In this study we outline the techniques used to transform multibeam acoustic data into spatial layers that
can be used for predictive habitat modelling. The results allow us to identify multibeam attributes which
may act as potential surrogates for environmental variables that influence biodiversity and define which
variables may be reliable for predicting the distribution of species in temperate waters. We explore
a method for analysing the spatially coincident multibeam bathymetric and backscatter data from
shallow coastal waters to generate spatial data products that relate to the classes derived from fine-scale
visual imagery obtained using an autonomous underwater vehicle (AUV). Classifications of the multibeam
data are performed for substrate, rugosity and sponge cover. Overall classification accuracies for
the classes associated with substratum, rugosity and sponge structure were acceptable for biodiversity
assessment applications. Accuracies were highest for rugosity classes at 65%, followed by substratum
classes at 64% and then sponge structure classes at 57%. Random forest classifiers at a segmentation scale of 30 performed best in classifying substratum and rugosity, while K-nearest neighbour classifiers performed best for sponge structure classes, with no difference in accuracy between scale 30 and 60.
Incorporating backscatter variables using segmentation improved the overall accuracy achieved by the
best performing model by between 1% (rugosity) and 9% (substratum) above using topographic variables
only in the grid-based analyses. Results suggest that image-based backscatter classification show
considerable promise for the interpretation of multibeam sonar data for the production of substrate
maps. A particular outcome of this research is to provide appropriate and sufficiently fine-scale physical
covariates from the multibeam acoustic data to adequately inform models predicting the distribution of
biodiversity on benthic reef habitats.