Studying human–nature relationships through a network lens: A systematic review

1Leibniz Centre for Tropical Marine Research (ZMT), Bremen, Germany; 2University of Bremen, artec Sustainability Research Center, Bremen, Germany; 3Institute of Environmental Systems Research (IUSF), Osnabrueck University, Osnabrueck, Germany; 4Department of Biology and Ecology of Fishes, LeibnizInstitute of Freshwater Ecology and Inland Fisheries (IGB), Berlin, Germany; 5Posgrado en Ciencias del Mar y Limnología, Universidad Nacional Autónoma de México, Unidad Académica Mazatlán, Mazatlan, Mexico; 6German Federal Agency for Nature Conservation, Isle of Vilm, Putbus, Germany and 7Department of Biosciences, Centre for Ecological and Evolutionary Synthesis (CEES), University of Oslo, Oslo, Norway


| INTRODUC TI ON
Human activities and environmental change alter ecological systems (Halpern et al., 2008;Kappel, 2005), often with unpredictable consequences for delivering ecosystem services essential to societal well-being and development world-wide (MEA, 2005;Rockström et al., 2009). This is aggravated by cumulative impacts of anthropogenic stressors progressing with a continuously growing human population (Giakoumi et al., 2015;Halpern et al., 2015Halpern et al., , 2019. Understanding and predicting the consequences of environmental change and/or of management intervention has increasingly received scientific interest in recent years. Modelling approaches to predict responses of ecosystems to anthropogenic influences include mechanistic models, statistical models and machine learning approaches (Schuwirth et al., 2019), although the underlying model assumptions and uncertainties are often enough not specified, impeding the effective use for decision-making (Gregr & Chan, 2015). In addition, ecological and social systems interact in complex and dynamic ways at different geographical and temporal scales forming interlinked social-ecological systems (SES; Berkes, Colding, & Folke, 2003;Folke et al., 2010;Hughes, Carpenter, Rockström, Scheffer, & Walker, 2013).
The dynamic processes in the ecological and social (sub-)systems and their interlinkages produce outcomes at the larger SES level, which, in turn, influence and change the subsystems and their components (Brondizio, Ostrom, & Young, 2009;Ostrom, 2009). This is why understanding these complex interdependencies in SESs is critical to developing effective strategies for steering towards more sustainable and resilient human-nature relationships (Bodin, Robins, et al., 2016;Yletyinen, Hentati-Sundberg, Blenckner, & Bodin, 2018).
Network analysis (NA) is a powerful tool for investigating relational structures and processes (Janssen et al., 2006). The elements of interest are represented as nodes (also called vertices or actors) and the interaction(s) between them as links (also called edges, ties or arcs; see Figure 1). The analysis of relational structure has a long research tradition in various academic fields, including social sciences (i.e. social network analysis [SNA], cf. Burt, 1992;Emirbayer & Goodwin, 1994;Granovetter, 1973;Moreno, 1934-but consider Freeman, 1996, for an overview of even earlier steps towards SNA) and ecology (ecological network analysis [ENA], cf. Fath et al., 2019;Fath & Patten, 1999;Finn, 1976;Hannon, 1973;Ulanowicz, 1980Ulanowicz, , 1983. Social systems may be understood as networks in which nodes commonly represent individual persons or collective actors (Lusher, Koskinen, & Robins, 2013;Wasserman & Faust, 1994).
In recent years, NA has increasingly been used for studying diverse phenomena at the human-nature interface, for example, to explore the impact of natural resource use (by humans) on a specific ecological system (e.g. Baird, Fath, Ulanowicz, Asmus, & Asmus, 2009;Bodin & Prell, 2011;Fath, Scharler, Ulanowicz, & Hannon, 2007;Heymans, Coll, Libralato, Morissette, & Christensen, 2014;Rocchi, Scotti, Micheli, & Bodini, 2017;Villasante et al., 2016). Drawing on graph theory, NA is a systematic approach to organizing, categorizing and quantifying the various components of a predefined system based on empirical data (Lusher et al., 2013;Wasserman & Faust, 1994), and is-in theory-well-suited to investigate relationships and system structures within complex SESs (Janssen et al., 2006). In the context of environmental governance, network analysis is often derived from the assumption that network structures help explaining the K E Y W O R D S complexity, connectivity, natural resource management, network analysis, social-ecological modelling, social-ecological network, social-ecological system F I G U R E 1 Example of the graphical representation of a directed network showing relationships (called links, edges, arcs or ties) between five hypothetical entities (called nodes, vertices or actors; figure first published in Kluger, Kochalski, Müller, Gorris, & Romagnoni, 2015) effectiveness of governance output. For example, more centralized networks were discussed to facilitate effective and quick responses to high-risk environmental governance challenges such as invasive species eradication (Lubell, Jasny, & Hastings, 2016), however seem to be less robust to changing socio-political circumstances (Gorris & Glaser, in press). To give some examples, Marín, Gelcich, Castilla, and Berkes (2012) related co-management performance to social networks (reflecting social capital) of smallscale fishers involved in co-management regimes of benthic resources in Chile. Partelow and Nelson (2020) used NA to study the evolution of self-organized institutions for the governance of the dive tourism sector on an Indonesian island, analysing how the social collaborative networks of dive shop owners stipulated collective action for the emergence of adaptive environmental governance. A steadily growing number of studies uses the notion of social-ecological networks (SENs; cf. Sayles et al., 2019). However, the integrated analysis of social and ecological system components remains challenging (Bodin, Robins, et al., 2016;Cumming, Bodin, Ernstson, & Elmqvist, 2010), because of, for instance, the different conceptualizations of nodes and links in the social and ecological realm  and differing terminology and scopes .
In this research, we review studies carried out in the context of natural resource management that use network analysis of empirical data with the aim to (a) offer a systematic overview of existing approaches to conceptualize and analyse human-nature relationships and (b) to examine their potentials and challenges for practical use in natural resource management. We use a typology of network studies based on the degree to which both realms (human, nature; represented by ecological and/or human (societal, institutional) actors, as well as possible links between them) are integrated in the network conceptualization. Subsequently, we conduct a structured, purposive review of existing network studies, and systematize the studies based on the typology. The specific focus of this effort lies on synthesizing how networks are conceptualized, what means of analysis are applied and in what environmental setting (i.e. ecosystem type) the studies are embedded in. After presenting the results, we then discuss the comparative challenges and potentials of the different network analytical approaches in the context of the ongoing scientific discourses on networks and natural resource management.
The study complements recent efforts that push methodological and theoretical advancements in SEN analysis with the aim to allow for more integrated network research (see e.g. Barnes et al., 2019;Bodin, Robins, et al., 2016;Bodin & Tengö, 2012;Sayles et al., 2019). The discussion of the results of our study, however, emphasizes also the benefits of the diversity of partially articulated (sensu Sayles et al., 2019) network conceptualizations and approaches for tackling various possible challenges for natural resource management in the context of complex human-nature relationships.
This diversity allows researchers to select an approach that is best suited for a particular case study or research context, and to address the specific questions at hand. Our article particularly aims to assist researchers interested in studying human-nature relationships through a network lens who are new to the topic, by adding an even broader view to the perspective than presented by the abovementioned authors, as to be able to choose the most suitable approach for their study by discussing potentials and challenges of the different approaches and show-casing previous studies.
Moreover, we take stock of the empirical network studies in the literature on natural resource management and discuss potentials and pathways for further research.

| T YPE S OF NE T WORK CON CEP TUALIZ ATIONS
Countless possibilities to conceptualize SES as networks exist and there are numerous ways to conceptualize links within and between social and ecological system components. For example, some may categorize a link as ecological if it entails the exchange of a natural resource (e.g. the selling of fish or sharing of wood; Baggio et al., 2016), or as social if it describes an anthropomorphized action (e.g. cohesion among dolphins, e.g. Wiszniewski, Lusseau, & Möller, 2010).
However, general archetypes of network conceptualizations can be derived from the degree to which social and ecological components and relationships are included. Following Bodin and Tengö (2012), a SES can be thought of as consisting of both social and ecological entities (i.e. nodes) representing the social and ecological SES subsystems. These two units may be connected by three types of links: (a) those that connect social nodes (SS links), (b) those that connect ecological nodes (EE links) and (c) links that connect social and ecological nodes (SE links; for a schematic representation see Figure 2). Hence, we define the different types of links based on the nodes they interrelate, not on the nature of the link itself. Acknowledging that there are other concepts, based on our definition, social-ecological edges only describe those links crossing the human-nature interface.
F I G U R E 2 Theoretical conceptualization of a directed network consisting of social (in yellow) and ecological (in green) nodes connected by three types of links: social-to-social (SS, in yellow), ecological-to-ecological (EE, in green) and social-ecological (SE, in grey) Consequently, three different archetypes of network conceptualizations emerge ( Figure 3): • Type I considers one type of nodes (either from the social or from the ecological realm) and one type of link (either SS or EE interactions). Since only one realm (social or ecological) is represented, SE links are not incorporated.
• Type II integrates two types of nodes (from both the social and from the ecological realm) and two types of links (SS and SE, or EE and SE links). Although the interaction of one realm with certain actors from the respectively other dimension is conceptualized, no further links between the nodes of that other realm are considered.
• Type III comprises two types of nodes (from both the social and from the ecological realm) and three types of links (SS, EE and SE) between these actors.
While Type I integrates only entities and their interactions from one realm (either the ecological or social dimension), Type II and III integrate   Table 1). Only peer-reviewed journal articles published in English until 31/12/2019 were considered, that is, excluding book chapters, editorial material, pure reviews and books. In total, the search retrieved 656 articles (Table 1). Full records of each search string were downloaded as individual BibTex files that were joined in the R environment (R Core Team, 2019) using the bibliometrix package (Aria & Cuccurullo, 2017). Graphics presenting results (i.e. Figures 5 and 6) were created using the r packages readxl (Wickham & Bryan, 2019) and ggplot2 (Wickham, 2009), while for Figure 6, additionally the package dplyr (Wickham, François, Henry, & Müller, 2019) was used.

| ME THODS
All articles were reviewed and included in the analysis if they (a) applied a network approach to empirical data and (b) focused on humannature relationships. For this, the notion of 'human-nature relationship' was defined in a broad sense, that is, studies were included if they explicitly aimed to improve management or governance of the natural environment (ecosystems, natural resources, SES, etc.), to understand human impacts on ecosystems, to analyse implications of ecosystem change on humans or offered explicit policy implications relevant for at least one of these three research areas. These criteria did not have to be explicitly stated within the study, but had to be identifiable for coders.
Since this study focused on network approaches based on graph theoretical foundations, studies using other theoretical concepts for studying networks (e.g. neural networks, Bayesian networks) were excluded.
Theoretical work was only considered if it contained tests of respective assumptions on empirical datasets. Following these criteria, 138 of the 656 studies initially encountered were classified as relevant ( Figure 4).
The selected articles were then reviewed with regard to the type of nodes and links, and how these were conceptualized, based on the described typology (Section 2, Figure 3). Moreover, the study system (terrestrial, freshwater, marine) was recorded, as well as other characteristics related to the case studies (e.g. country and continent, year of publication). The coding was first tested in a pilot classification of F I G U R E 3 A typology of network studies analysing human-nature relationships in social-ecological systems based on different levels of integration of the social (yellow) and/or ecological (green) realm. EE, ecologicalto-ecological links; SE, social-ecological links; SS, social-to-social links (for further conceptualization of links compare Section 2 and Figure 2) five articles done by all five authors, with codes per category emerging from the data itself. After having discussed the procedures and results of the pilot, all authors classified a comparable number of randomly assigned articles. Reliability and homogenization of coding between articles was achieved through cross-checking between authors and consensus-building discussions. The compiled data reviewed and analysed in this study have been made publicly available on the Zenodo Digital Repository . Type I study (1% of all Type I) used two networks with ecological and social nodes, respectively, while analysis was conducted separatedly.

| Quantitative overview
Since five studies used stylized networks for empirical analysis (i.e. they did not use a specific case study), 133 studies were classified

656
TA B L E 1 Search strings identified through several rounds of discussion among co-authors and then applied for the retrieval of literature from the ISI Web of Knowledge database. Peerreviewed journal articles published in English, that is, excluding book chapters, editorial material, pure reviews and books were considered, resulting in a total of 656 articles (Figure 4). Full records of each search string were downloaded as individual BibTex files that were joined in the R environment (R Core Team, 2019) using the bibliometrix package (Aria & Cuccurullo, 2017 Records assessed for eligibility: n = 656 for the system type in which the data were collected (Figure 6c). The majority of studies focused on land-based systems (64%, n = 85 of these 133 studies), followed by marine (24%, n = 32) and freshwater (9%, n = 12) ecosystems; only 3% (n = 4) of the studies included data from more than one system type ( Figure 6c).

| Type I network studies for SES research
Type I studies were defined to look at one node type, either from the social or from the ecological realm, and the links (either SS or EE) between them (compare Figures 2 and 3). The literature search identified 107 studies that corresponded to the Type I category (compare Figure 6a; Table 2). Most of these studies represented resource users or actors from the social realm involved in governing environmental problems as nodes. The idea of 'social capital' represented the most commonly used theoretical concept in the social setting and was applied, for instance, in the context of managing marine resources in a Kenyan fishing community (Bodin & Crona, 2008), or among fisher cooperatives in Chile to assess post-disaster trajectories (Marín, Bodin, Gelcich, & Crona, 2015). Especially the two aspects 'collaboration' and 'infor-    Sayles and Baggio (2017) output (Bodin, Sandström, & Crona, 2016). Another work described how segregation patterns in terms of information exchange emerging from ethnic clustering among fishers was correlated with environmental outcome related to shark bycatch rates (Barnes et al., 2016).
Studies from the ecological realm mainly analysed food webs (i.e. trophic links), for example focusing on the impact of fisheries on the marine food web (Rocchi et al., 2017), on the analysis of the role of keystone species, and/or the impact of invasive species on ecosystems (Ortiz, Rodriguez-Zaragoza, Hermosillo-Nuñez, & Jordán, 2015;Vasas & Jordán, 2006 (2015), as a third methodological example, constructed two separate networks comprising of social and ecological components, respectively, and then compared the findings of structural analysis with a spatial reference.
The latter approach could be understood as what Sayles et al. (2019) conceptually term a non-articulated SEN.

| Type II network studies for SES research
Type II networks were defined to integrate two different types of nodes, that is, from both the social and the ecological realm. Links thus represent either SS and SE edges, or EE and SE interactions, while no direct links between the nodes from the respectively other realm are incorporated (compare Figures 2 and 3). In all, 16 studies fell into this category (compare Figure 6a;  Outeiro, 2015) or ecological variables as defined by the SES framework of Ostrom (2009;e.g. Delgado-Serrano et al., 2015). Studies then looked at the interaction between resources and institutions in a resource management network, for example, through management actions (e.g. Alonso Roldán et al., 2015;Barfuss et al., 2017;Beilin, Reichelt, King, Long, & Cam, 2013;Leventon et al., 2017).
Other work studied the impact of fishing (i.e. with fishers representing social nodes) on food web dynamics (Dimitriadis et al., 2016;Levine et al., 2015). Haak, Fath, Forbes, Martin, and Pope (2017) Beilin et al., 2013). Only one study incorporated a temporal component into their analysis (Barfuss et al., 2017) while most other studies used network data to represent a status quo.
Showing the diversity of applications, the studies' results related to a variety of purposes including the aim to solve questions on collaboration between institutions in resource management, describe effects of actor interactions in networks, identify the most critical variables in a SES and management units to be targeted and enhance understanding of processes and interactions in SESs (see also Table 2).

| Type III network studies for SES research
Type III network studies were defined to consider two node types, that is, entities from both the social and the ecological realm, as well as links between social actors (SS), between ecological units (EE) and bridging the human-nature interface (SE; cf. Figure 3). In all, 15 studies fell into this category (cf. Figure 6a; Table 2

| Understanding human-nature interaction
In an ever-interconnected world, with multiple anthropogenic pressures driving environmental and resource degradation (Giakoumi et al., 2015;Halpern et al., 2008), it becomes imperative to study and understand these complex dynamics. Tackling these problems becomes an ever important endeavour if conceptualizing human-nature interactions in the context of complex SES (Berkes & Folke, 1998). While many different methodological approaches may exist to model SES, including mechanistic and statistical models (Schuwirth et al., 2019), modelling such complex systems requires to include non-linear feedbacks, adaptive processes, different time scales and spatial characteristics, as well as risks and uncertainties, while holding a clear knowledge of the key components of a specific problem (Levin et al., 2013). These aspects pose a challenge to most methodological approaches, but a basic first step should be the understanding of key elements and their interactions. This is why this study was based on two straightforward premises: (a) humans and nature are closely connected in SES (Ostrom, 2009) and (b) network analysis is exceptionally well-suited to understanding relational data (Wasserman & Faust, 1994). Our review indicated the increasing recognition of NA for describing human-nature relations since the mid-2000s, with both the annual number of studies and the level of integration continuously increasing ever since ( Figure 5).
Other studies exploring temporal patterns of research output related to a specific topic similarly reported increases in publication numbers in recent decades, which can indicate the progressing maturation of a scientific field (increased scientific and/or public interest leading to an intensification of research efforts) but also reflects general trends (acceleration of cooperation and publication processes). This exponential increase in publications makes systematic reviews and meta-analysis necessary tools to generate evidence-based practice and to resolve seemingly contradictory research outcomes in the respective fields (Gurevitch, Koricheva, Nakagawa, & Stewart, 2018), as well as to identify gaps in knowledge for the guidance of future research (Castellanos-Galindo et al., 2020). For example, systematic reviews have covered the topic of marine climate change research (Pedersen et al., 2016), sea grass ecology (Duarte, 1999), coral reef management (Comte & Pendleton, 2018)

| Potentials and challenges of the different network study approaches
Network analysis assumes, generally speaking, that there are properties and processes emerging from this global view that could not be seen if system components were studied individually, and separately. This consideration calls for the construction of networks as comprehensive as possible, that is, explicitly including in its analysis as many components of the network as possible.
Type III network studies-representing social and ecological nodes and the articulated links between both node types (i.e. SS, SE, EE links)-offer therefore a viable pathway for research focusing on these emergent properties at the SES level that can only be explained through the interactions of its parts (Ostrom, 2009). It is, without doubt, methodologically and conceptually interesting to construct Type III networks to advance theory development based on the direct integration of social and ecological entities in the network conceptualization Bodin et al., 2019).
The potential to include direct connections and feedback loops to capture the relationship within and across both realms is especially appealing. This allows to assess effects and repercussions for both realms simultaneously and theories from both the social and natural sciences can be integrated and tested using the same dataset and methodology (Bodin, Robins, et al., 2016;Guerrero et al., 2015). An in-depth discussion of the potential of the existing conceptualizations and analytical approaches of Type III networks is found in Sayles et al. (2019). However, it is important to emphasize that Type III network studies face also challenges, and a broad discussion of differently articulated SENs-as done in the present work-is of great value to compare the respective potential and limitations of different Types. For example, Type III networks require high amounts of data and high conceptual effort for constructing the network. The identification of relevant variables and network boundaries becomes even more challenging when one has to choose from a rich pool of possible social, ecological and socio-ecological variables (Bodin & Tengö, 2012 Section 4.4). However, we would argue that the application of Type III studies is-due to the abovementioned aspects-still rather academic.
In contrast, Type I and Type II network conceptualizations present other advantages while facing challenges too. Clearly, both types are not as elegant since they do not include social and ecological variables as 'equal' partners in the conceptual and mathematical formulation of the network. Yet, Type I approaches have the strong advantage of building on long research traditions in ecological (ENA) and social (SNA) network analyses. Numerous theoretical assumptions have been developed in these lines of research that can be operationalized for the context of understanding human-nature relationships (see e.g. Bodin & Crona, 2008;Bodin & Prell, 2011;Fath et al., 2007;Ulanowicz, 2004). This provides for enhanced trans- foundations of the Type I studies than the Type III studies can. For example, when human interaction with a food web is conceptualized as resource extraction (e.g. fishing), then social-to-ecological links are described in the same unit (biomass/energy flow) as ecological-to-ecological links among the biological actors, which facilitates quantitative analysis. Hence, as mentioned above, we argue that especially Type II network research offers an underexplored potential and ample scope for future studies.
In general, the specific research question should drive the decision of how much integration is actually necessary. Approaches that direct their attention to network complexity with, for example, multi-layer network coupling spatial patches, are well described; for example in Pilosof, Porter, Pascual, and Kéfi (2017), who describe multiple approaches for multi-layer networks. Nonetheless, their focus is chiefly on ecological aspects, which would classify all of the mentioned networks into our Type I, irrespective of their complexity.
Other multi-layer networks (e.g. ing networks to study social-ecological interdependencies, which is an especially difficult task for Type III network studies. Given the existing barriers and challenges for the development of Type III studies, a detailed quantitative analysis of one realm using a Type I or II approach might, in comparison, be more suitable and/or relevant to questions related to the impact evaluation of (expected) changes.

| Conceptual and methodological challenges for integrating social and ecological realms
One basic assumption to systematically revise and categorize were conceptualized as to cross the human-nature interface, that is, connecting one social and one ecological node (cf. Section 2).
Alternative approaches could engage, in contrast, with the nature of links. In this case, an ecological link might be anything that is derived from nature-be it physically (e.g. the movement of organisms, flow or sharing of biological resources) or theoretically (e.g. ecosystem services). This would imply that two social nodes (e.g. two fishers, or a retailer and a customer) could be connected by an ecological link (e.g. biomass of fish sold). Accordingly, two ecological vertices (e.g. individuals or groups of mammals) could be interconnected by a social action (e.g. the transfer of knowledge). This intriguing approach, however, was not followed in our work due to the difficulty to standardize the intrinsically subjective definitions of social edges, relating to different perspectives on anthropomorphized, human-centred concepts, for example, on what a social action entails. Since the purpose of this review was to present a general conceptualization aimed to engage with an as broad audience as possible, the discussion of these concepts would go beyond the scope of the present work.
In our literature review, most studies that fell into Types I and II analysed primarily social nodes (Figure 6b). This can likely be ex-  Table 1). Hence, although the graph theoretic foundations of network research do offer common ground for a joint scientific terminology (Janssen et al., 2006), there is room for improvement in terms of building a common language that could better integrate ecological research in SEN research. This 'disciplinary fragmentation' (sensu Gregr & Chan, 2015) is also common to other scientific topics. As an example, Gregr and Chan (2015) found in their review only a 5% overlap of papers being captured by different search strings related to (marine) ecosystem modelling tools, as well as a lack of cross-referencing. We would similarly argue that while adding the disciplinary foci and methods to SES research might help in deepening the understanding of behaviour and dynamics of single system components, true social-ecological approaches should, however, necessarily be of inter-and transdisciplinary nature.
Hence, a common language-to which we hope to have contributed with the present work-is an indispensable first step to structuring a joint research agenda.
In terms of methodology, a certain degree of overlap was identified with respect to topics and methods covered by the different Types. For example, analysis of the network topology is widely applied in all three categories, for example, the analysis of node removal was used to assess network-wide impact in Type I (e.g. Rocchi et al., 2017) but also Type III (e.g. Ortiz & Levins, 2017;Zador et al., 2017) cases.
Multi-layer network approaches were found for Type I (e.g. Prager & Pfeifer, 2015), Type II (e.g. Geier et al., 2019;Haak et al., 2017) and Type III (e.g. Sayles & Baggio, 2017) studies. Some authors even combine approaches, for example node removal on multi-layer networks (Baggio et al., 2016). In terms of topics addressed, the foodweb effects of human action (represented as fishing or harvesting resources) were assessed using Type II (e.g. Dimitriadis et al., 2016) and Type III but also using Type I (Rocchi et al., 2017) network approaches; while issues of spatial misfits were addressed using Type I (Easdale et al., 2016), Type II (Bergsten, Galafassi, & Bodin, 2014) or Type III (Bodin, Robins, et al., 2016;Ernstson et al., 2010). This exemplifies that several topics can be successfully approached by a range of network parametrization possibilities representing different levels of social-ecological integration, and also by using different analytical techniques. However, a particular challenge lies in the varying conceptualizations of nodes and links used in different NA studies because this complicates the comparison across cases, even though this theoretically represents one of the strengths of network analysis . Similarly, the scale at which the social and ecological units that are included as nodes are defined (e.g. local to global), further hampers useful cross-case comparisons.
Our results show that the majority of case studies, for which networks (of all Types I-III) were constructed, were from the terrestrial context ( Figure 6c). An increasing focus on marine SESs, however, is especially important considering that 40% of the world's population lives within 100 km of the coast with ever-growing demands for marine biotic resources and increasing pressure on coastal marine ecosystems (UN, 2017). Studies addressing marine systems were proportionally more represented in Types II and III than in Type I, possibly hinting at that these systems hold high potential for implementation of Type III studies, maybe because of the data availability, or because of the different characteristics of human-nature interaction at sea compared to terrestrial human-nature interactions.
However, a large number of marine food-web models (i.e. networks of Type I) already exists but has not been captured by our literature review, rendering this comparison difficult. Based on these food webs, the connection to the social realm can conceptually relatively easy be included using fishing or other types of resource use for representing the social dimension (i.e. for construction networks of Type II or III). An example for this is the recent study by Kluger, Scotti, Vivar, and Wolff (2019) presenting a multiplex, multi-layer SEN for a SES in which small-scale fisheries and aquaculture represent important contributions to local livelihoods. For this, social science data collection to study the dynamics within the fisheries value chain were combined with existing food webs (based on Kluger et al., 2016). In addition to the importance of an increased focus on coastal marine SES, the comparison across a wide range of different system types (terrestrial, marine, freshwater) holds strong potential to identify common relational features, but has not been opted for in the reviewed studies.

| CON CLUS I ON S AND OUTLOOK
In an increasingly interconnected world, the understanding of  Yletyinen for her constructive feedback on an early version of the manuscript, as well as two anonymous reviewers and the editor for their constructive critiques that have all helped to improve the article. Open access funding enabled and organized by ProjektDEAL.

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to declare. substantially to the writing process.

DATA AVA I L A B I L I T Y S TAT E M E N T
The compiled data reviewed and analysed in this study have been made publicly available on the Zenodo Digital Repository through