A Method for Palaeoclimatic Reconstruction in Palynology Based on Multivariate Time-Series Analysis

摘要

A method based on multiple regression, canonical correlation analysis and the Kalman filter is developed with a view to producing a reliable reconstruction of climate from pollen spectra. The study region is the southern France, where traditional analytical techniques are unsatisfactory due to the heavy influence of human activity and the existence of a complex climate. The first stage of the proposed method is the same as the standard techniques in which a spatial network of 36 sites with recent pollen and climate data were analyzed by regression techniques. But since the recent data are frequently subjected to disturbance by such factors as human activity or the particularities of the local vegetations, a second, independent analytical stage was added, in which pairs of fossil sites were compared using ancient data exclusively. In this way a model of the climatic forcing can be obtained, with the above-mentioned factors left out of account. The reconstructions obtained from the first two stages were then compared in the third stage of the method, and the result is a corrected reconstruction, from which it appears that at 14 000 BP annual temperature was 9oC cooler than present, with a 4°C increase between 13 500 and 11 500 BP followed by a 2oC cooling at 10 500 BP. The Holocene temperature maximum occurred between 8000 and 6000 BP. when the temperature was 1°C warmer than present.