New methods for the analysis of an important historic forest data set

Use of the General Land Office (GLO) Bearing Tree (BT) data in describing pre-sttlement forest conditions now has an 80+ year history and a long bibliography. Nevertheless, methods for evaluating biases in these data, for both species composition and structure, remain only partially developed and without a definite statistical basis, limiting their usefulness. I have addressed this issue by developing two new methods to estimate historic tree densities. Method one requires only the original data, and involves a thorough statistical analysis of detectable surveyor biases. Method two involves revisiting original BT sampling locations and dendrochronologically evaluating uncertainties in the original data. This latter method is particularly relevant wherever data quality issues exist, or where direct forest change estimates are desired. Each method has been evaluated using BT data, the first using statewide data from Minnesota, and the second from selected sites in the central Sierra Nevada, California. Analyses of the Minnesota data (1^st method) show that the bias generated by choosing non-random bearings from survey point to tree was common, particularly in high-density forests, but was present in low-density forests also, and can underestimate density by a factor of 10 or more. It also indicates that the magnitude of surveyor bias was a function of stand density. Conversely, angle sectors with missing trees were common in oak savannah/prarie ecotone regions and lead to overestimates of a similar magnitude. Neither bias has ever been accounted for in previous studies. Biases regarding tree size or species are more difficult, but not impossible, to detect and correct. Analyses of the Sierran data (2^nd method) show that surveyors were unbiased in their BT selection about 80% of the time, choosing the closest available tree in a given angle sector. Thus, the original data can at least sometimes be analyzed without extensive statistical analyses.