Feeding global aquaculture

The farming of many aquatic species relies on feed, which incurs social and environmental impacts. An issue that has raised much attention is the reliance on capture fisheries for reduction into fish meal and oil. “Reduction fisheries”—industrial seine fleets targeting small pelagic fish such as anchoveta, whiting, and sardine—account for an estimated average of one-sixth of the mass of the global marine catch and can comprise nearly one-third in some years (1–3). Approximately 70% of this biomass is processed into aquaculture feed, with the remaining 30% being used for other animal feed, supplements, and cosmetics (4). Reduction fisheries have global ecological impacts (5), affecting the structure of exploited populations (6) and reducing food availability for predators, such as fish, seabirds, and marine mammals (7). Reduction fishing also diverts millions of tons of food-grade fish (e.g., anchovies and sardines) and nutrients from countries with high rates of hunger to farmed aquatic animals (e.g., salmon and shrimp) intended for luxury markets (8–11).

The fish-in:fish-out (FI:FO) metric was developed to quantify the reliance of aquaculture on captured fish. FI:FO is distinct from the feed conversion ratio (FCR) metric, which divides dehydrated feed inputs by wet fish outputs. FI:FO reconstructs a “live weight equivalent” from reported feed use, which approximates the biomass of wild fish consumed, then divides it by farmed output to estimate the ratio of fished biomass inputs to farmed fish biomass outputs for a given farm, farmed species group, or the aquaculture sector as a whole (12). FI:FO quantities should therefore reflect integrated average wild fish utilization across all stages of the farmed fishes’ life cycles.

The FI:FO metric has been through several iterations. Revisions have been made to account for “residual” and “recovered” oil embedded in fish meal or coming from the same fish that were made into fish meal, though debates persist surrounding how to apportion these subtractions between divisions of the aquaculture industry farming different animal taxa (henceforth: “species groups”). A method emerged, established by Hardy in 2009 (13), which was used by Naylor et al. (14) in 2021 to analyze Food and Agriculture Organization of the United Nations (FAO) reports and industry surveys and compare FI:FO ratios from 1997 and 2017 across the top 11 feed-consuming species groups. These methods yielded results that showed that the intensity of wild fish use in fish and crustacean farming fell by 85% over 20 years.

However, translating processed feed inputs into a live weight equivalent of wild fish requires making approximations for proportions of wild fish incorporated in aquaculture feeds and assumptions regarding how they are reduced and processed. Naylor et al. (14) compiled a comprehensive estimation of these feed reduction parameters, concluding that aquaculture feed, on aggregate, consisted of approximately 7% wild fish in 2017. Yet, the proprietary status of feed manufacturing requires taking these data from voluntary industry disclosures, which are difficult to validate. We compiled additional feed composition datasets obtained using survey, projection, or metastudy for similar time frames by the FAO (15), Monterey Bay Aquarium (MBA) Seafood Watch (16–25), and Pahlow et al. (26). A wider range of source-independent estimates may help gauge uncertainty, as well as risk.

With respect to methods for calculating FI:FO, common conventions in scholarly literature omit two substantial sources of wild fish. First, recovered oil (oil produced during fish meal processing) is typically calculated assuming no losses and subtracted entirely (13, 14), but industry reports indicate that an estimated 27% of this oil is not used by aquaculture, but instead allocated to human supplements, terrestrial animal feeds, cosmetics, and other uses (27). Second, all meal and oil from trimmings—parts of marine animals’ bodies that are removed during processing—are conventionally classified as by-products and therefore subtracted on the basis that they are not wild fish inputs. Furthermore, whole fish are increasingly classified as “trash,” particularly when caught while targeting other species, and reported entirely as by-product (28, 29). However, industry reports estimate that roughly two-thirds of trimmings and by-products (henceforth: “trimmings”) are sourced from wild-captured fish (27).

Beyond inputs, reduction fisheries incur high rates of fishing mortality due to “slipping” (30–36)—when vessels leave purse seine nets slightly open to allow unwanted catch to escape, which results in estimated postrelease fish mortality rates ranging from 38% to more than 99% in trials (36–42). The collateral mortality involved in slipping, along with additional “bycatch” from fisheries from which trimmings are sourced, has not been considered in FI:FO formulations to date. Last, we adjust the formula to approximate the efficiency of the aquacultural feeding process, as opposed to total industrial inputs over outputs, considering that more than a quarter of farmed fish and crustacean biomass is derived from systems that do not use compound feeds (henceforth: “unfed”).

It is also important to note that any metric for assessing the reliance on fish meal and oil is not a comprehensive metric for assessing the sustainability of aquaculture feed because these calculations do not include the reliance of aquaculture on terrestrial feed inputs (43). Terrestrial feed production is linked to generation of emissions and effluent, as well as freshwater consumption (26), land use changes (44), and opportunity costs (45). Here, we aim to account for previous omissions and trade-offs to provide a more comprehensive environmental evaluation of the feed requirements for global aquaculture. These methods can be used to inform future research for more comprehensive life cycle assessments.

The impact of aquaculture on wild fish is greater than commonly cited. Of the four feed composition datasets presented here, the parameters used by Naylor et al. (14) for the most referenced calculation of FI:FO (14) is also the dataset that demonstrates the lowest use of fish in global aquaculture feeds. In addition, if we take a broader and ecologically oriented approach to calculating the biomass of fish killed for aquaculture feed, a substantial fraction of the impact of aquaculture feed on marine food webs has been previously overlooked.

Figures used for feed composition data have extensive ramifications, as the FI:FO method of assessing fish use intensity is highly sensitive to errors, assumptions, and underreporting. Minor variations in parameters can have compounding effects on metrics. The accuracy of available data is difficult to evaluate; thus, the scenarios selected here incorporate the widest available range of peer-reviewed literature, offering a representative range of estimates. By replicating this exercise in scenarios, we aim to portray a less precise, yet therefore more accurate picture of the fish extraction impact of aquaculture.

Many additional forms of uncertainty still remain, and policy should reflect the risks of these still poorly characterized resource use requirements. Almost all parameters used in this study and others are self-reported by industry, taken or projected from voluntary surveys, as aquaculture feed corporations are generally not required to disclose feed formulations and often classify them as proprietary information.

Disclosure and inspection policies can both enhance transparency. Stable isotope analysis has begun to help disaggregate nutrient composition of feeds by using isotopic signatures to partition fractions of ingredients in aquaculture feed, such as crops, manure, and wild fish (53, 54). Public and private policy interventions should recognize that opacity in feed formulation and use represents a barrier to the sustainability of the sector. If corporations voluntarily disclose and independently verify feed compositions, or are compelled to do so legally, the methods proposed by researchers could be evaluated and this opacity would be reduced.

Uncertainties in feed formulation are particularly consequential with regard to the use of fish oil. Less oil is extracted from fish than meal, which means variations in fish oil content are more consequential in calculating live weight equivalents of captured fish (13). Therefore, estimates for farmed species groups such as salmon that use oily carnivorous fish feed have higher variability and greater uncertainty.

In addition, the reduction efficiency of fish oil, in particular, can vary widely both between years and within years, because the body fat of small pelagic fishes fluctuates across species, seasons, and age demographics. Oil yield falls when, for instance, the anchoveta catch includes a high proportion of juveniles or has undergone an El Niño event. Since 2000, captured anchoveta have become smaller on average, with the highest size classes all but disappearing (55). It may be that the 7.6% fish oil reduction efficiency calculated for 2017 in this study and possibly even the 5% used by Naylor et al. (14) are overestimates going forward. While fluctuations caused by fish life cycles are difficult to capture, calculating reduction efficiencies from weighted averages of species in reported catch (data S1) can help quantify these variables more accurately than assigning constant values in future research.

Evaluating alternative parameters for feed composition, we find that a relatively small (17%) variation in proportion of marine-sourced feed more than tripled the final figures for aggregate efficiency of fish use. Pahlow et al. (26), the most fish-intensive feed scenario, is the only dataset obtained by metastudy, as opposed to surveying aquaculture feed manufacturers, although these data are still mainly from industry. These sources indicate higher proportions of fish inclusion than others, but are less up to date as a result of scarcer data. However, the MBA data are the most contemporary and yield the second highest FI:FO quotients. Last, the FAO dataset demonstrates how a small degree of underreporting might have compounding ramifications for the efficiency of wild fish use in fish and crustacean farming.

Comparing results calculated by Naylor et al. (14) to those derived from the same data using our adjusted calculations, conventional methodological assumptions tend to lower FI:FO ratios. While our calculations indicate that the efficiency of oil extraction from wild fish may be underestimated by Naylor et al. (14), attributed in part to the high reported use of sardines in the study year, the deduction of all trimmings does not accurately portray the reliance of aquaculture on wild capture. Industry disclosures suggest a full two-thirds of these trimmings are cut from wild fish, while by some definitions these cuts are classified as by-products, they are only one processing stage removed from wild catch and trimmings can comprise up to 70% of an animal’s biomass (56). Furthermore, as demand for aquaculture feed has risen, so has the economic value of trimmings, such that they may constitute a substantial proportion or even majority of a wild animal’s sale value, contributing to demand for their capture (57). Last, the “by-products” category includes millions of tons of whole fish, which are classified as trash due to being nontarget species. Therefore, we do not exclude wild fish trimmings from wild fish inputs.

Expanding further, collateral fishing mortality represents another substantial ecological impact of feed manufacturing that has gone largely unaccounted for due to the standard of limiting scope to fish inputs. Bycatch in fisheries not only includes other fish species, but also can include seabirds, turtles, marine mammals, and invertebrates. Moreover, the definition of bycatch fails to include slipping mortality because these fish are never hauled onto deck. While the convention is to quantify fish inputs, it may be more ecologically appropriate to consider fishing mortality. Therefore, we include, yet explicitly disaggregate, collateral mortality when evaluating the efficiency of fish and crustacean farming, as it is ultimately a question of the impact of the industry on wild fish populations.

Although the impacts of feed manufacturing are considerable, they represent a fraction of the environmental impacts of aquaculture as a whole. Unaccounted wild fish mortality extends far beyond feed, including abandoned gear (58, 59), processing collateral (60), the capture of cleaner fish (61), and “ranched” species that cannot be bred in captivity, such as tuna and eel (62). Indirect mortality and broader environmental impacts also include transmission of pathogens (63), application of pesticides and antibiotics (64, 65), outbreaks of introduced species (66, 67), domestication of wild species (68), eutrophication (69, 70), deoxygenation (71), chemical pollution (72, 73), stream diversion (74), tidal disruption (75), coastal degradation (76), and greenhouse gas emissions—which may be underaccounted for (77) and are a priority for future research.

In terms of efficiency, the inclusion of unfed fish in the “fish out” denominator can give the impression that FI:FO estimates the efficiency of the fish feeding process, when it is actually an analysis of the industry at large, of which more than a quarter of production comes from unfed, primarily traditional, and commercially unavailable systems. Therefore, to provide estimates for average feeding efficiency, we have also provided calculations excluding fish farmed in unfed systems from FI:FO. As opposed to total industrial inputs over outputs, the aggregate FI:FO ratio of fish and crustacean feeding processes is more than a third higher: ranging from 0.51 to 1.56 before considering collateral mortality (Fig. 2).

Last, the impact of terrestrial feeds have been largely overlooked in discussions on aquaculture sustainability. Adopting the most widely cited scenario for efficiency gains in wild fish use requires that crop use would need to outpace the growth of aquaculture production. This illustrates that over time, aquaculture outputs do not decouple from inputs, but rather inputs shift from marine to terrestrial impacts.

While crop demand for aquaculture feed can be extrapolated across a time series, land and water use estimates are expressed here only as a snapshot, due to increases in crop yield efficiency over recent decades. However, the increase in the aquaculture sector’s total utilization of feed crops under the Naylor&al scenario—468% overall and 77% on a per-unit basis—has likely outpaced improvements in crop yields (tons of crops produced per hectare). Between 1997 and 2017, global maize and soy yields increased by 38 and 32%, respectively (78). This suggests that the land requirements demanded by aquaculture production have increased per unit of production and overall.

Considering marine and terrestrial inputs combined, these findings reiterate that fish and crustacean farming does not, on net, produce calories or protein (79). Retention of dietary nutrients in feed is less studied and more variable, but also a net loss (80). While some analyses have examined nutrient retention in a FI:FO framework, nutrient availability in terrestrial feed inputs must also be considered to make congruent comparisons with other food sectors. Future efforts should analyze the net micronutrient benefits and losses across various aquaculture species groups globally. Although aquaculture can provide concentrated sources of deficient nutrients in some contexts (81), it can diminish nutritional quality in others (82), and reduction fisheries remain a notable driver of malnutrition (11, 83).

While FI:FO is informative in some contexts, it compares a small fraction of inputs to total outputs and omits the impacts of shifting to terrestrial feeds (84, 85). The methods we provide do not equate to a full life cycle assessment (LCA), but provide a more accurate quantification of fish extraction and crop cultivation impacts, which is a prerequisite to more accurate LCAs in future research.

A wider set of criteria for sustainable policies and investments, beyond FI:FO, include numerous marine and terrestrial impacts of food production. Many alternative protein options have low terrestrial impacts and minimal marine impacts (86), but micronutrient differences remain a concern and nutritional profiles are in need of improvement. More broadly, this review widens the scope of sustainability considerations for aquaculture inputs, including bycatch, slipping mortality, and terrestrial impacts. However, it does not evaluate downstream environmental impacts, broader ecological impacts, or impacts on labor, public health, and animal welfare (87). Sustainable policies and investments into food production should be evaluated and compared across as many of these impacts as possible.

The expanded view of feeding global aquaculture offered here suggests that common sustainability accounting methods have been too narrow, overconfident in their precision, and overly optimistic. Both marine and terrestrial impacts are still highly uncertain, but these revised estimates suggest that the environmental impacts of this sector, in its current form and structure, are sufficiently large that directives to expand this sector on sustainability grounds should be reconsidered.

Salmon silhouettes were illustrated by The Noun Project contributor B. Kaito. Additional species silhouettes in the Supplementary Materials were illustrated by contributors Laymik, Rfourtytwo, S. Modgekar, Vrijlansier, Jasminova, and Y. Matra, used under creative commons license.

Funding: Funding for this research for S.R. and M.N.H. was provided by the GRACE Communications Foundation. J.J. and P.M. received no funding.

Author contributions: The research question and methods were conceived by M.N.H. and J.J. S.R. performed the literature review for all scenarios and model parameters. The methods were refined to account for additional mortality sources, fish oil and trimmings estimates, and crop feed sources by S.R., P.M., and J.J. S.R. and M.N.H. performed the scientific data analysis. S.R. created all visuals and summary tables, and formatted the Supplementary Materials. M.N.H., P.M., and J.J. supervised the scientific research. All authors contributed to writing and revising the manuscript.

Competing interests: The authors declare that they have no competing interests.

Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.