Things are not as good as they seemed. Some of the datasets ony cover a few years. Others are longer, but very spotty. However, as they say, "Needs must when the devil drives". Here is what all of the different stations look like:

Figure 4. Plot of all stations on the West Antarctic Peninsula. You can see the difficulty in determining an average temperature change over the area. Some stations swing quite widely, while others show much less variation.

Are the winters warming? Well … obviously, it depends on exactly which datasets you use to create your area average. Do we include the spotty orange dataset that starts about 1986, or not? What about the blue datasets that only exist for the sixties and seventies? Based on these decisions, our answers will be different.

Next I looked at the major datasets. As you know, there are several temperature datasets that cover the globe. For the land alone, we have the CRUTEM3, GISS 250 km, and GISS 1200 km land datasets. The two GISS datasets use the same surface stations. However, they differ in that they extrapolate the temperature of empty gridcells using all relevant stations within either a 250 km or a 1,200 km radius respectively.

All of these are available at KNMI, which is an outstanding resource. Here are the month-by-month trends for each of those datasets:

Figure 5. Month-by-month and annual ("Ann") trends for the air temperatures (land only) for the area of the West Antarctic Peninsula outlined in yellow in Figure 1.

There are several interesting things about this graph. First, a simple average of all of the stations ("All Station Average") gives results that are broadly similar to the CRUTEM results. I assume that this is because of the general similarity in the climate zones of the 30 temperature stations around the peninsula, which allows for a direct average rather than the more sophisticated methods (anomalies or first differences) as used in the global datasets.

Next, in several months there is a difference of a full degree (per fifty years) in the trends of the CRUTEM and the GISS datasets. The various datasets are often claimed to be in good agreement. But this is only globally. When we get down to a gridcell-by-gridcell and month-by-month comparison of the trends, they are often quite different.

Since they are (presumably) using the same basic data (the 30 land stations), this is odd. Note that the annual trends are in reasonable agreement, but the monthly trends differ … why should that be?

The effects of the GISS algorithm for filling in the empty gridcells are also curious. Depending on the extrapolation radius chosen (250 km or 1,200 km) they differ by up to a half a degree in fifty years.

Finally, none of the datasets show a temperature rise of 6°C in fifty years in any month, as the Science paper claims. My bad number alarm was accurate. So I'm in mystery about where that claim might come from. August has the highest trends, at three to four degrees depending on the dataset chosen. But that's a long way from six degrees.

Now, it is often said that the warming of the Peninsula is due to warming of the surrounding ocean. So I decided to take a look at that as well. Here are the same datasets, showing both the land and the ocean:

Figure 6. Land and ocean temperature trends for the area outlined in yellow in Fig.1

Here, the differences between the datasets are larger. For the first five months of the year the CRUTEM+HADSST dataset shows a much smaller trend than GISS, up to a a degree and three quarters smaller. The rest of the year, the datasets are much closer than in the first five months. Why would they be different in part of the year, and not the rest of the year?

In addition, this dataset makes it unlikely that the ocean is driving the warming. The trends including the ocean are almost all either the same or smaller than the land-only trends. This is particularly true of the CRUTEM vs CRUTEM+HADSST datasets.

Finally, I took a look at a shorter period, from 1979 to 2009, so that I could compare trends from the ground-based datasets with the UAH MSU satellite based dataset. Here is that data:

Figure 7. Ground and satellite data compared for the area outlined in yellow in Fig. 1. Note that these are thirty year trends rather than fifty year trends, as shown in the other figures.

Here, things get markedly odd. The satellite data shows cooling in about half the months. The overall annual satellite trend is … zero. Go figure. We see much greater differences between the ground based sets. The GISS peak warming is no longer in August, but in May. None of this makes a whole lot of sense … but there it is.

Final conclusions?

First, once again some mainstream climate scientists are exaggerating. There is no dataset in which we see a WAP air temperature rise of 6°C in fifty years as claimed in the science paper.

Second, although it is widely claimed that there is good agreement between the various ground based datasets, as well as between the ground and satellite data, in this case we see that they are all quite different. Not only the amplitude, but in many cases the sign of the trend is different between ground and satellite data. The CRU/Hadley dataset varies from the GISS datasets. In all, there is not a whole lot of agreement between any pair of datasets.

All of which makes it very difficult to come to any conclusions at all … except one.

My only real conclusion is that it would be nice if we could get some agreement about one of the most basic data operations in the climate science field, the calculation of area averages of temperatures from the station data, before we start disputing about the larger issues.


The surface temperature data stations used for Figures 3 and 4 are identified in the GISS dataset as:

Teniente Matienzo (Ant South A
Base Almirante Brown
Dest. Naval Melchior
Palmer Station
Bernado O'Higgins
Dest. Naval Decepcion Sout
Deception Is. S Atlanti
Base Esperanz
Hope Bay
Base Arturo P
Centro Met.An, Marsh
Admirality Bay
Rothera Point