Sedimentology and sample analysis
Limestone samples were examined in hand sample and thin section, with specific focus on the taxonomic identification of corals, coralline algae and large benthic foraminifera. The spatial context of dive samples was established from the video observations (i.e. precise sample location and whether broken off or loose). Each sample was then examined for evidence of orientation in both hand sample (e.g. discolouration and excessive growth on one side) and thin section (e.g. geopetals, growth direction, sorting, etc.). Thin section analysis was used to identify common assemblages of component grains and fossils that were then used to define sedimentary facies. These facies were compared with the modern environmental conditions of component biota to interpret their palaeoenvironments (e.g. [Adey et al., 1982], [Cabioch et al., 1999], [Fletcher et al., 2008], [Grigg, 1981], [Grigg et al., 1981], [Hallock, 1984], [Maragos, 1977], [Murray, 1991], [Renema, 2006], [Verheij, 1993] and [Veron, 2000]). The stratigraphic relationships of the facies within each sample were also recorded using cross-cutting and superposition principles to determine the sequence of palaeoenvironmental changes.
Statistical analysis of facies composition
Statistical analysis of the most common but compositionally diverse limestone facies was carried out. Microscope point counting of individual grains was used to examine variation in its components with respect to both depth and sample location on different terraces. This count used a Prior Model G point counter on 1 mm rows at 50 μm steps (selected from average grain size) and a 300 point sample size (for feasibility and uniformity). Forty-eight samples were selected and grains were categorised in 20 classes and prepared in matrices of sample number versus classification population. This data was analysed using Dissimilarity Matrices and Principle Components Analysis (Jongman et al., 1995) in the statistical software packages SPSS 14.0 and PcOrd. Multivariate differences between populations were tested using Multi-Response Permutation Procedure (MRPP), which is analogous to a non-parametric Manova (Zimmerman et al., 1985). To minimise the chance of a Type I error, a Bonferroni Correction was applied to the data to produce a significance level of 0.0083.