Automated biostratigraphic correlation of palynological records on the basis of shapes of pollen curves and evaluation of next-best solutions

We present a fully automated method to correlate palynological records on basis of the shapes of individual pollen curves. To this end the pollen curve between two successive samples, named “object”, is described by the following characteristics: (1) the length of the interval between the samples, (2) the average pollen percentage, or score, of the curve within the interval, and (3) the slope of the curve within the interval. The latter two characteristics, after having been transformed to an ordinal scale comprising three classes, are used to determine the dissimilarity between the objects. Dissimilarities are expressed as cost values, and are calculated on the basis of a simple look-up table. The lengths of the intervals are not incorporated in the dissimilarity measure but are used as weight factors instead. As we do not charge costs for compressing the stratigraphical columns, they can be squeezed and stretched to a high degree in order to find the best correlation. The best correlation between two pollen curves is defined as the correlation for which the costs of aligning all objects of both curves sum to the minimal total costs. Our method is not only applicable to the curves of a single pollen taxon but also allows the incorporation of the curves of several taxa into a single analysis. Finally, in addition to the optimal alignment, we determine next-best correlations as well. In order to visualize the entire set of best and next-best correlations, we introduce a so-called cost-landscape. The x- and y-axes of the cost-landscape correspond to the depth axes of the two pollen profiles, whereas the total costs of the alternative alignments are depicted as elevation. The optimal correlation appears as a river in the cost-landscape, while aspects of the river bank reveal the fit of alternative matchings. We use a Geographic Information System (GIS) to explore the cost-landscape. As an illustration of the methodology we apply our matching procedure to a palynological data-set that covers the Holocene period in northern Europe. These data enable us to compare the matching outcome with other correlation techniques, like correlation by zones and slotting. We discuss several possible improvements of the procedure.