Eigenvector Analysis of Reconstructed Holocene July Temperature Departures over Northern Canada

• Published in: Quaternary Research
• Main Authors: Andrews, John T., Diaz, Henry F.
• Format: Article in Journal/Newspaper
• Language: English
• Published: Cambridge University Press (CUP) 1981
• Subjects: Canada; Kay; Keewatin
• Online Access: http://dx.doi.org/10.1016/0033-5894(81)90017-x; http://api.elsevier.com/content/article/PII:003358948190017X?httpAccept=text/xml; http://api.elsevier.com/content/article/PII:003358948190017X?httpAccept=text/plain; https://www.cambridge.org/core/services/aop-cambridge-core/content/view/S0033589400021852

Abstract July temperatures for the past 6000 yr at 11 sites in northern Canada have been predicted by transfer-function equations. Normalized departures from the mean of each time series at 250-yr intervals are analyzed by principal component (eigenvector) analysis. An initial analysis included 9 sites and the first three principal components accounted for 85.7% of the variance. Maps of the loadings on the principal components show broad spatial coherence on all three components. Temporal coefficients (principal component scores) illustrate major regional and local midsummer temperature variations. An additional 2 sites were then included but the spatial pattern of the loadings remained essentially unchanged. A further test of this approach, with a view toward predicting paleoclimates of northern regions, was to use the spatial coefficients (loadings) to estimate the July temperature departures at an “unknown” site (Long Lake, Keewatin). This reconstruction compares favorably with an independent transfer-function reconstruction (Kay, 1979). Power spectrum analysis of the significant principal component scores (temperature departures) over the 6000 yr showed that the temporal fluctuations associated with the first three principal components follow a “red noise” spectrum, indicative of strong persistence in the reconstructed climatic records. The scores on the fourth principal component approximate a “white noise” spectrum. A peak in power between 2000 and 3000 yr occurs in the variance spectrum of the second principal component (significance 10%). We conclude that eigenvector analysis of Holocene paleoclimatic data has considerable power and may be useful for identifying regional and local climatic variations.