Thanks.

w.

Are you comparing temperatures to the mean of all the observations or just the observations for the same month? At Realclimate, they are comparing the most recent April temperature to the average of all April temperatures, not the temperatures from all months in all years. Obviously, you have to adjust for seasonal observations.

If so, perhaps both of you could provide some evidence for your claim? Also, to clarify the question, perhaps both of you could define what you mean by a “better proxy”? It’s a harder question than it appears, as the tree rings’ value as a proxy depends heavily on which month of the year, or the past year, we are discussing.

Finally, the paper I have been discussing on bristlecones shows the results for all of the high altitude bristlecone sites studied. Using the “response function” as a measure of how good a proxy is, the results are mixed. At some sites, bristlecones are a better proxy for temperature than moisture, and at some sites it’s the other way around. There is no site which is what I’d call a “good” proxy site for either one.

Go figure …

w.

Dano, if you ever find some scientific issue to comment on, please do.

You missed it, willis.

The scientific issue, willis, was that SS asserted that the trees upthread were a better proxy for moisture than temperature.

Again, the issue was that the assertion was the trees upthread were a better proxy for moisture than temperature.

To restate: the scientific issue was the value of moisture vs temp in certain tree rings. That was the scientific issue.

I, therefore, told SS to write up his ground-breaking finding to share with the dendro community.

I’d be happy to hear, for example, if you find anything wrong with my analysis.

It looks like you’ve crunched some numbers and that they disagree with numbers given elsewhere. You’ve done an “audit” that you claim is a…er…tipping point. Skimming your post for 30 seconds, I have nothing to add, except that I’m not as qualified as you, apparently, to judge statistical interpretation. I presume your comment is being looked over.

Best,

D

I see my statement might be misinterpreted. I should have been more accurate. What I meant was that we would expect to find 3 or 4 months with a sigma of 2.7 or greater. Upon investigation, I find that there are in fact four such months in the dataset

All the best,

w.

I’d be happy to hear, for example, if you find anything wrong with my analysis.

Thanks,

w.

Let me know, willya, when your groundbreaking paper gets accepted? Thanks!

Best,

D

Since there are 1108 data points in the full dataset, we’d expect to find 3 or 4 months like April 2006.

Do you find 3 or 4 months like April 2006 in the dataset?

The difference is quite significant. Here is the number of standard deviations (sigma) away the 1961-1990 mean, and away from the dataset mean:

Month__________ ’61-’90_______Full Dataset
Dec 2005__________2.2__________1.9
Jan 2006__________2.7__________2.4
Feb 2006__________1.8__________1.3
Mar 2006__________0.7__________0.5
Apr 2006__________4.6__________3.5

Every one of these is smaller.

Then they say:

The April mean temperature is almost 5 standard deviations above the mean, a “5 sigma event” in statistical parlance. Under the assumption of stationary ‘normal’ statistics, such an event is considered astronomically improbable (less than 1 in 10^6),

Well, this is a little wonky as well. While a 5 sigma event is less than 1 in 10^6, they don’t show a 5 sigma event, they show a 4.6 sigma event (less than 1 in 10^5). Using the real figures, however, the number is less than 10^3. And this is using normal i.i.d. statistics.

The lag 1 autocorrelation of the statistics is a bit hard to calculate, due to missing values. Interpolating the values is difficult as well, as an entire year is missing during the time of the changeover between the stations. However, the values do not change much from the missing data (~0.01-0.02). Here, I have used the more conservative figures. The effect of the autocorrelation is to increase the standard deviation, so the values have smaller “sigma” statistics. Here is the complete table:

Month__________ ’61-’90____Full Dataset____Full Dataset w/autocorrelation
Dec 2005__________2.2__________1.9__________1.5
Jan 2006__________2.7__________2.4__________1.8
Feb 2006__________1.8__________1.3__________0.9
Mar 2006__________0.7__________0.5__________0.4
Apr 2006__________4.6__________3.5__________2.7

Note that the largest sigma in the bunch represents a one in 288 odds of occurence (less than 1 in 10^2). Since there are 1108 data points in the full dataset, we’d expect to find 3 or 4 months like April 2006.

It is also worth noting that the December, February, and March temperatures were not the warmest in the record for those months. The March temperature has been exceeded no less than 27 times in the past …

w.