Small‐scale spatial autocorrelation in plant communities: the effects of spatial grain and measure of abundance, with an improved sampling scheme

Abstract Questions How does the spatial structure of plant communities vary with the spatial grain and with the measure of species presence used? How can communities most efficiently be sampled for spatial autocorrelation? Location Four communities – riverbed, bog, ultramafic shrubland/herbfield and forest – in southwest N ew Z ealand. Methods Each site was sampled over an extent of ca. 120 m at seven spatial grains, from 0.0025 to 25 m 2 , using an innovative triangular sampling scheme. At the 1‐m 2 grain, species abundances (local frequencies) were recorded, as well as presence/absence. Results The percentage variation in species composition explained by distance, i.e. by spatial autocorrelation, was higher at larger grain. However, it reached a maximum of only 15%. The nugget – the Y‐intercept of the dissimilarity/distance relation – has been seen as a measure of randomness in community composition. It was generally about 0.5 dissimilarity on a 0–1 scale, although values in the range 0.7–0.8 were found at smaller grain sizes in the forest. The 90% distance, i.e. the distance at which dissimilarity reaches 90% of its final value, was interpretable only for the two sites where spatial autocorrelation was strong, but gave realistic estimates. Unsurprisingly, some parameter estimates were unrealistic when the fits were poor. Abundance information added nothing to the ability of distance to predict dissimilarity. Conclusion The strength of spatial autocorrelation rose with increasing grain, to a low value but one congruent with the few comparable studies in the literature. That is, control of species composition seemed to be at the larger grain sizes sampled, rather than at a very fine scale. Strong spatial autocorrelation has been reported only over large extents, over environmental heterogeneity and/or when examining one guild within a community. The nugget was generally somewhat lower than other values in the literature, indicating less randomness. The lack of increased spatial community predictability when including species abundances conforms to the majority of previous studies, suggesting that the primary community control is on the presence of species, not their abundance. However, the differences in spatial autocorrelation between the four sites sampled emphasize that comparative studies using consistent methods are needed. The triangular sampling scheme used here was rapid, accurate, and efficient in its distribution of distances.