Microeconomic adaptation to severe climate disturbances on Australian coral reefs

Coral reefs are increasingly affected by climate-induced disturbances that are magnified by increasing ocean temperatures. Loss of coral reefs strongly affects people whose livelihoods and wellbeing depend on the ecosystem services reefs provide. Yet the effects of coral loss and the capacity of people and businesses to adapt to it are poorly understood, particularly in the private sector. To address this gap, we surveyed about half (57 of 109) of Australian reef tourism operators to understand how they were affected by and responded to severe impacts from bleaching and cyclones. Reef restoration and spatial diversification were the primary responses to severe bleaching impacts, while for cyclone-impacts coping measures and product diversification were more important. Restoration responses were strongly linked to the severity of impacts. Our findings provide empirical support for the importance of response diversity, spatial heterogeneity, and learning for social-ecological resilience. Supplementary Information The online version contains supplementary material available at 10.1007/s13280-022-01798-w.


Introduction
We undertook an exploratory study to empirically assess adaptation to severe climate disturbances on Australian coral reefs by tourism operators. We focused on four primary research questions: (1) how did tourism operators in Australia respond to severe climate-related disturbances, specifically the coral bleaching events in 2016 and 2017 and severe cyclones in 2011 and 2017? (2) How applicable is the microeconomic adaptation framework developed by Bartelet et al. (2022a) towards adaptation to climate change by coral reef

Response clusters
We now calculate the partial correlations between the adaptive responses that operators adopted in response to climate disturbances. These partial correlations reflect whether particular responses were more frequently implemented together than others. We used Spearman's Rank correlation because our responses are measures on a binary scale.
We found eight positive partial correlations between our individual adaptive responses that were significant at a p-level of 5% ( Figure 1). Based on these significant associations, we decided to make some changes to the a priori classification of adaptive response as proposed in Table 3. Most notably we decided to merge the adaptive responses of operational change, product diversification, and livelihood diversification into a combined adaptive response cluster linked to changes in 'operating model' because they were all linked to responses on the business and operational side. Compared to our a priori categorization, we classified 'spatial diversification' as a separate adaptation cluster because it was frequently implemented and not significantly associated with any of the other adaptive responses.
We found that the adaptive responses of 'monitoring (reefs and/or climate)' and 'restoration' were significantly correlated, although our a priori classification had defined monitoring as a protective measure. We used the monitoring and restoration responses as separate responses in our consequent analysis because these were each implemented by a relatively large fraction of operators. In accordance with our a priori classification, the adaptive responses of 'relief measures' and 'support-seeking' were significantly correlated.
Finally, one of the adaptive responses that was mentioned as other response by 16% of the participants was 'visitor education', i.e. informing and educating visitors about the causes and consequences of the climate disturbances. We merged the visitor education response with 'climate action' because they were significantly associated and because visitor education could potentially have an effect on future carbon emissions similar to a company taking climate action itself.

Predictor data preparation
For the predictors, we transformed the age of the company representative into a binary predictor (older versus younger), the business type (snorkel vs. scuba), and the company size (# passenger seats on boats) into a categorical predictor. We measured the number of passenger seats using nine multiple-choice options that ranged from '0-10 seats' to '>500 seats'. Through visual inspection of the data, we identified three clusters that we consequently categorized as small (<20 seats), medium (20-200 seats), and large (>200 seats). We included company size as a categorical rather than an ordinal predictor because the effects were not ordered linearly for all response models. We used small-sized companies as the reference group.

Small
Because all our predictors are on a binary scale, we standardized our only non-binary predictor (disturbance severity) using z-scores, by subtracting the mean and dividing by twice the standard deviation (Gelman, 2008). Dividing by twice the standard deviation standardizes each variable to have a mean of '0' and a standard deviation of '0.5'; this technically standardizes all predictors on a binary scale. Coefficients for continuous predictors from the Bayesian models are now directly comparable and should be interpreted as the effect of a one-standard deviation change in the predictor variable on the response variable. , psgseats_cat = factor(psgseats_cat, levels = c('small', 'medium', 'large')), z.dist_severity = (dist_severity-mean(dist_severity))/(2*sd(dist_severity)) ) ggplot(data_modified, aes(x=scuba_binary)) + geom_bar()