An evaluation of statistical techniques for discriminating Picea glauca from Picea mariana pollen in northern Alaska

Abstract

Two statistical procedures, linear discrimination and maximum likelihood discrimination, were evaluated for use in estimating percentages of Picea glauca and P. mariana pollen in lake sediments of northern Alaska. Each procedure is based on a comparison of the dimensions of unclassified Picea pollen grains with those of reference pollen of each species. The reference pollen collection in this study consisted of 675 P. glauca and 600 P. mariana grains, representing 51 trees at 24 sites in Alaska. The reference collection was divided into two approximately equal parts; one was used to derive the discriminant functions of each technique and the other was used to test the accuracy of each function on populations of known composition. The maximum likelihood procedure produced unbiased estimates across the entire range of test populations (P. glauca: P. mariana ratios ranging from 5:95 to 95:5). The linear discrimination estimates were strongly biased in favor of the less frequent species in populations having less than 25% of either species. We applied both techniques to measurements of Picea pollen in surface sediments of 62 lakes in the boreal forest of northern Alaska. Picea pollen percentages estimated by maximum likelihood discrimination more closely revealed the distribution of Picea trees in the landscape.