To support coastal planning through improved understanding of patterns of biotic and abiotic surrogacy at broad scales, we used Gradient Forest Modelling (GFM) to analyse and predict spatial patterns of compositional turnover of demersal fishes, macro invertebrates and macroalgae on shallow temperate Australian reefs. Predictive models were first developed using environmental surrogates with estimates of prediction uncertainty, and then the efficacy of the three assemblages as bio-surrogates for each other was assessed.

Data from underwater visual surveys of subtidal rocky reefs were collected from the south-eastern coastline of continental Australia (including South Australia and Victoria) and northern coastline of Tasmania. These data were combined with 0.01°-resolution gridded environmental variables to develop statistical models of compositional turnover (beta diversity) using GFM. GFM extends the machine learning, ensemble tree-based method of Random Forests (RF), to allow the simultaneous modelling of multiple taxa. The models were used to generate predictions of compositional turnover for each of the three assemblages within unsurveyed areas across the 6600 km of coastline in the region of interest.

The most important predictor for all three assemblages was variability (measured as standard deviation from measures taken interannually) in sea surface temperature. Spatial predictions of compositional turnover within unsurveyed areas across the region of interest were remarkably congruent across the three taxa. However, the greatest uncertainty in these predictions varied in location between the different assemblages. Pairwise congruency comparisons of observed and predicted turnover between the three assemblages showed that invertebrate and macroalgal biodiversity were most similar, followed by fishes and macroalgae, and lastly fishes and invertebrate biodiversity, suggesting that of the three assemblages, macroalgae would make the best bio-surrogate for both invertebrate and fish compositional turnover.