From: Tim Osborn <t.osborn@uea.ac.uk>
To: "Gustafson, Diane" <DGustafs@nas.edu>
Subject: Re: Proxy time series
Date: Wed, 29 Mar 2006 14:36:50 +0100
Cc: Keith Briffa <k.briffa@uea.ac.uk>

<x-flowed>
Dear Diane / Mike / NRC Committee,

At 22:18 28/03/2006, Gustafson, Diane wrote:
>Dear Tim:
>
>Our National Research Council Committee on Surface Temperature 
>Reconstructions has been considering your paper with Keith Briffa 
>published in a recent issue of Science.  Could you please elaborate 
>on your criterion for selecting the proxy time series included in 
>the analysis.  We are interested in how you computed the correlation 
>between the proxy time series and local temperature time series.  Is 
>the correlation based on filtered or detrended time series?  How 
>would you counter the potential criticism that your selection method 
>tends to favor proxy time series that show a strong 20th century warming?
>
>It would be most helpful for us if you could reply in time for us to 
>consider your response at our meeting tomorrow morning.  Thanks in 
>advance for your help.
>
>Mike Wallace

We (Tim Osborn and Keith Briffa) will first respond to these specific 
questions about our recent Science paper.  In addition, copied below 
are some further comments by Keith Briffa on issues related to 
tree-ring proxy records, that may be of interest to the committee.

The primary purpose of our paper was to implement an alternative, and 
possibly complementary, method of proxy-data analysis to the methods 
used in most previously published reconstructions of past NH 
temperature variations.  We did not want to introduce an entirely new 
selection of proxy records (even if this were possible), because that 
would obscure whether differences in our conclusions, compared with 
published work, arose from our method or a different selection of 
proxy records.

We decided, therefore, to make use of as many of the individual 
records used in almost all the previously published NH temperature 
reconstructions, excluding any records for which an indication of at 
least partial temperature sensitivity was lacking.  So, very low 
resolution records for which comparison with instrumental 
temperatures is problematic were excluded.

We used records specifically from Mann and Jones (2003) and Esper et 
al. (2002).  In addition we included records from Mann et al. (2003), 
which I think just adds the van Engelen documentary record from the 
Low Countries in Europe, because the others were already in the Mann 
and Jones set.  We excluded duplicates, and our paper explains which 
series we used where duplicates were present.  We did not average the 
Tornetrask, Yamal and Taimyr tree-ring records as done by Mann and 
Jones, because we could see no reason not to use them as individual series.

The series used by Mann and Jones had already been correlated with 
their local instrumental temperatures -- using decadally-smoothed, 
non-detrended, values -- so we accepted this as an indication of some 
temperature sensitivity.  For the other series, we calculated our own 
correlations against local instrumental temperatures, trying both 
annual-mean or summer-mean temperatures.  In our paper's 
supplementary information, we state that we used the HadCRUT2 
temperatures for this purpose, which combines land air temperatures 
with SST observations.  In fact, we used the CRUTEM2 land-only 
temperature data set for this purpose.  These should be identical 
where the proxy locations are not coastal.  For these correlations, 
we did not filter the data, nor did we detrend it, and we used the 
*full* period of overlap between the proxy record and the available 
instrumental record.

We excluded records that did not show a *positive* correlation with 
their local temperatures.  The remaining set includes most of the 
long, high resolution records used by others, such as Moberg et al., 
Crowley and Lowery, Hegerl et al., Mann, Bradley and Hughes, etc. as 
well as by Mann and Jones and Esper et al.

The final question, regarding the selection method favouring records 
that show a strong 20th century warming trend, is a more 
philosophical issue.  As stated above, we did not actually use 
strongly selective criteria, preferring to use those records that 
others had previously used and only eliminating those that were 
clearly lacking in temperature sensitivity.  To some extent, 
therefore, the question is then directed towards the studies whose 
selection of data we used.  Certainly we did not look through a whole 
host of possibilities and just pick those with a strong upward trend 
in the last century!  And we don't think the scientists whose work we 
selected from would have done this either.  There are very few series 
to choose from that are >500 years long and are from proxy 
types/locations where temperature sensitivity might be expected.  It 
would be entirely the wrong impression to think that there are 140 
such a priori suitable possible series, and that we picked (either 
explicitly or implicitly) just those 10% that happened by chance to 
exhibit upward 20th century trends.

The correlation with local temperature is an entirely appropriate 
factor to consider when selecting data; these could be computed using 
detrended data, though for those that we calculated, our use of 
unfiltered data means that the trend is unlikely to dominate the 
correlation.  One would need to inspect the trend in the temperature 
data at each location to evaluate how much influence it would have on 
the results; but in locations where a strong upward trend is present, 
it would be right to exclude proxy records that did not reproduce it, 
though also correct that a proxy shouldn't be included solely on the 
basis of it having the trend, especially where the proxy resolution 
is sufficient to test its ability to capture shorter term fluctuations.

Finally, note that our method has not selected only those records 
with a strong 20th century warming trend.  Of the 14 proxies selected 
(see our figure 1), 7(!) do not have strong upward 20th century 
trends: Quebec, Chesapeake Bay, W Greenland, Tirol, Tornetrask, 
Mangazeja, and Taimyr.  Our method gives equal weight to all records, 
so it should not be biased towards a single record, or a small number 
of records, that do show strong upward trends.

Here are the additional comments on tree-ring issues:

I would also like to take the opportunity, if you will allow, to 
comment briefly on some reports that have reached me concerning the 
contribution made by Rosanne D'Arrigo to your Committee. Apparently, 
this is being interpreted by some as reflecting adversely on the 
validity of numerous temperature reconstructions that involve 
significant dependence on tree-ring data. This is related to 
Rosanne's focus in her presentation on the apparent difference 
between measured temperatures and tree growth in recent decades  - a 
so-called "divergence" problem.
  First let me make it clear that as I did not attend the Committee 
meeting I am not able to comment specifically on the details of 
Rosanne D'Arrigo's actual presentation, though I am aware of her 
papers with various co-authors related to this "divergence" in the 
recent (circa post 1970 ) trends in tree-growth and temperature 
changes as recorded in instrumental data, at near tree-line sites in 
the Canadian Arctic. There are also other papers dealing with 
'changing growth responses' to climate in North American trees.

I have co-authored a paper in Nature on the reduced response to 
warming as seen in tree-ring densitometric data at high-latitude 
sites around the Northern Hemisphere, increasingly apparent in the 
last 30 years or so.


First, it is important to note that the phenomena is complicated 
because it is not clearly identifiable as a ubiquitous problem. 
Rather it is a mix of possible regionally distinct indications, a 
possible mix of phenomena that is almost certainly in part due to the 
methodological aspects of the way tree-ring series are produced. This 
applies to my own work, but also very likely to other work.

The implications at this stage for the 'hockey stick' and other 
reconstructions are not great. That is because virtually all long 
tree-ring reconstructions that contribute to the various 
reconstructions, are NOT affected by this. Most show good coherence 
with temperature at local levels in recent decades. This is not true 
for one series (based on the density data). As these are our data, I 
am able to say that initial unpublished work will show that the 
"problem" can be mitigated with the use of new, and again 
unpublished, chronology construction methods.

In the case of the work by Rosanne and colleagues, I offer my 
educated opinion that the phenomenon they describe is likely also, at 
least in part, a chronology construction issue. I am not saying that 
this is a full explanation, and certainly there is the possibility of 
increased moisture stress on these trees, but at present the issue is 
still being defined and explored. As the issue needs more work, this 
is only an opinion, and until there is peer-reviewed and published 
evidence as to the degree of methodological uncertainty , it is not 
appropriate to criticize this or other work . For my part, I have 
been very busy, lately with teaching and IPCC commitments, but we 
will do some work on this now, though again lack of funds to support 
a research assistant do not help.

The matter is important but I do not believe that the facts yet 
support Rosanne's contention, in her Global Biogeochemical Cycles 
paper (Vol. 18, GB3021, doi:10.1029/2004GB002249, 2004) that an 
optimum physiological threshold has been consistently exceeded at a 
site in the Yukon. This conclusion should certainly not be taken as 
indicating a widespread threshold exceedence.

It was my call not to "overplay" the importance of the divergence 
issue, knowing the subtlety of the issues, in the fortcoming IPCC 
Chapter 6 draft. We did always intend to have a brief section about 
the assumption of uniformitarianism in proxy interpretation , 
including mention of the possible direct carbon dioxide fertilization 
effect on tree growth (equally controversial), but it is likely to 
conclude that here as well , there is no strong evidence of any major 
real-world effect. This and the divergence problem are not well 
defined, sufficiently studied,  or quantified to be worthy of too 
much concern at this point. The uncertainty estimates we calibrate 
when interpreting many tree-ring series will likely incorporate the 
possibility of some bias in our estimates of past warmth, but these 
are wide anyway. This does not mean that temperatures were 
necessarily at the upper extreme of the reconstruction uncertainty 
range 1000 years ago, any more than they may have been at the bottom. 
The real problem is a lack of widespread (and non-terrestrial) 
proxies for defining the level of early warmth, and the vital need to 
up-date and study the responses of proxies in very recent times.

Best regards,

Tim Osborn and Keith Briffa

--
Professor Keith Briffa,
Climatic Research Unit
University of East Anglia
Norwich, NR4 7TJ, U.K.

Phone: +44-1603-593909
Fax: +44-1603-507784

http://www.cru.uea.ac.uk/cru/people/briffa/



Dr Timothy J Osborn
Climatic Research Unit
School of Environmental Sciences, University of East Anglia
Norwich  NR4 7TJ, UK

e-mail:   t.osborn@uea.ac.uk
phone:    +44 1603 592089
fax:      +44 1603 507784
web:      http://www.cru.uea.ac.uk/~timo/
sunclock: http://www.cru.uea.ac.uk/~timo/sunclock.htm

</x-flowed>

