MULTIVARIATE, AUTOCORRELATION AND SPECTRAL ANALYSES OF A POLLEN PROFILE FROM SCOTLAND AND EVIDENCE FOR PERIODICITY

A vertical core through a Holocene peat in Scotland has been divided o ptimally into segments by multivariate and geostatistical analyses of the pollen of twenty important interrelated taxa. Pollen counts were c onverted to principal components (from the correlation matrix), and th ey were strongly correlated with depth. Their variograms were used to explore the stratigraphical structure in the core. The overall form of the variograms was monotonically increasing, but with a wave of lengt h approximately 32 cm superimposed on it. Power spectra of the princip al components, computed after filtering to remove long-range trend, ha d strong spectral peaks, with frequencies corresponding to the 32 cm f ound in the variograms which suggest that the pollen assemblage change s cyclically. The data were transformed to canonical variates, and the core was segmented using these variates by (a) minimizing the within- segment variation on average (global optimization), and (b) by maximiz ing the Mahalanobis distances between adjacent segments (local optimiz ation). The results of the two methods were similar. When matched to t he data they showed that distinct types of pollen assemblage, from woo dland and more open grassland, alternate in a quasi-cyclic way. When c onverted to time the period was approximately 800 years.