Fine-scale benthic biodiversity patterns inferred from image processing

Despite potentially considerable advantages over traditional sampling techniques, image-derived indices of habitat complexity have rarely been used to predict patterns in marine biodiversity. Advantages include increased speed and coverage of sampling, avoidance of destructive sampling, and substantially reduced processing time compared to traditional taxonomic approaches, thus providing a starting point for more detailed analysis if warranted. In this study, we test the idea that the mean information gain (MIG) and mean mutual information (MMI), two indices of image heterogeneity that we derived from photographs of marine benthic assemblages, represent good preliminary predictors of biodiversity patterns for 133 benthic invertebrate and algal taxa on jetty pylons in Gulf St Vincent, South Australia. Both MIG and MMI were spatially structured, with evidence of among-site differences that were also evident in the benthic data. When combined with information on the spatial structure within the dataset (site and depth), MIG and MMI explained ∼35% of deviance in invertebrate species richness, ∼43% in Shannon's evenness and up to 50% of dissimilarity in species composition. This explanatory power is of a similar magnitude to many other, less readily available, surrogate measures of biodiversity. These results corroborate the idea that indices of image heterogeneity can provide useful and cost-effective complements to traditional methods used for describing (or predicting) marine epibiota biodiversity patterns. This approach can be applied to many case studies for which photographic data are available, and has the potential to result in substantial time and cost savings.