Comparison Of Different Distance Measures For Cluster Analysis Of Tree-Ring Series

Abstract

Sixty individual ring-width series of oak (Quercus robur L.) from six sites in the northwestern Iberian Peninsula, ranging from 50 to 120 years, were grouped using hierarchical cluster analysis with different types of distance measures. Euclidean distances as well as other linkage distances based on statistics used to crossdate tree-ring series (Gleichläufigkeit and coefficient of correlation with its corresponding t-value) were compared. In addition, a new distance measure based on a corrected inversion of the Student’s t is proposed in the present paper, which takes into consideration the number of years used for series comparison. The Euclidean distances, commonly used in ecological analyses, inefficiently identified homogeneous units of trees based on their ring-width patterns. Among crossdating statistics, the correlation coefficient was more effective than Gleichläufigkeit, but the most satisfactory results were obtained when 1/t was used as distance measure. Finally, these methods of cluster analysis have been implemented into a computer program for future use of the dendrochronological community.