QUANTITATIVE BIRCH ( BETULA L.) POLLEN SEPARATION BY ANALYSIS OF SIZE FREQUENCY DATA

S ummary The Gordon—Prentice maximum likelihood method for estimating proportions of morphologically overlapping pollen taxa in a mixture requires knowledge of the underlying frequency distributions. This requirement is hard to fulfil when size is the discriminating character. However, if normality can be assumed the underlying distributions can be inferred from the data. This indirect approach ‐ due to Usinger ‐ permits reliable separation of Betula species in northwest European pollen diagrams. Techniques for separating mixtures of known distributions and for estimating underlying normal components are compared and assessed. For illustration, birch pollen size data from a Late Weichselian/early Holocene sequence in Finnmark, Norway (H. C. Prentice, unpublished) are subjected to a simple graphical technique to derive (two) normal components and then to the Gordon‐Prentice method and an analytical least‐squares method to estimate proportions. The inferred distributions are broadly consistent with Usinger's for Betula nana and Betula pubescens. The results show an Allerød‐Younger Dryas fall in the pollen percentage of Betula nana , a Betula pubescens rise at the beginning of the Holocene (interpreted as long‐distance pollen transport), and a subsequent major Betula nana expansion concurrent with the regional Ericales maximum.