Classificatory discriminant analysis of pollen data in northeastern Italy - I. Numerical method

Abstract

The paper describes the application of a numerical method (classificatory discriminant analysis) to a palynological data set, The method is based on classical multivariate analysis techniques and is intended to provide the objective dating, along the radiocarbon timescale, of undated pollen levels. Observations are classified into one of a number of pre-defined groups on the basis of one or more numerical variables; for each observation a discriminant criterion is developed using either a linear of a quadratic discriminant function; the discriminant criteria are then applied to a second data set and are used to classify data devoid of the classificatory numerical variable. Techniques estimating the probability of correct classification are described. In the present example radiocarbon dating is the classificatory variable the discriminant criteria derived from radiocarbon-dated pollen levels are used to objectively place undated pollen levels along the radiocarbon timescale. The technique is applied to a pollen data set from north-eastern Italy, made up of published pollen diagrams developed in the period 1930-5-1986; two subsets can be identified (labelled "pre-1958" and "post- 1958") on the basis of the ~3-fold increase in the number of pollen-identified species after 1958. In the post-1958 set a number of radiocarbon dated levels area available discriminant criteria are derived from these and applied to all undated levels contained in both the pre-1958 and post-1958 subsets. This positions all pollen levels along the radiocarbon timescale. Implications of the application of this technique for palynology are diverse: pollen diagrams devoid of radiocarbon dating can be integrated with radiocarbon- dated pollen diagrams. Current gaps in the spatial coverage of palynological information can be (objectively) filled by pollen diagrams developed before, or devoid of, radiocarbon dating.