Systematic Review of Multi-Species Models in Fisheries: Key Features and Current Trends

Abstract

In the context of ecosystem-based fisheries management (EBFM), multi-species models offer a potential alternative to traditional single-species models for managing key species, particularly in mixed-fishery settings. These models account for interactions between different species, providing a more holistic approach to fisheries compared to traditional single-species management. There is currently no comprehensive list or recent analysis of the diverse methods used to account for species interactions in fisheries worldwide. We conducted a systematic review to objectively present the current multi-species models used in fisheries. The systematic search identified 86 multi-species models, which were then evaluated to assess their similarities. Employing a clustering analysis, three distinct groups were identified: extensions of single-species/dynamic multi-species models, aggregated ecosystem models, and end-to-end/coupled and hybrid models. The first group was among the most diverse, owing to their ability to integrate biological components, while maintaining an intermediate level of complexity. The second group, primarily defined by the EwE method, features an aggregated biomass pool structure incorporating biological components and environmental effects. The third cluster featured the most complex models, which included a comprehensive representation of size and age structure, the ability to incorporate biological components and environmental effects, as well as spatial representation. The application of these methods is primarily concentrated on small pelagic and demersal species from North America and Europe. This analysis provides a comprehensive guide for stakeholders on the development and use of multi-species models, considering data constraints and regional contexts.

1. Introduction

Traditionally, marine fisheries management has relied heavily on single-species population models to guide the development of measures aimed at ensuring the sustainability of fisheries resources [1,2,3]. These models typically focus on key population dynamics, including growth rate, recruitment, selectivity, biomass, natural mortality, and fishing mortality. The interplay of these factors determines the fluctuations in the population size of target species over time.

Beyond single-species population models, fisheries management increasingly employs more complex models, such as ecosystem models. These models aim to account for broader ecological processes, including biological interactions between species, which is a core aspect of multi-species models.

An ecosystem model (EM) is a framework that incorporates ecosystem components (e.g., species or functional groups) and ecological processes (e.g., predation, perturbations, and dispersal), using data to make inferences about specific elements or the entire ecosystem. These models, often visualized as networks, help predict outcomes of complex ecological interactions and are widely used in marine and terrestrial management [4]. Those EMs that include and focus on ecological processes associated with biological interactions (such as predator–prey dynamics and competition) may be called multi-species models (MSM), which are designed to investigate the responses between different species within an ecosystem or community. According to the previous literature, all MSMs are EMs, but not all EMs are MSMs. For simplicity, the term “MSM” was used to refer to the models analyzed in this work, as all the EMs considered addressed biological interactions between species or functional groups.

MSMs are capable of modeling trophic interactions, competition, predation, and other ecological dynamics [5,6]. While competition may be less relevant than predation, as changes in competition can be offset in the trophic structure, predation directly affects certain species through predator–prey interactions [7]. Other interactions modeled in MSMs include sources affecting natural mortality (M), dependence on predator/prey abundance on population growth [8,9], and reproductive success [10]. These factors are closely related to the predator–prey dynamics within the ecosystem.

The considerations incorporated in MSMs enable its use to advance towards Ecosystem-Based Fisheries Management (EBFM) [11]. EBFM embodies a holistic approach, emphasizing the interconnected and complex relationships between humans and marine resources [12]. However, a significant challenge for EBFM lies in integrating more ecosystem realism into the prevalent single-species management paradigm [13,14]. Single-species assessments and management remain the standard, while multi-species models are not routinely employed or widely accepted within the fisheries management context [15]. Progressing with these tools not only contributes to EBFM, but also aligns with the principles of the Ecosystem Approach to Fisheries (EAF) [11].

MSMs are primarily used for strategic advice in fisheries management, informing long-term decisions and bracketing a range of viable options. Tactical advice, such as setting quotas, is less common due to concerns about increased uncertainty resulting from the inclusion of additional ecosystem processes, and limited knowledge on marine ecosystem dynamics might result in elevated levels of uncertainty [1,2,7]. However, the use of MSMs has expanded to various other purposes. In the US Atlantic and the Irish Sea, these models have been employed to adjust single-species management reference points, considering multispecies interactions and enhancing ecosystem understanding [13,16,17,18]. Additionally, MSMs have been used to provide better estimates of natural mortality (M) for single-species stock assessments by accounting for changes in predation over time [19,20,21,22].

The complexity of MSMs can be defined by their features. These features include the model’s ability to fit age- or size/stage-composition data, the inclusion of biological interactions, the technical interactions (like bycatch and discards), the number of species or functional groups accounted for, and the incorporation of effects from biological components (e.g., abundance of plankton, birds, and mammals) and environmental effects (e.g., changes in salinity and temperature), whether these processes are explicitly modeled or not. The nature of the assumed functional response (i.e., the relationship between prey density and per-predator–prey consumption) can also lead to different population dynamics outcomes [23,24,25,26]. Another important aspect is the extent of spatial representation (implicit or explicit) in the model. Explicit spatial modeling allows understanding how individual species utilize various habitat niches and how spatial interactions within and across species influence management performance [27].

Given the inherent complexity and interconnectivity of ecosystems, as well as the limitations of single-species models in capturing interspecific interactions (e.g., predation and competition), several studies have previously been conducted to analyze the development of MSMs. Plaganyi [3] described the features of MSMs and categorized them into different groups. Skern-Mauritzen et al. [28] identified the contexts where ecosystem drivers were used to provide tactical advice for fish stocks managed by various bodies worldwide. Marshall et al. [29] focused their discussion on the inclusion of ecosystem considerations in US fisheries stock assessments. Perryman et al. [30] conducted a comprehensive search for ecosystem models that accounted for biological interactions, with a particular emphasis on those employed for Management Strategy Evaluation (MSE) [10,31].

While these studies have advanced our understanding of these tools, a systematic assessment of how MSMs are related to the geographical areas in which they are applied remains lacking. This research aims to document those applied to answer questions related to fisheries. Additionally, we explore the connections between the features of these models and the specific regions where they were applied. Keys aspects include their use by large marine ecosystems (LMEs), identification of prevalent species within LMEs, and keys features of each model category.

2. Material and Methods

2.1. PRISMA Method

A literature review on MSMs was conducted using Elsevier’s Scopus database (www.scopus.com, accessed on 23 November 2022). The review methodology followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [32,33]. PRISMA is a standardized framework that helps ensure clear and complete reporting of research methods and findings for systematic reviews and meta-analyses. The PRISMA approach involves the use of a checklist and a flow diagram to promote transparent and comprehensive presentation of the research process and results.

By adhering to the PRISMA guidelines, the systematic review of the MSMs literature aimed to provide a synthesis of the existing knowledge, and was focused on peer-reviewed journal articles published between 2010 and November 2022.

As Geary et al. [4] noted, in the past decade, the increased availability of comprehensive data has facilitated the implementation of complex ecosystem modeling, leading to a surge in the demand for these modeling approaches. Additionally, Plaganyi [3] extensively reviewed pre-2010 methods for assessing species interactions and their impact on marine fisheries management, Espinoza-Tenorio et al. [34] examined how these methods could be effectively be used in EBFM, and Fulton [35] discussed the more complex methods used in existing modeling frameworks, also pre-2010. These factors contributed to the decision to use 2010 as the starting year for the systematic review.

The search query in the Scopus database was formulated as follows: TITLE-ABS-KEY ((model*) AND (stock OR assessment OR fisher*) AND (multi-specie* OR multispecie* OR “multi species” OR multi-stock* OR multistock* OR “multi stock”)).

The subject areas included were “Agricultural and Biological Sciences”, “Environmental Science”, and “Earth and Planetary Sciences”. In this research, the explicit use of the term “multi-species” was crucial to identify MSMs that incorporated biological interactions, and those articles focused on multispecies aquaculture, trophic analysis, or that did not meet the established eligibility criteria, were excluded from the analysis.

The scope of this review is broad, aiming to encompass all articles that have utilized MSMs in the context of fisheries. It is important to note that the search strategy and database limitations may have resulted in the exclusion of some relevant articles from the literature. Additionally, while some applications of MSMs in fisheries may be reported solely in the grey literature, these sources were not considered for inclusion in order to maintain a focus on peer-reviewed evidence and simplify the analysis.

2.2. Eligibility Criteria

The selection and inclusion of scientific literature was guided by the following eligibility criteria: (1) the case study should incorporate one or more MSMs that specifically address questions related to fisheries, particularly those concerning management advice, (2) the case study must be related to marine fisheries, (3) the MSM must address predator–prey and/or competition dynamics, and (4) the MSM in question must differ from other MSMs already included in the review in terms of its parameterization, geographic area, and target species.

The last criterion was established to avoid duplication of references. In cases where different articles applied the same MSM in the same scenario without major changes, only the oldest article was included in the review. These eligibility criteria ensured that the selected literature focused on methods that were specifically developed for marine fisheries, with a clear emphasis on trophic interactions and biological dynamics. This approach facilitated a comprehensive and non-redundant synthesis of the current state of research on the use of MSMs in this context.

2.3. Features of Multi-Species Models

The articles reviewed described at least one multi-species model (MSM) and its various features based on the models’ ability to include: age/size-structured data, biological components, environmental effects, technical interactions, number of species or functional groups, type of functional response, and spatial representation. Notably, the inclusion of biological interactions was a requirement for the MSM analyzed in this paper. A full description of the features and conditional statements can be found in Table S1.

We considered five out of the six categories of MSMs, following the criteria established by Plaganyi [3], Espinoza-Tenorio et al. [34], and O’Farrell et al. [36], and including the categories defined by Fulton [35]:

• Conceptual and qualitative models: Such methods focus on identifying specific, large-scale stressors and their effects on natural systems, rather than assessing a particular species or multiple species [37,38]. These models do not incorporate the precise quantitative estimates of the magnitude or strength of the species interactions. Instead, these models only account for the qualitative sign (+, −, 0) of the species interactions [39]. These types of models were outside the scope of the present review and were, therefore, not included.
• Extensions of single-species models (EXT): They focus on the dynamics of a single species, but include the effects of interactions with different species as fixed effects. These models can explicitly include predation mortality, although it is usually treated as another type of fishery rather than being estimated as part of natural mortality [7,40].
• Dynamic multi-species models (DYN): These models use a limited number of species or functional groups that are likely to have relevant interactions with the target species. They are built upon single-species theory to understand the dynamics of multi-species fisheries, but do not address the ecosystem as a whole. They can include diverse environmental variables depending on the scenario [41,42,43].
• Aggregated ecosystem models (AGG): These models consider all trophic levels (producers and consumers) in the ecosystem to explore energy flow among the levels. They include both top-down and bottom-up processes, which allows for the development of trade-off relationships between prey harvest and predator biomass [44]. The most representative example is Ecopath with Ecosim (EwE) [45,46] and its spatial form, Ecospace. Ecopath creates a mass-balanced snapshot, while Ecosim uses Ecopath parameters as initial conditions to produce time-dynamic simulations [45,47].
• End-to-end models (E2E): These models track nutrient flows through the ecosystem components, simulating annual cycles of nutrients and feeding into a representation of lower trophic levels, higher trophic levels, and even anthropogenic effects. E2E models can be coupled to different submodels, such as NEMURO [48,49,50]. The use of E2E models has been focused on strategic management, such as performing management strategy evaluations (MSEs) [15]. In this context, models such as Atlantis [51,52] can be employed as operating models to represent the impacts of fishing and other anthropogenic effects and capable of simulating the full trophic spectrum and considering physical, biochemical, and human components in a spatially resolved area.
• Coupled and hybrid model platforms (C&H): These models incorporate interactions by coupling or combining different types of model platforms. Unlike EwE and Atlantis, this framework is specifically designed for the coupling or combination of diverse model types [35]. Individual-based (IBM) and agent-based models fall within this category, employing multiple submodels to integrate the complexity of individual behavior, which can influence system dynamics. Models such as OSMOSE (Object-oriented Simulator of Marine Ecosystems Exploitation) [53] can incorporate IBM-based age-structured fish or predator population and trophic interaction models, biogeochemical plankton production models, hydrodynamic and environmental models, habitat models, representations of human activities, and potentially even representations of their social and economic drivers [35].

The current work does not engage in a comparison of technical details, complete data requirements for each model, or model validation. This level of detail would not be beneficial for subsequent comparisons, especially for more complex MSMs. Highly complex approaches involve trade-offs that cannot be extrapolated beyond the specific sample system. In contrast, models that sacrifice some complexity allow for extrapolation to other ecosystems and situations [54,55].

The geographical distribution of the categorized MSMs was mapped based on their use frequency across Large Marine Ecosystems (LMEs) around the globe. LMEs were developed by the US National Oceanic and Atmospheric Administration (NOAA) to further ecosystem-based management and are closely linked to Regional Fisheries Bodies [56]. By mapping the MSMs according to LMEs, we simplified the visualization of their worldwide distribution, considering substantial variations in area coverage for each MSM. The map was generated using the QGIS software (https://www.qgis.org, accessed on 29 August 2024) [57]. A comprehensive analysis was conducted to identify the key commercial marine species by examining their frequency within the MSMs found. This analysis was performed based on LMEs, with particular attention to those LMEs containing the highest number of MSMs.

2.4. Clustering of Multi-Species Models

A clustering analysis was conducted to identify the features of the MSMs that best align with model categorizations and assess their similarity. Given the mixed nature of the data (qualitative and quantitative), Gower distances were used in a partitioning around medoids (PAM) clustering algorithm. A dimensional reduction was performed using t-SNE (t-Distributed Stochastic Neighborhood Embedding). t-SNE (t-Distributed Stochastic Neighbor Embedding) allowed visualizing the clusters identified by PAM, revealing the associated categories for each group. Notably, t-SNE preserves local structure in the data, meaning points close together in the visualization are more similar than distant points, indicating MSMs with similar features.

In t-SNE, perplexity is a parameter which influences the embedding and interpretability of visualizations [58]. Low perplexity values focus on local structure, resulting in smaller, tightly packed clusters [59]. Conversely, high perplexity values capture more global structure, leading to larger, more spread-out clusters. The perplexity value was optimized to best represent the underlying data structure.

Considering prior knowledge of MSM categories and the extensive variability in their features due to their diverse applications, the “model category” variable was included in the analysis alongside MSMs features. This approach constrained the clustering analysis, allowing for an assessment of key features characterizing each known category and validating existing differences between categories.

The analyses were conducted using R version 4.2.1 [60] with Packages Tidyverse [61], Rtsne [62], cluster [63], and factoextra [64].

3. Results

3.1. Identification of Multi-Species Models

The systematic review search formula yielded a total of 787 articles. From these, only 78 articles met the eligibility criteria described and were fully reviewed (Figure 1). A total of 86 MSMs were identified across the 78 articles reviewed (

Table 1. Multi-species models according to their category, name, LME, location, and references (n = 86).

id	Model Name		Large Marine Ecosystem (LME)	Specific Location	Reference
Extension of singles-species models
1.	MSSCAA	Statistical catch-at-age model with predation	Gulf of Alaska	Gulf of Alaska	[65]
2.	PS-mod	Schaefer surplus production model modified	Indonesian Sea	Java sea, Rembang Regency	[66]
3.	PS-mod	Schaefer surplus production model modified	Indonesian Sea	Java sea, Pati Regency	[67]
4.	PS-mod	Schaefer surplus production model modified	Indonesian Sea	Java sea, Semarang city	[68]
5.	PPI-CSA	Predation pressure index for Collie–Sissenwine catch-survey analysis	Northeast U.S. Continental Shelf	Gulf of Maine	[69]
6.	SPM-SH	Steele–Henderson surplus production model with predation	Northeast U.S. Continental Shelf	Northwest Atlantic coastal ecosystem	[44]
Dynamic multi-species models
7.	SMOM	Spatial multispecies operating model	Antarctica	Scotia Sea	[70]
8.	CHS	Age-structured dynamic model	Baltic Sea	Baltic Sea	[71]
9.	EE MSM	Ecological–economic multispecies model	Baltic Sea	Central Baltic Sea	[72]
10.	SMS	Stochastic multispecies model	Baltic Sea	Baltic Sea	[73]
11.	MSI-SOM	Multispecies interaction stochastic operative model	Baltic Sea	Baltic Sea	[74]
12.	MSPM	Multispecies stock production model	Baltic Sea	Baltic Sea	[75]
13.	Gadget	Globally Applicable Area-Disaggregated General Ecosystem Toolbox	Baltic Sea	Baltic Sea	[75]
14.	Gadget	Globally Applicable Area-Disaggregated General Ecosystem Toolbox	Baltic Sea	Baltic Sea	[76]
15.	Gadget	Globally Applicable Area-Disaggregated General Ecosystem Toolbox	Barents Sea	Barents Sea	[42]
16.	ASPM	Model of intermediate complexity for ecosystem assessment	Benguela Current	Benguela	[77]
17.	MSM	Multispecies size-spectrum model	Celtic-Biscay Shelf	Celtic Sea, 7e-j	[78]
18.	LeMans	Length-based multispecies analysis by numerical simulation	Celtic-Biscay Shelf	Irish Sea	[79]
19.	MBD	Multispecies biomass dynamics model	East Bering Sea	Bering Sea	[80]
20.	MDD	Multispecies delay difference model	Eastern Bering Sea	Eastern Bering Sea	[81]
21.	MBD	Multispecies biomass dynamics model	Eastern Bering Sea	Eastern Bering Sea	[81]
22.	CEATTLE	Climate-enhanced, age-based model with temperature-specific trophic linkages and energetics	Eastern Bering Sea	Eastern Bering Sea	[41]
23.	MS-ASA	Multispecies age-structured assessment model	Gulf of Alaska	Gulf of Alaska	[82]
24.	MS-PROD	Multispecies production model	Gulf of Alaska	Gulf of Alaska	[83]
25.	MSASA	Multispecies age-structured assessment model	Gulf of Alaska	Gulf of Alaska	[84]
26.	MICE-in-space	Spatio-temporal model of intermediate complexity for ecosystem assessments	Gulf of Alaska	Gulf of Alaska	[85]
27.	CEATTLE	Climate-enhanced, age-based model with temperature-specific trophic linkages and energetics	Gulf of Alaska	Gulf of Alaska	[86]
28.	MSVPA	Multispecies virtual population analysis	Humboldt Current	Southern Chilean waters	[87]
29.	LB-MSM	Length-based multispecies fisheries model	Indonesian Sea	Wakatobi National Park	[88]
30.	Gadget	Globally applicable area-disaggregated general ecosystem toolbox	Labrador-Newfoundland	Flemish cap	[89]
31.	GadCap	Globally applicable area-disaggregated general ecosystem toolbox	Labrador-Newfoundland	Flemish cap	[90]
32.	MICE	Model of intermediate complexity for ecosystem assessment	Mediterranean Sea	Pomo pits	[91]
33.	mizer	Multispecies size-spectrum model	No LME	Central and eastern tropical Pacific sea	[92]
34.	MICE	Model of intermediate complexity for ecosystem assessment	No LME	Moorea Island	[93]
35.	SMS	Stochastic multispecies assessment model	North Sea	North Sea	[94]
36.	mizer	Multispecies size-spectrum model	North Sea	North Sea	[95]
37.	T-ONS	Trade-offs North Sea model	North Sea	North Sea	[96]
38.	FishSUMs	Size-structured multispecies model	North Sea	North Sea	[97]
39.	FishSUMs	Size-structured multispecies model	North Sea	North Sea	[98]
40.	LB-MSM	Length-based multispecies analysis by numerical simulation modified	North Sea	North Sea	[99]
41.	AS-MSM	Multispecies age-structured population model	Northeast U.S. Continental Shelf	Georges Bank	[100]
42.	MSVPA-X	Extended multispecies virtual population analysis	Northeast U.S. Continental Shelf	Northeast U.S. Coast	[101]
43.	MS-PROD	Multispecies production model	Northeast U.S. Continental Shelf	Georges Bank	[83]
44.	MSSCAA	Multispecies statistical catch-at-age model	Northeast U.S. Continental Shelf	Georges Bank	[102]
45.	LeMans	Length-based multispecies analysis by numerical simulation	Northeast U.S. Continental Shelf	Georges Bank	[103]
46.	MSSCAA	Multispecies statistical catch-at-age model	Northeast U.S. Continental Shelf	Northwest Atlantic coastal ecosystem	[44]
47.	mizer	Multispecies size-spectrum model	Southeastern Australian Shelf	Australian Southern and Eastern	[104]
48.	mizer	Multispecies size-spectrum model	Yellow Sea	Haizhou Bay	[105]
49.	MSSM	Multispecies size-spectrum model	Yellow Sea	North Yellow Sea	[106]
50.	MSSM	Multispecies size-spectrum model	Yellow Sea	North Yellow Sea	[107]
51.	MSSM	Multispecies size-spectrum model	Yellow Sea	Haizhou Bay	[108]
Aggregated ecosystem models
52.	EwE	Ecopath with Ecosim	Baltic Sea	Baltic Sea	[75]
53.	EwE	Ecopath with Ecosim	Canary current	Canary Islands, El Hierro	[109]
54.	EwE	Ecopath with Ecosim	Celtic-Biscay Shelf	Bay of Viscay	[110]
55.	nGoM Ecopath	Ecopath with Ecosim	Gulf of Mexico	Gulf of Mexico	[111]
56.	EwE	Ecopath with Ecosim	Insular Pacific-Hawaiian	Puakō, Hawaii	[112]
57.	EwE	Ecopath with Ecosim	Mediterranean Sea	Greek Ionian Sea	[113]
58.	EwE-Ecospace	Ecopath with Ecosim and Ecospace	Mediterranean Sea	Gulf of Gabes	[114]
59.	EwE	Ecopath with Ecosim	Mediterranean Sea	Thermaikos Gulf	[115]
60.	EwE	Ecopath with Ecosim	Northeast U.S. Continental Shelf	Northwest Atlantic coastal ecosystem	[44]
61.	EwE-MICE	Intermediate complexity Ecopath with Ecosim	Northeast U.S. Continental Shelf	Northwest Atlantic coastal ecosystem	[44]
62.	EwE	Ecopath with Ecosim	Somali Coastal Current	Chwaka Bay	[116]
63.	EwE	Ecopath with Ecosim	Somali Coastal Current	Gazi Bay, Kenya	[117]
64.	EwE	Ecopath with Ecosim	Southeast U.S. Continental Shelf	Core sound	[118]
65.	EwE	Ecopath with Ecosim	Sulu-Celebes Sea	Danajon Bank	[119]
End-to-end models
66.	Atlantis	Atlantis	California Current	California currents	[120]
67.	Atlantis	Atlantis	Iceland Shelf and Sea, Faroe Plateau and part of the Greenland Sea	Arctic and Atlantic waters (Iceland)	[121]
68.	Atlantis	Atlantis	New Zealand Shelf	Chatham Rise	[122]
69.	Atlantis	Atlantis	Northeast U.S. Continental Shelf	NEUS (Gulf of Maine to Cape Hatteras)	[123]
70.	Atlantis	Atlantis	Norwegian and Barents Sea	Nordic and Barrent Sea	[120]
71.	Atlantis-SE	Atlantis	South-eastern Australian Shelf	Southern Australian waters	[124]
72.	Atlantis	Atlantis	South-eastern Australian Shelf	Southern Australian waters	[125]
73.	Atlantis-RCC	Atlantis	South-eastern Australian Shelf	Southern, southwest and Eastern central Australian waters	[126]
Coupled and hybrid model platforms
74.	OSMOSE-Coupled	Coupled end-to-end model for the southern Benguela	Benguela Current	Benguela	[127]
75.	IBM-Coupled	Coupled end-to-end model for the California Current system	California Current	California current	[50]
76.	OSMOSE-Coupled	Coupled end-to-end model for the Eastern English Channel fish community	Celtic-Biscay Shelf	Eastern English Channel	[128]
77.	MSSM-Coupled	Coupled multispecies size spectrum model for the eastern Bering Sea	East Bering Sea	Eastern Bering Sea	[129]
78.	OSMOSE-SoG	Coupled end-to-end model for the Strait of Georgia	Gulf of Alaska	Strait of Georgia, Canada	[130]
79.	OSMOSE-PNCIMA	Coupled end-to-end model for the Pacific North Coast Integrated Management Area	Gulf of Alaska	PNCIMA off western Canada	[131]
80.	IBM	Spatially explicit individual-based model	Gulf of Mexico	Gulf of Mexico	[132]
81.	OSMOSE-WFS	Coupled end-to-end model for the West Florida Shelf	Gulf of Mexico	West Florida Shelf	[133]
82.	OSMOSE-GoG	Coupled end-to-end model for the Gulf of Gabes	Mediterranean Sea	Gulf of Gabes	[134]
83.	OSMOSE-Coupled	Coupled end-to-end model for the Mediterranean Sea	Mediterranean Sea	Mediterranean Sea	[135]
84.	OSMOSE-GoG	Coupled end-to-end model for the Gulf of Gabes	Mediterranean Sea	Gulf of Gabes	[136]
85.	OSMOSE-MED	Coupled end-to-end model for the Mediterranean Sea	Mediterranean Sea	Mediterranean Sea	[137]
86.	SS-DBEM	Size spectrum dynamic bio-climate envelope model	Somali Coastal Current	EEZs Kenya and Tanzania	[138]

). Table S2 presents the 787 articles identified in the study.

The primary reason for rejecting articles was the first eligibility criterion. Several studies utilized MSMs for purposes not directly related to fisheries, such as exploring trophic relationships between species or the effects of environmental factors on marine communities.

3.2. Categorization and Geographical Distribution of Multi-Species Models

The majority of the 86 MSMs identified were classified as DYN models, accounting for 45 cases. This category included size-spectra based models, such as mizer [139]. It is important to note that these size-spectra models are capable of accounting for all trophic levels and the flow of biomass between them. They are also able to simulate both bottom-up and top-down interactions, sharing similarities with the AGG and E2E model categories.

For the AGG category, 14 cases were found, with the EwE model being the dominant one. The C&H category had 13 cases, which were primarily OSMOSE coupled with other submodels.

There were six EXT models, which were related to modified surplus production models. Lastly, eight cases using E2E models were identified, all of which involved the use of the Atlantis model.

Figure 2 shows that a total of 25 LMEs contain at least one study case of using MSMs, out of the 66 existing LMEs worldwide. Notably, almost half of the 86 MSMs found in this review (49%, or 42 MSMs) have been used in just 5 of these LMEs.

The Northeast U.S. Continental Shelf LME had the highest number of MSMs, with 11 reported. The Baltic Sea, Gulf of Alaska, and Mediterranean Sea LMEs each had 8 MSMs, while the North Sea LME had 7 MSMs.

It is worth noting that out of the 86 MSMs listed, only 2 could not be associated with a georeferenced LME. Most of the LMEs were linked to more than one MSM. However, there were still 12 cases where only a single MSM was found within an LME.

The MSMs consistently focused on a specific group of commercially important species within particular LMEs. When examining the 5 LMEs responsible for nearly 50% of the MSMs, the most frequently included species were:

-

Northeast U.S.: Atlantic cod (Gadus morhua), Atlantic herring (Clupea harengus), and Atlantic menhaden (Brevoortia tyrannus).

-

Baltic Sea: Atlantic cod, Atlantic herring, and sprat (Sprattus sprattus).

-

Gulf of Alaska: Arrowtooth flounder (Atheresthes stomias), Pacific cod (Gadus macrocephalus), and pollock (Theragra chalcogramma).

-

Mediterranean Sea: European hake (Merluccius merluccius), sardine (Sardina pilchardus), and European anchovy (Engraulis encrasicolus).

-

North Sea: Haddock (Melanogrammus aeglefinus), Atlantic herring, and Atlantic cod.

3.3. Multi-Species Models’ Clustering

Based on the results obtained from categorizing and characterizing the 86 MSMs (Table S3), 3 clusters were identified using a PAM clustering algorithm. These clusters were subsequently visualized using t-SNE (Figure 3).

The first cluster was composed of 49 MSMs, predominantly characterized by DYM (43 MSMs) and EXT (6 MSMs). This cluster exhibited a detailed age/size structure representation, largely driven by the DYM category, which represented age/length structures for most groups, unlike the majority of the EXT models that did not. The most common functional response within this cluster was the “Holling type II”. This cluster was defined by an average of eight species or functional groups

The second cluster can be defined as an AGG cluster, composed of 14 AGG models. This cluster was characterized by aggregated biomass pool structure, which also incorporated both biological components and environmental effects. The dominant functional response within this cluster was driven by the “Foraging arena” hypothesis, which is closely associated with Ecopath with Ecosim (EwE) method. The average number of species or functional groups included in this cluster was 34.

The third cluster was composed of 24 MSMs, primarily represented by E2E and C&H models. These featured a full size/age structure representation, incorporated biological components and environmental effects, and included spatial representation. The dominant functional responses in this cluster were “Holling type II” and “Opportunistic Predation”, with the latter being closely associated with OSMOSE models. Similar to the second cluster, this third cluster had an average of 35 species or functional groups. Additionally, two DYN models, GadCap (id 31) and MICE (id 34), were also allocated to this cluster (see Table 1 for details).

4. Discussion

4.1. Regional Interest in Multi-Species Models

The models compiled in this study are associated with 25 of the 66 Large Marine Ecosystems (LMEs) identified globally, representing 38% of the total. The distribution of the MSMs is primarily concentrated in North America and Europe (Figure 2). The disproportionate distribution of fisheries publications can be attributed to the concentration of research output from a small number of established fisheries science centers, primarily located in Europe, North America, Australia, and Japan [140,141]. This phenomenon is further reinforced by inter-institutional collaborations, as demonstrated by Syed et al. [140]. Moreover, Dornelas et al. [142] reported that biodiversity time series data for North America, Northern Europe, and the Australian coast span approximately 90 years, which may be related to their economic power, research investment, and scientific capacity. Furthermore, the prevalence of English in academic publishing skews research representation, potentially overlooking important insights from non-English speaking regions [143].

In the MSMs reviewed, the choice of species is not only based on their capture volume or commercial value. Instead, it is also influenced by their historical significance in research. These model organisms are well-studied, allowing for more straightforward exploration of biological and ecological questions [144].

Several MSMs were initially developed in Australia, including Model of Intermediate Complexity for Ecosystem assessments [43] and Atlantis [145]. These models evolved from earlier iterations such as Minimally Realistic Models [146] and Integrated Generic Bay Ecosystem Model [51]. Additionally, multispecies Antarctic models were developed and implemented in South Africa [147]. While South Africa, Australia, and New Zealand have a rich history in developing and implementing fisheries methodologies and considering environmental change impacts, our findings indicate that, from 2010 onwards, the utilization of these models in these regions is comparatively lower than in North America or Europe.

4.1.1. North Sea

According to Pope [96], stakeholders in the North Sea were concerned with achieving Fishing Mortality at Maximum Sustainable Yield (Fmsy) in a multi-species context, as well as implementing the landings obligation (discard ban). In response to these concerns, MSMs have been developed over the years for this LME.

In addition to haddock, herring, and Atlantic cod, the MSMs found in the review (Table 1) also included other species that constitute a large portion of the landings in the North Sea, such as whiting (Merlangius merlangus), Norway pout (Trisopterus esmarkii), and sandeel (Ammodytes tobianus). It is worth noting that the North Sea LME is home to more than 100 stocks [148].

The difficult recovery of the Atlantic cod population over the years was also a key driver for the use and development of MSMs in this LME. These models aimed to address the complex interactions between species and the potential impacts of climate change on the ecosystem [94,97]. The development and application of MSMs in the North Sea LME reflect the stakeholders’ concerns about achieving sustainable fisheries management, while also accounting for the multi-species dynamics and the effects of environmental changes on the ecosystem.

4.1.2. Baltic Sea

The Baltic Sea is characterized by a well-documented interaction between three commercially important species: Atlantic cod, sprat, and herring. These species have overlapping distributions, particularly in ICES subdivisions 25–29 and 32, and together, they account for approximately 95% of the total catches in the region [149]. The Baltic Sea is known to be susceptible to environmental changes [150,151,152], which further contributes to the relevance and exploration of MSMs in this LME.

4.1.3. Gulf of Alaska

The pollock fishery is the largest groundfish fishery in the Gulf of Alaska (GoA) [65,153]. Pollock, along with Pacific herring (Clupea pallasii), are major prey species for Pacific cod and arrowtooth flounder [82], making them key components of the GoA ecosystem and fisheries [154,155].

In trawl fisheries, species such as Pacific halibut (Hippoglossus stenolepis) are designated as prohibited species catch, with limited quota allocations [156]. However, this management approach has not led to drastic decreases in catch limits for other fish species. Halibut is an important predator in the GoA ecosystem, particularly for older individuals that have already entered the commercial fishery [85,86,157].

4.1.4. Mediterranean Sea

Traditionally, fisheries management in the Mediterranean Sea (MED) has relied on effort control and technical measures, such as area restrictions, minimum landing sizes, and gear limitations, with no catch limits and technical interactions in many fisheries [158]. This approach, combined with the complexity of managing the 23 countries surrounding the MED and the multiple anthropogenic and environmental stressors negatively impacting fish stock populations [159,160], has created a critical situation due to the current state of overexploitation of several commercial fisheries [161,162,163].

European hake is the species most frequently encountered in the MSMs of this LME. This is because European hake has a wide spectrum of prey, with smaller individuals (less than 12–13 cm total length), exhibiting no particular preference towards certain taxa, while larger individuals specialize in more energy-rich and larger prey, such as sardines, European anchovies, or even small hakes [164,165,166].

In contrast, sardine and European anchovy are prey for multiple predators, sustaining higher trophic levels and making them subjects of study to evaluate trophic dynamics in different scenarios for the region [167,168].

4.1.5. Gulf of Maine

The Gulf of Maine (GoM), particularly the Georges Bank, was the most frequently mentioned area for the U.S. East Coast Large Marine Ecosystem in the reviewed models. In this region, Atlantic cod (Gadus morhua) preys on Atlantic herring (Clupea harengus) and juvenile Atlantic cod [100,169]. Meanwhile, Atlantic menhaden (Brevoortia tyrannus) is preyed upon by several species, including large fishes such as striped bass (Morone saxatilis) [170] and Atlantic bluefin tuna (Thunnus thynnus) [171], as well as marine mammals like the bottlenose dolphin (Tursiops truncatus) [172].

In addition to supporting commercial fisheries, some of these predator species have significant value in recreational fisheries, ecotourism, and cultural heritage [44]. Organizations like the Atlantic States Marine Fisheries Commission (ASMFC) and the Northwest Atlantic Fisheries Organization (NAFO) have set ecosystem management objectives for certain species in this LME [44,173]. The research by Chagaris et al. [18] played a key role in establishing ecological reference points (ERPs) for Atlantic menhaden. Using a simplified ecosystem model, they linked menhaden dynamics with key predators like striped bass. These ERPs were adopted by the Atlantic Menhaden Management Board, supporting ecosystem-based fishery management.

4.2. Multi-Species Model Groups

The groups of models shown in Figure 3 incorporate different sets of features. Even within the same MSM category, there can be significant variation in the specific features that different MSMs utilize. In this review, only seven features were considered for each MSM. Note that MSMs GadCap (id 31; [89]) and MICE (id 34; [93]) were misclassified. This is confirmed by the negative silhouette widths (−0.307 and −0.067 respectively), suggesting their inappropriateness in the assigned clusters. Specifically, MICE (id 34) includes spatial representation, while GadCap (id 31) incorporates technical interactions, biological components, and environmental effects—all of which are typical of the E2E/C&H cluster.

GadCap, a multispecies model for the Flemish cap [89], was built using Gadget [42]. It aligns with the MICE [43] philosophy, balancing simplicity and detail while focusing on key species and interactions for effective fisheries management. Notably, GadCap utilizes adaptable tools integrating data for predictive analysis and can be refined as knowledge or data availability improves.

Similarly, the model developed by Lagarde et al. [93] adheres to the MICE principles. It evaluates the impact of different fishery management strategies on coral reef ecosystems, considering major ecological complexities like habitat dynamics, trophic interactions between fish, and environmental perturbations. This issue could be corrected modifying the number of clusters, but the current number was the one with lowest number of wrong observations; besides, there is no point in increasing the number of clusters if the objective is to simplify the grouping of these models. The first cluster of EXT/DYN models is the most diverse in terms of the features analyzed, with a high number of MSMs (49) accounted for in this cluster. Regarding the representation of age/size structured data, most of the EXT models do not have population structure, as they are built upon production models like those developed by Fogarty et al. [174] and Gaichas et al. [83]. These models require biological parameters and are structured according to time series of aggregated landings for a group of species.

On the other hand, dynamic multi-species models such as mizer and FishSums rely on size-structured data, while others are based on more traditional methods (SCA or VPA) depending on the availability of age-structured data. The lack of spatial representation also defines this cluster, which is logical since these MSMs are less complex and less data-intensive than the AGG, E2E, and C&H models.

The AGG cluster is clearly defined, as almost all of the models within it are EwE models, and their features do not vary significantly between them. EwE is one of the most widely used models for evaluating ecosystem impacts. In this review, only one article was found to have used EwE in conjunction with the Ecospace module (id 58; [114]). Instead of population-structured data, EwE utilizes trophically linked biomass pools that represent major ecosystem functional groups. These functional groups can be further divided into stages or “stanzas” for species that exhibit complex trophic ontogeny [45,46]. Craig and Link [175] showed several case studies demonstrating EwE’s utility in tactical decision making, such as managing nutrient runoff for coral reef preservation and evaluating sustainable yields in mixed fisheries. These applications show EwE potential to transition from strategic to tactical management advice in fisheries.

Despite the above, these models are heavily influenced by the input diet matrices, which are linked to the trophic levels considered in the simulations. Predation is a crucial component of ecosystem dynamics, and predator–prey functional responses can be used to represent predation in ecosystem models [176]. One key feature of this model cluster is the functional response based on foraging arena theory [177]. In this approach, prey biomass is formulated as flowing between invulnerable and vulnerable states, reflecting the spatial and temporal constraints on prey and predator activities [178]. This allows the model to represent diverse functional response forms, such as linear, hyperbolic, or sigmoidal.

Choosing the appropriate functional response model is crucial, as an erroneous choice can lead to either underestimating or overestimating the risk of stock impairment, rather than capturing the more realistic stock dynamics. This is because the model output is highly sensitive to the functional response used [65,101,179].

Complex ecosystem models, such as Atlantis, can incorporate multiple functional response models, allowing each species to be represented by the most suitable functional response based on predator behavior [120,180]. This flexibility in the functional response modeling helps to better capture the underlying predator–prey interactions. However, this is contingent upon the complexity of the model, as conducting a comprehensive sensitivity analysis to evaluate alternative forms of functional responses is challenging.

In the third cluster (E2E/C&H), the models typically include a more comprehensive representation of the ecosystem, incorporating environmental factors, nutrients, phytoplankton, and potentially zooplankton or filter-feeding groups. These models are generally more complex and detailed, with a large number of parameters (over 100 in some cases) that need to be calibrated. They also require extensive age-structured and spatially explicit data for the key species [35]. Due to this complexity, these models are primarily used in MSE frameworks, where the focus is not on finding a single optimal solution, but rather on evaluating the performance of alternative management approaches under a range of uncertainties [35].

Given the potential impact of climate change, which could reduce global fish biomass by up to 30% by 2100 under the RCP8.5 scenario [181], several of the revised MSMs [41,50,71,75,80,86,93,94,101,120,121,126,129,134,135,138] have addressed the effects of climate variability on the productivity of marine species. These models employ various approaches depending on the MSM category. A top-down approach, often associated with fisheries management, examines how variables such as temperature-dependent predation, foraging, metabolic, and growth rates influence the development of the target specie. This approach is commonly implemented in dynamic models or end-to-end models such as GADGET, MSSCAA, or Atlantis. Conversely, a bottom-up approach focuses on understanding how environmental drivers affect populations. This method is more prevalent in coupled and hybrid platforms, utilizing several submodels to capture the complexity of these interactions.

Ecosystem models face uncertainty from both parameters estimates and structural components, increasing with model complexity, but simpler models also have limitations like key process omission and inappropriate scales. In dynamic multi-species models, data-intensive requirements can be reduced by focusing on a smaller ecosystem subset and making assumptions about data representativeness, albeit at the cost of increased model uncertainty. Additionally, uncertainty representation can be improved through penalized likelihood functions, which estimate parameters and provide confidence intervals when fitting data. EwE static-flow limitations are offset by its ability to address parameter uncertainty. Atlantis’ exploitation model interacts with the biological model, providing ‘simulated data’ with realistic levels of measurement uncertainty in the form of bias and variance to the sampling and assessment submodel.

Species interaction rates can be inferred from diet data, which is arguably a critical input for MSMs. Diet data are required for EwE, Atlantis and also for dynamic multi-species models such as MSVPA. As mentioned earlier, OSMOSE model assumes that predation is an opportunistic process, where predators will feed on any available prey.

Validating ecosystem models is inherently challenging because creating ideal experimental setups for these complex models often requires a wide range of treatment and control scenarios, as well as significant temporal and geographic scope [4]. Given the inherent complexity of MSM category, conducting a comprehensive sensitivity analysis to assess alternative configurations may prove to be overly intricate. Specifically, cross-validating model predictions across multiple modeling approaches can be a potent tool for mitigating uncertainty in decision-making processes [182].

A potential limitation of this review is the use of only one search engine (Scopus). Other articles related to the topic could have been left out of the review, as they may not have been indexed in the Scopus database. The use of Scopus offers several practical advantages, including its comprehensive content coverage across a wide range of disciplines, indexing a larger number of unique journals compared to Web of Science, as well as its intuitive user interface that facilitates efficient literature searching and retrieval [183,184]. Future reviews on this topic will consider using other databases, such as Web of Science or Google Scholar.

5. Conclusions

Managing mixed fisheries effectively remains a significant challenge. While the need for improved approaches is clear, most frequent applications of multispecies and ecosystem approaches have been observed in the Northern Hemisphere, particularly North America and western Europe. These initiatives, driven by the goal of advancing towards the next generation of ecosystem-based fisheries management [185], have focused on culturally and commercially important fisheries, particularly those targeting small pelagic and demersal species. This geographic and species focus underscores the need to acknowledge the most significant predator–prey interactions, especially those involving small pelagic fish.

The available multi-species models have been developed and applied under diverse criteria and needs, often driven by their data requirements, particularly those related to predation (e.g., consumption rates, stomach contents, and spatial overlap). The three clusters identified in this work align with insights from previous studies [3,34,35,36] in terms of complexity scaling, where the more data-intensive MSMs are separated from the less complex models.

This work shows a sustained development and application of MSMs incorporating predation dynamics for fishery management and research. Our results serve as a comprehensive guide for stakeholders, decision makers, and scientists to review existing work related to multi-species models. It offers insights into their adoption in fisheries management and research, considering data availability and regional development.

As is well known, a balance between complexity and uncertainty is required, considering data constraints and the needs of stakeholders. The same is true for MSMs, especially for question-oriented MSMs (MICE), since their use is very broad and their features can shift, as was seen in this work, even when biological interactions are involved.