Here’s an example of how one can wander off curious little by-ways in trying to replicate Hockey Team materials. I’m working on the seemingly simple task of testing the correlations to gridcell temperature in Osborn and Briffa 2006. One of the series – a "quiet" series – is Fisher’s "West Greenland dO18 Stack". Also with Svalbard in the news so to speak, and Jean S posting up about Norwegian temperatures, I thought it would be interesting to look at some Greenland temperatures as well.
The sites used by Fisher are all about 71N, between 35 and 40W (one at 45W). But one you go to check the HadCRU gridcell, it doesn’t have any values. Nor do the cells to the left or right. The closest gridcell with values was 67N, 37W, one gridcell to the south. You’d think that this might have been what they used, especially since the sites are just a titch north of the gridcell border. But I had some other Hockey Team information to reconcile. Some time ago, I got information from Phil Jones on gridcell blocks used in Jones and Mann 2004 for correlation – here they based the comparison on some kind of average from all gridcells in the blocks 62.5-72.5N; 57.5W-42.5W, i.e. they did not use the closest block. It seemed odd, but, hey, it’s the Hockey Team.
When you look at a map of Greenland, the gridcells used in the Jones and Mann comparison are all on the west coast of Greenland, while there was another set of values for gridcells along the east coast of Greenland, with the center, where the ice cores are, lacking any values. In passing, the ice cores are all drilled from sites that are very high – about 3100 m.
So was there any difference in the temperature histories on the two sides of the island. Here I’ve plotted up the three gridcells with the most values on each side of the island, showing here the plot of monthly values, which contains some texture lost in the annual data. Take a look at the west coast sites in Figure 1 and the east coast sites in Figure 2. I’ve shown "trend" lines in each case. The data goes up to the end of 2004.
Figure 1. HadCRU2 temperature anomalies for three west coast Greenland gridcells.
Figure 2. HadCRU2 temperature anomalies for three east coast Greenland gridcells
A couple of features catch my eye. First, there isn’t a big trend on either coast. Second, the 1930s were warmer than the present, a point we’ve heard on many occasions. Third, the amplitudes of the east coast gridcells is lesas than the west coast gridcells. Fourth, there is little change in upspikes; in "warm" periods, what you see is more an absence of downspikes, which characterize the late 19th century. Any slight trend in a couple of gridcells is dependent on these 19th century downspikes. I’m not drawing any conclusions from it, but there is no obvious "Arctic amplification" here; in fact, there is none. So if one of the fingerprints is Arctic amplification, CSI could find no fingerprints at this crime scene. The Hockey Team must have wiped off the prints and "moved on".
I made annual averages of the west coast and east coast gridcells and these ended up being pretty similar and look more or less like the individual gridcells.
Next, I compared Fisher’s dO18 stack to the west coast gridcell average (using the presumed Jones and Mann comparand.) The top panel below shows the dO18 series from 8000 to the present and the bottom panel compares the dO18 re-scaled to have the same scale as the instrumental record in the 1800-2000 period.
Figure 3. Top – Fisher dO18 stack; bottom – black -Fisher dO18 stack rescaled; red- west coast Greenland gridcell average.
Again, there are lots of interesting features to this plot. The dO18 series has remarkably little centennial variation. It has no Little Ice Age or MWP. (It was incorporated as a canonical Hockey Team series by Bradley and Jones 1993, arguing against the Little Ice Age, and has been used in virtually every Hockey Team study since then.) No wonder Bradley and Jones like it. The only problem is that it doesn’t have a 20th century pulse either. But if you splice in some short HS-shaped series to force a blade (e.g. Jacoby NH or C England documentary, that type of thing), it’s a great series for making a hockey stick shaft, if htat’s what you want to do.
Fisher 1996 (NATO) noted this lack of centennial texture, but placed in a broader context. He noted that this set of cores (and also GISP2) had very little centennial variability throughout the entire Holocene. In this respect, they were very different from many other cores (e.g. Agassiz, Devon) which had very warm Holocene Optimum – significantly warmer than modern or medieval – and considerable centennial variability. Fisher noted the problem, but had no firm explanation. He posited that the GISP2 and the Fisher stack cores were very high and he speculated that their very coldness and remoteness made them less susceptible to change. As an analogy, think of a LIFO inventory – this is the inventory base that never gets changed. Anyway, Fisher had no explanation -he said, however, that, until you solved the big problem of Holocene lack of variability, it was mere trifling to to worry about centennial variability. Fisher, by the way, did not have a Hockey Team approach to data. I asked him in November 2003 about one of his studies, just after MM03. He inundated me with data; he sent me a diskette with lots of data and copies of numerous interesting papers.
Looking at the proxy and temperature average on a similar scale, one thought does occur to me. In the warm 1930s, the dO18 record continues to have major downspikes that are not present in the temperature record. Now in the monthly temperature histories, there were more 19th century downspikes than 20th century downspikes. The gridcell histories are based on more moderate coastal records, maybe the cold period downspikes continue to be prevalent at altitude. Just musing.
There are other factors in accumulation of annual dO18 records. There is far more variation within a single year in dO!8 levels from summer to winter than there is interannually. The majority of accumulation is in the summer. Fisher notes that individual stormy years (1916) show up in the record. To the extent that the dO18 is a temperature record of this high-altitude location, there is even less evidence of 20th century uniqueness in this record than in the coastal gridcells.
It has decent correlation to the gridcell average (.5), but would serve as a classic example of a proxy where most of the correlation comes from high-frequency scales, since it has negligible centennial variability. Briffa and others validate the use of this proxy because of its correlation (r) to gridcell temperature. Wait a minute, didn’t we just read the Hockey Team en masse saying that the r2 was an invalid statistic? And aren’t they using the correlation r here (and everywhere) to validate proxies? I guess it was the squaring process that made it invalid.