Potential natural vegetation and environment: a critique of Kusbach, Shaw & Long

Abstract

AbstractAimsRecent work by Kusbach, Shaw & Long (2015, Applied Vegetation Science 18: 158–168) critiqued a system of potential natural vegetation classification widely employed on coniferous forests of the Rocky Mountains in the US, arguing that the lack of fit by a statistical model of vegetation type to environment suggests that these potential natural vegetation types may not represent basic ecological units of land. The aims of this paper are to (1) critically examine the model of environmental variability employed in their analysis, (2) evaluate the appropriateness of the statistical methods used in their analyses, and (3) demonstrate that despite the limitations in their approach the models actually worked well and refute the conclusions reached by Kusbach, Shaw & Long.LocationInterior western USA including Utah, SE Idaho, W Wyoming and Colorado, N Arizona, NW New Mexico and E Nevada.MethodsI examined the methods and results presented in Kusbach, Shaw & Long with respect to (1) the suitability of the model of environmental biophysics, (2) the appropriateness of the statistical methods employed, (3) the significance of the results obtained, and (4) the logical validity of the interpretations presented by the authors. I estimated the statistical significance of their predictive model results using a cumulative binomial probability based on the number of sample plots/type and the observed number of correct predictions obtained in their analyses. I estimated the probability of obtaining within‐type dispersions in PCA using a Monte Carlo randomization technique.ResultsThe models of environmental biophysics employed in Kusbach, Shaw & Long are poorly developed and inadequate in representing the primary ecological factors structuring the distribution of vegetation in the study region. The statistical model employed assumes that potential natural vegetation types are simply categorical, when in fact they are instances of a multidimensional ordinal variable. The failure to weight errors resulting from the model presents a vastly exaggerated error distribution. However, despite the shortcomings in the model of environmental biophysics and the inappropriate statistical tests, their results are in fact highly significant and argue for conclusions diametrically opposed to those presented by the authors.