One of the concerns that people are increasingly expressing to us in reaction to demonstrations of methodology problems in Hockey Team multiproxy studies is more or less this: even if the methods used in these various studies are flawed from the point of view of a statistical purist, no one’s presented an alternative interpretation of the proxy data. Here’s a recipe for apple pie instead of cherry pie.
Hockey Team methodology in most of the studies is this: pick a small subset of all the proxies (which in their case are chosen so that the subset has a 20th century ramp e.g. bristlecones or so that there is a combination of mere noise plus a few with a 20th century ramp). Here I’ve done a little exercise – not to present an alternative view of climate history – but to show that, if, from the population of proxies, you try to make apple pie instead of cherry pie, you can do so.
Here I’ve picked 8 series from my files not randomly, but because I knew that they had elevated MWP values, scaled them and made an average (which is more or less what non-MBH98 Hockey Team methodology is.) If I wanted to change the scaling properties of the series, there are proxy series with whatever noise properties that you want. This is my first run. So it is picked , but not tuned. The number of series in Hockey Team subsets in the 11th century portion of Jones et al 1998 is only 3 and Moberg uses only 11 series for his low-frequency portion. I could add a couple and make 11 and it wouldn’t change the point. For example, the Iceland sea-ice used in Crowley; the GRIP borehole data …
6 of 8 selected series use data from canonical multiproxy studies. The Sargasso Sea, Conroy Lake and Indigirka series are used in Moberg (Sargasso also in Crowley). (Moberg only uses the high-frequency tree ring information from Indigirka, but obviously low-frequency tree ring information is regularly used in Hockey Team studies.) I’ve used the updated Polar Urals ring width series from Esper et al 2002. I’ve used a subset of the Yang composite – the subset without Thompson’s Dunde and Guliya ice cores, but the subset is the majority of the data. I’ve used Cuffey’s Greenland series, as archived by Alley . I’ve used Mangini’s speleothem data. I’ve used the Polar Urals treeline data digitized from Briffa et al . This is a little repetitive (but Hockey Team studies sometimes use two bristlecone/foxtail sites e.g. Esper, Osborn and Briffa). I don’t have a digitized bristlecone treeline series, but it would be similar.
So here are the raw series. Their inter-series correlations are pretty good by Hockey Team standards and I suspect that, even without tuning, surpass the performance of some selections in terms of containing a signal, especially the sought-after "low frequency" signal. Maybe I’ll work that up. All series have been smoothed with 40-year gaussian filters, a common Briffa technique.
Figure 1. Eight proxy series, all smoothed with 40-year gaussian filter.
Next here is a "reconstruction" using these proxies. The left panel shows the 2000-year history; the right panel shows the reconstruction fitted against the CRU instrumental record. The correlation is 0.51 and the correlation between smoothed versions is 0.81.
Figure 2. "Reconstruction. Right panel – reconstruction compared to instrumental. Vertical scale: deg C. basis 1960-1990 (from fit to CRU)
I haven’t tuned all the bells and whistles. For example, I haven’t done a calibration-verification exercise yet. But you’re starting off with something that you can tune to have a terrific RE value if it doesn’t already. Where you sometimes run into trouble on RE statistics is that the scale of the reconstruction overshoots in the verification period i.e. you’ve got the trend, but you lose the RE statistic because the reconstruction overshoots. You can tune this by adding in white noise series, of which there are an abundance in the proxies (see my post on Huybers #2 for this.)
I’m not saying that this is an alternative reconstruction of temperature. The point is that cherry pie is not only thing that you can make from the proxy orchard.