Quality of Historical Data on Midwestern Old-Growth Forests

Abstract

Quality of historical data on species composition in midwestern old- growth forests was assessed for its potential for monitoring long-term vegetation dy- namics. Vectors of compositional dynamics superimposed on ordinations revealed that perceived compositional change owing to sampling error may be similar in size to actual long-term changes. Other data quality problems may arise from big-tree bias, inexact overlap of stand boundaries, varying treatment of stand edges, and in- adequate descriptions of methods and results. To promote an improved long-term monitoring data base, a checklist is given of some commonly overlooked points of de- sign and publication of compositional studies of forests. In 1949 John Potzger received a letter from the editor of the American Midland Natu- ralist soliciting a paper from Potzger's then-recent study of woods near Versailles, Indi- ana (J. E. Potzger archive, Butler University Library, Indianapolis, Ind.). Potzger obliged and the paper was cheerfully accepted by the editor, with the caveat that the ta- bles of species composition should be deleted. Potzger disagreed, writing My experi- ence with omitted tables from published papers has been very unsatisfactory . . . in 25 years, or less, such data would in all likelihood not be available. Furthermore, he of- fered to pay additional typesetting charges to get the tables into print. Potzger's pub- lished tables (Potzger, 1950; and many other papers) have proven useful for later com- parisons. In contrast, most of his unpublished and field data are not present in his archive and are apparently lost forever. We hope that this paper will help spread Potzger's view toward the future. We have assembled published long-term compositional data from midwestern forests to examine the consistency and patterns of compositional change. This compilation re- vealed variation in quality and type of data. If, between two sample dates, composi- tional differences due to sampling error are larger than actual compositional changes, little confidence can be placed in the observed trends. This problem was assessed by comparing the size of long-term changes within stands, as shown by vectors of composi- tional change on an ordination, to compositional variation arising from sampling error. We also compare the influence of broad classes of study methods on the size of per-