New approaches to direct gradient analysis using environmental scalars and statistical curve-fitting procedures

Abstract

The conceptual framework of direct gradient analysis (DGA) is discussed in relation to the functional, factorial approach to vegetation. Both approaches use abstract simplified environment gradients with which to correlate vegetation response. Environmental scalars based on physical process models of environment and/or known biological growth processes can be incorporated to make analyses less location specific. An example of an environmental scalar (radiation index) for converting aspect and slope measurements to the more biologically relevant radiation input at a site is given.The problem of the shape of species response curves to environmental gradients is examined using a sample of I 286 plots from eucalypt forest in southern New South Wales. An important conclusion is that skewed or bimodal response curves may be due to unsatisfactory distribution of observations and/or unrecognized environmental factors. The use of Generalized Linear Modelling (GLM) as a method for providing a statistical basis for DGA is presented. Analyses using GLM, and presence/absence data are presented for a range of eucalypt species (Eucalyptus rossii, E. dalrympleana, E. fastigata etc.). Successful prediction of species distributions (realized niches) can be achieved with mean annual temperature, mean annual rainfall, radiation index and geology. Quadratic terms are required in many cases, indicating bell-shaped response curves. The major variability associated with species niches is shown to be related to a limited number (4) of environmental factors. DGA with biologically relevant scalars and appropriate statistical methods is suitable for studying many problems of species' realized niches and plant community composition. Preface

Weighted averaging, logistic regression and the Gaussian response model

“…Breslow & Day, 1980). Ignoring confounding variables may give, for example, spuriously bimodal response curves (Austin et at., 1984). The real power of logistic regression lies in the simultaneous analysis of the effect of several environmental variables, including potentially confounding variables (see Appendix).…”

“…A more general approach to statistical testing in GLIM is to compare the residual deviance of a model with that of an extended model (Austin et a!., 1984;Dobson, 1983). The additional terms in the model are significant when the difference in residual deviance is larger than the critical value of a chi-square distribution with k degrees of freedom, k being the number of additional parameters.…”

“…Logistic regression is a Generalized Linear Modelling technique (GLIM), and is the equivalent for presence-absence data of ordinary multiple and polynomial regression (Dobson, 1983;McCullagh & Neider, 1983). Austin, Cunningham & Fleming (1984) showed the usefulness of GLM and GLR in their study of the occurrence of a range of eucalypt species in relation to temperature, rainfall, radiation and geology. There is no good evidence for the exact shape of a species response c11rve; we shall show that GLR is a practical method.…”

Species Richness Pattern along Altitudinal Gradient in Central European Beech Forests

“…To allow for direct comparison for both layers, the relevé size was included in the regression models as a covariate. To account for the effects of aspect and slope which may be confounded with the effects of altitude, we calculated a xericity index from slope and aspect (Austin et al 1984), and included it into the regression models as a covariate. Similarly, cover of the tree layer was included into regression models as a covariate, because this characteristic had a relatively wider variability and could affect the outcomes of the analyses.…”

Risk mapping for sensitive species to underwater anthropogenic sound emissions: Model development and validation in two Mediterranean areas

“…On the other hand, directly derived topographic variables (slope, aspect, topographic position, or slope characteristics) are generated without much loss of precision and they could be used for a priori predictions needed when site-specific data are unavailable. Austin (1980Austin ( , 1985, Austin et al (1984), and Austin and Smith (1989) defined three types of ecological gradients, namely resource, direct, and indirect gradients. Resource gradients address matter and energy consumed by plants or animals (nutrients, water, light for plants, food and water for animals).…”