There has been a good deal of discussion regarding the correlation between temperatures at various locations throughout Antarctica. Several people have looked at the relationships between correlation and distance by creating graphs linking the two. One of the difficulties in interpreting these is that they are affected by a variety of factors, including the shape and topography of the continent and by the fact that the place is completely surrounded by a large pool of water.
It is informative to pick several locations and to see how the AVHHR series at that location is related to all other locations. I selected two points: the tip of the peninsula (Steig series 1) and the obvious interior point: the South Pole (grid point 1970 is the closest).
For a selected site, after calculating the 5509 correlations, we graph them using a color scale to represent the correlation (as usual red is positive and blue is negative and white areas represent zero correlation). The location of the grid site is represented by a green +. Keep in mind that these are correlations measuring relationships between temperatures at the grid point, not positive or negative trends.
First, we take the latest revelation from Steig, the cloud masked AVHHR data. The grid point is on the tip of the peninsula.
Several things stand out in the graph. Obviously, the region immediately adjacent to the grid point is strongly correlated, but what is somewhat surprising is that the correlation drops off fairly becoming negative while still in the Western Antarctic area. The relatively low correlation continues to the rest of the continent.
Next, we take the original reconstruction: ant_recon.txt. This was supposedly reconstructed from the previous data using RegEM and the manned surface stations:
The correlation has strengthened dramatically in the Western Antarctic so that now the pattern exhibited by the reconstruction at the tip of the peninsula seems to be reflected by the entire west. As wll, the Eastern portion has now become more strongly inversely correlated with the peninsula.
There were two other reconstructions done by Steig to demonstrate “robustness” of their procedure. One was to “detrend” the original reconstruction in the period from 1982 to 2006 presumably before redoing the RegEM. Using the data from that reconstruction we get:
As well, they apparently did a 3 PC reconstruction which I believe did not use RegEM. I tried to relate it to the PCs you get from the cloudmasked data set, but sofar I have been unsuccessful. Another puzzle to solve! Anyway:
The last three graphs are pretty much identical – to each other – and NOT to the AVHRR data purported to be the basis of the reconstructions. The robustness seems to be a result of whatever else was done by their less than amply described procedures.
The same process used above was repeated, this time choosing the South Pole location as the comparison grid point, In the same order:
This time the strong correlation in the region immediately adjacent to the pole has been stretched to most of Eastern Antarctica. My guess would be that this is due to the presence of a surface station at that spot correlation was in the original AVHRR or (whose record is fairly strongly correlated to the AVHRR data given by Steig). Whether that correlation is in the original satellite data or induced by procedures used to remove extreme values is unclear since those procedures have not been adequately described.
The R script for reproducing these results is found here. It is a Word document called corr_script.doc. I have put it in that format since WordPress screws up the quotes if I enter it into the post and it also won’t let me upload ordinary text files or files of most any other format.
Update: I have added two more plots to the post by request. These correspond to the grid point at the Casey station (coordinates: lat = -66.3, lon = 110.5) on the opposite side of the continent from the peninsula. The results:
Update 2: Due to further requests, I have added two more grid sites to the mix:
Grid site 817: Coordinates: (240, -80)
Grid site 3855: Coordinates: Vostok