TLDR: Welfare interventions in aquaculture don't reliably transfer across species, so each species requires independent research and implementation. I introduce "target populations" as a unit of analysis to quantify how many groups exist. From 2000 to 2023, the number of species comprising 85% of aquaculture production grew from 14 to 22. This means the structural cost of covering the same share of the industry has grown substantially, independent of how well any single intervention performs. This is a portfolio problem, not an intervention problem, and many major producing countries are trending towards further fragmentation.

Summary

In terrestrial agriculture, one species (the domestic chicken) accounts for 89% of all farmed vertebrates. That lack of species diversity is part of what made cage-free campaigns scalable and tractable: the same welfare issues recurred across farms, enabling one intervention to address the same problem everywhere, and similar farm contexts made that intervention feasible everywhere.

Aquaculture does not have this structure. 22 species (e.g. Mrigal carp, Atlantic salmon) comprise 85% of farmed aquatic vertebrates, and 260 more account for the rest. This diversity has not consolidated over time. The number of species comprising the top 85% of production has increased from 14 to 22 since 2000.

This matters because welfare interventions don't easily transfer across species. Even if some welfare problems fundamentally are shared (e.g. low oxygen is bad), in practice their biology and farming practices can be so different that an intervention developed for a shared problem in one species won’t work as effectively for others. Surface paddle aerators that improve oxygen for carp are useless for Atlantic salmon because they live deeper in the water column. Percussive stunners, even if validated for Atlantic salmon, will never work for tilapia because their skulls are too laterally compressed. The further apart two species are taxonomically, the less likely interventions carry over. At the distance between salmon and tilapia, about as far taxonomically as a chicken and a crocodile, we should assume that no intervention transfers, even if basic biological principles are conserved. 

Most cost-effectiveness analyses (CEAs) look at one intervention in isolation, or compare one intervention against another. That level of analysis is useful, but it cannot see the landscape as a whole. As the industry fragments across more species, covering the same share of the industry requires more distinct interventions. This is a structural overhead cost that grows with fragmentation regardless of how many individuals any one given intervention reaches. 

To make this argument, I use a unit of analysis that captures when two groups of animals actually require independent interventions, and when they don't. That unit is a target population. This post develops that framework, applies it empirically to global aquaculture data from 2000 to 2023, and draws out what the results mean for movement strategy.

What is a target population?

A target population is the group of animals an intervention can actually benefit: those who share both the welfare problem the intervention addresses and the operational context that makes it feasible. If welfare problems or operational contexts differ enough that the intervention no longer applies or no longer meaningfully reduces suffering, those animals constitute a distinct target population requiring independent assessment. 

Species distance is one dimension

What causes welfare problems and operational contexts to differ? The further apart two species are taxonomically, the less they share in common physically (salmon's torpedo-shaped body vs tilapia's deep, laterally compressed form), and behaviorally (salmon migrate thousands of kilometers; tilapia establish small territories). These differences affect what welfare problems exist and what interventions are likely to transfer.

Species distance can be underestimated as a boundary because welfare problems appear to be universally shared (e.g. low oxygen is bad, handling causes fear and distress). This intuition is reasonable but misleading. The threshold at which a problem causes harm differs across species, and establishing where that threshold lies requires empirical work. For example, lowering stocking density is used as a way to decrease crowding stress. While this principle is generally true, schooling species like milkfish prefer proximity, while low densities cause aggression in Atlantic salmon by allowing territorial behaviors that higher densities suppress.^[1] This shifting goldilocks zone appears in dissolved oxygen, feed composition, and water quality parameters more broadly.

A potentially more serious problem is that research on these parameters optimize for production rather than welfare per se. Growth rate and survival are not the same as welfare outcomes. This has clearly happened in terrestrial systems (Frankenchicken), but is less recognized as an issue in aquaculture. ^[2] Shared biology establishes that a problem matters across species but does not necessarily reduce the work required to understand how it matters to the species. 

Table 1. Evolutionary distance comparisons for aquatic and terrestrial farmed species

Other dimensions matter too

Species is not the only source of divergence. The actual unit of analysis is a tuple: (Species, Production System, Procedures & Technology, Local Context). As diversity increases along any of these dimensions, the number of distinct target populations tends to increase sometimes multiplicatively, though not all theoretical combinations exist in practice.

Table 2: Dimensions along which target populations diverge

Example: sea lice in Atlantic Salmon

An intervention that works for one combination of species, production system, technology, and local context can fail entirely for another not only because the welfare problem is different, but because the solution pathway is. Take for example, the issue of sea lice in Atlantic salmon. The viable interventions and effectively the number of individuals any intervention can reach, diverge based on operational context: 

Table 3: How operational constraints fragment the target population (sea lice interventions in Atlantic salmon)

This framework establishes when species, production system, technology, or local context diverge enough to require independent interventions. The answer to how many target populations exist is entirely intervention-dependent (a stunning intervention and a stocking density standard will carve the industry into different populations). I use species and species-country combinations as tractable proxies. This is a simplification that likely understates true fragmentation but is sufficient to show its direction and scale.

How many target populations are there in aquaculture?

Using FAO data from 2000-2023, and species and species-country combinations as proxies for target populations, here's what the industry looks like. ^[3]

Figure 1. The 22 species comprising 85% of global farmed aquatic vertebrates (2023), compared to terrestrial production at scale. Bubble size is proportional to number of individuals. Colors indicate taxonomic order-- 8 orders are present within the 85%. The remaining 15% spans 259 additional species not shown. 

If we factor in country contexts, the global aquaculture industry currently operates across 1337 active species-country combinations. Table 4 compares this fragmentation to chicken production. 

Table 4.

This fragmentation is not evenly distributed. To measure it at the country level, I use Effective Number of Species (ENS).  ENS is a measure of diversity that is calculated by exponentiating the Shannon index, which converts it into a more intuitively interpretable metric. It accounts for both how many species are farmed and how evenly production is distributed across them. ^[4]

Table 5. Species diversity patterns across representative countries among the top 20 aquaculture-producing countries (excluding Iran). Global ENS is 27.3

Taken together, these numbers point to a structural challenge for direct welfare interventions. An organization deploying a welfare intervention for one species reaches, in expectation, roughly 1/ENS of the farmed aquatic vertebrates in a given country (assuming no additional constraints from production systems or local context). Globally, with an ENS of 27, a single species-level intervention reaches roughly 4% of farmed aquatic vertebrates as a biological ceiling. This is radically different from terrestrial agriculture, where 89% of all farmed vertebrates are chickens, housed largely either as layer hens in battery cages or broiler chickens.

Trends in species diversity over time

Global Trends

From 2000 to 2023, the number of aquatic vertebrates farmed globally increased along three different measures of diversity: ^[5]

• Total species: 169 → 281
• Species comprising top 85% of production: 14 → 22
• Effective Number of Species: 16 → 27 

Country Trends

Among the top 20 aquaculture-producing nations, ENS trajectories (2021-2023) diverge: 10 countries show diversifying trends, 4 show consolidating trends, and 6 exhibit volatile patterns with no clear direction. 

• Strongly Diversifying: ENS increased by >0.1
• Weakly Diversifying: ENS increased by 0 to 0.1
• Weakly Consolidating: ENS decreased by 0 to -0.1
• Strongly Consolidating: ENS decreased by <-0.1 

Table 6. ENS trajectories of the 20 highest producing countries of vertebrates (2021-2023). ^[6]

Fragmentation is a portfolio problem, not an intervention problem

As the industry fragments across more species, covering the same share of it requires more distinct interventions. This is a portfolio problem: the movement's total coverage is a function not just of how well any single intervention performs, but of how many independent interventions the movement can sustain in parallel. Evaluating cost-effectiveness at the level of a single intervention cannot see this. A well-designed tilapia intervention reaches more animals today than it would have in 2000, because the number of farmed Nile tilapia increased from 3 billion individuals to 17 billion over that period. By that measure, fragmentation may not appear to matter. Instead of asking what one intervention can return, this analysis asks what it costs the movement to cover the industry as a whole.

Between 2000 and 2023, the aquaculture industry grew 3.4 times in individual count, but the number of species comprising 85% of the industry grew from 14 to 22, a 57% increase. Each of the 8 new species requires independent context specific welfare research, field studies, and implementation infrastructure that must take place before an animal in that population is meaningfully helped. 

Do outcome-based standards sidestep this cost?

Some regulatory instruments appear to sidestep this cost. Outcome-based standards like minimum dissolved oxygen thresholds, and prohibitions like banning CO₂ stunning, don't prescribe how producers must act. They shift the burden of finding a solution onto producers. But this relocates R&D costs rather than eliminating them, and it makes welfare outcomes hard to track (particularly where producers resist scrutiny of whether their chosen alternatives are actually welfare-positive). When chemical sea lice treatments were restricted for Atlantic salmon, farms shifted to thermal delousing and cleaner fish, both of which introduced serious welfare harms (thermolicers carry up to 19% mortality risk, and cleaner fish face significant welfare costs of their own). It seems possible that suffering from delousing efforts could have outweighed the suffering from the lice burden itself. 

Identifying those harms required new research cycles and vast monitoring to detect, and this happened in the species with the most developed welfare infrastructure in aquaculture. Where that infrastructure doesn't exist, the consequences of transferred interventions will be harder to detect. Assuming transferability without empirical validation means accepting outcomes that could be null, or worse than the status quo. 

What this means for the movement

Neither outcome-based standards nor prohibitions structurally resolve the costs that a diversifying industry creates. Within any given target population, a well-researched and implemented intervention can still meaningfully reduce suffering. But when evaluated at the portfolio level, fragmentation has made the task measurably more expensive per animal over time, with the trajectory pointing toward further difficulty for at least half of the top 20 producing countries.

Recommendations

This analysis shows three things: fragmentation in aquaculture is structurally important for animal welfare, it has increased substantially over the past two decades, and for a majority of major producing countries the trajectory points toward further diversification. Taken together, this constitutes a structural constraint on the cost-effectiveness of direct welfare interventions as a whole that is worsening over time.

1. Monitor the landscape. Species composition is only one dimension driving heterogeneity in welfare problems. Production systems and procedures are diversifying too, and these compound to increase the industry's overall number of target populations. Monitoring should therefore track not only what is being farmed but how, with what systems, and with what methods. I recommend treating this as a live strategic input for investment timing including earlier R&D investment in species expected to consolidate or scale significantly, and potentially avoiding investment in species whose production is likely to decline.
2. Explore preventative approaches. The movement has already demonstrated it can shape which species enter industrial production. Octopus farming bans in multiple jurisdictions prevented an entire taxonomic group from scaling, suggesting broader preventative strategies may be worth exploring. Two directions worth investigating:
   • Increasing the barrier to entry for novel species. EU regulations already require environmental risk assessment before novel species can be introduced for human consumption. A welfare-specific analogue doesn't yet exist, though EURCAW-Aqua's existing mandate to evaluate welfare concerns for novel species suggests the mechanism could be extended.
   • Intervening early in a candidate species' trajectory, while production is small and momentum limited, is likely cheaper than waiting until the species is entrenched. Given what this analysis shows about fragmentation's effect on direct interventions, species currently at the margins of production are worth monitoring, particularly if their trajectory suggests future scale.
3. Take the long-term portfolio question seriously. The structural trajectory identified here (more species means more target populations) suggests the movement should periodically reassess the balance between direct intervention and alternative approaches. At some threshold of fragmentation, demand reduction or other strategies may offer better expected returns.

Author Notes

I’ve also done work together with Shloka Janapaty thinking about how species diversification will change over time. This will be shared in a follow up post!

This work was supported by the EA Animal Welfare Fund. Views expressed are my own. 

I used AI assistance in drafting, editing, and writing the analysis code, which I supervised and reviewed. 

Thank you to the following individuals for their time and feedback on this work: Lauri Torgerson-White, Becca Franks, Sagar Shah, Hannah McKay, Ben Stevenson, Haven King-Nobles, Jeff Sebo, Evan Boyar, and Constance Li.

1. ^{^}

   See here for a discussion on the challenges of finding a "golden" stocking density.

2. ^{^}

   For Atlantic salmon, the fish species with arguably the richest welfare research, establishing a welfare indicator toolkit required evaluating over 90 candidate indicators, differentiated by production system and life stage. 

3. ^{^}

   See here for methods and data.

^{^}

For example, although Norway farms 6 different species, their production is dominated by Atlantic salmon, so their ENS is 1.4. Functionally their production is so monolithic that we can consider it as if they are farming 1.4 equally abundant species. Conversely, China has an ENS of 16.5, meaning their production fragments across populations equivalent to farming 16.5 species of equal size. Based on the ENS, this suggests a welfare intervention targeting one species could reach approximately 6% (1/16.5) of China's farmed animals on average, compared to approximately 70% (1/1.4) in Norway on average.

^{^}

The diversity reported here is the most conservative scenario. ~20% of tonnage reported to FAO has no species specificity. The conservative scenario allocates these individuals to the dominant known species within the most specific available taxonomic group, searching first within the reporting country and year, then the country's full production history, then the broader region, and finally globally. This minimizes diversity by concentrating unknowns into already-dominant species.

^{^}

 Countries marked with an asterisk show consistent directional trends across NEI allocation scenarios but differ in the magnitude of change, indicating moderate sensitivity to NEI allocation assumptions. See methods for more details.

Show all footnotes