Oecologia (Berlin) (1987) 71:229-232 Oecologia 9 Springer-Verlag 1987 Diversity of Eucalyptus species predicted by a multi-variable environmental gradient C.R. Margules, A.O. Nieholls, and M.P. Austin CSIRO, Division of Water and Land Resources, GPO Box 1666, Canberra, A.C.T. Australia 2601 Summary. Changes in species diversity are examined in rela- tion to a multidimensional environmental gradient using Eucalyptus species in south-eastern Australia. By fitting a generalized linear model, the response of the community parameter, species diversity, is shown to be related to three environmental variables, mean annual rainfall, mean annu- al temperature and a relative measure of solar radiation. The effects of rainfall and temperature were both statisti- cally significant and large, solar radiation was significant but small. However, the influence of the two major vari- ables was not independent but interacted in a complex way that prevents adequate description of species diversity as a function of either variable alone. Possible biological ex- planations of the complexity are discussed in terms of limit- ing conditions at low temperatures, and competition be- tween guilds of species at high temperatures and medium to high rainfall. Key words: Species diversity Rainfall - Temperature - Eucalyptus Abramsky and Rosenzweig (1984) studied the diversity of rodent species in relation to a gradient in rainfall, which they used as a surrogate for productivity. They found that the number of rodent species relative to increasing rainfall described a skewed peaked curve with the peak at low rain- fall levels. The same shaped response is predicted by Til- man's (1982) productivity/diversity model based on compe- tition between plants for limiting nutrients. Apart from graphical two-dimensional models (Peet 1978), previous models of species number in relation to environmental vari- ables have considered only one variable. Whilst single vari- ables may adequately predict species number where that particular variable is limiting, the different ways species can coexist are unlikely to be discovered by studying single variable gradients alone because over most of their ranges, species respond simultaneously to more than one variable. Here we extend the analysis to three environmental vari- ables using a very large data set and show how their simul- taneous effects can be tested. Using a different statistical procedure we conclude that the response curve of species number does not necessarily have the skewed shape postu- lated, but that the shape depends on the levels of other variables and the presence of other groups of species. Offprint requests to: C.R. Margules We have collected, and compiled from other sources, data from 4,977 plots on the canopy tree species present in a region of approximately 40,000 km a in south-eastern Australia (Austin et al. 1983 ; Austin et al. 1984). The region includes the coastline as well as Australia's highest moun- tain range and has marked gradients in mean annual rain- fall (50~2,300 mm) and mean annual temperature (3.5 ~ 16.8 ~ C). The plots in the data set vary in size from 0.02 to 0.25 ha and the majority of the tree species recorded belong to the genus Eucalyptus (76 species or 71%). Most other genera occur in so-called rainforest patches in moist, protected gullies at relatively high rainfall and temperature levels (Helman 1983). Estimates of mean annual rainfall and temperature for each plot were made using Laplacian smoothing spline functions (Hutchinson 1984; Adomeit et al. 1984), while a solar radiation index was calculated for each plot from slope aspect and latitude using the com- puter program SUNDAY (Austin et al. 1984). Statistical models relating the number of Eucalyptus species to the three environmental variables were fitted to these data using generalized linear modelling (McCullagh and Nelder 1983; Dobson 1983), and the computer package GLIM (Baker and Nelder 1978). Rainfall, temperature and solar radiation were chosen because they have been shown previously to be good predictors of the environmental distribution of cer- tain species of Eucalyptus (Austin et al. 1984). All the mod- els assumed a Poisson distribution of the number of species at each site, and the data were log transformed. Models were fitted for each variable (here called factors) singly. Each model includes an area co-variate because the different plot sizes have an effect on species number which has to be accounted for. Area is significant in all models. A statistically significant relationship was obtained with each single factor model though different response shapes were found for each factor (Fig. 1). Thus, interpretation of any one model in the absence of the other factors would be misleading. Table 1 shows the results of fitting a three- factor model including interactions and the procedure for obtaining a final 'best' model (Austin et al. 1984). The procedure was as follows. First, a maximal model defined as the main effects (each factor singly), plus all pairwise interactions and the covariate, plot size, was fitted. Then, the interactions and the covariate were removed, one at a time, and the significance of the associated change in deviance was tested. Non-significant terms were elimi- nated and a new model was fitted which included the only significant interaction, that between rainfall and tempera- ture, plot size, and each of the three main effects. The terms