Research article Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 © Author(s) 2022. This work is distributed under the Creative Commons Attribution 4.0 License. Global nutrient cycling by commercially targeted marine fish Priscilla Le Mézo1,2 , Jérôme Guiet3 , Kim Scherrer1 , Daniele Bianchi3 , and Eric Galbraith1,4,5 1 Institut de Ciencia i Technologia Ambientals (ICTA), Universitat Autonoma de Barcelona (UAB), Barcelona, Spain 2 Laboratoire de Météorologie Dynamique, ENS Ulm, Paris, France 3 Atmospheric and Oceanic Sciences, University of California, Los Angeles, CA, United States 4 Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain 5 Earth and Planetary Sciences, McGill University, Montréal, QC, Canada Correspondence: Priscilla Le Mézo (priscilla.le-mezo@lmd.ens.fr) Received: 21 September 2021 – Discussion started: 27 September 2021 Revised: 22 April 2022 – Accepted: 23 April 2022 – Published: 18 May 2022 Abstract. Throughout the course of their lives fish ingest particulate export flux and was generally more significant food containing essential elements, including nitrogen (N), in the low-export oligotrophic tropical gyres. Our study sup- phosphorus (P), and iron (Fe). Some of these elements are 100 kg ports a significant, direct role of the CTF10 g fraction of the retained in the fish body to build new biomass, which acts ichthyosphere in global nutrient cycling, most notably for Fe, as a stored reservoir of nutrients, while the rest is excreted which has been substantially modified by industrial fishing. or egested, providing a recycling flux to water. Fishing ac- Although we were not able to estimate the roles of smaller tivity has modified the fish biomass distribution worldwide species such as mesopelagic fish because of the sparsity of and consequently may have altered fish-mediated nutrient cy- observational data, fishing is also likely to have altered their cling, but this possibility remains largely unassessed, mainly biomass significantly through trophic cascades, with impacts due to the difficulty of estimating global fish biomass and on biogeochemical cycling that could be of comparable mag- metabolic rates. Here we quantify the role of commercially nitude to the changes we assess here. 100 kg targeted marine fish between 10 g and 100 kg (CTF10 g ) in the cycling of N, P, and Fe in the global ocean and its change due to fishing activity, by using a global size-spectrum model of marine fish populations calibrated to observations of fish 1 Introduction catches. Our results show that the amount of nutrients poten- 100 kg Nutrient elements are used by organisms to construct the tially stored in the global pristine CTF10 g biomass is gen- molecules they need to grow and metabolize but are often erally small compared to the ambient surface nutrient con- scarce, limiting growth rates (Moore et al., 2013; Sterner and centrations but might be significant in the nutrient-poor re- Elser, 2002). Plankton and bacteria dominate the cycling of gions of the world: the North Atlantic for P, the oligotrophic nutrients in the ocean (Sarmiento and Gruber, 2006), but an gyres for N, and the high-nutrient, low-chlorophyll (HNLC) increasing number of studies recognize the contribution of regions for Fe. Similarly, the rate of nutrient removal from animals to biogeochemical cycles (e.g., Wilson et al., 2009; the ocean through fishing is globally small compared to the Bianchi et al., 2013; Saba et al., 2021). Locally, it has been inputs but can be important locally, especially for Fe in the shown that marine animals can significantly impact the sup- equatorial Pacific and along the western margin of South ply and storage of nutrients with consequences for primary America and Africa. We also estimate that the cycling rate production (Cavan et al., 2019; Roman et al., 2014) and in- 100 kg of elements through CTF10 g biomass was on the order of teract with the cycling of elements through direct and indi- 3 % of the primary productivity demand for N, P, and Fe rect pathways (Vanni, 2002; Atkinson et al., 2017; Allgeier globally, prior to industrial fishing. The corresponding ex- et al., 2017). For instance, Leroux and Schmitz (2015) de- 100 kg port of nutrients by egestion of fecal matter by CTF10 g scribed a theoretical framework in which animals control the was 2.3 % (N), 3.0 % (P), and 1 %–22 % (Fe) of the total flux of nutrients up the trophic chain through predation and Published by Copernicus Publications on behalf of the European Geosciences Union. 2538 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish release of waste products while also affecting the cycling of cycling of N and P by fish in fished and un-fished coastal nutrients through non-consumptive effects (e.g., prey selec- sites of the Bahamas and the Caribbean, respectively. Lay- tion and stress induced in prey). In addition, since animals man et al. (2011) showed lower recycling rates of nutrients can swim and move in the water column, they are also able by fish in fished sites, due to biomass reduction and habitat to transport nutrients from one place to another, over dis- fragmentation. Beyond fish biomass reduction, Allgeier et al. tances that increase with animal size (Hall et al., 2007; Vanni, (2016) stressed the role of community size structure that, in- 2002; Roman et al., 2014). But thus far there has been little fluenced by fishing, also led to reduction in nutrient storage effort to estimate how the global fish population, which we and cycling. term “ichthyosphere”, influences large-scale nutrient cycling Although numerous works have identified significant in the ocean. local effects, little is known about the contribution of During their life cycle, fish assimilate, store, and recycle the ichthyosphere to nutrient budgets at the global scale. essential elements that they need to build their body tissues. Maranger et al. (2008) used global fish catch data to estimate This storage of nutrients within fish biomass is important for the total removal of N by commercial fisheries. They inte- human nutrition as wild-caught fish globally provide essen- grated a spatial component in their analysis by computing tial proteins and other micronutrients (Hicks et al., 2019). N removal in 58 large marine ecosystems (LMEs) and com- Apart from a dietary interest for humans, the reservoir of nu- pared it to fertilizer runoff in each off these LMEs. Moreno trients comprised by fish biomass can play a role in ecosys- and Haffa (2014) took a similar approach to estimate the tem function as a nutrient stockpile, the importance of which amount of Fe removed each year by fishing from 1950 to depends on how much nutrients are available otherwise (All- 2010. They also used biomass estimates from literature to geier et al., 2016). For example, the nutrients embedded in quantify the amount of Fe in the global fish biomass and fish biomass are not directly available for primary produc- the amount of Fe cycled by this biomass each year. How- ers, which can appear as a type of competition for resources ever, the total inventories and cycling rates have remained (Hjerne and Hansson, 2002). But at the same time, the cy- unquantified due to the lack of reliable global fish biomass cling of elements by fish acts as a source of nutrients to pri- and metabolism estimates. mary producers, as fish recycle elements through the excre- Recently Bianchi et al. (2021) used a dynamical global tion of dissolved bioavailable components. model of marine commercially targeted fish (CTF), con- The cycling of N and P by fish has often been studied in strained by observed fish catches, to estimate the total CTF freshwater systems, but little is known for the global ocean biomass and cycling rates and their distribution in the world’s (Schindler and Eby, 1997; Vanni, 2002; Vanni et al., 2006; oceans. Our study builds on Bianchi et al. (2021) by using an Griffiths, 2006). For instance, McIntyre et al. (2008) showed updated version of the model to estimate the role of CTF that fish are able to create hotspots of recycled nutrients in in the global cycling of the three most important growth- streams that could meet more than 75 % of the algal and mi- limiting nutrients: N, P, and Fe (Fig. 1). Section 2 lays out the crobial requirement for N, and Allgeier et al. (2014) showed method we use to investigate the nutrient dynamics for the that fish are a fundamental nutrient source to coral reefs. Ad- total CTF biomass for body sizes between 10 g and 100 kg, ditionally, fish egest particulate products that can be min- 100 kg hereafter CTF10 g , using model simulations for both a re- eralized and enhance productivity or that can sink to depth constructed state prior to industrial fishing (“pristine”) and and export elements (Davison et al., 2013; Saba and Stein- a simulated global peak catch. We then present and discuss berg, 2012), so that they are no longer available for pri- the results as a series of relatively self-contained sections by mary producers. As such, fish biomass can act as a bank ac- 100 kg topic. Section 3 details the total nutrient content in CTF10 g , count for nutrients, depositing nutrients when fish are feeding 100 kg rapidly and withdrawing nutrients to the water column when Sect. 4 focuses on nutrient cycling rates by CTF10 g , Sect. 5 100 kg metabolism exceeds predation. Finally, the egestion of par- discusses an extension of the CTF10 g estimates to all fish ticulate products by fish has been shown to modify the stoi- from 1 g to 1000 kg, and Sect. 6 presents the impact of fishing chiometry of biogenic particles, including dramatic changes 100 kg on CTF10 g (Fig. 1). Section 7 concludes the paper. of Fe : C, implying that egested material may also modify the relative availability of nutrients through the water column (Le Mézo and Galbraith, 2020). 2 Methods The amount of nutrients stored and cycled by fish can vary with different environmental and physiological factors, 2.1 Model description and simulations in space and time (e.g., Halvorson and Small, 2016; Prabhu We used an ensemble of simulations from the BiOeconomic et al., 2016; Francis and Côté, 2018; Czamanski et al., 2011; mArine Trophic Size-spectrum (BOATS) model (Carozza Allgeier et al., 2014). In addition to natural variations, an- et al., 2017), also used by Bianchi et al. (2021). thropogenic activities, mostly fishing, modify the storage and cycling of nutrients by the ichthyosphere. For instance, Lay- man et al. (2011) and Allgeier et al. (2016) analyzed the Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2539 been shown to limit fish harvests; Chassot et al., 2010; Stock et al., 2017), while inhabiting the same environment, which therefore makes them subject to the same basic ecological constraints. The constraints we apply in the model are the impacts of water temperature on growth, mortality, and phy- toplankton size, and net primary production. Although this biologically “coarse-grained” approach precludes resolution of species-level dynamics, it is solidly founded in bioener- getic principles and is well-suited to providing a global view of the entire ecosystem on long timescales, given that it is likely to be relatively robust under any changes in the distri- bution, abundance, or evolution of commercial stocks under historical fishing pressure (Guiet et al., 2020). Figure 2 pro- vides a schematic overview of the model structure. The BOATS model was deliberately developed to repre- sent marine organisms over the size range most heavily tar- geted by fisheries, since this is the range for which fishery data can offer useful constraints on the ecosystem function (Carozza et al., 2016, 2017). The starting point of 10 g coin- Figure 1. The ichthyosphere role in global nutrient cycling. The cides roughly with the weight of mature anchovy (approxi- diagram schematically indicates the cycling of nutrients (N, P, and mately 11 cm in length according to Pauly and Tsukayama, Fe) through the global fish biomass (solid arrow), the nutrients con- 1984), while above 100 kg the growing significance of mam- tained within the biomass, and the removal of nutrients from the mals in the ocean makes a strictly ectothermic model less ca- ocean through fishing (dashed arrow). fin and fout are the fluxes pable of capturing the full marine animal community (Hatton in and out of the global fish biomass, respectively. E indicates ex- et al., 2021). cretion (release of dissolved compounds), and F indicates egestion Fishing effort and catch are computed assuming open- (release of solid feces). access dynamics and based on the Gordon–Schaefer fishery economics model (Gordon, 1954; Schaefer, 1954) with in- creasing catchability over time due to technological progress 2.1.1 The model (Galbraith et al., 2017). The model represents fish on a two-dimensional grid, i.e., BOATS provides a global, size-based numerical simulation longitude and latitude, which is divided in regular 1◦ × 1◦ of commercially targeted marine organisms (including mol- grid cells. Thus, the model does not resolve the vertical di- luscs and crustaceans, here collectively termed “fish”) by mension but sums all ecosystem productivity and biomass coupling an ecological module with a fishery economics within the water column at each horizontal point. Given that module (Carozza et al., 2016, 2017). The ecological model the model does not resolve interactions in space between in- is based on processes derived from macro-ecological the- dividuals, this reduction in dimensionality does not – on its ory (Carozza et al., 2016), parameterized through a Bayesian own – introduce any bias. The model is forced by monthly Monte Carlo approach that optimizes model performance by climatologies of observed net primary production and sur- minimizing the discrepancy between observed and simulated face ocean temperature (Dunne et al., 2007). catch and biomass for globally distributed LMEs (Carozza Galbraith et al. (2019) hypothesize that fish growth is re- et al., 2017). Modeled fish are divided into three “super- duced under iron scarcity in the wild and demonstrate that the species” groups defined by the asymptotic mass of fish: implementation of a simple form of iron limitation of fish in small (0.3 kg), medium (8.5 kg), and large (100 kg). Indi- BOATS improves the simulated fish catch in oceanic regions vidual sizes are binned logarithmically into 30 mass classes known to have low iron concentrations. We thus used a dif- ranging from 10 g to 100 kg, with all three size groups start- ferent version of the model from Bianchi et al. (2021) that ing at 10 g (Carozza et al., 2016). The three groups are not includes an Fe limitation of fish growth, using surface nitrate intended to represent the entire marine ecosystem but rather concentrations as a proxy for iron limitation as described in the sum of all species that have been commercially harvested Galbraith et al. (2019). (and are therefore accounted for in harvest records and stock assessments, which are used to constrain the model). The un- 2.1.2 Simulations derlying philosophy of the model is that, although these very diverse species differ widely in their biological strategies, all We use the parameter sets selected in Bianchi et al. (2021) to are competing for food energy ultimately provided by the fix- run our updated Fe-limited model version. Briefly, from the ation of organic carbon through photosynthesis (which has Monte Carlo ensemble of 10 000 simulations, Bianchi et al. https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2540 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Figure 2. Schematic overview of the BOATS model. The red, green, and black arrows indicate dependencies of model components on external forcings. The top panel indicates the energetic limits of growth as a function of fish size, while the bottom panel illustrates the three size spectra of fish groups (for simplicity only the large group is represented), their internal dynamics, and the link to economics via harvest and the interactive effort. (2021) selected the parameter combinations (31 in total) that age mass ratio of nutrient to carbon, X : C. Prior work has best match the historical catch maxima across LMEs as re- suggested that body nutrient concentration of fish may be af- constructed by the Sea Around Us Project (SAUP) (Pauly fected by several factors such as body size, ontogeny, species, and Zeller, 2016), while simultaneously falling within the ac- sex, diet, temperature, or water nutrient concentration (e.g., ceptable bounds for the catch-to-biomass ratio as constrained Halvorson and Small, 2016; Prabhu et al., 2016; Allgeier by stock estimates (Ricard et al., 2012). We then perform et al., 2017), with species appearing to be the most impor- simulations with our Fe-limited version of the model using tant factor (Allgeier et al., 2020). Among these factors, our these 31 model parameter combinations. Each simulation in- model could potentially account for change during ontogeny cludes 200 years without catch to estimate pristine biomass as organisms grow in size, but analysis of the available data at equilibrium, followed by 220 years with fishing driven by (see Supplement) shows little to no systematic variation in the only increase in the catchability of biomass at 7 % yr−1 specific nutrient content with size. We thus assumed constant to reproduce the historical progression of the global fish- nutrient proportions throughout food webs. ery (Galbraith et al., 2017). The simulations slightly under- Although the model implicitly includes molluscs and crus- estimate the observed LME catch peak of ∼ 110 Mt yr−1 , taceans, they represent only a small proportion of the com- and the variation between LME peaks corresponds to ob- mercial catch between 10 g and 100 kg (from SAUP data in- servation with a squared Pearson product-moment correla- vertebrates comprised about < 14 % of total catch). Addi- tion coefficient r 2 = 0.42, p value < 10−9 , when averaged tionally, the measured nutrient content of molluscs and crus- across all ensemble members (Figs. S1 and S2 in the Supple- taceans falls within the uncertainty range around the mean ment). More details on the observational constraints and on value for fish (Tables S2–S3 in the Supplement). As a re- the Monte Carlo approach used here can be found in Bianchi sult, we did not attempt to account for invertebrates sepa- et al. (2021). From the 31 simulations of the ensemble, we 100 kg rately, but we applied the fish nutrient ratios to all CTF10 g analyze the global biomass and cycling rates at pristine state biomass. The Fe content of fish is poorly constrained (few and at the time of the global peak catch. whole body measurements), so we rather use the 95 % confi- dence interval from available data, which ranges between 10 2.2 Nutrient content of fish and 200 µmol Fe (mol C)−1 (Galbraith et al., 2019). To estimate the quantity of a nutrient X stored in fish biomass, we convert our modeled biomass estimates from wet weight to carbon (C) weight and multiply by an aver- Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2541 Based on these terms, we follow Bianchi et al. (2021) 100 kg to estimate the total flux of carbon through the CTF10 g based on the community average τ and growth rates Gi (C) of all simulated fish within the size spectrum. We calculate the ingestion for each fish within the size bins by dividing Gi (C) by τ (C) (which is 0.17 when averaged across the ensemble). We then subtract the growth Gi (C), to arrive at the carbon cycling rate, and sum over all fish (Fig. 3b). We note that at steady state for the community level there is no net growth since new production is balanced by predation, but by subtracting Gi (C) we ensure that there is no double- counting of internal cycling within the simulated size spec- trum through predation on simulated fish. Because the model mean predator–prey mass ratio for the ensemble is 0.6 × 104 , there is very little predation within the resolved size spectrum (there are only 4 orders of magnitude across the smallest to the largest size bin), so that the subtraction of Gi (C) could lead to a small systematic underestimate in our cycling es- timates. However, we also consider that predation by large predators not resolved by the model, including very large fish, marine mammals, or birds, will cause a portion of the resolved fish growth to be cycled by other organisms outside 100 kg of the CTF10 g . Any consequent underestimate is therefore likely to be significantly less than the value of τ (C) (i.e., sig- nificantly less than 17 %). 100 kg Thus, our carbon cycling rate equation for CTF10 g is Figure 3. Schematic of the flow of elements (a) within an individ- given by the output flux equation: ual and (b) through the size spectrum. I : ingestion; G: growth and reproduction; A: absorption; E: excretion; F : egestion (feces); α: X X fout (C) = (Ii (C) − Gi (C)) = (Ei (C) + Fi (C)) absorption efficiency, i.e., assimilation efficiency net of egestion; γ : i i somatic assimilation efficiency, i.e assimilation allocated to growth; X fin : flux entering the spectrum; fout : flux going out of the spectrum. = fin (C) − (Gi (C)). (1) i We can then use the input rate of carbon to the fish popu- 2.3 Nutrient cycling by fish lation, fin (C), to calculate the cycling terms for any nutri- ent element based on the stoichiometry of the ingested or- First, we define the terms we use by considering the fate ganic matter and element-specific absorption. We do so by of a nutrient element when ingested by an individual fish taking the nutrient X-to-carbon mass ratio within the prey, (Fig. 3a). A fish i ingests a mass flux of a nutrient, Ii (e.g., (X : C)prey , and the absorption efficiency of nutrient X, αX , g N d−1 ), of which a fraction α (between 0 and 1) is ab- to compute sorbed across the gut to produce an absorption flux, Ai , while the remainder is egested in the form of feces, Fi = (1 − α)Ii fin (X) = fin (C)(X : C)prey , X (Fig.3). Note that absorption efficiency here is defined as the E(X) = (αX Ii (X) − Gi (X)), difference between ingestion and egestion. Published esti- i mates of absorption efficiencies of fish are listed in Table 1. F (X) = fin (X)(1 − αX ), (2) A fraction of the element absorbed across the fish gut is then used for growth and reproduction, Gi = γ Ai = γ αIi , where fin (X) is the ingestion flux of nutrient X into all fish, where γ is the somatic assimilation efficiency. The trophic and E(X) and F (X) are the excreted and egested fluxes of efficiency, τ , is commonly defined as the ratio of the produc- nutrient X, respectively, for all fish. tion of new biomass to the ingestion rate and is equal to γ α. Note that we separate the ingested fraction, I , and egested The remainder is used for maintenance and excreted back to fraction, F , using a constant mean absorption coefficient for the water; Ei = (1−γ )αIi (Fig. 3a, b). We define the cycling each nutrient, αX (Table 1). rate of a nutrient through the fish as Ii −Gi = Ei +Fi , which As indicated by Eq. (2), the cycling rate we calculate can be thought of as the biologically processed outputs of the for an element X will be proportional to the average value fish. of (X : C)prey . To maintain tractability, we assumed that https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2542 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Table 1. Mean nutrient content in fish, zooplankton, and phytoplankton and mean absorption efficiencies (A) of N, P, and Fe for fish used in this study. % N in ww % P in ww C/N C/P Fe / C (mol C (mol N)−1 ) (mol C (mol P)−1 ) (µmol Fe (mol C)−1 ) Fish Global ± SD (range) 2.8 ± 0.4a 0.6 ± 0.2a 4.6a (3.4–6.4) 49a (29–82) 21c,e (10–200) Zooplankton Global ± SD (range or 1.4 ± 0.3b 0.14 ± 0.04b 4.7a (2.9–7.7) 140a (84–231) 30.6c (8.2/85, 4.1/248) Fe-poor/Fe-rich, low/high) Phytoplankton Global (Fe-poor/Fe-rich, From relationships in Galbraith and Martiny (2015) (Eq. 4) f 60.5 (5/92c , 2.13/258f ) low/high) C N P Fe Fish mean absorption 0.88a 0.86a,b 0.71a,b 0.24d efficiency (A) Data from a Czamanski et al. (2011) (percent converted back to wet weight using a 25 % dry weight in wet weight (Galbraith et al., 2019), geometric mean, and P absorption efficiency computed from the linear regression between predator and prey C / P ratio). b Schindler and Eby (1997). c Galbraith et al. (2019). d Thodesen et al. (2001). e Prabhu et al. (2016). f Moore et al. (2013). “ww” is for wet weight, Fe-poor and Fe-rich refer to the conditions in which the organisms lived, and low/high are the low and high estimates from gathered data in Galbraith et al. (2019). Standard deviations are the arithmetic standard deviations, associated with the arithmetic mean. Ratio mean values are geometric means, and the ranges are the 95 % confidence interval, except for fish Fe : C for which the range is estimated from Galbraith et al. (2019). zooplankton provide a representative indication of the stoi- 2.4 Primary producer demand, nutrient chiometry of the fish food source, and we used mean zoo- concentrations, export, and atmospheric deposition plankton N and P nutrient content to compute the cycling of these elements through the fish biomass (Table 1). For zoo- We compare the nutrient cycling by fish to the nutrient de- plankton Fe content, spatial variability appears to be more mand by primary producers, in order to provide a rough char- important than for N and P, given significant differences be- acterization of its magnitude. We use an averaged satellite- tween Fe-rich and Fe-poor regions (Table 1; Galbraith et al., based primary productivity (PP) (Dunne et al., 2007) to com- 2019). In order to be thorough, we computed Fe cycling in pute the PP demand for N, P, and Fe. We predict the C : P and three different ways based on the various computation of the C : N ratios in phytoplankton using empirical relationships Fe : C distribution of zooplankton: (1) we used a mean Fe : C with PO3− − 4 and NO3 surface concentrations as described in in zooplankton of 30.6 µmol Fe (mol C)−1 , (2) we used a spa- Galbraith and Martiny (2015): tial variation between Fe-rich and Fe-poor regions and used the Fe : C mean values in Fe-poor and Fe-rich regions, re- [NO− 3 ]surf (N : C)phyto = 12.5 % + 3 % , spectively, from Table 1, and (3) we used the same spatial 0.32 + [NO− 3 ]surf variation, but with the low and high Fe : C estimates of zoo- (P : C)phyto = 0.6 % + 0.69 %[PO3− (4) 4 ]surf , plankton from Table 1. For the spatial variations in Fe : C, we assumed that Fe-poor conditions are encountered in HNLC where nutrient concentrations are in µmol L−1 . regions, which are determined by a concentration of surface Similarly to zooplankton, the Fe : C of phytoplankton is nitrate [NO− −3 3 ] larger than 5 mmol N m . In order to take computed by allowing variation in stoichiometric ratios be- into account the gradient between these regions, we locally tween the mean value found in Fe-poor conditions and the weighted zooplankton Fe content using a Michaelis–Menten mean value found in Fe-rich conditions, or between the high function. and low phytoplankton Fe : C estimates (Table 1), using a Fe-rich or high  Fe-poor or low Michaelis–Menten equation analogous to Eq. (3). (Fe : C)zoo = (Fe : C)zoo + (Fe : C)zoo We then use the phytoplankton nutrient ratios to compute [NO− the export of nutrients from a satellite-based estimate of total 3 ]surf  Fe-rich or high −(Fe : C)zoo (3) C export (Dunne et al., 2007). 5 + [NO− 3 ]surf We use the World Ocean Atlas-observed PO3− − 4 and NO3 water concentrations (Garcia et al., 2013) and dissolved Fe concentrations simulated by the biogeochemical model Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2543 TOPAZ2 (Dunne et al., 2013) to compare the fish biomass dust greatly overwhelms the iron in fish (Mahowald et al., nutrient content to the surface ocean ambient concentrations 2009; Myriokefalitakis et al., 2016). of nutrients and to compute the stoichiometric ratios (e.g., in 100 kg In summary, the storage of nutrients by CTF10 g biomass Eq. 4). The TOPAZ2 model represents the cycles of differ- can be quite significant, compared to the dissolved nutrient ent elements from carbon to calcite and Fe with 30 different inventory of the euphotic zone, but only where ambient dis- tracers and the dynamics of three groups of phytoplankton. solved nutrient concentrations are low and/or fish biomass is Surface concentrations are computed using the 2002–2019 100 kg high. In these areas, CTF10 g biomass could be expected to annual mean euphotic depth from Aqua MODIS (2018). have a greater potential as a source (if stored nutrients are Finally, we compare the rate of nutrients removal by fish- made available) or a sink (if the nutrients cannot be used by ing at the time of global peak catch to current atmospheric primary producers) of nutrients. deposition fields of soluble N (Brahney et al., 2015) and Fe (Mahowald et al., 2009) (Fig. S3). The catch rate and its 3.1 Comparison to previous estimates spatial distribution are simulated by the coupled economic– ecological model and systematically differs from the actual Our estimates for the N and Fe contents of the ichthyosphere historical peak in that the model ensemble simulates higher differ from previous studies to some degree, which can be ex- catch rates in the open ocean than observed. plained by differences in the estimates of global fish biomass and/or uncertainty regarding fish nutrient contents (we were not able to find prior estimates for P). The amount of N stored within the global fish biomass has 3 Fish biomass: a living pool of nutrients been previously estimated to be about 23 Tg (Allgeier et al., 2017)1 , which is about 66 % less than our global pristine es- Our results show that the nutrients contained within timate of 68.7 ± 30.5 Tg N (Table 2). Their computation is 100 kg CTF10 g biomass represent a non-negligible proportion of based on an estimation of fish biomass of 0.9 Gt (Jennings 100 kg the ambient dissolved concentrations in areas where these et al., 2008) while our ensemble pristine CTF10 g biomass is 100 kg concentrations in seawater are low and/or where CTF10 g 2.5 ± 0.8 Gt (a smaller biomass estimate than that of Bianchi biomass is high. The highest amounts of N, P, and Fe in the et al. (2021) due to Fe limitation of fish), a difference of pristine fish biomass are located in the most productive re- biomass of 64 %. Additionally, we used a mean N content gions along the coasts, where most simulated fish biomass of 2.8 %, slightly higher than their value of 2.6 %, because occurs (Fig. 4a). Globally, the estimated pristine biomass of we used measurements made only on wild fish and did not 100 kg CTF10 g , which represents 2.5±0.8 Gt of wet biomass, con- try to account for all the catch diversity in organisms. To re- tains 69 ± 31 Tg of N, 15 ± 14 Tg of P, and 0.012–0.23 Tg of flect this uncertainty, we indicate the species-related uncer- Fe, of which about half is found in the large marine ecosys- tainty around the mean value we use, ±0.4 %, which covers tems (LMEs) (Table 2). Note that, in our computations, fish the value used in Maranger et al. (2008), for our calculations biomass is the only term that varies spatially since the nutri- in Tables 2 and 4. ent contents of fish, (X : C), are held globally constant (see For Fe, Moreno and Haffa (2014) estimated that the Methods section). global fish biomass stored between 0.07 and 0.7 Tg, which 100 kg The N content of CTF10 g biomass is locally on the same is roughly 3-fold higher than our range of 0.012–0.23 Tg of order of magnitude as ambient surface NO− Fe (Table 1). To compute these values, Moreno and Haffa 3 concentrations, exceeding 50 % of [NO− (2014) used an estimated fish biomass of 0.9–2 Gt, which 3 ] in the oligotrophic gyres where nutrient concentrations are low and in coastal shelves where is lower than our modeled estimate of 2.5 Gt for commer- large fish biomass accumulates prior to industrial fishing cially targeted fish only due to a conservative maximum es- 100 kg (Fig. 4b, Table 2). The amount of P in CTF10 g biomass timation (Wilson et al., 2009). However, they used a range of Fe : C values, 0.073–0.324 g Fe kg−1 of wet weight (ww) represents a high proportion of available PO3−4 in the North for ray-finned fish, that is about 3–12 times larger than Atlantic Ocean, which is relatively P-poor, exceeding 30 % our 10–200 µmol Fe (mol C)−1 range, equivalent to 0.006– in some areas (Fig. 4b). The ratio also exceeds 20 % in the 0.12 g Fe kg−1 ww (assuming 12.5 % C in ww). We are more western North Pacific and in a few locations such as in the confident in our compilation of Fe : C values, which is up- 100 kg Arabian Sea. CTF10 g biomass stores much higher Fe com- dated, more complete, and only based on peer-reviewed stud- pared to dissolved surface concentrations in the highly pro- ies (Galbraith et al., 2019). Nonetheless, the differences high- ductive subtropical front waters of the North Pacific, North lighted here, and the large range of estimates (low–high val- Atlantic, and Southern oceans, with relative values exceed- ing 50 % for the high-end estimate (Fig. 4d). Despite low sur- 1 It seems that there is a typo in Allgeier et al. (2017). Their face Fe concentrations in the Southern Ocean, the proportion estimate is based on the Jennings et al. (2008) wet biomass estimate 100 kg in CTF10 g is particularly low due to the low modeled fish of 9 × 108 t of fish and they used 2.6 % of N in fish; thus the total N biomass, while in the tropical Atlantic the input of Fe from harvest should be 23.4 Tg and not 233.4 Tg https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2544 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Figure 4. Modeled commercial pristine fish biomass mean nutrient content and relative to surface nutrient concentrations. (a) P content 100 kg (mmol P m−2 ), N content (mmol N m−2 ), and Fe content (low and high estimates, µmol Fe m−2 ) of the global pristine CTF10 g biomass, 3− and (b) N content relative to surface NO−3 concentrations (%), (c) P content relative to surface PO4 concentrations (%), and (d) Fe content (low and high estimates) relative to modeled surface dissolved Fe (%) from the TOPAZ model (Dunne et al., 2013). All surface concentrations are integrated over the 2002–2019 annual mean euphotic depth (Aqua MODIS, 2018). The N, P, and Fe content of fish is represented on a single map given that we used a globally constant nutrient ratio for each element, so that all spatial variation is caused by fish biomass. Table 2. Table of values in LMEs from the model ensemble simulations in the pristine state and at the global peak catch. This table contains 100 kg 100 kg integrated values of nutrient content in CTF10 g biomass (Tg) and the ratio of nutrient content in CTF10 g biomass with surface nutrient concentrations (%). N P Fe (low estimate) Fe (high estimate) Content (Tg) Global 68.7 ± 30.5 14.9 ± 13.6 (1.2 ± 0.4) × 10−2 0.23 ± 0.07 Pristine LME 37.4 ± 17.2 8.1 ± 7.5 (6.3 ± 2.1) × 10−3 0.13 ± 0.04 At global peak Global 26.2 ± 14.7 5.7 ± 5.8 (4.4 ± 1.9) × 10−3 (8.8 ± 3.8) × 10−2 catch LME 9.8 ± 5.4 2.2 ± 2.2 (1.6 ± 0.7) × 10−3 (3.3 ± 1.4) × 10−2 Content/surface concentration (%) Global 21.7 ± 9.6 7.1 ± 6.5 0.50 ± 0.16 9.9 ± 3.1 Pristine LME 73.6 ± 33.8 24.3 ± 22.5 1.4 ± 0.5 27.6 ± 8.7 At global peak Global 7.9 ± 4.4 2.3 ± 2.4 0.17 ± 0.07 3.4 ± 1.5 catch LME 19.1 ± 10.5 5.4 ± 5.4 0.29 ± 0.13 5.7 ± 2.4 Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2545 ues), highlight the large uncertainty on the Fe content of 4 Nutrient cycling by the commercial fish biomass whole fish and the great need for more measurements in this domain. Because they are capable of large-scale movement and al- Note that our modeled estimates are likely to be more re- ter the stoichiometry of particles, fish can have impacts on liable in LMEs since fish biomass in these regions is better nutrient cycling that differ from single-celled heterotrophic constrained by fish catch data. plankton. In this section, we gauge the potential scale of these impacts by estimating the rates at which fish cycle nutrients, 3.2 Nutrient content variations in fish what fraction of primary productivity (PP) this cycling rep- resents, and how much it can contribute to the export of nu- Many factors contribute to variations in fish body nutrient trients from the euphotic zone as sinking egested materials. concentrations, which we cannot model explicitly but in- Note that this is intended only to illustrate the potential mag- stead capture within our uncertainty estimates. Among the nitude, which could be built upon with coupled fish biogeo- different influencing factors, fish species is important; for in- chemistry modeling. stance bony fish contain larger quantities of P compared to The global cycling, i.e., excretion plus egestion, of other fishes (El-Sabaawi et al., 2016). Species can also vary 100 kg nutrients by the pristine CTF10 g biomass represents in terms of the size of storage components (e.g., Czamanski et al., 2011) and the number and size of bones in vertebrates, about 210 ± 113 Tg N yr , 15.6 ± 8.0 Tg P yr−1 , and 0.12– −1 which generally increases with adult size (Sterner and Elser, 0.77 Tg Fe yr−1 , of which about half is in the LMEs (Ta- 2002). Variations in body nutrient contents can also be ex- ble 3). Like nutrient storage, modeled cycling by fish is larger plained by sex and life stage, e.g., decreased Fe body con- where the biomass is higher (Figs. 5 and S5). The three dif- tent occurs in female rainbow trout during sexual maturation ferent ways of computing Fe cycling by the commercial fish (Shearer, 1984), though these would tend to average out at biomass show similar spatial patterns, but Fe cycling is larger the population level. Studies have also shown that ontogeny when using the weighted spatial variation between the low may affect nutrient content; for example juveniles have less and high Fe : C estimates in zooplankton (Fig. 5). The spa- P than adults (Pilati and Vanni, 2007) while Fe content varies tially weighted computations suggest the possibility that Fe 100 kg throughout the life cycle of salmon (Shearer et al., 1994). cycling by CTF10 g might be reduced in HNLC regions Because our model uses size to differentiate between fish, (principally in the Southern Ocean and the subarctic Pacific we analyzed aquatic animal body nutrient and body size data Ocean). from Vanni et al. (2017) for any existing relationship between size and nutrient content. We found no relationship between 4.1 Nutrient cycling by commercial fish relative to body N content and size and only a weak but significant re- primary production lationship between body P and size (Fig. S4). In this data 100 kg set, the changes of P content with size seem to be more re- The modeled N cycling by pristine CTF10 g biomass con- lated to the difference between vertebrates and invertebrates tributes on average to 2.2±1.2 % of the N demand of primary and to be significant for benthic organisms more than pelagic producers (PP) in LMEs (Table 3) and generally accounts for organisms. A recent study has indeed shown that the taxo- less than 5 % of the demand, except in some coastal areas nomic identity is prominent in driving nutrient content vari- where it can be as high as 14 % (Fig. 6a). Similarly, the mod- 100 kg ations compared to size (Allgeier et al., 2020). In addition, eled pristine P cycling by CTF10 g represents 1.2 % of the Hjerne and Hansson (2002) found no significant changes in total P demand in the LMEs (Table 3), less than 4 % of the the N and P content of fish with species (sprat and herring), global P demand with a larger contribution in the north and fish size, seasons, or different areas of the Baltic Sea, as did equatorial Atlantic coastal regions, and larger than 6 % con- Griffiths (2006) for the P content of different fish species in tributions in some coastal areas (Fig. 6b). The high-end es- lakes. The scant available data for Fe do not allow us to draw 100 kg timate of the Fe cycling by CTF10 g relative to PP demand conclusions on the variations in Fe content with size. for Fe is slightly more significant than the ratios for N and P, Although nutrient ratios show many fascinating variations as it represents up to 13 % of the PP demand for Fe in some between species, these are small relative to variations in fish coastal areas and less than 10 % everywhere else (Fig. 6e). biomass density in the ocean. Thus, most of the spatial vari- Our global estimates are broadly consistent with the or- ations in the nutrient content of fish are likely to be due to der of magnitude influence of fish estimated by prior lo- variations in fish abundance rather than species assemblages, cal studies. For all the coral reefs in the ocean, Allgeier and the uncertainty related to nutrient ratios is likely to be et al. (2014) estimated that the total fish community sup- small compared to the uncertainty on fish biomass. plies about 1.2 Tg N yr−1 , which is about 0.6 % of our global pristine estimate (0.8 % compared to global cycling at peak catch) or 1.2 % of the LME pristine estimate (2.1 % of our 100 kg N cycling by CTF10 g biomass at global peak catch) (Ta- ble 3)). This is consistent with the fact that coral reefs cover https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2546 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Table 3. Table of values from the model ensemble simulations in the pristine state and at the global peak catch at both the global scale and in 100 kg LMEs. This table contains integrated values of the amount of nutrients cycled by the CTF10 g biomass (Tg yr−1 ), the ratio of this cycling 100 kg with the global primary producers’ demand for these nutrients (%), the amount of nutrients egested by the CTF10 g biomass (Tg yr−1 ), and its ratio with the exported nutrient quantities (%). N P Fe (low estimate) Fe (high estimate) Total cycling (Tg yr−1 ) Global 210 ± 113 15.6 ± 8.0 0.12 ± 0.25 0.77 ± 0.19 Pristine LME 101 ± 55 7.5 ± 3.9 0.06 ± 0.12 0.36 ± 0.09 At global peak Global 145 ± 88 10.8 ± 6.3 0.08 ± 0.18 0.56 ± 0.17 catch LME 56 ± 35 4.2 ± 2.5 0.03 ± 0.07 0.21 ± 0.07 Cycling/PP demand (%) Global 1.5 ± 0.83 0.91 ± 0.47 0.16 ± 0.33 2.7 ± 0.66 Pristine LME 2.2 ± 1.2 1.2 ± 0.6 0.26 ± 0.53 4.0 ± 1.0 At global peak Global 1.2 ± 0.75 0.76 ± 0.44 0.12 ± 0.26 2.2 ± 0.68 catch LME 1.5 ± 0.9 0.85 ± 0.5 0.15 ± 0.33 2.6 ± 0.8 Egestion (Tg yr−1 ) Global 29.4 ± 15.9 4.5 ± 2.3 (9.1 ± 18.9) × 10−2 0.59 ± 0.14 Pristine LME 14.1 ± 12.3 2.2 ± 1.1 (4.4 ± 9.1) × 10−2 (2.7 ± 0.7) × 10−1 At global peak Global 20.4 ± 12.3 3.1 ± 1.8 (6.3 ± 13.5) × 10−2 0.42 ± 0.13 catch LME 7.8 ± 4.9 1.2 ± 0.7 (2.4 ± 5.2) × 10−2 (1.6 ± 0.5) × 10−1 Egestion/export (%) Global 2.3 ± 1.2 3.0 ± 1.5 1.1 ± 2.3 21.7 ± 5.3 Pristine LME 2.1 ± 1.1 2.6 ± 1.3 1.1 ± 2.3 20.1 ± 5.0 At global peak Global 2.1 ± 1.2 2.7 ± 1.6 1.0 ± 2.1 19.5 ± 6.0 catch LME 1.6 ± 1.0 2.1 ± 1.2 0.8 ± 1.7 15.7 ± 5.0 about 255 000–600 000 km2 (Spalding and Grenfell, 1997), nutrient concentrations are combined and when nutrient in- which is about 0.4 %–0.8 % of the LME area. Hernández- puts from anthropogenic sources are low. Our results indeed León et al. (2008) estimated that zooplankton supply about suggest an increased contribution of fish cycling when these 1780 Tg N yr−1 worldwide, representing 12 %–23 % of the conditions are combined. However, the strength of this con- requirements of phytoplankton and bacteria. This estimated tribution also depends on the timing between the release of zooplankton cycling is about 11 times our modeled recy- nutrients by fish and the primary producers’ demand for these 100 kg nutrients. cling rate by the pristine CTF10 g biomass and 6–12 times 100 kg 100 kg For Fe, CTF10 g cycling represents a more important what CTF10 g cycling could provide for primary producers. fraction of the PP demand compared to N and P, but large Given that the biomass distribution of all marine organisms uncertainties remain in its computation. At the global scale, has a slope of 0 (Hatton et al., 2021) (Fig. S7), and assum- Moreno and Haffa (2014) estimated that the amount of Fe ing that zooplankton mass ranges over 12 orders of magni- excreted by the commercial marine fish biomass ranged be- tude, zooplankton biomass would be expected to be roughly tween 0.4–1.5 Tg Fe yr−1 . Our modeled estimate range of 3 times the biomass of the total fish biomass in the 10 g to 0.12–0.77 Tg Fe yr−1 , based on the high and low estimates 100 kg size range. In addition, due to their higher metabolic (Figs. 6e and S6), is lower but overlapping. The difference and ingestion rates (Maldonado et al., 2016; Griffiths, 2006), can in part be attributed to the lower Fe : C values we used zooplankton cycling rates are likely to be much higher than for zooplankton (Table 1) and highlights the uncertainty on fish cycling rates, also exemplified by our modeled cycling the Fe cycling computation (Fig. 5). In the Southern Ocean, size spectrum which has a negative slope (Fig. S8). McIn- whales have been shown to contribute to a maximum of tyre et al. (2008) showed that fish excretion can be important 0.2 %–0.3 % of the phytoplankton demand for Fe in a pre- in supplying N and P to primary producers when conditions whaling ecosystem and no more than 0.03 %–0.04 % in a of high fish biomass and high PP demand or low ambient Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2547 100 kg berg, 2012). Figure 6c–d, f quantify how much CTF10 g egestion may contribute to the export of N, P, and Fe to depth using the mean absorption efficiencies in Table 1 and assuming exported nutrient ratios, without fish, are on av- erage equal to those of phytoplankton. For all the nutrients, 100 kg CTF10 g -mediated export accounts for a larger part of the export in the warm, low-export regions of the world oceans, i.e., the tropical gyres, where it can contribute up to 50 % of the exported Fe for the high-end estimate (Fig. 6f), 6 % of the exported N, and 10 % of the exported P (Fig. 6c, d). 100 kg Globally, modeled pristine CTF10 g biomass egests 29.4 ± 15.9 Tg of N, 4.5±2.3 Tg of P, and 0.009–0.59 Tg of Fe each year, which on average roughly accounts for 2.3 ± 1.2 %, 3.0 ± 1.5 %, and 1.1 %–22 % of the export of N, P, and Fe, respectively, out of the euphotic zone (Table 3). These results are in agreement with Davison et al. (2013), who showed that the contribution of mesopelagic fish to the carbon export (via respiration, excretion, egestion, and death) is higher in regions where the total export is small. However, Davison et al. (2013) also showed that locally, in the Cali- fornia Current, the active transport of C by mesopelagic fish alone, which we do not model, represents about 15 %–17 % 100 kg Figure 5. Iron cycling by CTF10 g biomass computed using (a) a of the total carbon export at depth. In their modeling study, Aumont et al. (2018) estimated that, globally, diurnal verti- mean Fe : C in zooplankton, (b) a spatial variation between zoo- plankton Fe : C mean values in Fe-poor and Fe-rich areas and based cal migration of epipelagic organisms (all migrating fish and on [NO3 ] concentrations, and (c) the same spatial variation but us- zooplankton) contributes to the flux of carbon to depth of ing the low and high estimates of zooplankton Fe : C from data. (For about 18 % of the passive flux. Fish egesting and respiring at more details see Methods.) depth transport significant amounts of carbon and thus also transfer nutrients from the surface to deeper layers, a pro- cess that would have increased the contribution of fish to the post-whaling ecosystem, making their contribution negligi- export of nutrient if represented in the model. ble compared to that of zooplankton (> 70 % for micro- More than the fish contribution to total export, their ef- zooplankton only) (Maldonado et al., 2016). With a mod- fect on particles may be most relevant for stoichiometric eled contribution of about 0.05 %–0.5 % of the phytoplank- changes, especially for Fe. Indeed, since the absorption effi- ton demand for Fe in the Southern Ocean, modeled pristine ciency of Fe is smaller than that of C, the Fe : C in feces will 100 kg CTF10 g cycling coherently might be able to sustain a larger be greater than in the ingested particles, and exported fecal part of primary productivity than the current whale popula- material will have a greater Fe : C than biogenic sinking par- tion but still far less than zooplankton as discussed before for ticles made of phytoplankton aggregates or dead organisms N cycling. (Le Mézo and Galbraith, 2020). This is potentially impor- Finally, fish movements also allow the transport of nutri- tant for mesopelagic organisms feeding on sinking material ents laterally in the ocean, constituting a sink of nutrients in light of the possible Fe limitation of marine animals (Le where the fish forage and a source of nutrients where the fish Mézo and Galbraith, 2020; Galbraith et al., 2019). excrete, egest, or die (Vanni et al., 2013; Francis and Côté, 2018), an effect we do not explicitly include here. 5 Extended size spectrum and total fish biomass 4.2 Nutrient export by feces 100 kg Until now we have considered CTF10 g as represented Our results show that fish fecal material has the potential to explicitly by the BOATS model, which ignores non- affect the distribution of nutrients within the water column, commercially targeted fish as well as the small and large especially in regions of low export intensity. Egested nutri- ends of the fish size range. We provide a rough estimate of ents are integrated into fecal pellets that sink out of the sur- how the total fish biomass within a more inclusive marine 1000 kg face layer and are recycled at greater depths than if bound size spectrum spanning 1g larvae to 106 g sharks (F1 g ) to smaller particles (Wotton and Malmqvist, 2001; Turner, would compare to our estimates in two steps. First, we take 100 kg 2015), especially fish fecal pellets that can sink faster and the Bianchi et al. (2021) estimate that the CTF10 g is sup- deeper than marine snow and phytodetritus (Saba and Stein- ported by roughly half of the total primary production that https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2548 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Figure 6. Total nutrient cycling and egestion relative to export. Ratio (%) between the amount of nutrients cycled by the modeled pristine 100 kg global CTF10 g biomass and estimated primary producers’ demand for (a) N and (b) P, ratios of the modeled amount of nutrients egested by 100 kg the pristine CTF10 g biomass and export at the base of the euphotic zone of (c) N and (d) P, and high-end estimate of the ratio (%) between 100 kg the modeled amount of (e) Fe cycled by the modeled pristine global CTF10 g biomass and estimated primary producers’ Fe demand, and 100 kg (f) Fe egested by the pristine CTF10 g biomass and Fe export at the base of the euphotic zone. The high-end estimates are obtained using Fe cycling computed from the weighted spatial variation between the low and high Fe : C values of zooplankton and the weighted spatial variation between the averaged Fe : C ratios of phytoplankton in Fe-poor and Fe-rich conditions. could be available to this size range (best estimate 58 %), Johnston, 1999), we expect higher cycling of nutrients by from which we calculate that if the non-commercial frac- the extended size spectrum than would be implied by the tion were included the total biomass estimate would be about larger biomass alone. In the pristine ocean modeled here 5.2 Gt. We then consider that the expanded size range in- by BOATS, the C cycling rates decrease with size follow- cludes 6 orders of magnitude compared to the 4 orders of ing a slope of −0.37 mol C m−2 g−1 (Fig. S8). Based on this magnitude of the standard BOATS size range. In BOATS, slope, and the biomass slope of 0, we would expect that the the size spectrum of abundance has a slope of about −1, giv- 1000 kg CTF1 g would have cycled about 2.4 times more C in the ing the biomass size spectrum a slope of 0 (Fig. S7) con- 100 kg pristine ocean compared to the CTF10 g , with the 1–10 g sistent with the observed Sheldon spectrum (Hatton et al., size range cycling 40 % more C than the rest of the spectrum 2021). Thus, by extension, we estimate that the biomass con- 1000 kg tains about 1.5 times (6 orders of magnitude versus 4 or- combined. Taken all together, the total cycling by F1 g 100 kg ders of magnitude) more biomass, giving a total of 7.8 Gt would have been a factor of 4.8 greater than the CTF10 g 1000 kg of wet biomass. Overall, we estimate that F1 g exceeds rates discussed throughout the paper above. Assuming the 100 kg additional organisms eat prey with similar body nutrient con- CTF10 g by a factor of 3.1. tents as the organisms already modeled, all of the nutrient The extended size spectrum contains more biomass, but it cycling rates presented above would be increased proportion- also contains more small fish than the standard size range ally. of BOATS, a combination that would change the cycling rates of the total fish biomass. Since smaller fish are shown to have higher cycling rates than large ones (Clarke and Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2549 6 Fish catch: anthropogenic extraction of nutrients equatorial Pacific, along the western margins of Africa and from the ocean South America, and in the Arabian Sea, where catch is high and deposition is low (Fig. S3). However, most of N supply to As fishing activity represents a direct removal of nutri- the surface ocean occurs through vertical diffusion and mix- ents from the ocean, we estimated how much nutrients ing of the upper layers (Sarmiento and Gruber, 2006), which were extracted at the global peak catch and how these likely accentuates the fact that N extraction by fishing is in- extraction rates would compare to nutrient inputs to the significant at the global scale. ocean. Globally, we estimate that modeled fishing activity removes 5.4 ± 0.7 Tg N yr−1 , 1.2 ± 0.3 Tg P yr−1 , and 0.09– 6.2 Phosphorus 1.8 × 1010 g Fe yr−1 from the ocean at the time of global peak catch, of which a little less than 50 % is in LMEs (Fig. 7, Similarly to N, our estimate of P extraction by fishing is co- Table 4). Although our model is calibrated to agree reason- herent with previous work and shows that it is very small ably well with observed catch maxima in LMEs over the compared to inputs of P to the ocean and resupply from ver- years 1950 to 2010 (Pauly and Zeller, 2016; Bianchi et al., tical mixing and diffusion in the water column. We estimate 2021), the total catch varies between ensemble members and that the amount of P removed by fishing at the global scale tends to be overestimated in the open ocean, even with the amounts to 1.2 ± 0.3 Tg P yr−1 , of which 0.6 ± 0.2 Tg P yr−1 Fe limitation we added in this study, for reasons that remain occurs in the LMEs (Table 4). Huang et al. (2020) estimated unclear. In addition, the catch estimate we present is at the that wild and aquaculture fisheries, including finfish, crus- time of the global peak catch in idealized simulations, rather taceans, and molluscs, represented 1.1 Tg of P in 2016, which than being historically accurate. Consequently, the simulated is superficially similar to our estimate. However, our esti- average catch at global peak (196.2 ± 57 Tg wet weight) is mated catch of 196.2 ± 57 Tg of wet weight is larger than higher than the peak catch estimated from fishery observa- the global amount of catch they used of 169 Tg, even though tions (130 ± 65 Tg), so our estimates may exceed actual wild they considered aquaculture in addition to wild captures. Our capture extractions by 50 %, with the most significant overes- catch estimate at global peak catch overestimates the high timates in the open ocean (see Bianchi et al., 2021, for more seas catch as discussed at the beginning of this section and details). in Bianchi et al. (2021). Additionally, our estimate is solely based on fish P content (Table 1), which may slightly over- 6.1 Nitrogen estimate the amount of P extracted by fishing activity since crustaceans and molluscs have lower P content than finfish 100 kg Our estimate of N extraction from the CTF10 g catch is (Huang et al., 2020). consistent with previous computations, and the differences Contrary to N, the modeled removal of P from harvest mostly reflect the fish biomass estimates. For example, us- would largely exceed the atmospheric deposition of soluble P ing catch data, Maranger et al. (2008) estimated the amount as P inputs to the ocean mostly occur through riverine inputs of N returned to land via fishing in LMEs to be about (Table 4). Consequently, catch transfers P from the ocean to 0.9 Tg N yr−1 in the 1960s and 2.3 Tg N yr−1 in 2000. Our land whereas P supply to the ocean is mostly occurring in estimate of the N content of fish catch at global peak catch the coastal areas, with possible impacts on the P budget of in LMEs is 2.8 ± 0.4 Tg N yr−1 , which is slightly larger than the open ocean (Huang et al., 2020). But similarly to N, ver- Maranger et al. (2008) estimate in 2000. Allgeier et al. (2017) tical diffusion and mixing of the upper layers supply P to the also estimated the amount of N globally harvested to be about surface ocean in quantities that most likely render P extrac- 2.072 Tg N yr−1 using the Maranger et al. (2008) 2.6 % N tion by fishing relatively insignificant. content and the FAO catch data. The FAO data only contain reported catches and consequently is lower than the SAUP 6.3 Iron data used in Maranger et al. (2008) and used to calibrate BOATS, which explains part of the difference. Fe extraction by fishing activity is within the range of pre- Our study framework allows the spatial comparison of ex- vious estimates, but large uncertainties remain due to uncer- tracted nutrients to nutrient inputs, which was not the case tainty regarding the Fe : C of fish. Moreno and Haffa (2014) in previous work. For N, even though N extraction by fish- investigated the extent to which commercial catch has glob- ing can be significant locally compared to N deposition at ally translocated Fe from the ocean to land. They estimated the surface, N extraction is negligible compared to the other the global rate of translocation of Fe to be between 0.007 sources of N to the surface layers. Figure 8a compares the and 0.03 Tg in 2010. Our modeled global range of Fe re- amount of N extracted by fishing to the modeled soluble at- moval is about 0.0009–0.018 Tg Fe yr−1 (Fig. 7, Table 4). mospheric N deposition from Brahney et al. (2015). Glob- Our lower estimated values can once again be explained by ally, fishing removal of N is smaller than current modeled the difference in the Fe : C ratios used for fish. Even though atmospheric deposition of soluble N, with the higher values, our estimate is lower, it shows that locally Fe extraction can up to more than 60 % of the N deposition, in the southern be significant compared to Fe inputs from dust deposition. https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2550 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish Figure 7. Distribution of the modeled amount of N (g N m−2 yr−1 ), P (g P m−2 yr−1 ), and Fe ( µg Fe m−2 yr−1 ) extracted from the ocean at the time of the global peak catch. The two color bars for Fe represent the low and high estimates based on the 95 % confidence interval for Fe : C values in fish (10–200 µmol Fe (mol C)−1 ). Table 4. Table of values in LMEs from the model ensemble simulations in the pristine state and at the global peak catch. This table contains integrated values of the amount of nutrients removed by fishing in LMEs (Tg yr−1 ) and the ranges of values of the global nutrient inputs to the ocean from the literature (Tg yr−1 ). N P Fe (low estimate) Fe (high estimate) Catch (Tg yr−1 ) Global 5.4 ± 0.7 1.2 ± 0.3 9.1 × 10−4 1.8 × 10−2 LMEs 2.8 ± 0.4 0.6 ± 0.2 (4.7) × 10−4 (9.3) × 10−3 N P Fe Global inputs to the ocean (Tg yr−1 ) Soluble deposition 16–63a,c,e,h,j 0.1–0.5a,b,c,d,i,j 0.6–13.4b,c,f,g,k Rivers 80h 0.93–48i 0.08–0.09k N2 fixation 140h – – Iceberg melting – – 0.09–0.1k a Brahney et al. (2015). b Mahowald et al. (2009). c Okin et al. (2011). d Myriokefalitakis et al. (2016). e Fowler et al. (2013). f Ito (2015). g Wang et al. (2015). h Gruber and Galloway (2008). i Benitez-Nelson (2000). j Kanakidou et al. (2012). k Moreno and Haffa (2014). Although the high-end estimate of Fe extracted is globally spheric deposition, which represents 48.5 Tg P yr−1 (Tables 4 small relative to modeled soluble Fe deposition from Ma- and S1), then the extracted P flux represents 2.5 % of the howald et al. (2009), it reaches values larger than 100 % global input flux (1.2 % over the LMEs). For N, global catch in the coastal eastern equatorial Pacific and in some other represents about 2 % of the combined N inputs from atmo- coastal areas such as western South Africa, northern Europe, spheric deposition (49.6 Tg N yr−1 ), rivers (80 Tg N yr−1 ), and Canada (Fig. 8b), where modeled Fe deposition is small and N2 fixation (140 Tg N yr−1 ) (Tables 4 and S1). and harvest is high (Fig. S3). Contrary to N and P, Fe has a Note that fish extracted from a given area may have for- much shorter residence time and thus is subject to local per- aged elsewhere, especially large fish able to undertake long- turbations, among which Fe extraction by fishing could be distance migrations like tuna, salmon, or sharks (e.g., Afonso important. et al., 2017; Gresh et al., 2000). Consequently, the ratios between extracted nutrients and nutrient deposition may be 6.4 Local and time-dependent nutrient budgets over- or under-estimated, thus over- or under-estimating the role of fishing as a local sink of nutrients (Vanni et al., 2013). Nutrient budgets are subject to perturbations in space and in In addition, the relative timing of fishing effort along with time that can modify the relative strength of the nutrient ex- phytoplankton growth, nutrient input seasonality, and resi- traction by fishing activity. Some local nutrient budgets have dence times may also modify the importance of fishing activ- been investigated to compare the amount of nutrients ex- ity as a sink of nutrients (e.g., Francis and Côté, 2018; Vanni tracted by fishing to the nutrient loads (e.g., Hjerne and Hans- et al., 2006). son, 2002). If we were to do similar budgets, at the global scale, assuming all P inputs come from rivers and atmo- Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2551 high fish biomass, and low export production. The industrial fish catch generally represents a small extraction of nutrients globally compared to external inputs, though it removes a significant amount of P from the open ocean compared to ex- ternal inputs (mainly riverine). In general, local cycling of N and P by fish is less significant than Fe cycling by fish because N and P are resupplied globally through large-scale circulation processes, while Fe cycling is more local and sus- ceptible to perturbations through rapid scavenging for exam- ple. In addition, poor absorption of Fe by fish leads to an enrichment of the Fe content of fecal matter. 100 kg Globally, nutrient cycling by the modeled CTF10 g biomass is small compared to primary producers’ demand for these nutrients. The highest contributions are found close to the coasts where fish biomass and productivity demand are high. Fish egestion of nutrients via faecal pellets is likely to comprise the largest fraction of the sinking flux in regions of low export production, i.e., the subtropical gyres. Fecal Figure 8. Ratio (%) between simulated extracted nutrients and cur- pellets may also significantly impact the stoichiometry of rent aeolian soluble nutrient inputs at the surface of the ocean for (a) N and (b) Fe. “low” and “high” refer to the use of 10 or sinking particles, especially for Fe, with consequences for 200 µmol Fe (mol C)−1 in fish, respectively. mesopelagic organisms. Our study provides a first global glimpse of nutrient cy- cling by the ichthyosphere. Although the contribution of 100 kg 6.5 Reductions in nutrient cycling caused by fishing CTF10 g simulated by our model tends to be on the or- der of only a few percent of total surface nutrient budgets Fishing has had a dramatic influence on nutrient cycling by and fluxes, we estimate that these would be a factor of 3 or 100 kg CTF10 g , as it has permanently removed a large amount of more larger for the 1 g to 1000 kg range, including all fish. biomass, especially in the large size classes. In our ensemble Furthermore, the role of fish in shaping the ecosystem pro- of simulations, the cycling rates decrease by about 30 % for cesses through top-down pressure on their prey may be of the three elements considered at the time of the global peak similar or greater magnitude than the quantities estimated catch (Table 3), due to the global reduction in fish biomass here (Frank et al., 2005; Baum and Worm, 2009; Hessen of about 60 %. Since large fish are heavily targeted by fish- and Kaartvedt, 2014; Kavanagh and Galbraith, 2018). Fish eries, the size spectrum of CTF appears truncated at a larger can also be highly relevant at the local scale, for example size class at the time of the global peak catch compared to a by changing the vertical distribution of Fe in the water col- pristine state (Fig. S7). This reduction of the mean commu- umn as mentioned above, and as highlighted by many stud- nity size enhances the cycling of elements as smaller animals ies on fish in coral reefs. Unresolved factors, such as fish tend to have higher metabolic rates (e.g., Schmitz et al., 2010; migrations, would alter our results and the sensitivity of fish- Vanni, 2002), so that the reduction in overall cycling rate is mediated nutrient cycling to warming and deoxygenation due roughly half the reduction of biomass. to climate change (e.g., Lefort et al., 2015; Lotze et al., Finally, as the size classes below 10 g are not resolved 2019). There remains much to be learned about the role of (Fig. 2), we are not able to account for biomass changes fish in global nutrient cycling. that fishing might induce through trophic cascades, or how it reverberates up to fish through food supply (Dupont et al., 2022), which has the potential to further modify fish- Code and data availability. Model outputs and mediated nutrient cycling. code used to analyze them can be found here: https://doi.org/10.5281/zenodo.6538917 (Le Mézo et al., 2022). 7 Conclusions Supplement. The supplement related to this article is available on- In this study, we estimate the amount of N, P, and Fe con- line at: https://doi.org/10.5194/bg-19-2537-2022-supplement. 100 kg tained in and cycled by the global CTF10 g biomass, both in its pristine state and at the time of the global peak catch. The overall contribution of this commercial ichthyosphere to Author contributions. PLM, EG, and DB designed the study. KS provided the updated model version. JG performed the simulations oceanic nutrient cycling is relatively small but is more sig- nificant in regions of low ambient nutrient concentrations, https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2552 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish and the model–data comparison. PLM made the figures and wrote USA, https://doi.org/10.5067/AQUA/MODIS/L3M/ZLEE/2018, the manuscript. All the authors revised the manuscript. 2018. Atkinson, C. L., Capps, K. A., Rugenski, A. T., and Vanni, M. J.: Consumer-driven nutrient dynamics in freshwater ecosystems: Competing interests. The contact author has declared that neither from individuals to ecosystems, Biol. Rev., 92, 2003–2023, they nor their co-authors have any competing interests. https://doi.org/10.1111/brv.12318, 2017. Aumont, O., Maury, O., Lefort, S., and Bopp, L.: Evaluating the Po- tential Impacts of the Diurnal Vertical Migration by Marine Or- Disclaimer. Publisher’s note: Copernicus Publications remains ganisms on Marine Biogeochemistry, Global Biogeochem. Cy., neutral with regard to jurisdictional claims in published maps and 32, 1622–1643, https://doi.org/10.1029/2018GB005886, 2018. institutional affiliations. Baum, J. K. and Worm, B.: Cascading top-down effects of chang- ing oceanic predator abundances, J. Anim. Ecol., 78, 699–714, https://doi.org/10.1111/j.1365-2656.2009.01531.x, 2009. Benitez-Nelson, C. R.: The biogeochemical cycling of phos- Acknowledgements. We thank the reviewers and the editor for their phorus in marine systems, Earth Sci. Rev., 51, 109–135, very helpful comments on the manuscript and all the members of https://doi.org/10.1016/S0012-8252(00)00018-0, 2000. the BIGSEA group for the multiple exchanges on the subject. Bianchi, D., Galbraith, E. D., Carozza, D. A., Mislan, K. A. S., and Stock, C. A.: Intensification of open-ocean oxygen de- pletion by vertically migrating animals, Nat. Geosci., 6, 1–5, Financial support. This research has been supported by the Euro- https://doi.org/10.1038/ngeo1837, 2013. pean Research Council (ERC) under the European Union’s Hori- Bianchi, D., Carozza, D. A., Galbraith, E. D., Guiet, J., and De- zon 2020 research and innovation programme (BIGSEA, grant Vries, T.: Estimating global biomass and biogeochemical cy- no. 682602), the California Ocean Protection Council (grant cling of marine fish with and without fishing, Sci. Adv., 7, 41, no. C0100400), NASA (grant no. 80NSSC21K0420), the French https://doi.org/10.1126/sciadv.abd7554, 2021. ANR project CIGOEF (grant no. ANR-17-CE32-0008-01), and Brahney, J., Mahowald, N., Ward, D. S., Ballantyne, A. P., and Neff, the Extreme Science and Engineering Discovery Environment J. C.: Is atmospheric phosphorus pollution altering global alpine (XSEDE, grant no. TG-OCE170017). Lake stoichiometry?, Global Biogeochem. Cy., 29, 1369–1383, https://doi.org/10.1002/2015GB005137, 2015. Carozza, D. A., Bianchi, D., and Galbraith, E. D.: The ecologi- Review statement. This paper was edited by Kenneth Rose and re- cal module of BOATS-1.0: a bioenergetically constrained model viewed by Jacob Allgeier and Emma Cavan. of marine upper trophic levels suitable for studies of fisheries and ocean biogeochemistry, Geosci. Model Dev., 9, 1545–1565, https://doi.org/10.5194/gmd-9-1545-2016, 2016. Carozza, D. A., Bianchi, D., and Galbraith, E. D.: Formulation, General Features and Global Calibration of a Bioenergetically- References Constrained Fishery Model, PLOS ONE, 12, e0169763, https://doi.org/10.1371/journal.pone.0169763, 2017. Afonso, A. S., Garla, R., and Hazin, F. H. V.: Tiger Cavan, E. L., Belcher, A., Atkinson, A., Hill, S. L., Kawaguchi, S., sharks can connect equatorial habitats and fisheries across McCormack, S., Meyer, B., Nicol, S., Ratnarajah, L., Schmidt, the Atlantic Ocean basin, PLOS ONE, 12, e0184763, K., Steinberg, D. K., Tarling, G. A., and Boyd, P. W.: The impor- https://doi.org/10.1371/journal.pone.0184763, 2017. tance of Antarctic krill in biogeochemical cycles, Nat. Commun., Allgeier, J. E., Layman, C. A., Mumby, P. J., and Rosemond, A. D.: 10, 1–13, https://doi.org/10.1038/s41467-019-12668-7, 2019. Consistent nutrient storage and supply mediated by diverse fish Chassot, E., Bonhommeau, S., Dulvy, N. K., Mélin, F., Watson, R., communities in coral reef ecosystems, Glob. Change Biol., 20, Gascuel, D., and Le Pape, O.: Global marine primary production 2459–2472, https://doi.org/10.1111/gcb.12566, 2014. constrains fisheries catches, Ecol. Lett., 13, 495–505, 2010. Allgeier, J. E., Valdivia, A., Cox, C., and Layman, C. A.: Fish- Clarke, A. and Johnston, N. M.: Scaling of metabolic rate with body ing down nutrients on coral reefs, Nat. Commun., 7, 1–5, mass and temperature in teleost fish, J. Anim. Ecol., 68, 893–905, https://doi.org/10.1038/ncomms12461, 2016. 1999. Allgeier, J. E., Burkepile, D. E., and Layman, C. A.: Animal Czamanski, M., Nugraha, A., Pondaven, P., Lasbleiz, M., Mas- pee in the sea: consumer-mediated nutrient dynamics in the son, A., Caroff, N., Bellail, R., and Tréguer, P.: Carbon, nitro- world’s changing oceans, Glob. Change Biol., 23, 2166–2178, gen and phosphorus elemental stoichiometry in aquacultured and https://doi.org/10.1111/gcb.13625, 2017. wild-caught fish and consequences for pelagic nutrient dynam- Allgeier, J. E., Wenger, S., and Layman, C. A.: Taxonomic ics, Mar. Biol., 158, 2847–2862, https://doi.org/10.1007/s00227- identity best explains variation in body nutrient stoichiometry 011-1783-7, 2011. in a diverse marine animal community, Sci. Rep., 10, 1–10, Davison, P. C., Checkley, D. M., Koslow, J. A., and Bar- https://doi.org/10.1038/s41598-020-67881-y, 2020. low, J.: Carbon export mediated by mesopelagic fishes in Aqua MODIS: NASA goddard space flight center, ocean ecology the northeast Pacific Ocean, Prog. Oceanogr., 116, 14–30, laboratory, ocean biology processing group, Moderate-resolution https://doi.org/10.1016/j.pocean.2013.05.013, 2013. Imaging Spectroradiometer (MODIS) Aqua Euphotic Depth Data; 2018 Reprocessing, NASA OB, DAAC, Greenbelt, MD, Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2553 Dunne, J. P., Sarmiento, J. L., and Gnanadesikan, A.: A synthesis of Griffiths, D.: The direct contribution of fish to lake global particle export from the surface ocean and cycling through phosphorus cycles, Ecol. Freshw. Fish, 15, 86–95, the ocean interior and on the seafloor, Global Biogeochem. Cy., https://doi.org/10.1111/j.1600-0633.2006.00125.x, 2006. 21, GB4006, https://doi.org/10.1029/2006GB002907, 2007. Gruber, N. and Galloway, J. N.: An Earth-system perspec- Dunne, J. P., John, J. G., Shevliakova, S., Stouffer, R. J., Krast- tive of the global nitrogen cycle, Nature, 451, 293–296, ing, J. P., Malyshev, S. L., Milly, P. C., Sentman, L. T., Ad- https://doi.org/10.1038/nature06592, 2008. croft, A. J., Cooke, W., Dunne, K. A., Griffies, S. M., Hallberg, Guiet, J., Galbraith, E., Bianchi, D., and Cheung, W.: Bioenergetic R. W., Harrison, M. J., Levy, H., Wittenberg, A. T., Phillips, P. J., influence on the historical development and decline of industrial and Zadeh, N.: GFDL’s ESM2 global coupled climate-carbon fisheries, ICES J. Mar. Sci., 77, 1854–1863, 2020. earth system models. Part II: Carbon system formulation and Hall, R. O. J., Koch, B. J., Marshall, M. C., Taylor, B. W., How, baseline simulation characteristics, J. Climate, 26, 2247–2267, L. M. T., Koch, B. J., Marshall, M. C., and Taylor, B. W.: https://doi.org/10.1175/JCLI-D-12-00150.1, 2013. How body size mediates the role of animals in nutrient cycling Dupont, L., Le Mézo, P., Clerc, C., Maury, O., Aumont, O., Ethé, in aquatic ecosystems, The structure and function of aquatic C., and Bopp, L.: High trophic level feedbacks on ocean biogeo- ecosystems, Cambridge University Press New York, NY, 286– chemistry under climate change, submitted, 2022. 305, https://doi.org/10.1017/CBO9780511611223, 2007. El-Sabaawi, R. W., Warbanski, M. L., Rudman, S. M., Hovel, R., Halvorson, H. M. and Small, G. E.: Observational field studies are and Matthews, B.: Investment in boney defensive traits alters not appropriate tests of consumer stoichiometric homeostasis, organismal stoichiometry and excretion in fish, Oecologia, 181, Freshw. Sci., 35, 1103–1116, https://doi.org/10.1086/689212, 1209–1220, https://doi.org/10.1007/s00442-016-3599-0, 2016. 2016. Fowler, D., Coyle, M., Skiba, U., Sutton, M., Cape, J. N., Reis, S., Hatton, I. A., Heneghan, R. F., Bar-On, Y. M., and Galbraith, E. D.: Sheppard, L., Jenkins, A., Grizetti, B., Galloway, J. N., Vitousek, The global ocean size spectrum from bacteria to whales, Sci. P., Leach, A., Bouwman, L., Butterbach-Bahl, K., Dentener, F., Adv., 7, 1–9, https://doi.org/10.1126/sciadv.abh3732, 2021. Stevenson, D., Amann, M., and Voss, M.: The global nitrogen Hernández-León, S., Fraga, C., and Ikeda, T.: A global cycle in the 21th century, Philis. T. R. Soc. Lon. B, 368, 1621, estimation of mesozooplankton ammonium excretion https://doi.org/10.1098/rstb.2013.0164, 2013. in the open ocean, J. Plankton Res., 30, 577–585, Francis, F. T. and Côté, I. M.: Fish movement drives spatial and https://doi.org/10.1093/plankt/fbn021, 2008. temporal patterns of nutrient provisioning on coral reef patches, Hessen, D. O. and Kaartvedt, S.: Top-down cascades in lakes and Ecosphere, 9, e02225, https://doi.org/10.1002/ecs2.2225, 2018. oceans: Different perspectives but same story?, J. Plankton Res., Frank, K. T., Petrie, B., Choi, J. S., and Leggett, W. C.: Trophic 36, 914–924, https://doi.org/10.1093/plankt/fbu040, 2014. cascades in a formerly cod-dominated ecosystem., Science, 308, Hicks, C. C., Cohen, P. J., Graham, N. A. J., Nash, K. L., Alli- 1621–1623, https://doi.org/10.1126/science.1113075, 2005. son, E. H., D’Lima, C., Mills, D. J., Roscher, M., Thilsted, S. H., Galbraith, E. D. and Martiny, A. C.: A simple nutrient- Thorne-Lyman, A. L., and MacNeil, M. A.: Harnessing global dependence mechanism for predicting the stoichiometry of ma- fisheries to tackle micronutrient deficiencies, Nature, 574, 95– rine ecosystems, P. Natl. Acad. Sci. USA, 112, 8199–8204, 98, https://doi.org/10.1038/s41586-019-1592-6, 2019. https://doi.org/10.1073/pnas.1423917112, 2015. Hjerne, O. and Hansson, S.: The role fish and fisheries in the Baltic Galbraith, E. D., Carozza, D. A., and Bianchi, D.: A cou- sea nutrient dyanmics, Limnol. Oceanogr., 47, 1023–1032, 2002. pled human-Earth model perspective on long-term trends Huang, Y., Ciais, P., Goll, D. S., Sardans, J., Peñuelas, J., Cresto- in the global marine fishery, Nat. Commun., 8, 1–7, Aleina, F., and Zhang, H.: The shift of phosphorus transfers https://doi.org/10.1038/ncomms14884, 2017. in global fisheries and aquaculture, Nat. Commun., 11, 1–10, Galbraith, E. D., Le Mézo, P. K., Solanes Hernandez, G., Bianchi, https://doi.org/10.1038/s41467-019-14242-7, 2020. D., and Kroodsma, D.: Growth limitation of marine fish by low Ito, A.: Atmospheric Processing of Combustion Aerosols as a iron availability in the open ocean, Frontiers in Marine Science, Source of Bioavailable Iron, Environ. Sci. Technol. Lett., 2, 70– 6, 1–13, https://doi.org/10.3389/fmars.2019.00509, 2019. 75, https://doi.org/10.1021/acs.estlett.5b00007, 2015. Garcia, H. E., Locarnini, R. A., Boyer, T. P., Antonov, J. I., Bara- Jennings, S., Mélin, F., Blanchard, J. L., Forster, R. M., Dulvy, nova, O. K., Zweng, M. M., Reagan, J. R., Johnson, D. R., N. K., Wilson, R. W., Jennings, S., Dulvy, N. K., Wil- Mishonov, A. V., and Levitus, S.: World ocean atlas 2013, Vol- son, R. W., Melin, F., Blanchard, J. L., Forster, R. M., ume 4, Dissolved inorganic nutrients (phosphate, nitrate, sili- Dulvy, N. K., Wilson, R. W., Mélin, F., Blanchard, J. L., cate), NOAA institutional repository, NOAA atlas NESDIS, 76, Forster, R. M., Dulvy, N. K., and Wilson, R. W.: Global- https://doi.org/10.7289/V5J67DWD, 2013. scale predictions of community and ecosystem properties from Gordon, H. S.: The Economic Theory of a Common-Property Re- simple ecological theory, P. Roy. Soc. B, 275, 1375–1383, source: The Fishery, Education + Training, 62, 124–142, 1954. https://doi.org/10.1098/rspb.2008.0192, 2008. Gresh, T., Lichatowich, J., and Schoonmaker, P.: An Esti- Kanakidou, M., Duce, R. A., Prospero, J. M., Baker, A. R., Benitez- mation of Historic and Current Levels of Salmon Produc- Nelson, C., Dentener, F. J., Hunter, K. A., Liss, P. S., Ma- tion in the Northeast Pacific Ecosystem: Evidence of a howald, N., Okin, G. S., Sarin, M., Tsigaridis, K., Uematsu, M., Nutrient Deficit in the Freshwater Systems of the Pacific Zamora, L. M., and Zhu, T.: Atmospheric fluxes of organic N Northwest, Fisheries, 25, 15–21, https://doi.org/10.1577/1548- and P to the global ocean, Global Biogeochem. Cy., 26, 1–12, 8446(2000)025<0015:AEOHAC>2.0.CO;2, 2000. https://doi.org/10.1029/2011GB004277, 2012. Kavanagh, L. and Galbraith, E.: Links between fish abundance and ocean biogeochemistry as recorded in marine sediments, PLoS https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022 2554 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish ONE, 13, 1–22, https://doi.org/10.1371/journal.pone.0199420, terns of oceanic nutrient limitation, Nat. Geosci., 6, 701–710, 2018. https://doi.org/10.1038/ngeo1765, 2013. Layman, C. A., Allgeier, J. E., Rosemond, A. D., Dahlgren, C. P., Moreno, A. R. and Haffa, A. L. M.: The Impact and Yeager, L. A.: Marine fisheries declines viewed upside down: of Fish and the Commercial Marine Harvest on Human impacts on consumer-driven nutrient recycling, Ecol. the Ocean Iron Cycle, PLoS ONE, 9, e107690, Appl., 21, 343–349, https://doi.org/10.1890/10-1339.1, 2011. https://doi.org/10.1371/journal.pone.0107690, 2014. Lefort, S., Aumont, O., Bopp, L., Arsouze, T., Gehlen, M., and Myriokefalitakis, S., Nenes, A., Baker, A. R., Mihalopoulos, N., and Maury, O.: Spatial and body-size dependent response of marine Kanakidou, M.: Bioavailable atmospheric phosphorous supply to pelagic communities to projected global climate change, Glob. the global ocean: a 3-D global modeling study, Biogeosciences, Change Biol., 21, 154–164, https://doi.org/10.1111/gcb.12679, 13, 6519–6543, https://doi.org/10.5194/bg-13-6519-2016, 2016. 2015. Okin, G. S., Baker, A. R., Tegen, I., Mahowald, N. M., Den- Le Mézo, P. K. and Galbraith, E. D.: The fecal iron pump : impact tener, F. J., Duce, R. A., Galloway, J. N., Hunter, K., Kanaki- of animals on the Fe stoichiometry of marine sinking particles, dou, M., Kubilay, N., Prospero, J. M., Sarin, M., Surapip- Limnol. Oceanogr., 66, 1–39, 2020. ith, V., Uematsu, M., and Zhu, T.: Impacts of atmospheric Le Mézo, P., Guiet, J., Scherrer, K., Bianchi, D., and Galbraith, nutrient deposition on marine productivity: Roles of nitro- E.: Global nutrient cycling by commercially targeted marine gen, phosphorus, and iron, Global Biogeochem. Cy., 25, 1–10, fish (Le Mézo et al., 2022, Biogeosciences), Zenodo [data set], https://doi.org/10.1029/2010GB003858, 2011. https://doi.org/10.5281/zenodo.6538917, 2022. Pauly, D. and Tsukayama, I.: On the seasonal growth, monthly Leroux, S. J. and Schmitz, O. J.: Predator-driven elemen- recruitment and monthly biomass of Peruvian anchoveta (En- tal cycling: the impact of predation and risk effects graulis ringens) from 1961 to 1979, in: Proceedings of the ex- on ecosystem stoichiometry, Ecol. Evol., 5, 4976–4988, pert consultation to examine changes in abundance and species https://doi.org/10.1002/ece3.1760, 2015. composition of neritic fish resources, San José, Costa Rica, 18– Lotze, H. K., Tittensor, D. P., Bryndum-Buchholz, A., Eddy, T. D., 29 April 1983, FAO fishery report 291, edited by: Sharp, G. D. Cheung, W. W., Galbraith, E. D., Barange, M., Barrier, N., and Csirke, J., vol. 3, 987–1004, 1984. Bianchi, D., Blanchard, J. L., Bopp, L., Büchner, M., Bulman, Pauly, D. and Zeller, D.: Catch reconstructions reveal that global C. M., Carozza, D. A., Christensen, V., Coll, M., Dunne, J. P., marine fisheries catches are higher than reported and declining, Fulton, E. A., Jennings, S., Jones, M. C., Mackinson, S., Maury, Nat. Commun., 7, 10244, https://doi.org/10.1038/ncomms10244, O., Niiranen, S., Oliveros-Ramos, R., Roy, T., Fernandes, J. A., 2016. Schewe, J., Shin, Y. J., Silva, T. A., Steenbeek, J., Stock, C. A., Pilati, A. and Vanni, M. J.: Ontogeny, diet shifts, and nu- Verley, P., Volkholz, J., Walker, N. D., and Worm, B.: Global en- trient stoichiometry in fish, Oikos, 116, 1663–1674, semble projections reveal trophic amplification of ocean biomass https://doi.org/10.1111/j.0030-1299.2007.15970.x, 2007. declines with climate change, P. Natl. Acad. Sci. USA, 116, Prabhu, A. J. P., Schrama, J. W., and Kaushik, S. J.: Mineral require- 12907–12912, https://doi.org/10.1073/pnas.1900194116, 2019. ments of fish: A systematic review, Rev. Aquacult., 8, 172–219, Mahowald, N. M., Engelstaedter, S., Luo, C., Sealy, A., Ar- https://doi.org/10.1111/raq.12090, 2016. taxo, P., Benitez-Nelson, C., Bonnet, S., Chen, Y., Chuang, Ricard, D., Minto, C., Jensen, O. P., and Baum, J. K.: Examining P. Y., Cohen, D. D., Dulac, F., Herut, B., Johansen, A. M., the knowledge base and status of commercially exploited marine Kubilay, N., Losno, R., Maenhaut, W., Paytan, A., Pros- species with the RAM Legacy Stock Assessment Database, Fish pero, J. M., Shank, L. M., and Siefert, R. L.: Atmo- Fish., 13, 380–398, 2012. spheric Iron Deposition: Global Distribution, Variability, and Roman, J., Estes, J. A., Morissette, L., Smith, C., Costa, D., Mc- Human Perturbations, Annu. Rev. Mar. Sci., 1, 245–278, Carthy, J., Nation, J. B., Nicol, S., Pershing, A., and Smetacek, https://doi.org/10.1146/annurev.marine.010908.163727, 2009. V.: Whales as marine ecosystem engineers, Front. Ecol. Environ., Maldonado, M. T., Surma, S., and Pakhomov, E. A.: South- 12, 377–385, https://doi.org/10.1890/130220, 2014. ern Ocean biological iron cycling in the pre-whaling and Saba, G. K. and Steinberg, D. K.: Abundance, composition, and present ecosystems, Philos. T Roy Soc A, 374, 2081, sinking rates of fish fecal pellets in the santa barbara channel, https://doi.org/10.1098/rsta.2015.0292, 2016. Sci. Rep., 2, 1–6, https://doi.org/10.1038/srep00716, 2012. Maranger, R., Caraco, N., Duhamel, J., and Amyot, M.: Nitrogen Saba, G. K., Burd, A. B., Dunne, J. P., Hernández-León, S., Mar- transfer from sea to land via commercial fisheries, Nat. Geosci., tin, A. H., Rose, K. A., Salisbury, J., Steinberg, D. K., Trueman, 1, 111–113, https://doi.org/10.1038/ngeo108, 2008. C. N., Wilson, R. W., and Wilson, S. E.: Toward a better under- McIntyre, P. B., Flecker, A. S., Vanni, M. J., Hood, J. M., Taylor, standing of fish-based contribution to ocean carbon flux, Limnol. B. W., and Thomas, S. A.: Fish distributions and nutrient recy- Oceanogr, 9999, 1–26, https://doi.org/10.1002/lno.11709, 2021. cling in streams: can fish create biogeochemical hotspots?, Ecol- Sarmiento, J. L. and Gruber, N.: Ocean biogeo- ogy, 89, 2335–2346, 2008. chemical dynamics, Princeton University Press, Moore, C. M., Mills, M. M., Arrigo, K. R., Berman-Frank, https://doi.org/10.1515/9781400849079, 2006. I., Bopp, L., Boyd, P. W., Galbraith, E. D., Geider, R. J., Schaefer, M. B.: Some aspects of the dynamics of populations im- Guieu, C., Jaccard, S. L., Jickells, T. D., La Roche, J., portant to the management of the commercial marine fisheries, B. Lenton, T. M., Mahowald, N. M., Marañón, E., Marinov, Math. Biol., 53, 253–279, https://doi.org/10.1007/BF02464432, I., Moore, J. K., Nakatsuka, T., Oschlies, A., Saito, M. A., 1954. Thingstad, T. F., Tsuda, A., and Ulloa, O.: Processes and pat- Biogeosciences, 19, 2537–2555, 2022 https://doi.org/10.5194/bg-19-2537-2022 P. Le Mézo et al.: Global nutrient cycling by commercially targeted marine fish 2555 Schindler, D. E. and Eby, L. A.: Stoichiometry of Fishes and Their Vanni, M. J., Boros, G., and McIntyre, P. B.: When are fish sources Prey : Implications for Nutrient Recycling, Ecology, 78, 1816– versus sinks of nutrients in lake ecosystems?, Ecology, 94, 2195– 1831, 1997. 2206, 2013. Schmitz, O. J., Hawlena, D., and Trussell, G. C.: Predator con- Vanni, M. J., McIntyre, P. B., Allen, D., Arnott, D. L., Benstead, trol of ecosystem nutrient dynamics, Ecol. Lett., 13, 1199–1209, J. P., Berg, D. J., Brabrand, Å., Brosse, S., Bukaveckas, P. A., https://doi.org/10.1111/j.1461-0248.2010.01511.x, 2010. Caliman, A., Capps, K. A., Carneiro, L. S., Chadwick, N. E., Shearer, K. D.: Changes in Elemental Composition of Hatchery- Christian, A. D., Clarke, A., Conroy, J. D., Cross, W. F., Culver, Reared Rainbow Trout, Salmo gairdneri, Associated with D. A., Dalton, C. M., Devine, J. A., Domine, L. M., Evans-White, Growth and Reproduction, Can. J. Fish. Aquat. Sci., 41, 1592– M. A., Faafeng, B. A., Flecker, A. S., Gido, K. B., Godinot, C., 1600, https://doi.org/10.1139/f84-197, 1984. Guariento, R. D., Haertel-Borer, S., Hall, R. O., Henry, R., Her- Shearer, K. D., ÅSgård, T., Andorsdöttir, G., and Aas, G. H.: Whole wig, B. R., Hicks, B. J., Higgins, K. A., Hood, J. M., Hopton, body elemental and proximate composition of Atlantic salmon M. E., Ikeda, T., James, W. F., Jansen, H. M., Johnson, C. R., (Salmo salar) during the life cycle, J. Fish Biol., 44, 785–797, Koch, B. J., Lamberti, G. A., Lessard-Pilon, S., Maerz, J. C., https://doi.org/10.1111/j.1095-8649.1994.tb01255.x, 1994. Mather, M. E., McManamay, R. A., Milanovich, J. R., Morgan, Spalding, M. D. and Grenfell, A. M.: New estimates of global D. K., Moslemi, J. M., Naddafi, R., Nilssen, J. P., Pagano, M., and regional coral reef areas, Coral Reefs, 16, 225–230, Pilati, A., Post, D. M., Roopin, M., Rugenski, A. T., Schaus, https://doi.org/10.1007/s003380050078, 1997. M. H., Shostell, J., Small, G. E., Solomon, C. T., Sterrett, S. C., Sterner, R. W. and Elser, J. J.: Ecological stoichiometry: the biology Strand, Ø., Tarvainen, M., Taylor, J. M., Torres-Gerald, L. E., of elements from molecules to the biosphere, Princeton univer- Turner, C. B., Urabe, J., Uye, S. I., Ventelä, A. M., Villeger, sity press, https://doi.org/10.1515/9781400885695, 2002. S., Whiles, M. R., Wilhelm, F. M., Wilson, H. F., Xenopou- Stock, C. A., John, J. G., Rykaczewski, R. R., Asch, R. G., los, M. A., and Zimmer, K. D.: A global database of nitrogen Cheung, W. W., Dunne, J. P., Friedland, K. D., Lam, V. W., and phosphorus excretion rates of aquatic animals, Ecology, 98, Sarmiento, J. L., and Watson, R. A.: Reconciling fisheries catch 1475, https://doi.org/10.1002/ecy.1792, 2017. and ocean productivity, P. Natl. Acad. Sci. USA, 114, E1441– Wang, R., Balkanski, Y., Boucher, O., Bopp, L., Chappell, A., E1449, https://doi.org/10.1073/pnas.1610238114, 2017. Ciais, P., Hauglustaine, D., Peñuelas, J., and Tao, S.: Sources, Thodesen, J., Storebakken, T., Shearer, K. D., Rye, M., Bjerkeng, transport and deposition of iron in the global atmosphere, At- B., and Gjerde, B.: Genetic variation in mineral absorp- mos. Chem. Phys., 15, 6247–6270, https://doi.org/10.5194/acp- tion of large Atlantic salmon (Salmo salar) reared in seawa- 15-6247-2015, 2015. ter, Aquaculture, 194, 263–271, https://doi.org/10.1016/S0044- Wilson, R. W., Millero, F. J., Taylor, J. R., Walsh, P. J., Christensen, 8486(00)00525-1, 2001. V., Jennings, S., and Grosell, M.: Contribution of fish to the ma- Turner, J. T.: Zooplankton fecal pellets, marine snow, phytodetritus rine inorganic carbon cycle, Science, 323, 359–362, 2009. and the ocean’s biological pump, Prog. Oceanogr., 130, 205–248, Wotton, R. S. and Malmqvist, B.: Feces in Aquatic Ecosys- https://doi.org/10.1016/j.pocean.2014.08.005, 2015. tems, BioScience, 51, 537–544, https://doi.org/10.1641/0006- Vanni, M. J.: Nutrient Cycling by Animals in Freshwa- 3568(2001)051[0537:FIAE]2.0.CO;2, 2001. ter Ecosystems, Annu. Rev. Ecol. Syst., 33, 341–370, https://doi.org/10.1146/annurev.ecolsys.33.010802.150519, 2002. Vanni, M. J., Bowling, A. M., Dickman, E. M., Hale, R. S., Higgins, K. A., Horgan, M. J., Knoll, L. B., Renwick, W. H., and Stein, R. A.: Nutrient cycling by fish supports relatively more primary production as lake productivity in- creases, Ecology, 87, 1696–1709, https://doi.org/10.1890/0012- 9658(2006)87[1696:NCBFSR]2.0.CO;2, 2006. https://doi.org/10.5194/bg-19-2537-2022 Biogeosciences, 19, 2537–2555, 2022