ï~~2002
THE MICHIGAN BOTANIST
151
identified 3,989 witness trees. John Randall, Edward Albertson, and Clarence
O'Seeley surveyed Oscoda County from 1838 to 1840 and identified 4,033 witness trees. The survey instructions from the Surveyor General's Office indicated
that witness tree selection was not biased by size or species as a result of survey
instructions nor were the data manufactured by fraudulent surveyors because in
two subsequent resurveys (1884 and 1908) the remains of many of the original
witness trees were relocated (Clasen 1836; Lyon 1846).
To evaluate the presettlement species associations, the Cole/Hurlbert C8 association coefficient was calculated from the paired witness tree data on species
that occurred at a minimum of 20 locations in each of the two counties (Hurlbert
1969, Anderson & Anderson 1975). In the witness tree data, frequency varied by
species and the Cole/Hurlbert C8 association coefficient allowed valid comparisons across species of different frequencies because it was frequency independent. Survey points with only one witness tree were eliminated from the analysis. We constructed a dissimilarity matrix from the C8 values and performed
cluster analysis (Bastow et al. 1990) using multivariate analysis program,
SAMPL (W. Myers, University Park, PA). SAMPL derived the first cluster formation by linking nearest neighbors; the program then evaluated the second
closest neighbors for all species; this occasionally resulted in one species being
identified within two of the initial clusters. If this occurred, these clusters were
consolidated based on the linkage of second closest neighbors. After the linkage
of second closest neighbors, the cluster analysis stopped.
Soils and topography
Contingency table analysis was used to assess the importance of edaphic factors on presettlement forest composition (Strahler 1978). We pooled the witness
tree data from both counties for this analysis to allow inclusion of minor species
without encountering problems associated with small predicted values (Steel et
al. 1997). All witness trees were transcribed onto the USGS topographic quadrangles. Three topographic positions were identified: depressions, plains, and
uplands. Depressions were areas lower than the average elevation and typically
had swampy or boggy conditions. Plains were flat topographic areas and uplands
were isolated ridges or rolling hills. A species by topographic position matrix
was developed. A G-statistic was calculated for each species where G2 = 2,
(observed * In (observed / expected)). These values were examined for significance at oa = 0.05. For species with significant relationships between species
presence and topographic position, corrected standardized residuals were calculated by Ei / (Vi) A 0.5. Where
(observed - expected)
Eij =
(expected) ^ 0.5
and
Vij((observed column total - expected) x observed row total)
((observed row total - expected) x observed column total).
