Spatial Non-Stationarity And Anisotropy Of Compositional Turnover In Eastern Australian Myrtaceae Species

Knowledge of species compositional turnover, the rate of change in the number of species shared between locations along geographic and environmental gradients, is important for conservation planning. Spatially global models relating species and environmental turnover are well established. However, to date there has been no explicit assessment of the effects of geographical variation in parameters (spatial non-stationarity) and directional dependence (anisotropy) on these models. Such processes are well known to affect other geospatial analyses. Here, we assess how these affect the shape, goodness of fit and composition of species turnover-environment relationships. We use the eastern Australian distribution of species in the Myrtaceae, an important family of vascular plants in the region. We obtained distribution data for Myrtaceae species from herbarium records and corresponding environmental attributes (mean of 1 km gridded cell values aggregated to 10 km grid cells). Species compositional turnover was quantified using the Sorensen pairwise dissimilarity index. The turnover-environment relationship was analysed using generalised dissimilarity modelling (GDM), a purpose-designed statistical regression technique. The data were divided into three sets of spatially local subsamples: 27 rectangular east-west-aligned coastal-inland bands, 8 north-south coastally aligned bands and 12 symmetrical omnidirectional blocks. A separate GDM was fitted to each spatially local subsample. The results display marked evidence of spatial variation in the shape, goodness of fit and composition of local species turnover-environment relationships, with this variation appearing strongly directional. The observed spatial structure of local biodiversity-environment relationships, expressed as species compositional turnover, is unsurprising considering both the steep east-west environmental gradients associated with Australia's eastern ranges and the known decreasing sampling intensity in the same direction. Local spatial non-stationarity and anisotropy are expected outcomes irrespective of the chosen turnover index, study taxa or statistical model used. Currently, it is difficult to separate genuine biogeographic effects from data bias. Future analyses could better account for the observed spatial structure of both biodiversity-environment relationships and data bias by incorporating directionality in data subsampling strategies and explicit biogeographic predictors in model design.