Wilson et al 2007

Mar 24, 2007 – 10:59 AM

Rob Wilson has referred us to Wilson et al 2007. In addition to being an example of site selection, Wilson et al 2007 uses a type of principal components on a tree ring network – something that should be of interest to many CA readers – and an interesting illustration of non-Mannian statistical methods within the tree ring community.

Update (2016):  In comments to this post, there was considerable criticism of the failure of the authors to archive data at the time of publication, with one of the authors taking offence.  Unfortunately, as critics had feared, the missing measurement data was not archived until five years later (2012), with one site-species dataset still not archived.  Continue reading →

By Stephen McIntyre | Posted in Divergence, Multivariate, Noamer Treeline, Proxies | Tagged , | Comments (105)

Latitudinal Treeline

Mar 23, 2007 – 1:03 PM

A couple of days ago, I canvassed the ITRDB data bank for all tree ring series with values later than 1998, noting that there were many new series. Mann has justified the reliance on proxies not updated since 1980 and earlier, and thus not calibrated against recent warmth, on the basis that it is very expensive and time-consuming to do this – a justification which I believe to be laughable on its face. Since white spruce (PCGL) was used in treeline studies in the past, I did a quick collation of new PCGL series as a species identified previously as a temperature proxy to see whether ring widths for new contributions to the ITRDB data bank reflect recent warmth, as they should if tree ring widths were linearly correlated with temperature, as Mann and others have hypothesized.

I didn’t do any sub-analysis to see if the specific white spruce sites were “temperature sensitive” for which Mike Pisaric and Rob Wilson criticized me. My response – and that of other readers of those threads – was primarily to point out that this is what was done in MBH and other studies, which Pisaric and Wilson had never spoken against. Realistically, they aren’t going to speak out against Mann’s use of precipitation proxies in a temperature study so there’s not much point belaboring the matter further.

So let’s go back and re-visit their criticism of how I presented the information. They said that the PCGL series used in my average should not be used in an average because they are not “temperature” proxies but moisture proxies, as anyone can tell by looking at their location, and thus the average was meaningless. As an exercise, I’ve plotted the location of the PCGL proxies with post-1998 values in the map below, colored according to 5 author groups. I’ve also marked ( with a + sign) the 11 “treeline” sites of the original Jacoby and D’Arrigo 1989 study and joined all except the Gaspé site with a line (the Gaspé site is labelled both because it is anomalously located for both the Jacoby and D’Arrigo 1989 “treeline” study and heavily weighted in MBH98. There are many new Jacoby-D’Arrigo sites that have not been archived.

The Meko sites are shown in blue and, as Mike Pisaric says, are well away from latitudinal tree line. I’ve marked with smaller dots the location of all sites in the North American ITRDB data set (actually they go even further south), so you see that a PC1 taken from the MBH98 dataset includes almost no latitudinal treeline sites (and none earlier than about 1500). The Wilmking sites along the Brooks Range, Alaska look as though they must be pretty close to the latitudinal treeline, despite Rob Wilson’s caveats against Wilmking data. Sites in Alaska and along the Rockies may also be taken at altitudinal treelines so this has to be kept in mind when one looks at the map.

My objective was to see what relevant fresh data showed. Looking at this map, my conclusion is that the new Wilmking and Lloyd data from the Brooks Range, Alaska and the Jacoby site in Labrador are the only fresh data that are close to the latitudinal tree line and, accordingly, that examining this particular subset is the best implementation of the implicit Wilson-Pisaric criteria that can be done under current circumstances. There is a considerable amount of presumably relevant Jacoby-D’Arrigo data which is unarchived. But as long as it is unarchived, there’s not much that third parties can say about (and, in my opinion, until the data is archived, the study should not be cited by IPCC or others.)

wilson10.gif
Colored dots – sites with data from 1998 and later; black dots- ITRDB sites. + – Jacoby and d’Arrigo 1989 sites.

By Stephen McIntyre | Posted in Jacoby, Noamer Treeline | Tagged | Comments (29)

Wilson, Pisaric and Gaspé

Mar 22, 2007 – 2:35 PM

Two respected dendrochronologists wrote in criticizing my recent post averaging new white spruce chronology contributions to the ITRDB data bank. [Update: see post here with further thoughts on the location of this site. Next post in category] Rob Wilson wrote, using an uncharacteristic Gavin-esque sigh, as follows:
Continue reading →

By Stephen McIntyre | Posted in Jacoby, Noamer Treeline | Tagged , , | Comments (67)

Phil Jones and Sitka, Alaska

Mar 22, 2007 – 10:01 AM

I wa reading through Rob Wilson’s article on Gulf of Alaska ring widths and the following sentence caught my eye:

Mean temperature data for Sitka (1832—1887), that are not included in the GHCN archive, were also obtained (Phil Jones personal communication).

You may recall CRU’s refusal letter to Willis Eschenbach, one of a number of recent formal refusals by CRU to even identify their sites, much less provide their data. In that letter, CRU said:

We have sent all our data to GHCN, so they do, in fact, possess all our data.

Did Phil Jones send all his data to GHCN or not? If he did, why isn’t it included in the GHCN archive? There is 19th century data for Sitka in the GHCN archive: so what’s the difference between this data and Rob’s data? If it’s different, what’s the reason for the difference? If it’s the same, what did Jones mean when he told Rob Wilson that he had provided him with data that was “not included in the GHCN archive”? Rob criticizes me for being confused. Sometimes it’s hard not to be.

I’ll add that Rob Wilson has nothing to do with CRU refusing to disclose what stations they used, and, I’m sure that, if it were up to him, CRU would disclose this information even to people who aren’t “friends”.

By Stephen McIntyre | Posted in Surface Record | Comments (6)

NOAA Gridded Data

Mar 21, 2007 – 2:35 PM

We’ve discussed CRU and GISS gridded data, but many of the recent news stories about the “warmest winter” come from NOAA gridded data (for example here and here) , which seems to be gaining a little market share of news attention for gridded data.

I’ve started taking a look at the data. Given the intransigence of Phil Jones and CRU in refusing to disclose their station selection and methodology and the fact that NOAA is presumably subject to the U.S. Data Quality Act, there may be some advantages to trying to figure out how NOAA gets its results.

I’ve done a first pass in trying to replicate an individual gridcell and have replicated some features and not others. Maybe others will have some ideas.
Continue reading →

By Stephen McIntyre | Posted in Surface Record | Tagged , | Comments (37)

Al Gore on C-Span

Mar 21, 2007 – 7:25 AM

CSPAN webcast of House Energy Subcmte hearing on Global Climate Change starring Al Gore – starting 9:30 ET today.

By Stephen McIntyre | Posted in AIT | Comments (131)

The Briffa-Osborn Variance Adjustment

Mar 20, 2007 – 2:49 PM

UC inquired about the variance adjustment in Osborn et al (Dendrochronologia 1998), which is used in many Team publications. The number of series in many reconstructions declines as you go back in time. If you take an average of standardized series (the CVM method), the variance over an early time interval will be larger than the variance in a later time period.) The BO variance adjustment was used originally in proxy reconstructions but this procedure or a variant seems to have been introduced into some of the CRU temperature gridcell series as well. The adjustment is described as follows:

Each regional mean thus obtained tended to have greater variance during years when few chronologies were available to contribute to the average; this effect was corrected for by scaling by the square root of the effective number of independent samples available in each year.

First they state

Let’s assume that one starts with a set of series all standardized to 0 mean and sd -1. Then if \overline{X} is the average of n series with a mean correlation \overline{r} ,

(1) Var (\overline{X}) = \frac{1+ (n-1) \overline {r}}{n}

If the series are uncorrelated (\overline{r} =0 , the variance goes down to 1/n;

(2) Var (\overline{X}= \frac{1}{n}

whereas if the series are perfectly correlated ( \overline{r}=1 ), the variance stays at 1. They assert:

“an artificial signal will be introduced into the variance of Xbar if the sample size varies through time.”

Comparing (1) and (2), they define the “effective independent sample size” as follows:

(4) n'= \frac{n}{1 +(n-1) \overline{r}} ;

They express (1) as follows

(5) Var(\overline{X})=\frac{1}{n'}

When one thinks about this, this is a very odd terminology. This is measuring not so much the “independent sample size” as the relative lack of coherency in the sample – but let’s proceed, holding this thought. They make the unsurprising observation: “If \overline{r} is low, variance will increase strongly as n falls below 10.” They observe that \overline{r} in western U. S. confiers are about 0.6; in eastern US deciduous hardwoods about 0.3 and as low as 0.2 in deciduous European sites; and from 0.28 to 0.71/.74 for Siberian RW and MXD sites. They illustrate (Fig 2a) the average of 8 sites in S Europw where variance increases pre-1750 as n decreases (\overline{r} is only 0.07).

They go on to say:

The method presented here is theoretically based …Equation 4 provides the time-dependent effective sample size if supplied with the time-dependent available sample size. We would then expect the variance of the mean timeseries to vary according to equation (5). If we adjust the mean timeseries by

(6) Y(t) = X(t) \frac{1}{\sqrt {n('t)}}
then we would expect the variance Var(Y) to be independent of sample size (but would still have any real variance signals that are present in the data)”

They go on to discuss a couple of variations, where \overline{r}latex varies with time, but the idea is the same. Briffa and Osborn do not provide any third-party statistical references for this procedure.

Here is a function to implement the BO adjustment. rbar0 can be a time series or a constant.

#rbar0 is a vector of length of the total series, calculated in various ways externally
bo.adjust< -function(js.mean,rbar0){
count.eff<- count/(1+(count-1)*rbar0);#
NN<-max(count,na.rm=T)
count.eff.max<- NN/(1+(NN-1)*rbar0) ;#
var.adj<-sqrt ( count.eff/count.eff.max) ; #equation 7 of Osborn et al.
bo.adjust<-js.mean*var.adj
bo.adjust
}

I’ve included a script here illustrating the use of this method in attempting to replicate the archived version of Jones et al 1991. I can more or less replicate a smoothed version of the archived reconstruction, but the difference between my attempt to replicate Jones et al 1998 and the archived version can be up to 0.5 deg C in individual years. (And we’re told that these reconstructions are accurate to within a couple of tenths of a degree or so.)


Top – comparison of emulation to archived as smoothed; bottom – difference between emulation and archived version.

As to the Briffa-Osborn adjustment itself, if you have series with relative little inter-series correlation, one expects the variance to increase by reason of the Central Limit Theorem. Does the Briffa-Osborn adjustment do anything other than disguise this? I think that someone on the Team needs to prove the validity of the methodology statistically. Of course no one on the Team bothers. They just advocate a recipe and then assert it.

By Stephen McIntyre | Posted in Jones et al 1998, Statistics | Comments (16)

Up-to-Date White Spruce Ring Widths

Mar 19, 2007 – 9:24 PM

“Bring the proxies up to date” was the title of one of my earliest posts. Michael Mann had explained that doing so required the use of heavy equipment (like tree ring borers) and travel to out-of-the way sites such as Bishop, California or even Niwot Ridge, a full 45 minute drive from UCAR world headquarters in Boulder CO. As a result, Mann explained that few proxies were available after 1980 and it was therefore necessary to keep using bristlecone and other series ending in 1980 or so. [See next post in sequence.]

Obviously, given the warm temperatures in the 1990s and 2000s, up-to-date proxies reaching 1998 and later offer an ideal opportunity to test the validity of tree ring proxies out-of-sample. I’ve done a quick calculation of contributions to the North American tree ring data base at WDCP; I counted no fewer than 250 sites where there is data for 1998 or later.

The species with the most new sites is Douglas fir (PSME); there are also many new white spruce (PCGL) site – a species held to be a temperature proxy. There are 48 sites where new measurements have been archived, many due to the work of the Jacoby group in Alaska and northern Canada. Jacoby’s archiving is frustratingly incomplete however. Also presumably to be annoying, virtually none of the Jacoby chronologies are archived with the measurements. (This is the opposite of Briffa for Yamal, Taymir and the Tornetrask update, where the RCS chronologies are archived, but not the underlying measurement data.) The Team never makes it easy. In the new data, there are no bristlecone (PILO, PIAR) or foxtail (PIBA) sites with archived data for 1998 and later, although we know that Hughes carried out measurements at Sheep Mountain in 2002. (One of my standing predictions is that Sheep Mountain bristlecone ring widths did not go off the charts during the warm 1990s.)

Even without any Jacoby chronologies, there are still 14 PCGL chronologies reaching 1998 or later. The figure below shows the average of chronologies (the average of available chronologies.) I have trouble discerning a HS in this average. (I didn’t use Mannian principal components.)


Figure 1. Average of 14 North American PCGL chronologies

If the proxy hypothesis is that PCGL ring widths increase with increasing temperatures, I would say that these non-Jacoby chronologies are evidence against the hypothesis. Indeed, one might even be inclined to say that the average of these 14 chronologies is inversely related to temperature, as ring widths were low in the warm period in the late 1930s and increased to higher levels in the mid-1960s when temperatures were supposedly cooling.

Update: Rob Wilson and Mike Pisaric have observed that the Meko white spruce chronologies in northern Alberta are considered to be precipitation proxies rather than temperature proxies and argued that this post is “flawed!” In fairness, everything in this post is expressed conditionally: “if the proxy hypothesis is that PCGL ring widths increase with increasing temperatures”, then these particular updated chronologies are not evidence for that proposition. If they are not believed to be temperature proxies, then they are likewise not evidence for the proposition either. The logic may be a little subtle for a climatologist. I am quite prepared to accept their opinion that the white spruce chronologies from northern Alberta are not temperature proxies. However, as I point out in some subsequent point, applying the same logic requires dendroclimatologists to take a stand against some of the proxies used in multiproxy studies.

Update2: Discussion of Alberta sites by Meko here

By Stephen McIntyre | Posted in Noamer Treeline, Post-1980 Proxies | Comments (106)

The Team and Pearl Harbor

Mar 18, 2007 – 9:12 PM

One of the Team’s more adventurous assumptions in creating temperature histories is that there was an abrupt and universal change in SST measurement methods away from buckets to engine inlets in 1941, coinciding with the U.S. entry into World War II. As a result, Folland et al introduced an abrupt adjustment of 0.3 deg C to all SST measurements prior to 1941 (with the amount of the adjustment attenuated in the 19th century because of a hypothesized use of wooden rather than canvas buckets.) At the time, James Hansen characterized these various adjustments as “ad hoc” and of “dubious validity” although his caveats seem to have been forgotten and the Folland adjustments have pretty much swept the field. To my knowledge, no climate scientist actually bothered trying to determine whether there was documentary evidence of this abrupt and sudden change in measurement methods. The assumption was simply asserted enough times and it came into general use.

This hypothesis has always seemed ludicrous to me ever since I became aware of it. As a result, I was very interested in the empirical study of the distribution of measurement methods illustrated in my post yesterday, showing that about 90% of SST measurements in 1970 for which the measurement method was known were still taken by buckets, despite the assumption by the Team that all measurements after 1941 were taken by engine inlet.

(Note – May 2008): see current discussion here and Dec 2007 post here Continue reading →

By Stephen McIntyre | Posted in SST, Surface Record | Tagged , , , | Comments (93)

Unthreaded #7

Mar 18, 2007 – 5:19 AM

Continued from Unthreaded #6

By John A | Posted in Unthreaded | Comments (430)