From the UNIVERSITY OF OTAGO and the “one tree doesn’t work anymore” department:
Uncertainties in tree-ring-based climate reconstructions probed
Current approaches to reconstructing past climate by using tree-ring data need to be improved on so that they can better take uncertainty into account, new research led out of New Zealand’s University of Otago suggests.
Tree growth rings are commonly used as climate proxies because they can be well-dated and the width of each ring is influenced by the climatic conditions of the year it grew in.
In a paper appearing in the Journal of the American Statistical Association, statistics and tree ring researchers from Otago, the US and UK examined the statistical methods and procedures commonly used to reconstruct historic climate variables from tree-ring data.
The research was led by Dr Matthew Schofield of Otago’s Department of Mathematics and Statistics. His co-authors on the paper are departmental colleague Professor Richard Barker, Professor Andrew Gelman of Columbia University, Director of the Tree Ring Lab at Columbia Professor Ed Cook, and Emeritus Professor Keith Briffa of the University of East Anglia, UK.
Dr Schofield says that their approach was to explore two areas where currently used approaches may not adequately account for these uncertainties. The first area involves the pre-processing of tree-ring data to remove non-climate related factors believed to be largely unrelated to climate effects on tree growth. Such factors include tree age, as the older a tree gets the less wide its rings tend to grow.
“This is convenient to do and the resulting tree-ring ‘chronologies’ are treated as relating to only the climate variables of interest. However, it assumes perfect removal of the non-climatic effects from the tree-ring data and ignores any uncertainty in removing this information,” Dr Schofield says.
The second area of uncertainty the researchers studied involves the particular modelling assumptions used in order to reconstruct climate from tree rings. Many of the assumptions are default choices, often chosen for convenience or manageability.
“This has made it difficult to evaluate how sensitive reconstructions are to alternate modelling assumptions,” he says.
To test this sensitivity, the researchers developed a unified statistical modelling approach using Bayesian inference that simultaneously accounts for non-climatic and climatic variability.
The team reconstructed summer temperature in Northern Sweden between 1496 and 1912 from ring measurements of 121 Scots Pine trees.
They found that competing models fit the Scots Pine data equally well but still led to substantially different predictions of historical temperature due to the differing assumptions underlying each model.
While the periods of relatively warmer and cooler temperatures were robust between models, the magnitude of the resulting temperatures was highly dependent on the model being used.
This suggests that there is less certainty than implied by a reconstruction developed using any one set of assumptions.
Since the press release didn’t link to the paper or give the title, I’ve located it and reproduced it below.
A Model-Based Approach to Climate Reconstruction Using Tree-Ring Data
Quantifying long-term historical climate is fundamental to understanding recent climate change. Most instrumentally recorded climate data are only available for the past 200 years, s o proxy observations from natural archives are often considered. We describe a model-based approach to reconstructing climate defined in terms of raw tree-ring measurement data that simultaneously accounts for non-climatic and climatic variability. In this approach we specify a joint model for the tree-ring data and climate variable that we fit using Bayesian inference. We consider a range of prior densities and compare the modeling approach to current methodology using an example case of Scots pine from Torneträsk, Sweden to reconstruct growing season temperature. We describe how current approaches translate into particular model assumptions. We explore how changes to various components in the model-based approach affect the resulting reconstruction. We show that minor changes in model specification can have little effect on model fit but lead to large changes in the predictions. In particular, the periods of relatively warmer and cooler temperatures are robust between models, but the magnitude of the resulting temperatures are highly model dependent. Such sensitivity may not be apparent with traditional approaches because the underlying statistical model is often hidden or poorly described.
The full paper is here (open access, my local backup here::Schofield-cook-briffa-modeling-treering-uncertainty (PDF)) and the supplementary information is here:.uasa_a_1110524_sm7673 (PDF)
At the end of the paper they say this, which is worth reading:
Message for the paleoclimate community
We have demonstrated model-based approaches for tree-ring based reconstructions that are able to incorporate the assumptions of traditional approaches as special cases. The modeling framework allows us to relax assumptions long used out of necessity, giving flexibility to our model choices. Using the Scots pine data from Tornetr¨ask we show how modeling choices matter. Alternative models fitting the data equally well can lead to substantially different predictions. These results do not necessarily mean that existing reconstructions are incorrect. If the assumptions underlying the reconstruction is a close approximation of reality, the resulting prediction and associated uncertainty will likely be appropriate (up to the problems associated with the two-step procedures used). However, if we are unsure whether the assumptions are correct and there are other assumptions equally plausible a-priori, we will have unrecognized uncertainty in the predictions. We believe that such uncertainty should be acknowledged when using standardized data and default models. As an example consider the predictions from model mb ts con for Abisko, Sweden. If we believe the assumptions underlying model mb ts con then there is a 95% probability that summer mean temperature in 1599 was between 8.1 ◦C and 12.0 ◦C as suggested by the central credible interval (Figure 4(a)). However, if we adopt the assumptions underlying model mb ts spl pl we would believe that the summer mean temperature in 1599 may have been much colder than 8.1 ◦C with a 95% credible interval between 4.1 ◦C and 7.8 ◦C. In practice, unless the data are able to discriminate between these assumptions (which they were not able to do here as shown in Section 6), there is more uncertainty about the summer mean temperature in 1599 than that found in any one model considered. We believe that such model uncertainty needs to be recognized by the community as an important source of uncertainty associated with predictions of historical climate. The use of default methods makes evaluation of such uncertainty difficult.