Evidence for managing herbivores for reef resilience

Abstract

Herbivore management is an important tool for resilience-based approaches to coral reef conservation, and evidence-based science is needed to enact successful management. We synthesized data from multiple monitoring programs in Hawai'i to measure herbivore biomass and benthic condition over a 10-year period preceding any major coral bleaching. We analysed data from 20 242 transects alongside data on 27 biophysical and human drivers and found herbivore biomass was highly variable throughout Hawai'i, with high values in remote locations and the lowest values near population centres. Both human and biophysical drivers explained variation in herbivore biomass, and among the human drivers both fishing and land-based pollution had negative effects on biomass. We also found evidence that herbivore functional group biomass is strongly linked to benthic condition, and that benthic condition is sensitive to changes in herbivore biomass associated with fishing. We show that when herbivore biomass is below 80% of potential biomass, benthic condition is predicted to decline. We also show that a range of management actions, including area-specific fisheries regulations and gear restrictions, can increase parrotfish biomass. Together, these results provide lines of evidence to support managing herbivores as an effective strategy for maintaining or bolstering reef resilience in a changing climate.

Keywords: Hawai‘i; coral; human impacts; resilience-based management.

Figure 1.

Coefficients from a Bayesian hierarchical model estimating the relationship between total herbivore biomass and drivers that include metrics of fishing (red), land-based pollution (yellow), oceanography (blue) and habitat (grey). Circles are medians and horizontal lines correspond to 50% intervals (thick lines) and 95% intervals (thin lines) of posterior distributions of each coefficient. We interpret evidence of a relationship between the driver and herbivore biomass if either or both intervals do not overlap zero. noncomm, non-commercial fishing; OSDS, on-site sewage waste-disposal systems; SD, standard deviation; Max, maximum; Freq, frequency; chl-a, chlorophyll-a.

Figure 2.

(a) Posterior estimates of herbivore biomass summarized across moku (land divisions). Bars are means and error bars are 50% intervals of posterior predictions for all 100 m pixels in each moku, and thus are post-stratified estimates that account for the relative distribution of habitat and variation in other predictors. Moku with mean values in the upper quartile (top fourth) are colored turquoise, and moku with mean values in the lower quartile (bottom forth) are colored red. Numbers correspond to labels at the bottom of (b) and to numbers on map (inset) (a larger map of moku boundaries with labels is provided in electronic supplementary material, figure S6). (b) Violin plots of per cent potential biomass for all 100 m pixels in each moku estimated as the expected biomass from model in figure 1 divided by the potential biomass when fishing is minimized and all other predictors are held constant, with threshold (80%) below which benthic condition is predicted to decrease (from figure 3b). Moku with greater than 44% of area under the threshold are coloured yellow in b and the coastline is coloured yellow in the inset.

Figure 3.

Relationship between herbivore functional group biomass and benthic condition. (a) Coefficients from a Bayesian hierarchical model estimating the relationship between biomass of three herbivore functional groups (scrapers, grazers, browsers), measures of oceanographic conditions, pollution and depth and rugosity as predictors and the log-ratio of calcified benthic cover (coral + coralline algae) and macroalgal cover. Circles are medians and horizontal lines correspond to 50% intervals (thick lines) and 95% intervals (thin lines) of posterior distributions of each coefficient. (b) Probability from logistic regression (black line) that benthic condition is unaffected by difference in herbivore biomass between expected and potential levels with fishing minimized as a function of percentage of potential herbivore biomass (expected/potential) for each 100 m pixel in the study domain. Pixels were classified as either unaffected (1) or affected (log(calcified:macroalgae) increased) (0). (c) Density distributions of percentage potential biomass values plotted for pixels in which benthos is unaffected by change in biomass with fishing minimized (blue) and for pixels where benthic condition is reduced given minimized fishing (red). Yellow horizontal and vertical lines in (b) and (c) show where the probability of benthic change of 0.99 (y-axis) corresponds to percentage potential herbivore biomass (x-axis = 80%).

Figure 4.

Trends in parrotfish biomass across management interventions at three locations in Hawai‘i. Shapes are means and lines are 95% intervals from bootstrapped predictions from a binomial-gamma hurdle model that combine predictions of probability of the presence (binomial component) and biomass when present (gamma component). (a) Comparison of three marine reserves (circle), and eight non-reserve sites (diamond) on the island of Maui based on surveys conducted in 2018 and 2019. (b) Comparison before and after establishment of Kahekili Herbivore Fisheries Management Area on Maui established in 2009, before data are from 2008 and 2009, and after data from 2011 to 2015. (c) Comparison of sites in the West Hawai‘i Regional Fisheries Management Area before and after a ban on SCUBA spearfishing took effect in 2013 with before data from 2007 to 2010 and after data from 2016 to 2019, with comparisons split between 10 sites not in protected areas ('out'), 13 sites inside Fishery Management Areas (aquarium collecting prohibited and other gear restrictions in some areas, ‘FMA’), and 2 sites in Marine Life Conservation Districts (no-take areas, ‘MLCD’). There were few or no observations without parrotfishes inside the no-take areas in West Hawai‘i so means and intervals are from a gamma model rather than a hurdle model for the ‘MLCD’ portion of (c).