An unconventional approach to ecosystem unit classification in western North Carolina, USA

Abstract

We used an unconventional combination of data transformation and multivariate analyses to reduce subjectivity in identification of ecosystem units in a mountainous region of western North Carolina, USA. Vegetative cover and environmental variables were measured on 79 stratified, randomly located, 0.1 ha sample plots in a 4000 ha watershed. Binary transformation of percent cover followed by direct and indirect ordination indicated the 185 inventoried species were associated primarily with soil A-horizon thickness, soil base saturation, and aspect. Redundant cluster analyses, consisting of divisive and agglomerative methods for multivariate classification of core plots, followed by selective discriminant analysis of remaining non-core plots, indicated that the continuum of vegetation and environment could be grouped into five ecosystem units. Approximately 20 herbaceous, shrubs, and tree species and several soil and topographic variables were highly significant discriminators of ecosystem units. We also demonstrated that redundant cluster analysis may be used to subdivide ecosystem units into subunits of uniform understory composition and associated environment. Validation and refinement of classification units, linkage with faunal biological components, and arrangement into landscape areas suitable for resource management is needed before field application.

Introduction

The Wine Spring Creek watershed, located in the Nantahala National Forest of western North Carolina, is the focus of intensive investigations to gain a better understanding of ecological relationships for purposes of ecosystem management. The Wine Spring Creek ecosystem management project was begun in 1993 to establish a scientific basis for management of physical and biological resources (Swank et al., 1994). Working in the nearby Great Smoky Mountains National Park, Whittaker (1956)established that vegetation patterns in this region consist of species responding individually to a complex of temperature and moisture gradients associated primarily with elevation and landform. Classification of this continuum of vegetation and environment into ecosystem units – “units that can be distinguished by major differences in physiography, soils, and vegetation” (Barnes et al., 1982) and recur in a predictable pattern on the landscape – is needed for planning and resource management guidelines (Kimmins, 1991).

Others have studied selected portions of the Wine Spring Creek drainage to determine vegetation and environment relationships. DeLapp (1978)described the extent and characteristics of high elevation northern red oak (Quercus rubra) stands. Hedman and VanLear (1995)reported that riparian vegetation in the vicinity intergraded between mixed mesophytic and eastern hemlock (Tsuga canadensis) forests with understories typically dominated by rhododendron (Rhododendron maximum). In unpublished reports, Hedman (1993)qualitatively described riparian vegetation along major streams and Baker and VanLear (Undtd)characterized stream-side areas dominated by rhododendron. In a preliminary reconnaissance-type classification, based only on arborescent vegetation and topographic variables, McNab and Browning (1993)identified six ecological types associated with three perceived moisture regimes in each of two altitudinal zones. Using this preliminary classification, Elliott and Hewitt (1997)reported on species diversity in the ecological-types characterized by vegetative communities dominated by an overstory of northern red oak and shrub understory of flame azalea (Rhododendron calendulaceum). Additional investigation should extend vegetation–site relationships to include all flora and soil variables.

Several classification studies using all flora and soil variables have also been conducted within a 70 km radius of the study area (Gattis, 1992; Moffat, 1993; Patterson, 1994), where climate, geology, soils, landforms, vegetation and disturbance mechanisms are generally similar. Classification units identified in these studies varied from 6 to 10, and only about half of the units agreed in composition among studies, even though field and analytical methodologies were similar. Since classification is primarily a subjective process (Sokal, 1974) it is uncertain if the unique units identified in each study are real, represent artifacts of the field data set resulting from sample plot location and techniques, or resulted from arbitrary decisions made during the analysis. Researchers are faced with many subjective decisions when conducting classification studies, beginning with field location of sample plots and method of quantifying vegetation. Data analysis brings a host of other decisions concerning not only the choice of legacy or contemporary multivariate techniques, but also seemingly mundane questions of rare species retention or exclusion (including definition of rare), recognition (and deletion) of outlier samples, and input data order (Tausch et al., 1995). These and other subjective decisions may cause other researchers using the same dataset to arrive at different results when conducting similar classification studies.

The primary objective of our study was to group the biological and physical components of the study area into relatively homogeneous ecosystem units based on widely used methods of multivariate analysis. Since the results of this study will probably provide a basis for hypothesis testing by researchers and may be applied to resource management by managers, we used conservative methodology for identifying ecosystem units. The risk in this approach is that a minor but real ecosystem unit may have been sampled but not detected in the analysis. However, we also reduce the possibility of identifying a false unit that is primarily an artifact of the dataset. A principal focus of this paper is the exploratory and unconventional use of a combination of multivariate classification methods to achieve objective, reproducible results for identifying portions of landscapes with similar ecological potential.