Towards process-oriented management of tropical reefs in the anthropocene

Tropical reefs and the fish relying on them are under increasing pressure. Shallow-reef fish provide important ecological information in addition to sustaining fisheries, tourism and more. Although empirical metrics of fish biomass are widely used in fisheries management, metrics of biomass production—how much new biomass is produced over time—are rarely estimated even though such production informs potential fisheries yields. Here we estimate fish standing biomass (B), biomass production (P, the rate of biomass accumulation) and biomass turnover (P/B ratio, the rate of biomass replacement) for 1,979 tropical reef sites spanning 39 tropical countries. On the basis of fish standing biomass and biomass turnover, we propose a conceptual framework that splits reefs into three classes to visualize ecological and socio-economic risk and help guide spatial management interventions (for example, marine protected areas) to optimize returns on conservation efforts. At large scales, high turnover was associated with high human pressure and low primary productivity, whereas high biomass was associated with low human pressure and high primary productivity. Going beyond standing fish biomass to consider dynamic ecological processes can better guide regional coral reef conservation and sustainable fisheries management. Sustaining tropical reefs in the Anthropocene is a vital challenge. This study proposes guiding regional conservation on the basis of ecological processes and finds that the biomass and productivity of reef fish provide complementary information for management.

Tropical reefs and the fish relying on them are under increasing pressure. Shallow-reef fish provide important ecological information in addition to sustaining fisheries, tourism and more. Although empirical metrics of fish biomass are widely used in fisheries management, metrics of biomass production-how much new biomass is produced over time-are rarely estimated even though such production informs potential fisheries yields. Here we estimate fish standing biomass (B), biomass production (P, the rate of biomass accumulation) and biomass turnover (P/B ratio, the rate of biomass replacement) for 1,979 tropical reef sites spanning 39 tropical countries. On the basis of fish standing biomass and biomass turnover, we propose a conceptual framework that splits reefs into three classes to visualize ecological and socio-economic risk and help guide spatial management interventions (for example, marine protected areas) to optimize returns on conservation efforts. At large scales, high turnover was associated with high human pressure and low primary productivity, whereas high biomass was associated with low human pressure and high primary productivity. Going beyond standing fish biomass to consider dynamic ecological processes can better guide regional coral reef conservation and sustainable fisheries management.
Severe impacts on ecosystem functioning are among the most widespread ecological consequences of the Anthropocene, occurring as a direct result of resource overexploitation, habitat degradation and defaunation [1][2][3] . Tropical reefs are no exception, with degraded states and processes projected to worsen in the long term due to escalating human pressures [4][5][6] . Given the diverse and essential contributions to people provided by tropical reefs, such degradation will have overwhelming consequences for human livelihood and welfare, primarily in developing countries 7 .
Marine protected areas (MPAs) are a key management tool to counteract deleterious human impacts on tropical reefs [8][9][10] . MPAs, and in particular no-take or fully protected marine reserves (which ban extractive activities), can increase the abundance and biomass of fish species within their boundaries 11,12 and even benefit fisheries by improving catches in adjacent areas through the spillover of both adults and larvae [13][14][15] . In some cases, MPAs can improve human well-being, but in others they can be deleterious to people and become a source of conflict 16 . As alternatives, partially protected areas, locally managed MPAs and other effective area-based conservation measures 17 can maintain the provision of livelihoods to people 18 but sometimes to the detriment of ecological benefits 19 .
A critical challenge for marine conservation is therefore to sustain nature's contributions to people while protecting marine biodiversity by preventing ecological decline. Thus, to be sustainable and fair, Article https://doi.org/10.1038/s41893-022-00981-x on maximum length and temperature ranged from 0.02 to 8.71 (Supplementary Fig. 2), decreased according to the maximum observed body size and increased with temperature, as expected (Supplementary Fig. 3).

Reef fish biomass, biomass production and biomass turnover
We calculated fish standing biomass on the basis of our predicted growth trajectories, daily fish biomass production and biomass turnover (P/B ratio) for each transect, and we averaged these metrics across transects within each site, resulting in values for 1,979 sites spanning 39 countries. Across the 1,979 reef sites, fish standing biomass ranged from 1.36 g m −2 to 816 g m −2 , and fish biomass production ranged from 0.002 g m −2 d −1 to 2.74 g m −2 d −1 . Although biomass outliers (>99% of all values) were removed, sites with very high biomass still remained. These sites were mainly characterized by observations of important schools of large fish (for instance, a school of 120 Naso hexacanthus that were 40 cm long on one transect). Fish biomass turnover (the ratio between standing biomass and biomass production) ranged from 0.018% per day to 0.635% per day. Overall, standing biomass and biomass production exhibited a strong positive correlation (r = 0.76, P < 0.001). However, we observed only a weak, albeit significant, negative relationship between fish standing biomass and biomass turnover (r = −0.15, P < 0.001), confirming that these two latter metrics provide different and complementary information relevant to fisheries potential and management on tropical reefs ( Supplementary Fig. 4).

Distribution of reef sites in management classes
Of the 1,979 sites, our thresholds resulted in 90 sites (4.5%) classified as low standing biomass/turnover, where both fish biomass and biomass turnover were below 25% of the observed values. We identified 91 high biomass sites, with fish biomass above 95% and turnover values below 75% of the observed values, and 495 high biomass turnover sites, where biomass turnover was above 75% of the observed values. Almost every class was present across all sampled geographical areas, with no area showing sites of only one class ( Fig. 4 and Supplementary Fig. 5).
We calculated the mean relative percentage of each class as the average percentage of biomass in each of the six trophic-level bins (using trophic-level values from FishBase 37,38 ) (Fig. 2). Low turnover/ biomass sites were dominated by intermediate trophic levels (3/3.5), mainly benthic invertivores and planktivores, and had the lowest proportion of biomass at low trophic levels (2/2.5, mostly algal farming fish) and higher trophic levels (3.5-5). In contrast, high turnover sites had the highest relative biomass of primary consumers compared with sites in the other classes (2/2.5, mostly scraping and browsing herbivores). High turnover sites and high biomass sites both had higher proportions of biomass in the top trophic levels (predators) than sites in the low biomass/turnover class (Fig. 2).
For each site, we calculated the mean relative biomass and fish body size per diet category. Overall, mean carnivore and herbivore fish sizes were significantly higher in high biomass sites than in other sites (P < 0.001). The overall mean fish size regardless of diet was significantly lower in high turnover sites than in other sites (P < 0.001), and mean carnivore and herbivore fish sizes were also significantly lower in high turnover sites than in others (P < 0.001). The relative biomass of herbivores was significantly higher in high turnover sites than in other sites (P <0.05, Fig. 2).

Environmental and socio-economic drivers
The mean accuracy of our random forest model (Methods) classifying reef sites into management classes using nine environmental and socio-economic variables was 73% over 100 iterations. Sea surface temperature, human gravity (a proxy of human impact calculated as the ratio between population estimates of a place and the squared travel time to the reefs 9 ) and primary productivity had the strongest predictive capacity, with relative contributions (evaluated by permutation; Methods) management actions must be both context-specific and inclusive 18 , which entails a trade-off between objectives such as maintaining fisheries, ecosystem functioning and biodiversity 20,21 . Even though these objectives are not systemically antagonistic 20,21 , they are rarely considered together since MPAs may not be successful across both biodiversity and socio-economic objectives 22 . A range of management options and fisheries restrictions adapted to the local context may thus be necessary to reach several Sustainable Development Goals synergistically. However, the scientific basis or rationale to define and classify local contexts where different management options can be the most appropriate is unclear.
Here we suggest that examining tropical reefs along gradients of both fish biomass and biomass turnover (that is, the ratio between biomass production (P) and standing biomass (B), or P/B ratio) can inform on which reefs are at risk both ecologically and socially and can guide possible regional-scale conservation strategies on tropical reefs. Indeed, fish standing biomass is closely related to several ecological states and processes and has also often been used to manage coral reef fisheries 23,24 . However, fisheries yield also depends on the dynamics of biomass production (g m −2 d −1 ), a flow-based rate measuring biomass accumulation in a place over time, and biomass turnover (P/B ratio), the rate at which biomass is produced and stored 25 . Both metrics can provide complementary information for improving the local management of resources as they inform on the stability of fisheries yields, but they have less often been calculated due to data deficiency [26][27][28] . Furthermore, the relationship between biomass and biomass production is generally nonlinear, which may lead to increased biomass turnover with increasing fishing pressure across some range of biomass 29 . Indeed, a given reef can exhibit high turnover with a low to intermediate biomass, suggesting that fishing activities could be sustainable and that partial protection (for example, restrictions on fishing gears or effort) could promote a compromise between exploitation and conservation 30,31 . Other reefs can host a very high fish biomass but with low biomass turnover. In this case, if these reefs are not already under some kind of protection, preserving the ecological states, processes and contributions associated with high fish biomass by placing a no-take or no-entry MPA may be appropriate when socially feasible or acceptable 24,32 . Using 1,979 tropical reef sites from a global database on fish community composition and size structure, we calculated fish biomass, biomass production and biomass turnover ( Supplementary  Fig. 1). We split reef sites into three different management classes: high standing biomass, high biomass turnover and low standing biomass/ turnover ( Fig. 1). More precisely, we split reef sites by lower (25%) and upper (75%) quartiles for low versus high biomass turnover, and by 25% and 95% percentiles for low versus high biomass, on a logarithmic scale for both metrics. We then used environmental and human variables to model and understand the roles of social and environmental contexts in determining which of these classes each site belonged to.

Prediction of species growth rate
We used standardized and quantitative reef fish counts from the Reef Life Survey (RLS) database 33 , among which we selected tropical and subtropical reefs (showing a minimum monthly sea surface temperature of 17 °C) spanning all tropical ocean basins ( Supplementary Fig. 1). Following a previously published procedure, we used a broader definition of 'tropical' oceans, including locations between 17 °C and 20 °C where tropical species are present. Even though these included subtropical reefs, we hereafter refer to these reefs as tropical 34 . We used an existing framework and its associated rfishprod package to estimate fish growth trajectories (K max ) using fish individual body size and sea surface temperature (the two drivers of fish growth variation across reef fish species 35 ) and mortality rates using fish individual size, maximum size and growth coefficients 36 , for the 1,400 species in our database 25 . The predicted growth coefficients (K max ) for the 1,400 species based Article https://doi.org/10.1038/s41893-022-00981-x of 24.6%, 14.7% and 12.7%, respectively. Other parameters such as the human development index (a measure of socio-economic development that incorporates wealth, life expectancy and education 39 ), degree heating weeks and the presence of non-governmental organizations (acting as potential levers towards the achievement of conservation targets) also influenced the classification, with relative contributions of 11.4%, 10.2% and 10%, respectively (Supplementary Table 1 and Supplementary Fig. 6). Surprisingly, dependency on marine ecosystems and MPA management had weak predictive capacity (with relative importance of 7.6% and 5.7%, respectively; Supplementary Figs. 6-8).
We conducted a sensitivity analysis by modifying these thresholds by ±5% and ±15%. We also conducted a sensitivity analysis by removing transects that had individuals larger than 95% of all other individuals and transects that had schools with a number of individuals higher than 95% of all other schools, to account for potential high variability between transects. While necessarily changing the number of sites in each management class, all these sensitivity analyses did not change model performance or variable importance in subsequent analyses (see below, Supplementary Figs. 9 and 10 and Supplementary Table 1).
Partial dependence plots, showing the marginal effect of each covariate on the predicted outcome (that is, its independent effect if all other variables were held static), revealed that sites with high fish biomass were characterized by low human gravity and intermediate to high primary productivity (Fig. 3). Even though high biomass sites were characterized by overall low human gravity, the levels of human gravity and dependency on marine resources were highly variable among management classes (Fig. 4). High turnover sites were characterized by high temperature, high human gravity and low primary productivity. Of note, the low primary productivity observed in the high turnover class may be linked to biases in the method used to estimate primary We used the rfishprod package to estimate growth trajectories and mortality rates for the 1,400 fish species in the RLS database, and then to estimate standing biomass (g m −2 ), biomass production (g m −2 d −1 ) and biomass turnover (the P/B ratio multiplied by 100, expressed as % per day) for the 1,979 sites spanning 39 countries (Supplementary Fig. 1) 25,33,35,68 . M is fish body mass in grams, L is fish body length in centimeters, a is the coefficient of the power function, b is the allometric coefficient and z is the species-specific mortality coefficient. b, According to fish standing biomass and biomass turnover, we classified the reef sites into three management classes and one 'base' class: high biomass (green), high turnover (dark blue), low biomass/ turnover (orange) and mid-range sites (light blue, the base class). Tropical reefs in our case include those with a monthly sea surface temperature >17°C 34 . Panel a adapted with permission from ref. 25 , Springer Nature Limited.
Article https://doi.org/10.1038/s41893-022-00981-x production given that it does not account for spatial subsidies, which have been shown to drive biomass production 40 . In contrast, low biomass/turnover sites were characterized by high human gravity, high primary productivity and colder temperatures (Fig. 3).

Discussion
Process-based metrics such as biomass production provide important information on fisheries yields and biomass build-up and consequently need to be incorporated into decisions about where and how to manage tropical reefs 5 . Several studies show the decoupling of biomass turnover, a dynamic ecosystem rate, from standing stock biomass, which presents new opportunities to improve and adapt conservation strategies 29,30,41 . We propose a conceptual framework accounting for both fish standing biomass and biomass turnover to guide the incorporation of ecological processes into the management of tropical reefs. High biomass sites are characterized by more biomass in higher trophic levels, a higher relative biomass of upper trophic levels, and higher fish size of lower and upper trophic levels (Fig. 2), consistent  with the U-shaped trophic structure of vertebrate assemblages 42 . This U-shape ensures a high energy transfer efficiency (that is, a high proportion of production of lower trophic levels is converted into production at the upper trophic levels), a critical factor shaping marine ecosystem functioning but also sustainable fishing rates 43 . These sites are mainly located in areas with low human gravity, confirming their status as intact reefs with low anthropogenic pressures. We suggest that these remote sites with high fish biomass, if they are not already under some kind of protection, can be seen as the locations most suited to establishing no-entry or no-take marine reserves. Prioritizing conservation efforts towards areas with low economic and social costs is an increasingly used policy to rapidly and easily achieve conservation goals such as Aichi target 11, which aimed to protect "at least 17% of terrestrial and inland water areas and 10% of coastal and marine areas" by 2020, or Sustainable Development Goal 14, which aims to "conserve and sustainably use the oceans, seas and marine resources for sustainable development" 44 . These so-called 'residual' conservation efforts have come under criticism because they target locations where human impacts are already low and reducing resource extraction is assumed to make little difference in both ecological and socio-economic outcomes 44 . However, when considering multiple ecological goals simultaneously (fisheries, biodiversity and ecosystem functioning), marine reserves can provide substantial gains, including in areas with moderate to low human impact 21 . Furthermore, high fish biomass is shown to support key ecosystem functions (such as predation) and associated contributions to humanity, which justifies the urgent protection of these sites 23,24 . Although these high biomass reefs are in remote locations far from human uses, they can constitute important sources of larvae and juveniles that allow the replenishment of depleted stocks in other areas through dispersal corridors, if sufficiently connected 45 . Furthermore, the accumulation of large individuals of commercial species inside well-protected MPAs can benefit fisheries outside MPA boundaries through adult spillover, thereby contributing to food security and human livelihoods 14 .
In contrast, low biomass/turnover sites not only have little biomass overall but also lack a solid base at the bottom of the food web, with a low biomass of primary consumers (herbivorous fishes, Fig. 2), suggesting that benthic primary productivity may be low or that fishing pressure may be extremely high. These sites also exhibit a high relative biomass of mid-trophic-level generalists, consistent with high fishing pressure which can induce the replacement of herbivorous fish by herbivorous sea urchins 37 . These sites are at risk of ecological collapse given their low biomass as well as their low rate of biomass renewal; they should therefore benefit from conservation efforts when possible. However, it remains unclear how these sites would react to protection given their low regenerative and reproduction potential. Previous studies show that heavily depleted sites may need 35 years on average to recover 23 , and recovery could be impaired by low biomass turnover, meaning that protection efforts on these sites may only promote uncertain long-lasting benefits. However, further research is needed to understand the consequences of this low biomass/turnover state for local fisheries and how these sites can react to fishing restrictions to provide locally informed management decisions. Indeed, given that conservation operates on a tight budget 46 , the strategic placement of MPAs must avoid conservation failures, which may lead to an erosion of trust, increasing risk of conflicts 15,46 and hindering future conservation endeavours 47 . However, this class exhibits high inter-site variability, with certain sites hosting higher fish size of carnivores, which may explain their low turnover (Fig. 2). This calls for detailed analyses of the trophic and size structures of these sites before considering appropriate management measures. The high biomass turnover sites are usually located in areas with high human pressure, concurring with previous findings that turnover is positively associated with exploitation rate 29 (Fig. 3). Indeed, this suggests that fishing pressure removes large species contributing to high biomass and therefore increases biomass turnover, boosted by small species. This widespread exploitation of reefs is altering species composition and ecosystem functioning towards novel ecosystems 48 . However, several studies report that fisheries catches have maintained or even increased in heavily fished sites 29 . These artisanal fisheries are mostly sustained by an increase in herbivorous fish biomass such as parrotfish (Scaridae) or rabbitfish (Siganidae), which show high growth rates and therefore contribute to high biomass production 29,49 . In accordance with these previous observations, this management class was characterized by a higher relative biomass of herbivorous fishes than of the other management classes, as well as overall smaller fish size (Fig. 2). High biomass turnover has been identified as a compensatory ecological mechanism whereby fish populations characterized by high growth rates (such as herbivores) positively respond to size-selective fishing gears and biomass collapse through increased biomass turnover 29,41 . However, this compensatory mechanism may be only short term, as intensive degradation may eventually also lead to a decrease in herbivore biomass 49 . Indeed, we observe a lower mean fish size at lower and upper trophic levels than in other levels, suggesting that intensive degradation disrupted the U-shaped trophic structure of these high turnover sites (Fig. 2) 42 . Accounting for and monitoring reef fish biomass production and turnover appears key to understanding   Given the highly variable socio-economic context of these high turnover sites (Fig. 4), our framework suggests locally informed spatial management, where protection intensity needs to consider local ecological and social contexts 18 . When human gravity and dependence on marine ecosystems are high, we suggest concerted and co-managed local restrictions on these sites (for example, relating to industrial activities or destructive fishing practices) through the establishment of partially protected MPAs, locally managed MPAs or other effective conservation measures that can still sustain local fisheries and other well-being outcomes 15,18 . Other effective conservation measures may be of particular relevance here given their importance for Indigenous and Community Conserved Areas and ecosystem-based management 50 . In most instances, partial protection would not achieve the same ecological outcomes as full protection that is complied with 19,51 . However, in many coral reef countries, compliance with fully protected MPAs is low 52 , so better ecological outcomes can come from partial protection 53 . Additionally, socio-economic outcomes from partial protection (such as sustaining livelihoods and cultural connections) are mainly positive 19,54 , and this type of management may enable traditional practices 55 and improve the food security of local communities 15 . We argue that these protection measures should be prioritized where dependency on reefs is high and when the total exclusion of extractive activities is not feasible or socially appropriate. In some instances, access restriction can be combined with no-take areas, whereby access restrictions can act as the incentive for establishing no-take areas when communities might not otherwise be willing to give up areas for conservation 54,55 .
According to our thresholds, mid-range sites do not fit in any of the management classes and exhibit neither high biomass turnover nor high biomass. Here management should focus on avoiding the shift towards the ecological trap of low fish biomass and biomass turnover. A range of management strategies may help achieve this goal depending on local socio-economic, cultural and environmental contexts.
Overall, our framework and associated results remain conceptual and, given their global scale, do not necessarily represent any real-life implications or guides regarding specific sites that were studied. Our framework is intended as a heuristic means of considering both stable-state and dynamic measures of reef functioning to inform management. Our results may serve as a demonstration that both metrics are differently affected by environmental and socio-economic parameters, which in turn shape the state of reefs worldwide. Applications for management should be refined at a local scale, tailoring each strategy to local priorities, needs and contexts, while complementing this framework with other important measures of tropical reef state such as habitat complexity, size structure and number of endemic species 56,57 . Furthermore, we acknowledge that our defined classes are based on thresholds that are, to some extent, arbitrarily chosen. Yet, our sensitivity analysis shows that changing thresholds did not change model performance or variable importance. Finally, even though migratory fish made up only a small portion of the studied species, we ignored potential gains and losses in biomass production due to migratory fish. Planners will always be faced with hard choices 21 , but clearer trade-offs and frameworks may assist them in achieving biodiversity conservation priorities while considering socio-economic constraints. The decoupling between fish biomass and biomass turnover and the contrasting information they provide allow for context-dependent local management and provide crucial information for MPA placement or restrictions and the ecological and socio-economic outcomes they aim to achieve.

RLS
The RLS database is a global dataset of underwater visual transects that has monitored the abundance and size distribution of temperate and tropical fish species globally. According to the RLS protocol available at https://www.reeflifesurvey.com, a survey consists of a 50 m transect line on which two methods are applied. Method 1 monitors fish composition, abundance and size classes observed in two five-metre-wide by five-metre-high bands. Method 2 monitors invertebrates and cryptic fishes in two one-metre-wide by two-metre-high bands on either side of the transect line. We restricted our analyses to warm-water reefs by selecting sites that experience minimum monthly sea surface temperatures of 17 °C 34 . Although the present study focused on the tropics, whose sea surface temperature limit is usually set at 20 °C, we decided to use a broader definition of tropical oceans, thereby including locations where species of tropical affinity are present. We removed sharks, rays, seahorses, cryptobenthic fish families and fishes smaller than 5 cm, given the unreliable estimates of small and cryptic assemblages via underwater visual censuses 58,59 . Finally, we removed biomass outliers by filtering out transects with biomass and biomass turnover values 99% higher than all other values. In this study, we worked at the site scale, averaging our metrics across all transects in a given site, which may lead in some cases to variance due to different sample sizes (Supplementary Table 2). However, averaging values between transects at the site level allows us to limit sampling bias at the transect scale. At this global scale, there is a temporal gap between all the studied sites that cannot be avoided. However, within each site, all transects were carried out the same day, so the averaged values are only spatial and not temporal.

Fish biomass production and biomass turnover
The sum of all individual growth trajectories determines the energy flux at the community level, leading to changes in biomass production. Estimating growth trajectories at the individual level is thus the first step to estimating biomass production and establishing predictions at ecosystem levels. We used the abundance, body size and temperature reported by the surveys to estimate standing biomass, growth trajectories and mortality using a previously published framework and its accompanying package, rfishprod (v.0.01) 25 .
Briefly, we modelled growth trajectories using the two main drivers of reef fish growth, body size and temperature 35 . For this, we used the species-specific maximum length and the temperature at which the observations were made. We then obtained species-specific lengthweight relationships using parameters obtained from FishBase 38 . From this information, we computed a species-specific growth coefficient, K max , a standardized growth parameter under a von Bertalanffy growth model. This coefficient represents the von Bertalanffy growth coefficient (K) scaled to apply to a theoretical population where L inf , the species-specific asymptotic maximum length, is equal to the species L max , the maximum observed length, and has previously been used to estimate growth trajectories 4,25,29,41 . Biomass production, expressed as the accumulated biomass of an assemblage in a day, was obtained by subtracting the modelled weight gained by a fish after one day from the observed fish weight, calculated using the length-weight coefficients. Daily mortality rates were estimated stochastically by calculating the probability of a fish dying, taking into account individual growth trajectories, size and age, and removing fish that were expected to die, using previously published predictive models and the rfishprod package 25,36 . We estimated standing biomass and biomass production at the transect level by summing both metrics for each transect, and we calculated biomass turnover, the ratio between biomass production and standing biomass (P/B ratio) 25,26,60,61 . Site-level information was obtained by averaging all metrics across all transects of a given site.

Management classes
We divided the sites into three classes of standing biomass (<25%, 25-95% and >95%-low, medium and very high biomass) and biomass turnover (<25%, 25-75% and >75%-low, medium and high biomass turnover) to partition tropical reefs into three different management Article https://doi.org/10.1038/s41893-022-00981-x classes: (1) high biomass sites, (2) high biomass turnover sites and (3) low biomass/turnover sites. Sites that fit in none of these three management classes were classified as mid-range sites. These thresholds were defined on the basis of quantiles of observed data, except for high biomass, where we used the 95% threshold to define sites with high biomass 21 . For each of these classes, we calculated the relative biomass of each trophic-level group, following a previously published procedure 37 , and we calculated relative biomass and mean fish size per site and diet (herbivore, invertivore, omnivore, planktivore and carnivore) 42 . We statistically compared these metrics between management classes using the one-way analysis of variance test implemented with the aov function in R 62 (version 4.2.1), and we performed multiple pairwise comparison using Tukey multiple comparison of means with the TukeyHD function in R. We conducted a sensitivity analysis to test the influence of the choice of these thresholds on model performance and the influence of environmental and socio-economic covariates on the management classes by changing the thresholds determining the different classes to ±5% and ±15%. To account for high size variability between transects, we conducted a secondary sensitivity analysis by removing individuals whose size was higher than 95% of all other individuals and removing schools for which the number of fish was higher than in 95% of all other schools. The results show that the model performance and outcomes are robust to the choice of the thresholds (Supplementary Methods). We then modelled the effects of human and environmental variables on these classes on the RLS dataset.

Modelling management classes
We selected a set of four environmental and five socio-economic variables with the potential to explain the standing biomass and biomass turnover of fish communities. The environmental variables included net primary production, mean sea surface temperature, degree heating weeks and depth. The socio-economic variables included human gravity (a proxy of human impact considering human population size and accessibility to the reef 9 ) and management, as well as dependency on marine ecosystems, the human development index and the number of non-governmental organizations at the country scale. Since human rights and environmental protection are interdependent, we also included violence and freedom of speech at the country scale 63 . The human development index and dependency on marine ecosystems inform on the socio-economic context, which is a key correlate to illegal fisheries, overexploitation and the establishment of protected areas 5,46,[63][64][65] . Although a proxy of fisheries management agencies or cooperatives would have been more related to fisheries management, finding standardized data at this scale was challenging; we therefore chose the number of non-governmental organizations, which acts as a potential lever towards the achievement of conservation strategies 66 . To select only uncorrelated explanatory variables, we first computed the pairwise correlation between variables within each group of variables. For each pair of variables with a correlation higher than 0.7, we removed the variable involved in several high correlations and kept the other variable, until all pairwise correlations were less than 0.7. The uncorrelated selected variables were depth, primary productivity, human gravity, management, sea surface temperature, degree heating weeks, the human development index and the number of non-governmental organizations. All selected variables were log-scaled prior to analysis to reduce skew and improve model fit. A full description of the covariates, data sources and rationale can be found in the Supplementary Methods.

Modelling drivers of management classes
To understand the potential drivers of these management classes, we examined how key environmental and socio-economic drivers were related to the probability that a site was in any of the management classes. To model management classes according to our nine selected drivers, we used a random forest classification algorithm using the ranger package (v.0.12.1) 67 . We tested our model performance using 100-fold cross-validation. The drivers' importance was estimated using permutation, which is calculated using the percent increase in the mean square error after randomly permuting a given predictor for every tree in the model, averaging the error of the models and then determining the difference relative to the accuracy of the original model. The model parameters were num.trees = 1,000 and mtry = 3.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The biomass production and turnover data that support the findings of this study are available on Figshare with the identifier https://doi. org/10.6084/m9.figshare.20368932.v1.