Recent studies have provided a better understanding of the structure of mutualistic and antagonistic interactions in crop, non-crop and other habitats around the world. Collectively, this work has provided new insights into a number of ecological and evolutionary problems (e.g., community assembly and trait evolution) and has also indicated how to harness ecosystem services; mostly pollination and natural pest-control (Supplementary Table S1). In agro-ecosystems, the challenge ahead is to describe these systems in ways that the full suite of important direct and indirect interactions that occur within the landscape are incorporated (Hutchinson et al., 2019). Indeed, current assessments of changes in farm management overlook important processes derived from interacting species, such as changes in trait distributions at the species-level may be compensated for at higher levels of biological organisation (Ma et al., 2019). Furthermore, a wide range of other interactions that may impact the design of agro-ecosystems have been less well studied, including spillover of pollination from non-crop to crop plant species and beneficial interactions between soil and plant communities. Although we can continue to obtain insights from studying these interactions in isolation, theory predicts that just a few links connecting elements may markedly change the dynamics of wider systems (Watts and Strogatz, 1998).

Conceptually, merging ecological networks of different interaction types is state-of-the-art. In an agricultural context, multiple direct and indirect interaction types with crops, both above and below-ground, can be quantified (Fig. 1). The simultaneous quantification of the multiple ecosystem services provided by biodiversity can provide insights on: (i) trade-offs in crop and non-crop management practices; and (ii) the often-unexpected dynamical consequences of coupling ecological interactions (e.g., emergent effects, indirect effects which transcend interaction networks and plasticity in the role of species within multilayer networks); both of which facilitate the task of improving sustainable yields. Pocock et al. (2012) were the first to show the usefulness of a ‘network of ecological networks’ at a farm-scale by analysing how different interconnected groups of organisms respond to species loss. This work enhanced our understanding of the robustness and resilience of combined groups of plants and animals to environmental change, hence identifying vulnerable groups and demonstrating novel methods to restore ecosystem functioning based in identifying important plants and habitats.

Networks at the field or habitat scale are inherently nested within a wider landscape of complex interactions (Evans et al., 2013). These networks connect habitats at a range of scales, with connections resulting either from the movement of individual organisms in space (i.e., the utilisation of multiple habitats by the same organism or long-distance dispersal among populations), by the use of different habitats by different populations of the same species, or by the transfer and flux of materials and energy across the landscape through existing networks of species (Massol and Petit, 2013). Research on meta-communities shows that, due to the inherent heterogeneity of ecological systems across space, the network of networks affects the response of both the overall network and its subnetworks to perturbations (Holyoak et al., 2005).

Several attempts have been made to understand the interactions between networks at the landscape scale in agriculture. These studies have looked at the shared interactions between habitats in order to understand the potential benefits of, for example, shared parasitoids of pest herbivores in both crop and non-crop habitats (Derocles et al., 2014). They draw on how concepts such as apparent competition, spillover effects, and top-down/bottom-up effects proliferate into crop systems, potentially improving yields. For instance, the potential of top-down effects, otherwise known as trophic cascades, to regulate pests may be achieved by manipulating habitat heterogeneity. Habitat heterogeneity causes changes in the feeding behaviour of generalist predators, leading to the presence of predators feeding on a greater number of prey taxa (Staudacher et al., 2018). Generalist predators, in turn, are known to shift ecosystem states through exerting predation pressure on pest species (e.g., fruit-eating birds in orchards), generating an increased productivity of plants in natural systems and crops (Shave et al., 2018).

There are a range of methods for linking and analysing these spatial networks in agricultural regions. The concept of meta-networks, an extension of previous meta-population, community and ecosystem theory, has recently been developed to link previously isolated habitat-specific networks (Emer et al., 2018). Furthermore, spatial networks including ecological, social and management systems, have been proposed (Dee et al., 2017). This field is developing rapidly, yet further advances are required before we fully understand how networks at intermediate, landscape scales interact with one another to affect ecological processes (Poisot et al., 2015). Initial studies constructing and investigating networks at broad geographical scales (e.g., Great Britain; Redhead et al., 2018) demonstrate the potential for constructing spatially connected networks at broad scales using combinations of methods, both new and existing. We explore some of the new developments and advances in this area of network ecology in subsequent sections (see Connecting the agricultural landscape using networks).

Ecological networks are yet to be fully explored at the global scale, but there are efforts to collate international datasets to help answer macro-ecological questions. Nevertheless, advances in network science more generally are increasingly being applied to agricultural systems data, in particular the global trade of goods, services (Anderson, 2010) and ecosystem services (Silva et al., 2021). An appreciation of multiple scales is particularly important in agricultural systems, especially considering that the movement and trade of agricultural goods, as well as current and future global environmental changes and biological variation at broad scales, influence the agricultural processes occurring at smaller spatial scales (e.g., landscapes and fields). These studies are particularly important when considering the risks presented by economic and environmental shocks (Challinor et al., 2016). For example, an analysis of networks of trade in cereal crops from 1986 to 2013 showed that three entangled, time-invariant sub-networks were responsible for the majority of trade, but there were also transient subnetworks that displayed exponential growth over intermediate timescales (Dupas et al., 2019). Transient structures, although contributing relatively little to the total trade of cereals, may present a buffer against perturbations or shocks in the short-term. Little, however, is known about the resilience of these networks — a substantial gap in our understanding of agricultural networks and their ability to respond to future scenarios (but see Puma et al., 2015). Attempts have also been made to link together agricultural trade and social networks. For example, a triadic analysis framework successfully classified countries based on their level of connectivity to trade partners and indicated a distinct development trajectory associated with an increase in the levels of global agricultural imports and exports for a country (Shutters and Muneepeerakul, 2012).

Much can be learned from other disciplines where network approaches have been used to examine a range of global phenomena, including banking systems (May and Arinaminpathy, 2010) and disease transmission (Pastor-Satorras et al., 2015). From an agricultural perspective it may be possible to link together processes at the global scale (i.e., trade and community) to those at field and landscape scales. For example, changes in the international trade of pesticides and fertilisers will affect access to these products by farmers, in turn enhancing or restricting their application at landscape and field scales, and thus altering the structure and function of ecological networks at local scales through the effects of either higher or lower levels of chemical application.

A greater volume of data on different agricultural networks is required across many different network types in agro-ecosystems, including ecological, sociological and economic. Here, we focus on ecological networks due to recent developments surrounding the collection of species-interaction data.

Comprehensive and detailed ecological networks can be constructed using a range of techniques. Yet recent advances in DNA-based network construction methods provide unprecedented opportunities to scale-up the construction of highly resolved ecological networks (Derocles et al., 2018; Evans et al., 2016; Srivathsan et al., 2021). These techniques are particularly applicable to network construction where interactions may be cryptic, infrequent, short-lived or difficult to observe (Vacher et al., 2016). Furthermore, they allow the inclusion of evolutionary information (e.g., phylogenetic data), that is particularly important in the context of using networks predictively for restoration purposes (Raimundo et al., 2018). Within agro-ecosystems, examples of potential applications of eco-evolutionary networks include: linking above- and below-ground networks (Toju and Baba, 2018), incorporating crop viruses and pathogens into ecological networks and investigating the role of intra-specific competition. Although DNA-based methods show great promise, with low per sample costs (<10 cents; Srivathsan et al., 2021) and the ability to process large numbers of samples rapidly, we recognise these methods have limitations (Cuff et al., 2022) and their accessibility varies geographically. A range of alternative methods for constructing networks exist and are suitable for creating agro-ecosystem networks. For a more comprehensive review of novel methods of constructing replicated species interaction networks over large spatial scales see sections in Windsor et al. (in revision).

An increase in the amount of available network data, combined with a focus on the functional implications of such interactions (e.g., understanding the wider role of soil microbiota in agro-ecosystems), will enable a better understanding of the responses of agricultural networks to natural or anthropogenic perturbations. Such knowledge provides options for management in extensive sustainable systems and/or building resilience into those food systems which appear at risk of negative effects (see below).

Inference networks, constructed from co-occurrence data and/or existing information on species interactions (e.g., Pocock et al., 2020), provide a potentially valuable resource for investigating networks over broad spatial and temporal scales. Indeed, this method of network construction is gaining momentum when community data is generated using environmental DNA (eDNA) methods, although co-occurrence is not necessarily evidence of ecological interactions (Blanchet et al., 2020). These methods have numerous applications, including reconstruction of historical species interaction networks, construction of networks in remote regions with poor interaction datasets and development of networks across broad spatial and temporal scales that allow for testing global environmental change hypotheses. Furthermore, the potential application to existing long-term biomonitoring projects is an area of interest (Bohan et al., 2017), particularly in the context of examining pest and beneficial insect interactions in agro-ecosystems (Petsopoulos et al., 2021).

We currently have limited knowledge regarding how biotic interactions between plants and below-ground micro-organisms influence ecosystem productivity and fluxes of matter and energy through trophic levels, limiting our ability to understand how agro-ecosystems respond to environmental change. Again, DNA-based methods combined with network analysis offer new ways to overcome some of the hitherto difficulties in studying more complex plant–microbe interactions (Bennett et al., 2019) and unearthing a range of ecosystem services that can be harnessed (e.g., arbuscular mycorrhizal fungi and other root symbionts for plant growth). Indeed, agricultural plants do not only rely on direct defences when attacked, but they can also recruit pest antagonists such as predators and parasitoids, both above and below-ground, mainly via the release of volatile organic compounds (i.e., indirect defences). However, from an ecological network perspective it is mechanistically unclear how changes in one trophic level (e.g., soil arthropods) affect interactions in another (e.g., insect pollinators), and vice versa (Fig. 2), nor the potential trade-offs for focal plants (e.g., crops). As such, we do not know what drives the multidirectional fluxes between the above- and below-ground components of the agricultural networks at multiple scales. This advance is required to understand how these components of the system interact to alter productivity and ecosystem services within the agricultural environment.

Using non-crop habitats, for example field margins, hedgerows, uncultivated land and woodland, to influence the network structure of plant–animal networks in arable agriculture is a common technique to enhance the provision of ecosystem services in some regions of the globe. Assessments and management interventions, however, do not often consider the simultaneous provisioning of multiple ecosystem services (Gaba et al., 2015). It is nevertheless important to consider the potential synergies and trade-offs associated with decisions surrounding agricultural systems, especially considering that different non-crop habitats enhance different ecosystem services in a context-dependent manner (Albrecht et al., 2020).

Developing methods that allow for an understanding and management of multiple ecosystem services is an important challenge that needs to be addressed. This advance can be achieved using an adaptation of current network methods for selecting optimal mixtures of organisms to maximise the abundance and/or species richness of interacting organisms. For example, it is possible to use machine learning algorithms applied to network data to identify species contributing to different ecosystem services across habitats and determine optimal mixes of species to provide simultaneous ecosystem service provision (Windsor et al., 2021).

Although studies on multilayer networks are still in their infancy (Hutchinson et al., 2019), in agro-ecosystems there are several clearly defined ‘layers’ that can be incorporated into multilayer network frameworks. These include, but are not limited to; plant–herbivore, herbivore–parasitoid, herbivore–predator, plant–frugivore, plant–human (e.g., removal of weeds, diversifying plant species in field margins) and herbivore–human (e.g., pesticide use, enhancing the abundance of natural enemies of pest herbivores). Through incorporating this array of network types into one framework, we suggest that it is possible to gain a better understanding of the mechanistic processes through which management decisions at a range of scales alter the ecosystem services provided by the ecological networks embedded within the agricultural landscape. In turn, this would allow for an improved design of management strategies, incorporating all potential knock-on effects across the ecosystem and allowing for maximum ecosystem service provision whilst minimising environmental harm.

The current toolbox available for analysing ecological multilayer networks is limited to a small but growing number of analyses, such as qualitative measures of robustness (Pilosof et al., 2017). There are also challenges associated with linking together layers with different link information in a meaningful manner. For example, it is often not appropriate to combine individual weighted networks into a single weighted multilayer network as the methods used to collect data do not lend themselves to direct comparisons. There are also issues with linking large numbers of disparate layers in a parsimonious and computationally efficient way that fits within current mathematical frameworks. To address this challenge, we believe a bottom-up, plant-focused approach in cropped systems provides a suitable starting point to investigate the role of different interactions within agro-ecosystems. By altering plant communities there are a range of consequences that can influence even those organisms that are not directly connected to them in the network (Scherber et al., 2010). Leading on from this starting point, it would be possible to link in- and off-crop habitats as well as above- and below-ground components of the agricultural networks using multilayer networks. Doing so will allow for the development of management techniques that enhance the provision of ecosystem services through making use of ecological processes such as apparent competition (see A roadmap for using networks predictively in agriculture).

Most studies to date have analysed the structure of ecological networks (or a ‘snapshot’ of species interactions in time) and ignore important temporal dynamics. Yet the properties of ecological networks can vary drastically at a range of spatial and temporal scales (Poisot et al., 2015). To deal with the dynamic nature of ecological systems, a novel suite of network models has recently been developed, incorporating additional information on the phylogeny and traits of the organisms that comprise a given ecological network (Raimundo et al., 2018). These adaptive network models incorporate potential functional or co-evolutionary changes that might manifest themselves in response to changes in the wider network, adding greater detail — e.g., information on changes in the abundance and traits of organisms over time (Derocles et al., 2018). The aim of such methods is to allow for an improved understanding of how systems respond to extinctions and reductions, which is in stark contrast to classical approaches of robustness with a binary response of persistence or extinction based on whether all previously observed links are removed, e.g., complete disconnection. A remaining challenge surrounding adaptive networks, however, is that of parametrisation data which remains limited. DNA based methods may present a way forward providing an opportunity to gain information on the evolutionary history of organisms and useful information on potential interaction rewiring (see Collecting more network data to enable a whole system understanding of agro-ecosystems).

Further developing adaptive network models within agricultural systems provides significant potential, especially in research investigating how human management may affect other aspects of the agricultural system. Examples of potential research include understanding the response of network structure and function to pesticide resistance, climate change and reductions in pollinator diversity, as well as developing precision fertiliser application and investigating the optimal combinations of non-crop plants in organic agricultural systems. By developing these methods of analysis, we can gain more information about the ecological networks and suitable management practices can be developed to make use of eco-evolutionary principles (Loeuille et al., 2013). Ultimately this will allow adaptive networks to be used predictively for targeted management outcomes (Raimundo et al., 2018).

Currently, it is not possible to accurately identify, quantify and understand how management decisions at different scales affect the structure and dynamics of ecological and social networks, and thus how these processes impact upon agricultural productivity, sustainability and resilience, across the surrounding landscape (Massol and Petit, 2013). As such, a meta-network framework is required to understand the interactions between agricultural networks at the landscape scale. To date there are few studies investigating spatial meta-networks, yet this concept can be applied widely, across multiple network types (i.e., social and ecological), to understand how management or perturbations affect the provision of different ecosystem services at the agricultural landscape scale. Meta-network theory could provide a suitable framework within which agricultural questions at the landscape scale can be answered. This has been shown in previous work investigating plant–bumble bee interactions in forestry systems, which demonstrated the variable effects of habitat patch size and flight distances on the potential effectiveness of conservation and restoration strategies (Devoto et al., 2014).

Substantial field and laboratory work is required to upscale network assessments in agricultural environments to allow for an understanding of how spatial variation and heterogeneity in social and ecological processes alters the ecosystem services provided across the landscape. These efforts have started with theoretical models (Loeuille et al., 2013), but empirical (field-based and experimental) studies are now required to drive further advances in our understanding.

Agricultural systems are a combination of social and ecological networks (Fig. 2), with top-down effects of management altering the nature of ecological systems and networks (Lescourret et al., 2015). Yet, until recently these two interlinked networks have been analysed in relative isolation. New studies have sought to combine these (and multiple other) types of networks to provide insights into how the interface between management and biodiversity alters ecosystem functioning (Felipe-Lucia et al., 2021). These efforts have included agricultural environments (Hutchinson et al., 2019), and theory surrounding socio-ecological networks is sufficiently mature for actionable interdisciplinary research (Bodin et al., 2017). Putting multilayer networks to work in the context of agriculture may improve our understanding of the mechanistic basis through which management decisions translate to changes in production, or other services provided by the agro-ecosystems.

Examples at different scales demonstrate the importance of combining social and ecological networks in agriculture to understand where changes in social networks impact ecological networks, and vice versa. At the landscape scale, agricultural environments are composed of different types of farmers, ranging from family farmers through to intensive commercial farm managers, who form nodes within a farmer-biota network. In this example network, farmers are linked together through their effects on the wider biodiversity of the landscape, i.e., agricultural management by one farmer may affect biota at the landscape scale and in turn have impacts on other farmers. An example of this would be the application of pesticides reducing pollinator diversity and thus reducing the probability of pollination across other farms and crops — even those where pesticides are not applied. Across continental and global scales, ecological and social networks at local scales can be linked together through national and international trade networks, as has been the case in recent work investigating virtual pollination trade networks (Silva et al., 2021). In these networks, effects on pollinators at local scales will alter the yields of pollinator-dependent crops and thus have social and economic consequences in other regions of the globe with which those crops are traded. Linking together systems at large scales is particularly important (see Furthering global agricultural network analyses to help enhance the resilience of food systems).

Using the discourses surrounding ecosystem services and natural capital appears to have consolidated efforts linking social and ecological networks across other ecosystems. Such application has allowed for the fusion of information on management strategies and ecosystem service provision with detailed socio-ecological networks (Dee et al., 2017). For example, recent work that integrated ecosystem services into coastal salt marsh food webs found indirect risks to services through species loss, and that vulnerability across services was predictable (Keyes et al., 2021). Using an ecological network, this study links together social aspects of systems: (i) threats, the anthropogenic stressors imposed on the ecosystem by human actors (e.g., climate change or over-exploitation of species); and (ii) ecosystem services, the human benefits derived from the natural world. We contend that a similar ‘threats-network-ecosystem services’ advance is a tangible starting point for operationalising work in agro-ecosystems (Fig. 2), building on a bottom-up species-interaction approach. Here, using robustness analyses, it would be possible to examine how agricultural management and/or threats to species interacting with plants propagate through the network to impact multiple ecosystem services, and thus socio-economic activities.

Going forward, a significant challenge for the integration of social and ecological networks surrounds the collection of appropriate data. Specifically, collecting and collating data of the right kind (i.e., weighted links with comparable or interactable units), at the correct resolution (e.g., seasonal management decisions and knowledge sharing by farmers). Many methods exist for generating social data to construct networks; however, current methods are qualitative (i.e., using ecosystem service as a node linked to species without a measure of the effects of a species on service provision) and/or collect data in an inappropriate spatial or temporal resolutions for integration with ecological networks (Felipe-Lucia et al., 2021).

There appear to be three options for combining social and ecological networks: (i) include social networks in the framing of ecological network questions (i.e., use research questions driven by social networks — for example decision making and/or land management at the landscape scale); (ii) reduce the resolution (i.e., level of detail) of both social and ecological networks and use an existing analytical framework (e.g., Bayesian networks and game theory have previously been applied to multi-agent ecosystem service modelling; Mulazzani et al., 2017); or (iii) develop a completely new method/framework to link social and ecological systems.

Determining the resilience of global agriculture to perturbations, shocks and disturbances is crucial and a policy priority (Challinor et al., 2016), especially considering the global nature of agricultural food supply and the potential for future food shortages (Silva et al., 2021). Although data exist at the global scale, for example Food and Agriculture Organisation (FAO) crop and livestock products data (Food and Agriculture Organization of the United Nations, 2018), there remains limited network-based assessments of the global trade of agricultural products and research opportunities (Puma et al., 2015). Current data provide a strong basis for potential analyses as there is detailed information on the value of annual imports and exports of different agricultural products, as well as the quantity of products moving between countries. Using these data, understanding the robustness of the trade network, for example if a global health crisis prevents exports (Aday and Aday, 2020) or if climate conditions reduce production of agricultural goods within a geographic region, is a fruitful place to begin. Furthermore, building network-based analyses into current global and regional agricultural monitoring schemes (e.g., Group on Earth Observations Global Agricultural Monitoring [GEOCLAM], Anomaly Hot Spots of Agricultural Production [ASAP], Global Information and Early Warning System [GIEWS] and Famine Early Warning Systems Network [FEWS NET]; Fritz et al., 2019), would further enhance the ability to provide an early-warning system for global food systems and promote enhanced food security.