Fig. 5.

Odds ratios for the progression of bleaching categories from healthy to bleaching-related mortality. Odds ratios (with 95% confidence intervals) produced by the ordinal logistic regression model for (a) significant explanatory variables (depth, region and the different coral taxa), and (b) their interactions with depth (significant for region and coral taxa). OLR is an extension of binomial logistic regression in which multiple categories associated with an ordinal event are attributed an odds ratio statistic—the probability of the occurrence of that event based on a combination of one or more independent variables. Odds ratios for region refer to Far Northern GBR as the reference (combining Great Detached and Tijou) and for coral taxa to Acropora (corresponding to an odds ratio of one). An asterisk (*) marks categories or interactions for which log odds (and ratios) are significantly different from those of reference categories. Note the very narrow confidence intervals for depth in panel a, which do not overlap with the unit of odds ratio, therefore confirming the statistical significance of this variable. Location had no significant effect on the progression of bleaching as tested by OLR and individual locations were therefore grouped into two regions. The full model, including interaction effects between depth and each of the other two significant variables, region and coral taxa, explained 18.9% of the variation in the dataset. Depth, coral taxa and region explained 1.4%, 17.3% and 1.1% of the variation in bleaching response, respectively. The interactions depicted in panel b mean that the different coral genera do not respond homogeneously over depth, and that the effect of depth is not homogeneous between regions. See Supplementary Figure 4 for further modelling results