Research

INTRODUCTION

Food systems are complex social-ecological systems (Umetsu et al. 2014, Oteros-Rozas et al. 2019). Among the different types of food systems, urban food systems are of critical importance given the ever-increasing share of people living in cities. In particular, secondary cities across Africa and Asia face unprecedented rates of urbanization (Talukder et al. 2015). Besides issues of food security (Food and Agriculture Organization 2009, van Zutphen et al. 2021), urban food systems face problems of overweight and obese populations and high occurrence of infectious and food-related diseases (Ramankutty et al. 2018, Rohr et al. 2019, Alarcon et al. 2021). Successful management of urban food systems inherently requires transdisciplinary research (Schwarz et al. 2021) and is vulnerable to collective action problems (Bodin 2017). In particular, knowledge gaps can undermine successful design and adoption of interventions and policies that improve food system function. To govern such a complex social-ecological system successfully, different actors’ knowledge and interests must be considered, as actors may differ in their perspectives on the larger system and their place within it (Müller et al. 2020). Actors with different perceptions of the problems at hand will not learn from each other, coordinate successfully, or be able to address long-term problems collaboratively (Bodin 2017).

Scientific representations of urban food systems are increasingly sophisticated (Candy et al. 2015, Raza et al. 2020). Conceptual models are frequently used for graphical representations of social-ecological systems (Robinson 2006, Gray et al. 2015, Schlüter et al. 2019). A conceptual model abstracts a real-world system into a set of system components and their relations that is adequate for research while providing a scaffold for data analysis. In food systems, their formal representation often follows the semi-linear structure of a food supply chain model, with system activities beginning at agricultural production, progressing through processing and transport, toward consumption and waste, with additional factors and feedback loops added to reflect the complexity of the system (Candy et al. 2015, Alarcon et al. 2021). In contrast, food choice models typically focus on cognitive or behavioral processes that influence individual consumers’ choices of specific foods (Aertsens et al. 2009, Li and Kamargianni 2020). Although food availability is considered as part of these frameworks, the larger food supply chain and food production is represented primarily as an input to consumer choices, limiting the set of choices available and analysis of wider system dynamics. A common problem of these alternative conceptual representations of food systems is their translation into practice. In practice, food system actors and decision-makers typically rely on experience and context more than data, even when they state a desire to use scientific information (Kirchhoff et al. 2013). This gap is partly due to a perception among actors that scientific knowledge is too abstract and not sufficiently grounded in practical reality (Liu et al. 2008). Food systems research, in particular, has been criticized for its focus on high-level abstraction and its neglect of the practical aspects of decision-making (Kroll et al. 2019).

The process for designing a conceptual model of an urban food system is often guided by the implicit knowledge and mental references of the scientist developing the model (Liu et al. 2008). As such, actors from different parts of the urban food system, such as farmers, food retailers, government, and consumers, are rarely consulted directly at this early stage. The result is that food system actors often do not feel represented in the scientifically generated conceptual model (Allen et al. 2013). Consequently, actors’ low use of urban food systems research can be traced, in part, to the design process used to create the conceptual models (van der Zee and van der Vorst 2005). Specifically, this design process rarely represents the diverse knowledge, perspectives, and goals of actors within the urban food system (Kroll 2016, Holdsworth and Landais 2019, Alarcon et al. 2021). Without equal representation of all actor groups in the model design process, models can emphasize one set of information, priorities, preferences, and values over others and thereby recommend a set of decisions that ultimately contribute to systematic inequality (Bielicki et al. 2019).

Food system actors typically have different information and goals, depending on their sector (Howarth and Monasterolo 2016), and tend to make decisions in sector-specific silos (White et al. 2017). To achieve a more representative view that acknowledges the different forms of the food system for different actors, participatory and co-design approaches are suited to represent and engage the different actors. These approaches come in many forms; a typology of co-design approaches is provided by Chambers et al. (2021). Different types of co-design approaches have their own strengths and weaknesses and link to different objectives. Some aim to identify the actors involved in a particular system (Prell et al. 2009), whereas others are targeted at collaborative and social learning, seeking to encourage stakeholders to address conflict, think systemically, communicate openly, and learn from one another as they work through challenging, and sometimes conflictual, situations (Reed et al. 2010, Walker and Daniels 2019). In urban food systems, a representative approach that incorporates the perspectives of many different actors is difficult to implement in practice, given the complexity of the system and extreme differences in power and autonomy of the different actors. Typically, multiple perspectives on such a system can be synthesized in a conceptual model by adding every element that is considered important by any actor to the model and analyzing the resulting cause-and-effect relationships. However, this approach can quickly create an unmanageably large list of system elements and interactions. In a complex and dynamic system such as an urban food system, it is a non-trivial task to assemble variable relations into a single systems model. Extensive (participatory) research efforts have established lists of variables that could be related, in some degree, to food in cities (Battersby et al. 2009, Battersby 2013, Greenberg 2017), but with vague and sometimes contradictory information about how these variables are related. For example, the roles of supermarkets and other value chain actors are often mentioned as being important, but the ways in which these actors affect the food system as a whole are often not clarified. The task of sorting through these variables and identifying the most important ones is fraught with difficulty, politics, and potential for misunderstanding.

An alternative is to focus on causal maps of the system structure that are generated and synthesized through collaborative approaches (Lynam et al. 2012, Manuel-Navarrete 2015). These approaches build on advances in participatory modeling (Gray et al. 2015, 2018, Voinov et al. 2018), which aim to engage the implicit and explicit knowledge of stakeholders to co-create formalized and shared representations of reality (Jordan et al. 2018). This approach may assist food system actors and decision-makers to view scientific models and simulations as credible reflections of real urban food systems (van der Zee and van der Vorst 2005, Liu et al. 2008). Moreover, understanding the diverse perceptions of food system actors can help to identify opportunities for intervention and innovation.

An example of such a study for an urban food system looked at the availability of healthy foods in low-income urban communities in Baltimore, Maryland, USA (Mui et al. 2019). Researchers interviewed a group of food retailers, residents, neighborhood organizations, and city agencies and represented their perspectives as causal loop diagrams showing the dynamics of healthy and unhealthy food availability in the neighborhood. These diagrams were synthesized into one overall diagram representing healthy and unhealthy eating cultures, individual and community wealth, and social stability and support.

A similar study used collaborative modeling to build a systems model of the determinants of inequality in healthy eating in Australia (Friel et al. 2017). Collaborative causal models synthesized actor perspectives across a range of scientific and other knowledge. These food system models achieved for Australia and Baltimore cannot easily be transferred to the context of African secondary cities. There, informal systems operate alongside formal systems, socioeconomic inequalities result in a very diverse range of consumers and producers, and inequalities in infrastructure and local government services affect how food is used and waste is managed (Battersby et al. 2016, Kroll 2016, Crush and Young 2019).

Our objective was to test a widely applicable participatory approach to collect and synthesize perspectives from a wide range of food system actors on the causal structure of food systems. The approach is designed to be generally applicable; we illustrate it here with an application to an African urban food system in the city of Worcester, South Africa. We collect a set of food system models drawn by actors at different parts of the food system, compare the models to understand similarities and differences in actors’ implicit knowledge of urban food systems, and synthesize the models to create an overall representation of the urban food system in an African secondary city.

METHODS

Case study

Worcester is a rapidly growing secondary city located in the agricultural production region of the Breede Valley in the Western Cape province of South Africa. Breede Valley contributes substantially to South Africa’s exports from agri-processing, particularly wine, table grapes, and, increasingly, citrus (Cullis et al. 2018, Seeliger et al. 2018, Partridge et al. 2021). Urban expansion threatens productive croplands around the city and competes with agriculture for energy and scarce water resources. Food supply chains are globalized and consolidated, and most of the food produced in the Breede Valley region is exported (Rumble 2013). The area includes a rapidly urbanizing population facing dual challenges of hunger and diet-related non-communicable diseases (Balogun et al. 2015, du Plessis et al. 2016). Patterns of socioeconomic segregation and the apartheid spatial legacy further complicate the planning process for food system resilience. Residents of low-income areas and informal settlements have historically been under-represented in decision-making and planning processes (Kroll 2016). In 2012, the area experienced a protracted labor dispute that continues to affect relationships in the food system (Webb 2017).

Recruitment and stakeholder selection

The actor group was selected to include actors focused on multiple different elements of the food system: production, distribution, retail, and consumption. A small but diverse sample was preferred for our purpose of focusing on mapping relationships between actors using diagrams. While other elements of the food system might also be important, we selected these core areas to test our approach. Potential actors were identified through project partners in the Breede Valley Municipality and through a presentation to the Ward Committees of the southern neighborhoods of the city. Although the perspectives of these actors were not necessarily generalizable to all actors in the food system, this method captured a variety of perspectives and, importantly, included perspectives from actors who were traditionally disadvantaged in decision-making processes for the local area. Methods and instruments were approved by the Humanities and Social Science Research Ethics Committee of the University of the Western Cape (HS19/6/15).

Actor elicitation

Actors were invited to participate in a single-session interview and structured elicitation. Interviews were done individually or in very small groups of mostly two or three people. This approach was chosen, instead of a focus group approach with a mixed group of stakeholders, to ensure that each type of actor could express their perspective without being influenced by the very unequal positions the actors have within the food system.

We relied on user-centric design tools to maximize the participants’ legitimacy in the interview. Each interview began with a profile of the participants as end users of the research, including a list of their key strengths and challenges in the current food system, and a normative vision-setting exercise asking what their vision of success would be for the year 2050. The purpose of these exercises was to establish the participants as legitimate subject matter experts, regardless of their formal role description or qualifications.

Research participants, particularly those who do not have a lot of experience with scientific research, may sometimes misinterpret a request for information from a scientist as an evaluation or test and seek to provide a “correct” answer (Brown and Lamb 2019). To counter this potential issue, we used broad open-ended prompts (e.g., “Tell me everything you think about that topic”) to elicit richer, more accurate, and more coherent information than closed-ended questions. We avoided questions preceded by, “Do you know”, or similar prompts, as they could be interpreted as evaluating knowledge or memory rather than gathering information. Where detail was lacking, we used the participants’ own words to invite them to elaborate further. In general, we adopted a “funnel” interviewing protocol, beginning with open-ended prompts, followed by increasingly narrow questions and clarifications.

We aimed to create strong interviewer-participant relationships, including an atmosphere of safety, validating participants’ disclosures, and maintaining an open and friendly dialogue (Knox and Burkard 2009). Interviews were conducted with the assumption that all perspectives were valid, even (perhaps especially) if they contradicted existing scientific models, and that participants, in general, were not purposefully malingering or providing false responses. Additionally, when designing the interview, we paid special attention to the role of participants’ cultural backgrounds and values, especially cultural differences in communication styles, physical distance norms, and nuances of eye contact and gestural communication (LaFrance and Mayo 1978, Mandal 2014).

We tried to use elicitation methods that did not require the participants to have high scientific or technical literacy. Various methods for expert elicitation of systems models exist, but they typically require the participants to have some familiarity with either probability theory (Choy et al. 2009), logical rule-sets (Johnson et al. 2010), number literacy (Martin et al. 2012), computer literacy (Morris et al. 2014), or language skills. Instead, we asked participants to draw their mental representation of the food system on a large piece of paper. This type of graphic elicitation lowers the emphasis on language and correct wording, and instead allows participants to specify for themselves the order and representation of information (Bagnoli 2009, Jones et al. 2014). The drawing task becomes an additional tool through which the food system map becomes the record of the discussion and a further source for elaboration and questioning (Varga‐Atkins and O’Brien 2009).

Typically the respondents drew the diagram together, but they were also offered the option to draw their own diagram if they wished. If they were uncomfortable with drawing and writing, participants could dictate to the researcher who acted as a transcriber (with the participant able to view the drawing as it was being constructed and provide ongoing instructions and corrections). In practice, we found that participants with higher scientific literacy (government officials) tended to prefer to draw their diagrams themselves, whereas participants with lower scientific literacy (consumer representatives) preferred to dictate.

We then used stickers to allocate causal importance to the variables that the participants had previously identified in their systems diagrams. The purpose of this method was to avoid one person dominating the outcome in the small groups. Stickers are used as a practical alternative to number-based elicitation in Bayesian prior elicitation (Johnson et al. 2010) and multiple criteria decision analysis (Pictet and Bollinger 2008). In practice, we found that the sticker method was quick to explain and easily understood by participants. Combining the drawing and sticker methods allowed us to elicit equally detailed and accurate information from all participants, regardless of their literacy or experience with scientific research.

Comparison

Although we focused on qualitative interpretation and comparison of the actors’ diagrams, to make a reproducible and structured interpretation, we used network analysis methods to compare actors’ diagrams of the urban food system more formally. In this way, we used quantitative methods to support a qualitative interpretation (Scherp 2013). Network comparison metrics included: density, transitivity, clustering, centralization, and motifs (Zimmerman et al. 2018). Network density gives an index of the overall strength of variable relations across all possible causal ties in the network, where high density indicates that most nodes in the system are directly connected with one another. We used normalized density to allow comparison of networks of different size. Centralization is the extent to which the network is centred around a single node or set of nodes. Transitivity is the overall probability that adjacent nodes in the network are connected, where high transitivity indicates that most nodes in the system are connected through indirect causal relations. Clustering gives an index of the tendency toward dense local network communities, where high clustering indicates the likely presence of sub-networks of important processes. Motifs are significant recurring sub-graphs or patterns that reflect important features or properties of the system (Ferguson et al. 2020). To account for differences in the number of nodes and connections between actors’ food system diagrams, we normalized the density, centralization, and transitivity metrics.

In addition to the network comparison methods listed above, we compared actors’ identification of important nodes in the food system using the following metrics: degree, centrality, and rank. Node degree indicates the number of nodes to which a node is connected. High degree indicates that a node directly influences, and is influenced by, many other components of the network. Node centrality (measured by eigenvector centrality; Csardi and Nepusz 2006) indicates the degree of direct and indirect influence a node has on the whole system. High centrality indicates that a node indirectly influences, and is influenced by, many other components of the network. All network analyses were carried out using the R package igraph (Csardi and Nepusz 2006).

Synthesis

Food system diagrams were synthesized with a method for evidence synthesis using directed acyclic graphs (DAGs; Ferguson et al. 2020). The method involves four stages: mapping, translation, and integration (synthesis and recombination; Fig. 1).

1. Mapping: In this step, the hand-drawn food system diagrams provided by actors were translated into digital implied graphs (IGs). The outputs of the actor elicitation sessions were a set of system diagrams representing the food system from the perspective of diverse actors within it. Key variables in the food system were drawn as circles or points in the hand-drawn diagrams, and causal relations were drawn as arrows connecting the circles. Diagrams were hand-drawn by workshop participants and facilitators and subsequently digitized using specialized software for DAGs (Textor et al. 2016). Mapping produced one IG per participant group, corresponding to the food system perspective of the actor group under consideration (accurate or otherwise). The diagram is referred to as “implied” because it represents the implied structure of the food system according to that actor’s perspective. The IG acted as a structural template for translation in the following step.
2. Translation: In this step, the different IGs were examined for causal logic and translated into DAGs. DAGs allow the conceptual models of the food system to be used for causal inference (Pearl 1995). We examined all variables (nodes) in the IG and their connections (edges). Each edge was assessed using a sequence of causal criteria (Hill 1965). If all causal criteria were met, the edge was retained in the DAG; if any criteria were not met, the edge was deleted. The criteria were: temporality (cause must precede effect), face validity (the relationship must be plausible), recourse to theory (formal theoretical support for the relationship), and a thought experiment: whether we would expect to see different values for the outcome variable based on changes in the exposure variable. The output of the translation stage was a completed DAG in which all directed edges had been assessed and the IG had been altered accordingly. One DAG was created per participant, ready for synthesis in the subsequent step.
3. Integration—Synthesis: Causal maps were integrated by combining the directed edges and corresponding nodes from all DAGs. The output of this stage was an integrated DAG (I-DAG) that displayed the combined nodes and edges of all participant maps. We also used the network metrics to guide the synthesis: In combining nodes and edges from all stakeholder networks, the nodes with higher centrality in the input networks will also have higher centrality in the synthesized network.
4. Integration—Recombination and division: Nodes represent variables that have been identified by participants, and participants may identify the same or similar constructs as slightly different things. For example, one participant may identify “tukshops” while another identifies “spaza shops”; these are different terms that both refer to a small, informal retail location. Nodes were recombined if they (1) were both categories of another concept (e.g., “health inspectors” and “health regulations” could both be categories of health governance), or (2) had identical input and output edges, including directions. Nodes were divided if they were named similarly in multiple networks but had completely different input and output edges (e.g., “supermarkets” referring to both consumers’ food choices in the supermarket and the retail entity).

We used an iterative process to identify potential recombinations, wherein the full I-DAG was evaluated according to a suite of tests designed to evaluate the validity of expert-elicited complex system model structures (Pitchforth and Mengersen 2013). In particular, we used the “convergent validity” criterion to compare elicited network structures with existing models: for example, we expected that supply-focused network structures could resemble existing supply chain models (e.g., Candy et al. 2015, Zimmerman et al. 2018), whereas demand-focused network structures should resemble existing models of consumer choice (e.g., Aertsens et al. 2009, Li and Kamargianni 2020). Stakeholder reflection on the synthesized model is another possibility for validation, but was outside the scope of our study.

RESULTS

Actor elicitation

Participants included representatives of local government (four men, two women), large-scale commercial farmers (three men), a local community garden (two women), a local supermarket manager (one man), urban committee members (ward community representatives from low-income suburbs; one man and three women), and representatives of a very low income informal settlement on the outskirts of Worcester (two women; Table 1). An additional interview with a representative of the Worcester informal traders’ association was discarded due to low-quality data.

Local government representatives drew a conceptual model that focused primarily on food production, with the majority of nodes relating to agricultural inputs and processes (Fig. 2). The structure was similar to a supply chain diagram. Key parts of the system that were ranked as important or influential by local government representatives were: environment, water, genetics, climate, feed, regulations, infrastructure, soil, disease management, land use, technology, grading, farms, processing, retail, marketing, and consumer preferences.

Commercial farmers also drew a conceptual model that focused primarily on food production, with an additional focus on governance procedures (Fig. 2). This concept map was also structured similarly to a supply chain diagram. Key parts of the system that were ranked as important or influential by commercial farmers were: water, nursery, farm labor, machinery, information, governance, regulations, grading, processing, and consumer preferences.

The supermarket manager drew a conceptual model that focused on food production and retail, with a supply chain structure that included elements of governance and consumer choice. Key parts of the system that were ranked as important or influential by supermarkets were: health inspectors, staff training, processing, feed, genetics, water, and consumer preferences.

In contrast, community garden representatives drew a conceptual model that focused on the community garden, including inputs for the community garden and distribution of community garden produce in a similar format to a commercial supply chain. With this group, the ranking process was not completed because the model was constructed during an outdoor interview in the garden, with fewer data elicitation supplies.

Urban community representatives (“low-income consumers“) drew a conceptual model that focused primarily on consumer choice and emphasized a separation between the “formal” food system (supermarkets and formal stores) and the “informal” food system (tukshops or spaza shops). The resulting structure resembled a consumer choice diagram (Aertsens et al. 2009, Li and Kamargianni 2020). Key parts of the system that were ranked as important or influential by urban community representatives were: food prices, consumer preferences, quantities available at retail locations, time of purchase (regular daily purchases or less regular monthly purchases), information about food preparation and recipes, income and employment, municipal government services (including health regulations), culture, school meals, informal brokers, farms, home gardens, and the wealth and security of the urban area. Community representatives additionally emphasized the role of transport options, especially the expense and difficulty of travelling to supermarkets via minibus taxi.

Similarly, informal settlement representatives drew a conceptual model that focused primarily on consumer choice. However, in their model, they separated consumers into those who could afford to buy food, and those who obtained food from alternative sources such as soup kitchens or municipal waste. Key parts of the system that were ranked as important or influential by informal settlement representatives were: employment, household and municipal waste, soup kitchens, and informal stores. Consumers were additionally divided into race groups, with different race groups reported to have different dietary preferences and income options (e.g., recent immigrants were sometimes not eligible for unemployment grants and therefore had a limited set of available food choices). Another interesting difference with the other models was the explicit mention of the role of local informal brokers that connect producers to consumers.

Network comparison

The conceptual models that actors at different positions in the urban food system created reflect different mental models of the structure of the urban food system. These differences are reflected in the metrics calculated to compare the networks (Table 2). The network with the most nodes was the consumer network provided by community representatives, indicating that consumers saw the food system as highly complex and interdependent. The network with the highest density was the community garden network, but it also included relatively fewer nodes than other networks, indicating that community gardeners perceived the food system as small but with many direct causal connections. Commercial farmers reported the highest transitivity, indicating that farmers perceived the urban food system to be mainly connected through indirect, rather than direct, causal relations. Government representatives reported the most highly clustered network, indicating that government officials perceived the food system to be composed of many interrelated subsystems, including environment, land use, and regulations, as well as food production and distribution. In terms of centralization, the community garden network had the second highest centralization score after government; in contrast, the community representatives network had lower centralization.

All networks were clustered to some extent, indicating possible subnetwork structures that appeared in the diagrams as recurring patterns. These possible subnetwork structures were: supply chain motifs in the network models from government, farmers, and supermarkets, and consumer choice motifs in the network models from community gardens, consumers, and informal settlements.

Node comparison

Actors at different positions in the urban food system (government, farmers, retailers, and consumers) identified different nodes as important (Table 3). Unsurprisingly, actors tended to be knowledgeable about connections to their own position in the network, resulting in high degree and eigenvector centrality for nodes corresponding to the actor’s position in the urban food system. Nodes with high degree of mean centrality included: community gardens (as identified by community gardens), consumers (as identified by consumers), farmers (as identified by farmers), and supermarkets (as identified by supermarkets). However, unlike other stakeholders, residents of informal settlements did not see themselves as central to the urban food system.

Actors were also asked to rank important parts of the network where there was power or influence to make changes to the urban food system. Highly ranked nodes included: employment and soup kitchens (ranked by informal settlements), skilled farm labor (ranked by farmers), animal feed (ranked by supermarkets), animal and plant genetics (ranked by government), governance (ranked by farmers), health inspectors (ranked by supermarkets), regulations (ranked by governments and farmers), water (ranked by farmers and supermarkets), and food waste (ranked by informal settlements).

Synthesis

IGs were constructed from the actor maps reported above. Based IGs, the comparison step identified two recurrent motifs (Ferguson et al. 2020): supply systems from farms and the landscape surrounding the city, and demand systems of consumer food choices within the city. Thus, the integration step focused on identifying two separate but related submodels: a food supply model (of the supply chain format familiar to economics) and a consumer demand model (of the choice model format familiar to market theory; Fig. 3).

DISCUSSION

Having a full understanding of the urban food system and appreciating the actors’ different viewpoints and competing interests are vital to creating resilient and inclusive food systems in a growing and increasingly urban African population. Addressing these challenges requires understanding the functioning of the entire food system as well as how the food system is viewed by different actors, not only the most powerful, but also those whose perspectives may have been historically under-represented.

Similarities and differences with existing models of food systems

Because there were no prior formal models of the food system in the case study, we could not compare our model to existing local representations. However, our synthesized model (Fig. 3) differs from existing, more generic scientific models of the urban food system in several respects. We consulted actors embedded in the urban food system whose perspectives are not typically included in food system modeling, specifically, representatives of informal settlements, as well as farmers and food retailers. These actors reported additional informal, unregulated processes of food supply and food choice that were not previously documented in scientific research on food systems. These different viewpoints held by formal and informal food system actors may reflect the apartheid-era legacies of South Africa, with persistent inequalities in food access and urban form. They may also be found in other food systems that have similar characteristics.

Food systems are typically modeled and analyzed as single networks, particularly food supply chains or consumer choice models. Our study confirms that complex interdependent networks frequently contain network motifs: specific subnetworks that occur repeatedly (Bodin and Tengö 2012). We found both food supply chains and consumer choice models to be motifs. Therefore, rather than representing the food system by a single network, our synthesized model clearly distinguishes these subsystems.

The synthesized model includes a food supply chain that closely resembles existing models of food supply from a value chain perspective, such as the “food supply and environment” subsystem of the Australian healthy eating model (Friel et al. 2017). However, it also has some notable differences, particularly regarding the fate of agricultural produce not suited for export because of size, shape, or other blemishes. Farmers reported that a certain share of high-grade produce is reserved for international export markets, whereas medium-quality produce goes to local markets, and the lowest grade is picked up at the farm gate by local brokers. Thus, the share of fresh produce that is available for local markets depends on two co-occurring processes: existing pre-negotiated export agreements, and the quality of the harvest. The role of export agreements in taking high-quality local produce away from the local food system has been identified as problematic in previous research (Battersby 2017, Greenberg 2017). In contrast, our research suggests that while these export agreements are profitable for local farmers, they continue to supply the local food system with second-grade and third-grade produce.

Local informal brokers play a key linking role in the city’s food supply with the surrounding landscape. Brokers (sometimes called “bakkie traders” in reference to their vehicles) source produce that is graded below supermarket standard from farms surrounding the city and deliver it to urban street vendors and informal stores. As part of this process, brokers may make the food supply chain more efficient by taking produce that would otherwise go to waste and redistributing it to urban areas that are in short supply. Our synthesized model thus identifies brokers as uniquely important to the supply of fresh fruits and vegetables in the city. Existing models of urban food supply and demand do not typically consider these informal brokers, even though they may play a key role in countries in which there is opportunity for informal trade. In our research, we could only access one informal broker for a single poor-quality interview. Thus, even though we specifically tried to include the different actors in the value chain, we failed to obtain adequate information. This situation demonstrates the challenge in properly representing the perspectives of such actors, which are usually ignored completely in less consultative processes.

Food waste from the processing and retail parts of the food supply chain represents an important source of food for the poorest and most vulnerable city residents. The case study city, Worcester, hosts a large chicken processing plant, several supermarkets, and a large municipal landfill site (recently relocated). Replicating other research undertaken in the African context, we found that the poorest city residents, who cannot afford to buy food at informal stores, resort to scavenging from these various waste sites for food (Schenck et al. 2019). Food safety is thus a primary concern for these representatives, as shown in their interview responses, which include illness and hospital visits as part of the urban food system.

For urban consumers, neighborhood safety and security are perceived as key concerns that influence food purchasing decisions. This part of the model is similar to the “individual and community wealth” theme of the Baltimore, USA model of healthy food access (Mui et al. 2019). At the time of writing, some neighborhoods in Worcester were experiencing high levels of crime, violence, and gang activity. Representatives of this area told us that they rely on local spaza shops or tukshops for daily essentials because these informal stores sell food in smaller quantities than supermarkets. Carrying large quantities of food home from the store is perceived as a safety risk, and therefore, residents prefer to purchase food in quantities that can be safely concealed while walking home, such as a single teaspoon of sugar. Although previous research has identified small purchase quantities as a reason for purchasing from informal stores (Battersby 2012, Petersen and Charman 2018), the dominant assumption in these studies was that residents purchased small quantities because they could not afford larger quantities. In contrast, our research suggests that it is not price, but rather personal security, that drives urban consumers’ choice of small purchase quantities. The synthesized model includes transport method as an important part of consumer choices about food. Security, carrying capacity, and cost are salient aspects of the transport method: if residents have access to a car, they can transport larger quantities of food, in relative security.

The urban food system appears different to different actors, particularly for those with little power in the system and whose perspectives have not historically been well represented in scientific models (Bielicki et al. 2019). For example, in our study, the perspectives of government representatives and commercial farmers align well: both include supply chain motifs in their food system diagrams and focus heavily on agricultural production and inputs as important parts of the food system. In contrast, for urban residents, agricultural production and supply chains are viewed as less important than the immediate retail options (supermarkets and informal stores) and external restrictions on consumers’ abilities to purchase, transport, and store food. In particular, urban residents consider crime and safety as key elements of the food system, echoing previous research from low-income urban communities in the USA (Mui et al. 2019). Thus, if we had consulted only government stakeholders or farmers, our model might have produced recommended actions that had unintended consequences for urban residents.

Collaborative research

Scientific models tend to represent the formal, regulated, aspects of the urban food system because these aspects are well documented and well known to the government stakeholders who are typically consulted in food systems research. However, the practical reality of African urban food systems includes a large number of informal, semi-legal or illegal activities. We attempted an approach to capture some of these informal systems through interviews with a diverse range of actors and stakeholders. The sample size is relatively small, yet the results are illustrative that it is possible to represent these systems better with this approach.

We used a hybrid approach, including collaborative causal maps and directed acyclic graphs, to incorporate the perspectives of diverse actors across the African urban food system. This hybrid approach allowed us to isolate key causal variables and to avoid impractically large lists of system elements and counterfactuals. Causal inference logic allowed us to keep the system diagrams clear while maintaining transparency and reducing bias in the conceptual modeling process (Shrier and Platt 2008, Ferguson et al. 2020). Our intention with this approach was to retain the high-level explanatory power of scientific models of the urban food system while acknowledging the perspectives of the actors embedded within it.

Our methods built on existing participatory research, including decision-relevant scientific models (Acreman 2005, Brewer and Stern 2005, Feldman and Ingram 2009), stakeholder mapping (Cvitanovic et al. 2016), stakeholder workshops (Löschner et al. 2016), knowledge co-production (Frantzeskaki and Kabisch 2016, Adelle et al. 2020), and participatory conceptual models (Delgado et al. 2019, Steger et al. 2021). However, we took a specific approach that combined experiences from these existing approaches, targeted to the specific context of the food system. We found that it helped to rely on user-centric design tools that maximize the legitimacy of the participant, minimize the opportunities for response biases, and do not rely on deep scientific or technical literacy of the participants.

The synthesized model represents the viewpoints of a range of actors consulted for this research. By using the principles of causal logic and checking for convergent validity with existing models, we aimed to create the most accurate representation of the urban food system that we could. However, the procedure still fundamentally relied on interviews with actors who may have had incomplete, flawed, or inaccurate mental representations of the urban food system. We did not evaluate or test the accuracy of actors’ perceptions. Instead, we took the approach of expert systems models and accepted that the conceptual model represents “expert, not perfect, opinions” (Kadane and Wolfson 1998, Choy et al. 2009). Using a large sample size and data analysis (for example, from a consumer survey or value chain data) could help to assess the uncertainties and validate some of the model components.

Limitations

The potential benefits of conceptual modeling must be weighed against the cost and time required to gather the information. We collected data in interviews of one or a very small group of participants at a time, taking about one hour to complete each. Depending on the participant’s level of comfort and familiarity with research, this process may have been difficult for them. Furthermore, drawing the conceptual models is only the first step in creating a functioning causal model of the urban food system, with additional interviews and data required to test and parameterize the model, develop scenarios, and so on. If not approached carefully, this type of research could risk placing excessive burdens on participants’ time and attention.

Our study had a limited number of participants (18), although this is comparable to similar research (18 participants in Mui et al. 2019; 15 participants in Friel et al. 2017). Additionally, we oversampled consumers from poor parts of the city, partly by design; we aimed to capture viewpoints that may have been under-represented by previous research, but this may have affected the final synthesized diagram. Thus, further research is required to test and refine the model’s validity and completeness. Future studies could conduct additional interviews or add qualitative inquiry into the causal mechanisms proposed here. Our study should primarily be viewed as a “proof-of-concept” study. However, we consider that there are broadly generalizable lessons to be gathered from this research as it relates to larger issues of collaborative problems in social-ecological systems.

Lessons for collaboration in social-ecological systems

The problem of unintended consequences is further magnified by unequal distributions of power in the food system. Food systems models have been critiqued for not accurately representing the different ways that food systems work for the privileged and underprivileged, and for ignoring the agency of people in the system whose perspectives have not historically been considered by those in power (Kroll et al. 2019).

Due to the low cohesion among actors’ perceptions of the food system structure and the coexistence of multiple subsystems (such as the commercial farmer and urban community farm systems), an effective collaborative response to the food security issues in Worcester is currently unlikely. An effective response to a food and nutrition security crisis will require more coordination effort than just agreeing on a common goal. We suggest that this situation is not unique to Worcester and may affect many fast-growing urban centres in Africa. We further suggest that the COVID pandemic that took place shortly after the completion of this fieldwork demonstrates the importance of such coordination.

Individual actors may not anticipate all of the outcomes that might result from their decisions in a complex system (Ostrom et al. 1994). In a complex system that includes elements of society and the broader environment, actors may settle for nonrational decision methods such as heuristics and past experiences, without considering the system-wide consequences (Little 2012). The limited viewpoints of individual actors in the urban food system that we found may thus represent a serious threat to the link between society and the wider food system on which their livelihoods and lives depend. Improving collaborative potential in such complex social-ecological systems relies, at least in part, on bringing together different actors and encouraging the sharing of views and ideas (Bodin 2017). Building collaboration can be accomplished in several ways: by creating collaborative network structures that aid communication (Bodin 2017), by identifying key individuals who can bridge the gap between different actor groups (Folke et al. 2005), or by creating safe spaces and collaborative processes for information sharing (Pereira et al. 2018). Collaboration may also be improved by creating shared conceptual models, or mental models, to improve communication and shared representation of complex social-ecological systems (Delgado et al. 2019, Steger et al. 2021). An urban food system with high collaborative potential is less vulnerable to external shocks, less likely to deteriorate into crisis, and more likely to respond to surprises by creating space for renewal and novelty (Folke et al. 2005). The approach presented here may be a step contributing to enhancing collaboration.

CONCLUSION

Merely having a scientific representation of a system is no guarantee that the resource, in this case, food, will be managed effectively. We have demonstrated one option for creating better alignment between high-level scientific models of the urban food system and the varying real-world perspectives of actors embedded within it. The urban food system appears different to different actors, particularly for those with little power in the system and whose perspectives have not historically been well represented in scientific models (Bielicki et al. 2019). Our model synthesis aims to reduce the problem of unintentionally emphasizing one set of information and priorities over another.

Relatedly, decision makers may be more likely to implement the recommendations of food systems models that appear to be grounded in practical reality. Consequently, models created using a multistakeholder design process may produce insights that are more impactful on urban food systems than do models created using implicit knowledge and mental references of the scientist developing the model. Our study is an early attempt at creating such a model.

We have shown that misaligned perspectives and knowledge gaps exist between food system stakeholders. Such gaps may prevent actors from successfully coordinating to address long-term problems collaboratively. This situation is an especially urgent problem for urban food security, where cooperation is needed at city, regional, and global scales. Creating and maintaining effective collaboration between actors might be the only feasible way to address food security in Africa’s rapidly growing urban population. The synthesis of these diverging perspectives into a food system model may be a first step in that direction.

LITERATURE CITED

Acreman, M. 2005. Linking science and decision-making: features and experience from environmental river flow setting. Environmental Modelling and Software 20(2):99-109. https://doi.org/10.1016/j.envsoft.2003.08.019

Adelle, C., L. Pereira, T. Görgens, and B. Losch. 2020. Making sense together: the role of scientists in the coproduction of knowledge for policy making. Science and Public Policy 47(1):56-66. https://doi.org/10.1093/scipol/scz046

Aertsens, J., W. Verbeke, K. Mondelaers, and G. Van Huylenbroeck. 2009. Personal determinants of organic food consumption: a review. British Food Journal 111(10):1140-1167. https://doi.org/10.1108/00070700910992961

Alarcon, P., P. Dominguez-Salas, E. M. Fèvre, and J. Rushton. 2021. The importance of a food systems approach to low and middle income countries and emerging economies: a review of theories and its relevance for disease control and malnutrition. Frontiers in Sustainable Food Systems 5:642635. https://doi.org/10.3389/fsufs.2021.642635

Allen, E., C. Kruger, F.-Y. Leung, and J. C. Stephens. 2013. Diverse perceptions of stakeholder engagement within an environmental modeling research team. Journal of Environmental Studies and Sciences 3:343-356. https://doi.org/10.1007/s13412-013-0136-x

Bagnoli, A. 2009. Beyond the standard interview: the use of graphic elicitation and arts-based methods. Qualitative Research 9(5):547-570. https://doi.org/10.1177/1468794109343625

Balogun, T. A., M. J. Lombard, and M. McLachlan. 2015. The nutrient intake of children aged 12–36 months living in two communities in the Breede Valley, Western Cape province, South Africa. South African Family Practice 57(1):1-7. https://doi.org/10.1080/20786190.2014.980158

Battersby, J. 2012. Beyond the food desert: finding ways to speak about urban food security in South Africa. Geografiska Annaler: Series B, Human Geography 94(2):141-159. https://doi.org/10.1111/j.1468-0467.2012.00401.x

Battersby, J. 2013. Hungry cities: a critical review of urban food security research in sub-Saharan African cities. Geography Compass 7(7):452-463. https://doi.org/10.1111/gec3.12053

Battersby, J. 2017. Food system transformation in the absence of food system planning: the case of supermarket and shopping mall retail expansion in Cape Town, South Africa. Built Environment 43(3):417-430. https://doi.org/10.2148/benv.43.3.417

Battersby, J., R. Fincham, B. Frayne, and G. Haysom. 2009. Urban food security in South Africa: case study of Cape Town, Msunduzi and Johannesburg. Development Planning Division Working Paper Series 15. Development Bank of Southern Africa, Johannesburg, South Africa.

Battersby, J., M. Marshak, and N. Mngqibisa. 2016. Mapping the invisible. the informal food economy of Cape Town, South Africa. African Food Security Urban Network, University of Cape Town, Cape Town, South Africa and Balsillie School of International Affairs, Waterloo, Canada. https://doi.org/10.2307/j.ctvh8r2fq

Bielicki, J. M., M. A. Beetstra, J. B. Kast, Y. Wang, and S. Tang. 2019. Stakeholder perspectives on sustainability in the food-energy-water nexus. Frontiers in Environmental Science 7:7. https://doi.org/10.3389/fenvs.2019.00007

Bodin, Ö. 2017. Collaborative environmental governance: achieving collective action in social-ecological systems. Science 357:eaan1114. https://doi.org/10.1126/science.aan1114

Bodin, Ö., and M. Tengö. 2012. Disentangling intangible social-ecological systems. Global Environmental Change 22(2):430-439. https://doi.org/10.1016/j.gloenvcha.2012.01.005

Brewer, G. D., and P. C. Stern, editors. 2005. Decision making for the environment: social and behavioral science research priorities. National Academies Press, Washington, D.C., USA.

Brown, D. A., and M. E. Lamb. 2019. Forks in the road, routes chosen, and journeys that beckon: a selective review of scholarship on children’s testimony. Applied Cognitive Psychology 33(4):480-488. https://doi.org/10.1002/acp.3511

Candy, S., C. Biggs, K. Larsen, and G. Turner. 2015. Modelling food system resilience: a scenario-based simulation modelling approach to explore future shocks and adaptations in the Australian food system. Journal of Environmental Studies and Sciences 5:712-731. https://doi.org/10.1007/s13412-015-0338-5

Chambers, J. M., C. Wyborn, M. E. Ryan, R. S. Reid, M. Riechers, A. Serban, N. J. Bennett, C. Cvitanovic, M. E. Fernández-Giménez, K. A. Galvin, B. E. Goldstein, N. L. Klenk, M. Tengö, R. Brennan, J. J. Cockburn, R. Hill, C. Munera, J. L. Nel, H. Österblom, A. T. Bednarek, E. M. Bennett, A. Brandeis, L. Charli-Joseph, P. Chatterton, K. Curran, P. Dumrongrojwatthana, A. P. Durán, S. J. Fada, J.-D. Gerber, J. M. H. Green, A. M. Guerrero, T. Haller, A.-I. Horcea-Milcu, B. Leimona, J. Montana, R. Rondeau, M. Spierenburg, P. Steyaert, J. G. Zaehringer, R. Gruby, J. Hutton, and T. Pickering. 2021. Six modes of co-production for sustainability. Nature Sustainability 4:983-996. https://doi.org/10.1038/s41893-021-00755-x

Choy, S. L., R. O’Leary, and K. Mengersen. 2009. Elicitation by design in ecology: using expert opinion to inform priors for Bayesian statistical models. Ecology 90(1):265-277. https://doi.org/10.1890/07-1886.1

Crush, J., and G. Young. 2019. Resituating Africa’s urban informal food sector. Urban Forum 30:377-384. https://doi.org/10.1007/s12132-019-09374-4

Csardi, G., and T. Nepusz. 2006. The igraph software package for complex network research. InterJournal, Complex Systems 1695:1-9. https://igraph.org/

Cullis, J. D. S., N. Rossouw, G. du Toit, D. Petrie, G. Wolfaardt, W. de Clercq, and A. Horn. 2018. Economic risks due to declining water quality in the Breede River catchment. Water SA 44(3 July). https://doi.org/10.4314/wsa.v44i3.14

Cvitanovic, C., J. McDonald, and A. J. Hobday. 2016. From science to action: principles for undertaking environmental research that enables knowledge exchange and evidence-based decision-making. Journal of Environmental Management 183(3):864-874. https://doi.org/10.1016/j.jenvman.2016.09.038

Delgado, L. E., D. C. Pérez-Orellana, and V. H. Marín. 2019. Simplifying the complexity of social-ecological systems with conceptual models. Pages 15-32 in L. E. Delgado and V. H. Marín, editors. Social-ecological systems of Latin America: complexities and challenges. Springer, Cham, Switzerland. https://doi.org/10.1007/978-3-030-28452-7_2

du Plessis, L., M. G. Herselman, M. H. McLachlan, and J. H. Nel. 2016. Selected facets of nutrition during the first 1000 days of life in vulnerable South African communities. South African Journal of Child Health 10(1):37-42. https://doi.org/10.7196/SAJCH.2016.v10i1.984

Feldman, D. L., and H. M. Ingram. 2009. Making science useful to decision makers: climate forecasts, water management, and knowledge networks. Weather, Climate, and Society 1(1):9-21. https://doi.org/10.1175/2009WCAS1007.1

Ferguson, K. D., M. McCann, S. V. Katikireddi, H. Thomson, M. J. Green, D. J. Smith, and J. D. Lewsey. 2020. Evidence synthesis for constructing directed acyclic graphs (ESC-DAGs): a novel and systematic method for building directed acyclic graphs. International Journal of Epidemiology 49(1):322-329. https://doi.org/10.1093/ije/dyz150

Folke, C., T. Hahn, P. Olsson, and J. Norberg. 2005. Adaptive governance of social-ecological systems. Annual Review of Environment and Resources 30:441-473. https://doi.org/10.1146/annurev.energy.30.050504.144511

Food and Agriculture Organization. 2009. Declaration of the world summit on food security. WSFS 2009/2. Food and Agriculture Organization, Rome, Italy. https://www.fao.org/3/k6050e/k6050e.pdf

Frantzeskaki, N., and N. Kabisch. 2016. Designing a knowledge co-production operating space for urban environmental governance—lessons from Rotterdam, Netherlands and Berlin, Germany. Environmental Science and Policy 62:90-98. https://doi.org/10.1016/j.envsci.2016.01.010

Friel, S., M. Pescud, E. Malbon, A. Lee, R. Carter, J. Greenfield, M. Cobcroft, J. Potter, L. Rychetnik, and B. Meertens. 2017. Using systems science to understand the determinants of inequities in healthy eating. Plos One 12(11):e0188872. https://doi.org/10.1371/journal.pone.0188872

Gray, S., A. Voinov, M. Paolisso, R. Jordan, T. BenDor, P. Bommel, P. Glynn, B. Hedelin, K. Hubacek, J. Introne, N. Kolagani, B. Laursen, C. Prell, L. Schmitt Olabisi, A. Singer, E. Sterling, and M. Zellner. 2018. Purpose, processes, partnerships, and products: four Ps to advance participatory socio-environmental modeling. Ecological Applications 28(1):46-61. https://doi.org/10.1002/eap.1627

Gray, S. A., S. Gray, J. L. De Kok, A. E. R. Helfgott, B. O’Dwyer, R. Jordan, and A. Nyaki. 2015. Using fuzzy cognitive mapping as a participatory approach to analyze change, preferred states, and perceived resilience of social-ecological systems. Ecology and Society 20(2):11. https://doi.org/10.5751/ES-07396-200211

Greenberg, S. 2017. Corporate power in the agro-food system and the consumer food environment in South Africa. Journal of Peasant Studies 44(2):467-496. https://doi.org/10.1080/03066150.2016.1259223

Hill, A. B. 1965. The environment and disease: association or causation? Proceedings of the Royal Society of Medicine 58(5):295-300. https://doi.org/10.1177/003591576505800503

Holdsworth, M., and E. Landais. 2019. Urban food environments in Africa: implications for policy and research. Proceedings of the Nutrition Society 78(4):513-525. https://doi.org/10.1017/S0029665118002938

Howarth, C., and I. Monasterolo. 2016. Understanding barriers to decision making in the UK energy-food-water nexus: the added value of interdisciplinary approaches. Environmental Science and Policy 61:53-60. https://doi.org/10.1016/j.envsci.2016.03.014

Johnson, S. R., G. A. Tomlinson, G. A. Hawker, J. T. Granton, H. A. Grosbein, and B. M. Feldman. 2010. A valid and reliable belief elicitation method for Bayesian priors. Journal of Clinical Epidemiology 63(4):370-383. https://doi.org/10.1016/j.jclinepi.2009.08.005

Jones, N. A., H. Ross, T. Lynam, and P. Perez. 2014. Eliciting mental models: a comparison of interview procedures in the context of natural resource management. Ecology and Society 19(1):13. https://doi.org/10.5751/ES-06248-190113

Jordan, R., S. Gray, M. Zellner, P. D. Glynn, A. Voinov, B. Hedelin, E. J. Sterling, K. Leong, L. Schmitt Olabisi, K. Hubacek, P. Bommel, T. K. BenDor, A. J. Jetter, B. Laursen, A. Singer, P. J. Giabbanelli, N. Kolagani, L. B. Carrera, K. Jenni, C. Prell, and National Socio-Environmental Synthesis Center Participatory Modeling Pursuit Working Group. 2018. Twelve questions for the participatory modeling community. Earth’s Future 6(8):1046-1057. https://doi.org/10.1029/2018EF000841

Kadane, J., and L. J. Wolfson. 1998. Experiences in elicitation. Journal of the Royal Statistical Society: Series D (The Statistician) 47(1):3-19. https://doi.org/10.1111/1467-9884.00113

Kirchhoff, C. J., M. Carmen Lemos, and S. Dessai. 2013. Actionable knowledge for environmental decision making: broadening the usability of climate science. Annual Review of Environment and Resources 38:393-414. https://doi.org/10.1146/annurev-environ-022112-112828

Knox, S., and A. W. Burkard. 2009. Qualitative research interviews. Psychotherapy Research 19(4-5):566-575. https://doi.org/10.1080/10503300802702105

Kroll, F. 2016. Foodways of the poor in South Africa: how value-chain consolidation, poverty and cultures of consumption feed each other. PLAAS Working Paper 36. Institute for Poverty, Land and Agrarian Studies, University of the Western Cape and Centre of Excellence on Food Security, Cape Town, South Africa. http://hdl.handle.net/10566/4518

Kroll, F., E. C. Swart, R. A. Annan, A. M. Thow, D. Neves, C. Apprey, L. N. E. Aduku, N. A. F. Agyapong, J.-C. Moubarac, A. du Toit, R. Aidoo, and D. Sanders. 2019. Mapping obesogenic food environments in South Africa and Ghana: correlations and contradictions. Sustainability 11(14):3924. https://doi.org/10.3390/su11143924

LaFrance, M., and C. Mayo. 1978. Cultural aspects of nonverbal communication. International Journal of Intercultural Relations 2(1):71-89. https://doi.org/10.1016/0147-1767(78)90029-9

Li, W., and M. Kamargianni. 2020. An integrated choice and latent variable model to explore the influence of attitudinal and perceptual factors on shared mobility choices and their value of time estimation. Transportation Science 54(1):62-83. https://doi.org/10.1287/trsc.2019.0933

Little, A. 2012. Political action, error and failure: the epistemological limits of complexity. Political Studies 60(1):3-19. https://doi.org/10.1111/j.1467-9248.2011.00901.x

Liu, Y., H. Gupta, E. Springer, and T. Wagener. 2008. Linking science with environmental decision making: experiences from an integrated modeling approach to supporting sustainable water resources management. Environmental Modelling and Software 23(7):846-858. https://doi.org/10.1016/j.envsoft.2007.10.007

Löschner, L., R. Nordbeck, P. Scherhaufer, and W. Seher. 2016. Scientist–stakeholder workshops: a collaborative approach for integrating science and decision-making in Austrian flood-prone municipalities. Environmental Science and Policy 55(2):345-352. https://doi.org/10.1016/j.envsci.2015.08.003

Lynam, T., R. Mathevet, M. Etienne, S. Stone-Jovicich, A. Leitch, N. Jones, H. Ross, D. Du Toit, S. Pollard, H. Biggs, and P. Perez. 2012. Waypoints on a journey of discovery: mental models in human-environment interactions. Ecology and Society 17(3):23. https://doi.org/10.5751/ES-05118-170323

Mandal, F. B. 2014. Nonverbal communication in humans. Journal of Human Behavior in the Social Environment 24(4):417-421. https://doi.org/10.1080/10911359.2013.831288

Manuel-Navarrete, D. 2015. Double coupling: modeling subjectivity and asymmetric organization in social-ecological systems. Ecology and Society 20(3):26. https://doi.org/10.5751/ES-07720-200326

Martin, T. G., M. A. Burgman, F. Fidler, P. M. Kuhnert, S. Low-Choy, M. McBride, and K. Mengersen. 2012. Eliciting expert knowledge in conservation science. Conservation Biology 26(1):29-38. https://doi.org/10.1111/j.1523-1739.2011.01806.x

Morris, D. E., J. E. Oakley, and J. A. Crowe. 2014. A web-based tool for eliciting probability distributions from experts. Environmental Modelling and Software 52:1-4. https://doi.org/10.1016/j.envsoft.2013.10.010

Mui, Y., E. Ballard, E. Lopatin, R. L. J. Thornton, K. M. Pollack Porter, and J. Gittelsohn. 2019. A community-based system dynamics approach suggests solutions for improving healthy food access in a low-income urban environment. Plos One 14(5):e0216985. https://doi.org/10.1371/journal.pone.0216985

Müller, B., F. Hoffmann, T. Heckelei, C. Müller, T. W. Hertel, J. G. Polhill, M. van Wijk, T. Achterbosch, P. Alexander, C. Brown, D. Kreuer, F. Ewert, J. Ge, J. D. A. Millington, R. Seppelt, P. H. Verburg, and H. Webber. 2020. Modelling food security: bridging the gap between the micro and the macro scale. Global Environmental Change 63:102085. https://doi.org/10.1016/j.gloenvcha.2020.102085

Ostrom, E., R. Gardner, and J. Walker. 1994. Rules, games, and common-pool resources. University of Michigan Press, Ann Arbor, Michigan, USA. https://doi.org/10.3998/mpub.9739

Oteros-Rozas, E., A. Ruiz-Almeida, M. Aguado, J. A. González, and M. G. Rivera-Ferre. 2019. A social-ecological analysis of the global agrifood system. Proceedings of the National Academy of Sciences 116(52):26465-26473. https://doi.org/10.1073/pnas.1912710116

Partridge, A., T. Morokong, and A. Sibulali. 2021. Western Cape agricultural sector profile: 2020. Western Cape Department of Agriculture Division for Macro and Resource Economics, Elsenburg, South Africa. https://www.elsenburg.com/wp-content/uploads/2022/02/Western-Cape-Agricultural-Sector-Profile-2020.pdf

Pearl, J. 1995. From Bayesian networks to causal networks. Pages 157-182 in G. Coletti, D. Dubois, and R. Scozzafava, editors. Mathematical models for handling partial knowledge in artificial intelligence. Springer, Boston, Massachusetts, USA. https://doi.org/10.1007/978-1-4899-1424-8_9

Pereira, L. M., T. Karpouzoglou, N. Frantzeskaki, and P. Olsson. 2018. Designing transformative spaces for sustainability in social-ecological systems. Ecology and Society 23(4):32. https://doi.org/10.5751/ES-10607-230432

Petersen, L. M., and A. J. E. Charman. 2018. The scope and scale of the informal food economy of South African urban residential townships: results of a small-area micro-enterprise census. Development Southern Africa 35(1):1-23. https://doi.org/10.1080/0376835X.2017.1363643

Pictet, J., and D. Bollinger. 2008. Extended use of the cards procedure as a simple elicitation technique for MAVT. Application to public procurement in Switzerland. European Journal of Operational Research 185(3):1300-1307. https://doi.org/10.1016/j.ejor.2006.05.051

Pitchforth, J., and K. Mengersen. 2013. A proposed validation framework for expert elicited Bayesian networks. Expert Systems with Applications 40(1):162-167. https://doi.org/10.1016/j.eswa.2012.07.026

Prell, C., K. Hubacek, and M. Reed. 2009. Stakeholder analysis and social network analysis in natural resource management. Society and Natural Resources 22(6):501-518. https://doi.org/10.1080/08941920802199202

Ramankutty, N., Z. Mehrabi, K. Waha, L. Jarvis, C. Kremen, M. Herrero, and L. H. Rieseberg. 2018. Trends in global agricultural land use: implications for environmental health and food security. Annual Review of Plant Biology 69:789-815. https://doi.org/10.1146/annurev-arplant-042817-040256

Raza, A., E. L. Fox, S. S. Morris, R. Kupka, A. Timmer, N. Dalmiya, and J. Fanzo. 2020. Conceptual framework of food systems for children and adolescents. Global Food Security 27:100436. https://doi.org/10.1016/j.gfs.2020.100436

Reed, M. S., A. C. Evely, G. Cundill, I. Fazey, J. Glass, A. Laing, J. Newig, B. Parrish, C. Prell, C. Raymond, and L. C. Stringer. 2010. What is social learning? Ecology and Society 15(4):r1. https://doi.org/10.5751/ES-03564-1504r01

Robinson, S. 2006. Issues in conceptual modelling for simulation: setting a research agenda. In S. Robinson, S. Taylor, S. Brailsford, and J. Garnett, editors. Proceedings of the 2006 Operational Research Society Simulation Workshop. OR Society, Birmingham, UK.

Rohr, J. R., C. B. Barrett, D. J. Civitello, M. E. Craft, B. Delius, G. A. DeLeo, P. J. Hudson, N. Jouanard, K. H. Nguyen, R. S. Ostfeld, J. V. Remais, G. Riveau, S. H. Sokolow, and D. Tilman. 2019. Emerging human infectious diseases and the links to global food production. Nature Sustainability 2:445-456. https://doi.org/10.1038/s41893-019-0293-3

Rumble, J. 2013. Collaborative environmental governance in agriculture: a case-study from the Upper Breede valley. Thesis. University of Cape Town, Cape Town, South Africa. http://hdl.handle.net/11427/8472

Schenck, C. J., P. F. Blaauw, E. C. Swart, J. M. M. Viljoen, and N. Mudavanhu. 2019. The management of South Africa’s landfills and waste pickers on them: impacting lives and livelihoods. Development Southern Africa 36(1):80-98. https://doi.org/10.1080/0376835X.2018.1483822

Scherp, H.-Å. 2013. Quantifying qualitative data using cognitive maps. International Journal of Research and Method in Education 36(1):67-81. https://doi.org/10.1080/1743727X.2012.696244

Schlüter, M., B. Müller, and K. Frank. 2019. The potential of models and modeling for social-ecological systems research: the reference frame ModSES. Ecology and Society 24(1):31. https://doi.org/10.5751/ES-10716-240131

Schwarz, G., F. Vanni, and D. Miller. 2021. The role of transdisciplinary research in the transformation of food systems. Agricultural and Food Economics 9:35. https://doi.org/10.1186/s40100-021-00207-2

Seeliger, L., W. P. de Clercq, W. Hoffman, J. D. S. Cullis, A. M. Horn, and M. de Witt. 2018. Applying the water-energy-food nexus to farm profitability in the Middle Breede catchment, South Africa. South African Journal of Science 114(11-12). https://doi.org/10.17159/sajs.2018/5062

Shrier, I., and R. W. Platt. 2008. Reducing bias through directed acyclic graphs. BMC Medical Research Methodology 8:70. https://doi.org/10.1186/1471-2288-8-70

Steger, C., K. Gebrehiwot, S. A. Chengere, J. Marinkovich, B. W. Dullo, S. W. Zewde, and J. A. Klein. 2021. Mental models of a social-ecological system facilitate social learning among a diverse management team. Environmental Science and Policy 122:127-138. https://doi.org/10.1016/j.envsci.2021.04.006

Talukder, S., A. Capon, D. Nath, A. Kolb, S. Jahan, and J. Boufford. 2015. Urban health in the post-2015 agenda. Lancet 385(9970):769. https://doi.org/10.1016/S0140-6736(15)60428-7

Textor, J., B. van der Zander, M. S. Gilthorpe, M. Liśkiewicz, and G. T. H. Ellison. 2016. Robust causal inference using directed acyclic graphs: the R package ‘dagitty’. International Journal of Epidemiology 45(6):1887-1894. https://doi.org/10.1093/ije/dyw341

Umetsu, C., T. Lekprichakul, T. Sakurai, T. Yamauchi, Y. Ishimoto, and H. Miyazaki. 2014. Dynamics of social-ecological systems: the case of farmers’ food security in the semi-arid tropics. Pages 157-178 in S. Sakai and C. Umetsu, editors. Social-ecological systems in transition. Springer, Tokyo, Japan. https://doi.org/10.1007/978-4-431-54910-9_9

van der Zee, D. J., and J. G. A. J. van der Vorst. 2005. A modeling framework for supply chain simulation: opportunities for improved decision making. Decision Sciences 36(1):65-95. https://doi.org/10.1111/j.1540-5915.2005.00066.x

van Zutphen, K. G., D. Barjolle, S. van den Berg, B. Gavin-Smith, K. Kraemer, C. Musard, H. Prytherch, J. Six, S. Winter, and K. Woltering. 2021. Secondary cities as catalysts for nutritious diets in low- and middle-income countries. Food Systems Summit Brief. Center for Development Research (ZEF) and Scientific Group for the UN Food Systems Summit, Bonn, Germany. http://doi.org/10.48565/scfss2021-2y14

Varga-Atkins, T., and M. O’Brien. 2009. From drawings to diagrams: maintaining researcher control during graphic elicitation in qualitative interviews. International Journal of Research and Method in Education 32(1):53-67. https://doi.org/10.1080/17437270902759998

Voinov, A., K. Jenni, S. Gray, N. Kolagani, P. D. Glynn, P. Bommel, C. Prell, M. Zellner, M. Paolisso, R. Jordan, E. Sterling, L. Schmitt Olabisi, P. J. Giabbanelli, Z. Sun, C. Le Page, S. Elsawah, T. K. BenDor, K. Hubacek, B. K. Laursen, A. Jetter, L. Basco-Carrera, A. Singer, L. Young, J. Brunacini, and A. Smajgl. 2018. Tools and methods in participatory modeling: selecting the right tool for the job. Environmental Modelling and Software 109:232-255. https://doi.org/10.1016/j.envsoft.2018.08.028

Walker, G. B., and S. E. Daniels. 2019. Collaboration in environmental conflict management and decision-making: comparing best practices with insights from collaborative learning work. Frontiers in Communication 4:2. https://doi.org/10.3389/fcomm.2019.00002

Webb, C. 2017. Between precarity and paternalism: farm workers and trade unions in South Africa’s Western Cape province. Global Labour Journal 8(1). https://doi.org/10.15173/glj.v8i1.2842

White, D. D., J. L. Jones, R. Maciejewski, R. Aggarwal, and G. Mascaro. 2017. Stakeholder analysis for the food-energy-water nexus in Phoenix, Arizona: implications for nexus governance. Sustainability 9(12):2204. https://doi.org/10.3390/su9122204

Zimmerman, R., Q. Zhu, and C. Dimitri. 2018. A network framework for dynamic models of urban food, energy and water systems (FEWS). Environmental Progress and Sustainable Energy 37(1):122-131. https://doi.org/10.1002/ep.12699