Vegetation of the Obed Wild and Scenic River, Tennessee and a comparison of reciprocal averaging ordination and binary discriminant analysis

Forest vegetation was sampled in the Obed Wild and Scenic River in the Cumberland Plateau in Tennessee using 327 0.04 ha samples. Environmental variables were recorded for each of the plots. Cluster analysis using a minimum information technique (MINFO) and canopy importance values was used to classify 10 plant community types: river birch, beech-tulip poplar, tulip poplar, white oak, hemlock, sweet birch-hemlock-chestnut oak, chestnut oak-white oak, white pine-white oak-chestnut oak, white oak-scarlet oak, and Virginia pine types. Reciprocal averaging ordination identified two gradients, a moisture gradient arraying plant populations from xeric to mesic taxa and a second gradient separating mesic deciduous taxa from hemlock. Ordination scores correlated to environmental variables such that the first axis represented a gradient incorporating topographic position and soil moisture and the second gradient related to gorge width and probably protection. Disturbance history of the vegetation was reflected in the diameter-class distributions of arboreal species, but only the river birch, tulip poplar and sweet birch-hemlock-chestnut oak types appeared to be successional. Negative-exponential and negative-power regression models of the diameter distribution curves were statistically significant for each type and for the distributions of many individual canopy species.

Binary discriminant analysis (BDA) was evaluated using simulated coenocline data and field vegetation and environmental data obtained from the Obed River gorges. Using coenocline data, Haberman D-values became increasingly sensitive indicators of gradient location at higher beta diversity. However, both Q-mode and R-mode BDA were subject to distortion at higher beta diversities and lack of resolution at low beta diversities. Field data on the occurrence of canopy trees in plots from the Obed River gorges were analyzed using topo-graphic position, aspect, protection, slope angle, and horizontal plot shape. Some species had significant presence-absence patterns with each variable as indicated by Haberman D-values. Q-mode BDA revealed some useful species-environment patterns but was subject to quadratic distortion. Binary discriminant analysis's usefulness at the local scale is limited by lack of resolution at low beta diversity and distortion at high beta diversity.