Large-scale, long-term field data from the GBR Marine Park
The field data for CoTS, hard coral cover (here referred to as coral cover) and coral reef fish were obtained from the Australian Institute of Marine Science’s (AIMS) Long-Term Monitoring Programme (LTMP), while fisheries retained catch data were supplied by the Queensland Department of Agriculture and Fisheries (QDAF). The LTMP has been surveying CoTS populations and coral cover at reefs across the length and breadth of the GBR Marine Park since 198350 and has quantified the status and trend of benthic and reef fish assemblages since 1995. Specific examination of the effectiveness of zoning within the GBR Marine Park has also been undertaken24. The surveyed reefs are located within zones open to fishing (i.e. General Use, Habitat Protection and Conservation Park) and zones closed to fishing (i.e. Marine National Park Zones, Preservation and Scientific Research Zones) (Supplementary Table 1). The QDAF fisheries data comprise annual retained catch data from the Coral Reef Fin Fish Fishery including commercial, recreational (including charters) and Indigenous fisheries, as well as the Marine Aquarium Fish Fishery (Supplementary Data 1–3). Monthly catch return logbooks became compulsory for all trawlers and line fisheries on 1 January 198830. Retained catch data from each of these fisheries is collected separately and differently by QDAF (please see details below). Use of these data is by courtesy of the State of Queensland, Australia, through the Department of Agriculture and Fisheries.
For both the LTMP and QDAF data, the data sets are chronologically divided into report (LTMP) or financial (QDAF) years, respectively, from 01 July to 30 June. This means that, for instance, the second semester of 2017 belongs to the 2018 report or financial year. Hereafter we will refer to report or financial year as simply year. Below we explain each of these data sets in more detail.
LTMP CoTS and coral cover data
LTMP CoTS and coral cover data are available from 1983 to 2020. Both observed CoTS and coral cover data are based on field observations that employ manta tow surveys around the perimeter of each reef following AIMS’ Standard Operational Procedure51. Within this period, manta tows were conducted once per year but not all reefs were sampled every year. Briefly, manta tow surveys are a broad-scale technique that covers large areas of reef quickly and provides an assessment of broad changes in the distribution and abundance of corals and CoTS. During surveys, two boats each tow an observer clockwise and anti-clockwise around reef perimeters in a series of 2-min tows until they meet at the other end of the reef. Each observer records categorical coral cover (Supplementary Table 8) and the number and size of any CoTS observed (Supplementary Table 9) at the end of each 2-min tow51. Manta tow surveys are a non-targeting, rapid assessment method, and therefore it under-samples CoTS individuals that are <15 cm in total body diameter and also under-estimates true population numbers52. CoTS density at the individual manta tow level is measured as total number of individuals per 2-min tow. Coral cover at the individual manta tow level is measured as percentage coral cover category per 2-min tow. Categories (in quotes) were converted to proportions, using the mid-points of each category, for subsequent analyses as follows: “0” = 0, “1” = 0.05, “1L” = 0.025, “1U” = 0.075, “2” = 0.2, “2L” = 0.15, “2U” = 0.25, “3” = 0.4, “3L” = 0.35, “3U” = 0.45, “4” = 0.625, “4L” = 0.5625, “4U” = 0.6875, “5” = 0.875, “5L” = 0.8125, and “5U” = 0.9375. In our statistical analyses, we use both CoTS density and coral cover proportion at the individual manta tow level for a given reef–year combination, as well as averaged across manta tows for a given reef–year combination (please see statistical modelling below).
LTMP coral reef fish data
LTMP coral reef fish data are available from 1995 to 2020, with coral reef fish species surveyed following AIMS’ Standard Operational Procedure53. Here we focus on observations on 56 paired reefs surveyed biennially from 2006 to 2020. These paired reefs are surveyed as part of a dedicated study to examine the effects of the 2004 re-zoning of the GBR Marine Park, with one reef open to fishing (fished) and the other a no-take marine reserve where fishing was prohibited (unfished)41. Reef fishes are surveyed in a standard reef slope habitat on the northeast flank of each reef, which is oblique to the prevailing weather conditions and provides consistent relative exposure from which to draw meaningful spatial and zoning comparisons. At each reef, fishes are surveyed along five permanent 50 m transects in each of three sites (n = 15 transects reef−1 year−1). The start and end of each transect is marked with metal stakes, with smaller metal rods spaced every 10 m. Large mobile fishes (Acanthuridae, Chaetodontidae, Labridae (including scarine parrotfishes), Lethrinidae, Lutjanidae, Serranidae, Siganidae and Zanclidae are counted on 5 m wide belts (transect area = 250 m2). Total lengths (TL) of species targeted by fisheries are also recorded. In line with our hypotheses and for our statistical analyses, we retained monitoring information from paired fished and unfished reefs for the following taxa only: Labridae (wrasses), Lethrinidae (emperors), Lutjanidae (tropical snappers), and Serranidae (rockcods) (Supplementary Table 5). The use of standard operating procedures and the high level of training and experience of observers means that there is a high level of accuracy in the data, with little systematic observer bias in counts54.
QDAF fisheries data
Fisheries data for retained catches (in biomass) for the recreational charter and commercial line fisheries within the Coral Reef Fin Fish Fishery, as well as commercial net and trawl fisheries within the GBR Marine Park, were obtained from QDAF in September 2019. The focus was on obtaining retained catch data for individual coral reef fish species that are either known or likely consumers of different life stages of CoTS24, as well as for large piscivores that may influence CoTS population outbreaks through more complex trophic cascades35,36. Retained catch data from the commercial net and trawl fisheries were included based on advice from QDAF as both these fisheries have targeted and reported on some of these coral reef fish species. The commercial net fishery targets several finfish species, using a variety of different net fishing methods55. This fishery has reported annual retained catches for tropical snapper from 1989 onwards, and early records also include some retained catches (albeit relatively small) for coral trout and emperor. The commercial trawl fishery currently targets prawns, Moreton Bay bugs and scallop55, with early logbook records also reporting retained catches for coral trout, emperor, or tropical snapper.
Fisheries data for retained catches were obtained for each of these four fisheries at the smallest spatial (i.e. site, 6-by-6 nautical miles) and temporal (i.e. annual) resolution available, from 1988–1989 (i.e. reporting year 1989) to 2017–2018 (i.e. reporting year 2018). In addition, we also obtained annual total retained catch data for each of these fisheries at these spatial scales, comprising retained catches for all fin fish for charter and commercial line and net fisheries, as well as for prawns, Moreton Bay bugs and scallops for commercial trawl fisheries. For our analyses, the LTMP CoTS and coral cover data sets were aligned both spatially and temporally with this QDAF retained catch data set (please see response of CoTS density to fish biomass removal below).
The extracted QDAF retained catch data set was based on a list of 532 selected fish taxa, with each taxon linked to a unique CAAB code (Codes for Australian Aquatic Biota; https://www.cmar.csiro.au/caab/). QDAF advised that identification of individual species may not be completely reliable in their data set, hence retained catch data were also obtained or estimated for higher taxonomic levels (i.e. families). For example, data were obtained for popular species such as redthroat emperor (L. miniatus) and spangled emperor (L. nebulosus), as well as for all other emperor species, to estimate a total retained catch for the emperor family (Lethrinidae). We used the CAAB codes to assign each of the fish species caught by the recreational charter and commercial line, net and trawl fisheries to a coral reef fish family. In line with our hypotheses and for our statistical analyses, we focussed on the following six coral reef fish groups, namely, (1) Labridae (wrasses), (2) Lethrinidae (emperors), (3) L. miniatus and L. nebulosus (redthroat and spangled emperors), (4) Lutjanidae (tropical snappers), (5) Serranidae (rockcods) and (6) Plectropomus spp. and Variola spp. (coral trout) (Supplementary Table 2). Potential effects of other known predators of pelagic or benthic CoTS, including species of the families Ballistidae (triggerfish), Chaetodontidae (butterflyfish), Diodontidae (Porcupinefish), Haemulidae (grunters), Pomacentridae (damselfish) and Tetraodontidae (Pufferfish) (Table 1)24, could not be assessed as data on fisheries biomass for these groups were either not available or too limited for our analyses (Haemulidae).
For comparative purposes, annual retained catch data (in numbers) for the recreational and indigenous fisheries for the GBR Marine Park were obtained from QDAF in October 2018, the QFish website (https://qfish.fisheries.qld.gov.au/, accessed February 2019) and associated technical reports56,57,58,59, respectively. In addition, annual retained catch data (in numbers) for the Marine Aquarium Fish Fishery in the GBR Marine Park were provided by QDAF in September 2019 at the smallest spatial (i.e. site, 6-by-6 nautical miles) and temporal (i.e. annual) resolution available, from 1994–1995 (i.e. reporting year 1995) to 2017–2018 (i.e. reporting year 2018). These data were not included in our statistical analyses because (1) the retained catches are recorded in numbers and not in biomass, and (2) they are not available at spatio-temporal resolutions relevant for our study (recreational and Indigenous fisheries only).
Response of CoTS density to fish biomass removal
To determine whether CoTS densities can be predicted by the removal of fish biomass, we amalgamated the two quantitatively independent data sets, the AIMS LTMP CoTS and coral cover data set and the QDAF fisheries data set, at the smallest available spatial (i.e. fisheries logbook reporting site, 6-by-6 nautical miles) and temporal resolution (i.e. monitoring or reporting year). First, each LTMP monitoring location was mapped onto Google Earth and subsequently mapped onto the QDAF logbook reporting grids (30-by-30 nautical miles) and sites (6-by-6 nautical miles) for commercial fishing (see https://www.business.qld.gov.au/industries/farms-fishing-forestry/fisheries/monitoring-reporting/requirements/logbook-maps and Supplementary Fig. 1 for example). Each LTMP CoTS monitoring location visited from 1989 to 2019 was then given a unique grid–site code obtained from these logbook maps. This also showed, however, that some LTMP CoTS locations straddled more than one QDAF reporting grid and/or site. Consequently, the extent of each CoTS monitoring location, specifically the North–South and East–West extremes, was checked onto QDAF logbook maps. CoTS monitoring locations were subsequently given the unique grid–site code based on the largest area of reef covered by this code, or, if reef areas were similar across more than one grid and/or site, based on the length of reef perimeter surveyed. One reef was equally divided between two grid sites; QDAF fisheries data were only available for one of these and the reef was assigned to that grid–site code. Finally, locations and associated reef images on Google Earth were double-checked with those provided on the AIMS’ Reef Surveys public website (https://apps.aims.gov.au/reef-monitoring/). Any discrepancies were clarified with the LTMP team to ensure that LTMP monitoring locations were assigned to the correct Fisheries’ grid–site code.
A total of 2765 LTMP CoTS density records obtained from 1989 to 2019, comprising both CoTS density and coral cover percentage data, were assigned to a unique Fisheries’ grid–site code. Further examination of these assignments showed that 564 LTMP records had ≥2 observations in the same grid–site within the same year; these were addressed as follows. First, 142 of these 564 LTMP records comprised observations from different reefs in both fished and unfished zones. In these cases, the LTMP records obtained from reefs in unfished zones were removed (n = 68), as these reefs are closed to fishing and would not have been fished. Second, 422 of these 564 LTMP records comprised observations from multiple reefs either in fished or in unfished zones. In these cases, the LTMP records on CoTS density and coral cover percentage were averaged across multiple observations (mostly duplicate, but some triplicate and one quadruplicate observation) across fished or across unfished zones in the same grid–site within the same year. This resulted in a total of 2481 LTMP records assigned to a unique Fisheries’ grid–site code, comprising observations on CoTS density and coral cover percentage obtained from 1989 to 2019, and information on zoning status.
Following assignment of the LTMP records to a unique Fisheries’ grid–site code, we subsequently aggregated CoTS densities, for each unique site–year combination, to an average across manta tows conducted on fished reefs only and standardized to total number of individual CoTS per minute. Similarly, for each unique site–year combination, retained catches for coral reef fish were aggregated across commercial line, net and trawl and recreational charter fisheries to obtain an estimate of fish biomass removed. We note that these measurements are quantitatively independent because they were collected by different types of surveys (CoTS manta tows vs. fisheries retained catch reports), are within the same 6-by-6 nautical miles logbook reporting site (although not necessarily on the same reef) and are based on different units of measurement (CoTS density is based on time, whereas fish biomass removal is in kilos). Given these caveats, we assume that the data sets are reasonably comparable for the purposes of the large-scale and long-term trends that are of interest here; however, one should exercise caution when interpreting the results. The merged data set contained 19,239 paired CoTS–fisheries observations collected from 157 individual fisheries logbook reporting sites across the GBR from 1990 to 2018.
We analysed this data set using a Bayesian hurdle-gamma modelling approach, with the probability of CoTS density at time (t+x) being zero following a Bernoulli distribution with a logit link and the positive continuous outcomes being modelled using a Gamma distribution with a log link. We added coral cover (continuous: proportion) as a fixed effect for both the Bernoulli (zero) and Gamma (non-zero) components of the model in order to maintain the consistency with the two tests conducted under the role of zoning and coral cover on CoTS density. Fisheries in the GBR are only permitted in open reefs, so paired CoTS–fisheries observations where CoTS data came from closed zones only were removed and zoning status was not included as a covariate in this model. Because fish effects on CoTS density might be manifested after multiple time lags28, we ran six models for each of the six coral reef fish groups of interest, each corresponding to a (x) time lag between fish biomass removal and CoTS density from 1 through 6 years (i.e. 36 models in total). For the Gamma component only, and in order to examine whether CoTS densities within sites increase with increasing removal of coral reef fish following multiple time lags, we also added annual fish biomass removal (continuous: kg) at time (t) as a fixed effect; grid–site ID was added as having a hierarchical effect on the model intercept. We note that coral cover data were from the same years as CoTS density data (i.e. (t+x)). In Supplementary Method 2, we provide detailed model and algorithm fitting specifications, a power analysis to check predictive power of original model, posterior predictive checks, comparisons between prior and posteriors distributions, posterior distribution of model parameters and chain mixing trace-plots (Supplementary Figs. 4–40). We then determined the percentage of the posterior distribution of the slope between annual fish biomass removal at time (t) and CoTS density at time (t+x) that falls above 0.
Effects of no-take marine reserves on coral reef fish
To compare the biomass, density and size of the six fish groups on fished and unfished reefs, specifically those groups that influence CoTS densities, we used LTMP coral reef fish observations from 840 transects conducted on 56 paired fished and unfished reefs along the length of the GBR Marine Park between 2006 and 2020. All LTMP reef fish data from all species monitored within the families Labridae (wrasses—five species), Lethrinidae (emperors—16 species), Lutjanidae (tropical snappers—21 species) and Serranidae (rockcods—35 species) (Supplementary Table 5) were standardized by converting raw counts to densities per 1000 m2. Size was analysed at the site level as mean population total body length (TL cm). Standing biomass (kg) per 1000 m2 was calculated for each fish species from estimated fish lengths using published length–weight relationships60,61. Inferences about differences between no-take marine reserves and reefs open to fishing were based on Bayesian C.I.s calculated from parameters estimated from hierarchical models. Site-level mean population length was modelled following a gamma distribution, whereas density and standing biomass were modelled with a hurdle-gamma approach, similarly to described above for fisheries-dependent data. Each response was modelled as a function of zoning status (fished vs. unfished, categorical fixed effect), while accounting for the effects of site, reef pair ID, reef ID and year, following Emslie et al.41. In the hurdle-gamma models, the logistic component was modelled as a function of zoning status. In Supplementary Method 3, we provide detailed model and algorithm fitting specifications, posterior predictive checks, comparisons between prior and posteriors distributions, posterior distribution of model parameters and chain mixing trace-plots (Supplementary Figs. 41–60). Differences between values for unfished and fished reefs were then expressed as a percentage of the value on the fished reefs, such that a higher value in unfished compared with fished reefs would yield a positive difference, whereas a lower value would give a negative difference.
Effects of no-take marine reserves and coral cover on CoTS density
To quantify the effects of reef zoning and coral cover on CoTS density, we used CoTS density and proportional coral cover observations from 3358 manta tow surveys (i.e. reef–year combinations) conducted on 490 reefs along the length of the GBR Marine Park between 1985 and 2020. This data set contains 157,348 individual 2-min manta tows in which a total of 52,921 individual CoTS were recorded. Specifically, we employed a Bayesian hierarchical approach to model CoTS density per tow (individuals/2 min) following a negative binomial distribution. First, to determine whether CoTS densities are higher on fished reefs compared to unfished reefs we added zoning status (categorical: open, closed) as a fixed effect. Second, to determine whether CoTS densities change with coral cover we added tow-level coral cover (continuous: proportion) as a fixed effect. Reef–year combination (n = 3358) was added as an intercept-level hierarchical effect. In Supplementary Method 4, we provide detailed model and algorithm fitting specifications, posterior predictive checks, comparisons between prior and posteriors distributions, posterior distribution of model parameters and chain mixing trace-plots (Supplementary Figs. 61 and 62). We also tested for but found no evidence of zero inflation in the data (Supplementary Method 4 and Supplementary Fig. 63).
Testing hypotheses
For most models, we assess each hypothesis against the posterior distribution of its corresponding parameter in terms of percentage probabilities. For example, in the hurdle-gamma models described in the section above, we measure the percentage of the posterior distribution of the slope between annual fish biomass removal at time (t) and CoTS density at time (t+x) that falls above 0.
Reporting summary
Further information on experimental design is available in the Nature Research Reporting Summary linked to this paper.
Source: Ecology - nature.com
