A metacommunity‐scale comparison of species‐abundance distribution models for plant communities of eastern Australia

Various models of the species‐abundance distribution (SAD) have been proposed to fit empirically derived data however there is no general consensus as to which model provides the best fit. Further, the zero‐sum multinomial SAD model (ZSM) was proposed as a metacommunity model, yet it has not previously been fitted at the metacommunity scale. We note that SAD models based on compound lognormal distributions (such as the Poisson‐lognormal, PLN, and the negative binomial‐lognormal models, NBLN) can also be thought of as metacommunity models, and we compare these with the ZSM when fitted as metacommunity models to SADs of related communities. We collected five datasets in the Sydney Basin, eastern Australia, representing five different types of subtropical/temperate plant communities ranging from closed warm‐temperate rainforest to open wet sclerophyll forest to dry sclerophyll woodland. For each type of plant community, five local communities were identified across the Sydney Basin, and SAD data collected in five randomly placed 0.2‐ha quadrats at each local community. Analysis was performed at two levels: all abundance data from each local community were pooled across each metacommunity and analysed as a single pooled community; and a metacommunity model was fitted to all five local communities of a community type, simultaneously. For the pooled data, we considered the negative‐binomial (NB) and the log‐series (LS) models in addition to ZSM, PLN and NBLN. All five models performed similarly, however the LS had a better fit to three pooled communities and the ZSM and PLN to the remaining two communities. By contrast, the ZSM performed statistically better against the PLN and NBLN when considered as a metacommunity model. We conclude that the ZSM generally provides a more reliable null model for metacommunity abundance data than the lognormal model.