Tree Rings As Natural Recorders of Climatic Variability and Change: Their Properties, Strengths, and Limitations

Edward R. Cook

Lamont-Doherty Earth Observatory

Palisades, New York 10964

Climatic variability and change on decadal to centennial time scales is not as well understood as we would like in terms of its history and its causes. This knowledge gap is critically important to close if we are to understand more completely the degree to which human activities have resulted in the many anomalous patterns of climate variability being observed today. Long tree-ring records have played a significant, if sometimes controversial, role in closing this knowledge gap by providing exactly dated, annually resolved reconstructions of past climate (e.g., temperature, rainfall, drought) covering the past 500 to 1000 or more years. These reconstructions have in some cases been pivotal in testing various hypotheses regarding the relative contributions of natural and anthropogenic forcing. Yet given the great progress made over the past few decades in reconstructing climate from tree rings, there is still a significant amount of doubt and misunderstanding among some about how it is all done. Much of this doubt and misunderstanding relates to the fact that tree-ring analysis is largely a stage-wise set of statistical procedures applied to an ensemble of biological times series without any direct knowledge about how those time series developed through time. The ensemble in this case is a collection of cross-dated (i.e., exactly cross-matched in time) ring-width measurement series obtained from many trees of the same species growing in the same stand and, therefore, in the same environment, and the tree-ring chronology used for climate reconstruction is a mean-value function derived from the ensemble. Thus, the ensemble provides the basis for estimating the empirical uncertainty in any given tree-ring chronology used to reconstruct climate. This being the case, each stage of statistical processing used on the ensemble will have a different level of uncertainty, and some of it will be unquantifiable in any climatically meaningful way. Therefore, the estimated uncertainty in any tree-ring chronology is probably best regarded as a ?best guess? that is conditional on how the tree-ring data were processed. One might at least expect that minimizing the mean square error of the final chronology would be a reasonable goal. Paradoxically, that is not the case because doing so generally results in an unacceptable loss of climate variability at decadal to centennial time scales in the final chronology. These issues will be described and illustrated by example in the development of a 1000-year long temperature reconstruction from tree rings.