Using social network analysis to track the evolution of Pacific food system research collaborations over time

Climate change undermines the foundations of food and nutrition security, making it crucial to understand and improve the current research collaborations striving to fill knowledge gaps about the impacts on food systems. This study focuses on the network of research actors working on food systems and climate change in Pacific Island Countries and Territories (PICTs), a region particularly vulnerable to climate change, and actively integrating knowledge about climate impacts to inform adaptation planning. Specifically, this study looks at co-publication as a form of research collaboration and co-production, using network analysis to understand who are the prominent organizational actors driving knowledge development in the region. Overall, we found a distinct core of these organizations engaged in the research space, dominated in number by academic institutions largely based in Australia, the USA, and New Zealand, but knit together through the interactions of the regional government agencies of the Pacific Islands. Over time, the network of research collaboration on climate change and food systems in the region has grown and diversified. While regional Pacific institutions have remained central actors throughout, national and subnational actors are still in the minority, raising questions about efforts to decolonize research in the region. Furthermore, greater interaction between peripheral actors and the core (especially Pacific actors) could help integrate new research into Pacific knowledge bases and decision-making. Ultimately, when facing the novel conditions that climate change brings, combining the infusion of new knowledge and innovation with local expertise and ownership is critical.


Introduction
Climate change poses an existential threat to human populations, undermining the foundations of food and nutrition security.While the science is clear that climate change has a high likelihood of severe negative impacts on food systems (Fanzo et al. 2018;Mbow et al. 2019;Vermeulen et al. 2012), we still have much to learn and apply at regional and local levels.Food systems play a fundamental role in the well-being and prosperity of Pacific Island Countries and Territories, yet recent research has highlighted the vulnerabilities of food systems in the region, particularly when faced with compounding risks arising from climate hazards and extreme events, alongside other stressors like trade disruptions or a global pandemic (Davila et al. 2021).The system perspective required to successfully meet food security needs makes cross-pollination and collaboration critical to draw together different sectors and expertise.Furthermore, it is particularly critical that strong links between research and policy are made, and that findings can be applied to 153 Page 2 of 14 addressing climate challenges in practice.To strategically invest in filling important research gaps, it is crucial to capitalize on and improve the current landscape of research on climate change and food systems, build on existing collaborations, and identify the gaps in capacity.
This paper focuses on the network of actors researching food systems and climate change in Pacific Island Countries and Territories (PICTs), a region particularly vulnerable to climate change (Mycoo et al. 2022), that is actively integrating knowledge about climate impacts to inform adaptation planning (Mcleod et al. 2019).Specifically, this paper looks at co-publication as a form of research collaboration and co-production, to understand who are the prominent actors driving the field of knowledge in the region.By examining the characteristics of this co-authorship network over time, it is also possible to develop a sense of the strength of collaborative relationships.

A case for collaboration
Growth in co-authored papers and international collaborations over the last couple of decades speaks to the potential benefits to be gained through successful research partnerships (Bukvova 2010).If we consider research as knowledge creation, research collaboration networks can be critical for the cross-pollination of ideas and the strengthening of outputs (Huang 2014).Benefits discussed in the literature broadly include the following: the exchange of ideas across disciplines and contexts; access to resources, funding, and expertise; higher productivity and quality of results; and the capacity to address complex problems (Bukvova 2010;Huang 2014).Interactions between university, industry, and government actors are also offered as a key driver of innovation development (Rosenlund 2017).Similarly, more open forms of knowledge production, outside the confines of academia, are argued to be necessary for solving multi-faceted socio-environmental issues.It is important to recognize here that the knowledge creation in the published literature may not adequately represent "traditional knowledge," which is a growing focus in the Pacific, but not yet very visible in the region's published literature.
In this age of "complex global challenges," such as climate change, widespread environmental degradation, and infectious disease spread, calls for coordinated approaches across disciplines, sectors, and even geographies are growing stronger (Fransman et al. 2021; "Protect Precious Scientific Collaboration from Geopolitics" 2021).For many of these global issues and for climate change specifically, the move toward collaborative research is a product of necessity; the multiple scales, the urgency of the issue, and the need for novel solutions have been central reasons for the rise in this type of research (Cundill et al. 2019).Moreover, collaborative research that goes beyond academia can open space for non-Western scientific ways of knowing and yield user-driven outputs, which are increasingly important for justifying public funding for research.Finally, in a crisis discipline, where there are unequal impacts from climatic changes, collaborations have the potential to support vulnerable and low-income countries to build knowledge and capacity and strive toward more equitable forms of research ("Protect Precious Scientific Collaboration from Geopolitics" 2021).

Who drives the research agenda? Equity in the evidence
An emerging focus for research collaboration has been on justice-redressing inequitable distributions of certain knowledge, increasing ownership and mutual accountability, and directly benefiting those being assisted (Fransman et al. 2021).The decolonizing research agenda has also gained influence in some research and academic circles, questioning the dominant forms of academic research, how research problems are conceptualized, and by whom and who the research is for (Smith 2012).These visions have yet to be fully realized.Research partnerships pursued with aims of local benefits still have largely been between academic institutions of the Global North1 and practitioner or policy institutions operating in the Global South, creating the potential for power imbalances to arise.For instance, research institution partners tend to lead, regardless of which actors initiated the research, guiding the early stages of development and implementation of projects (Bukvova 2010).This supports how McCabe et al. (2021) characterize different types of inherent power dynamics that may lead to conflict.For instance, in those authors' assessment of Australian collaborative research projects, the power-related conflicts mostly stemmed from academic control of decisions, the processes set up for making the decisions, and the prevailing norm that academic knowledge outweighs that of practitioners'.These persistent power dynamics may be residual from the historic relationship between "researcher" and "researched," in which academic interests dictated the research terms, topics, funding, data collection, interpretation, and communication (Young and Freytag 2021).
Studies have also identified reciprocity and trust between academic and non-academic actors within the collaboration network as fundamental to successful collaborative research.

Page 3 of 14 153
Reciprocity emerges from positive and repeated relationships and can increase the value and relevance of what is achieved from partnerships (Young and Freytag 2021).Boundary organizations, which can broker or translate between academic and applied research, present opportunities to create this two-way relationship and can support evidence-based decision-making (Guston 2001;Cvitanovic et al. 2018).Sustaining relationships over time requires continually renegotiating expectations, benefits, and accountability (McCabe et al. 2021).Trust is also something that is built over time between partners, and is important for lowering barriers to collaboration between academia and industry (Rosenlund 2017).Repeated collaboration was also shown to be important for policy formulation in a study of multi-actor networks in Germany working on climate change adaptation (Schmid et al. 2016).Ultimately, collaborations-and associated levels of trust, dependency, and power-evolve over time.Upfront costs and investments in collaboration can be high to achieve trust and reciprocity, so maintaining these relationships over time is in all partners' interests (Young and Freytag 2021).

Collaboration networks on climate in the Pacific
This paper focuses on the 21 Pacific Island Countries and Territories (PICTs), which have demonstrated strong leadership in the climate change negotiations processes and a propensity for coalition and network building (Carter 2020).As part of international climate negotiations, the coalition of Pacific Island Nations spearheaded progress on more ambitious targets and clauses for loss and damage, while also relying on partnerships with regional organizations and non-governmental organizations (NGOs).Cooperation related to climate change is also increasingly seen as an issue of national and regional security among PICTs, with calls for greater engagement between governments and research institutions (Hauger 2015).One of the few collaboration network studies in the Pacific region employed social network analysis in islands affiliated with the USA to understand climate change practitioner collaborations and to guide future opportunities (Corlew et al. 2015).Based on communications between actors, that study depicted a well-connected network of practitioners.While climate action is critical, there are also significant gaps in the state of knowledge of sector-specific climate change impacts and who is involved in filling those.
In the Pacific region, there are mounting food and nutrition security concerns, growing dependence on food imports in the region, and constraints from urbanization (Andrew et al. 2022;Brewer et al. 2023), while up to 80% of Pacific Islanders simultaneously rely on local food production for meeting daily food or livelihood needs (Georgeou et al. 2022).Considering the centrality of food security issues and the established vulnerability of food systems to pressures from climate change, there is a strong impetus for further research cooperation and collaboration on the topic to inform decision-making and action in holistic and effective ways.Yet, in the PICTs, there have been no studies specifically examining research collaboration networks generally, nor those focused on climate change impacts in food systems.
This study aims to fill an important gap in our understanding of both food security and collaborative scholarship in the Pacific region by describing the co-authorship network for research on climate change and food systems within the PICTs.It follows the completion of the Pacific Climate Change Science Program (PCCSP) in 2011 (Power et al. 2011), looking at the landscape of research produced from 2010 onwards.This paper includes studies in one or more of the 21 PICTs.2

Methods
Building on the existing scholarship around the value and challenges of research collaborations, here we seek to examine what the co-authorship network structure can tell us about the nature of research collaboration on food systems and climate change in PICTs.Specifically, we analyze three aspects of the network: (1) Who are the most influential actors in the reported research network?(2) How centralized or diffuse is the structure of the reported research network?(3) How has the composition of the network evolved over time?
This study draws on data from a systematic review of climate change impacts on food systems in PICTs to map the co-authorship ties between different research actors.We applied social network methods to quantitatively and visually analyze the network and supported the implications of the findings with qualitative expert focus groups.

Data generation and collection
Network data was generated through a systematic review to identify both peer-reviewed and gray literature on climate change impacts in PICT food systems (methods are 153 Page 4 of 14 detailed in Trudinger et al. 2023).Studies were identified through keyword searches on Google Scholar, Web of Science database, Google web, and target searches of organizational websites and thus were restricted to those available online.Included studies needed to explicitly document one or more climate change impacts on one or more parts of the food system in at least one PICT.We then extracted information from 104 papers, including what aspects of the food systems were studied, what type of methods and data were included, the co-author affiliations, type of organization, and location, among other details.For the purposes of this analysis, research collaboration is operationalized using co-authorship (Bukvova 2010).Actors were classified into "types" based on their organizational affiliation: academic institution (i.e., university), non-academic research institution, bilateral donor (e.g., national aid agency), consultancy/private (e.g., private research and advising organization), government (e.g., ministry, department), regional and multilateral organization (i.e., institutions that transcend national borders but involve national governments), non-governmental organization (e.g., not-for-profit organization not affiliated with a government), and other.
While this study focuses on the co-authorship network arising from the systematic review, three workshops and a focus group (in January and February 2023) were also conducted with 12 experts across the region to support and complement the findings of the review (Table S1).Participants were intentionally selected to represent the Pacific's three sub-regions: Polynesia (Samoa), Melanesia (Vanuatu, Fiji), and Micronesia (Kiribati), and on the basis of their expert knowledge of climate change impacts and adaptation in food systems and regional context.At least one participant represented national government, academia, NGOs/ civil society, the private sector, and regional institutions.Participants were identified through consultation with the Secretariat of the Pacific Community (SPC) and the Australian Department for Foreign Affairs and Trade (DFAT) during a co-design process and through the systematic review itself.As such, the workshops and focus group provide narrative and context to complement the network analysis and reflexive insight from representatives within the research network.Participants were asked about gaps in the evidence to inform climate change adaptation in food systems and who the key players are in driving this research.The limited size and scope of these expert consultations mean this information was used primarily for illustrative or corroboratory purposes.
Human research ethics approval for the workshops/focus group was obtained through the ANU (approval #2020/685).The distribution of participants by type of organization and regional representation can be found in the Supplemental Material (Table S1).

Data analysis and visualization
We first summarized the review data, including the number of unique organizations, their geographic locations, and the number of studies in which different types of organizations were represented.
To address the question (1) "who are the most influential actors in the research network?," we identified the actors and organizations most well represented for the entire study time period (2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021)(2022).A two-mode network was constructed, where individual studies and the affiliations of authors were represented as two types of nodes, with co-authorship serving as the ties.If more than one author from the same institution authored a study, this was designated as a single tie.We calculated degree centrality scores, which is a direct measure of how many studies an actor authored, normalized by the number of potential studies (n=104).Values were normalized based on the number of shortest paths possible for each node type within the two-mode network (Borgatti and Everett 1997), and the network was mapped to visually depict the network structure.We then simplified the two-mode network to one-mode, in which organizations were represented as nodes and shared authorship on a study was represented as a tie.The tie was weighted based on the number of studies two nodes co-authored.Using the one-mode network, we calculated weighted betweenness centrality, which is measured as the number of shortest paths a node lies on, and is an indication of influence.
The second question deals with an overall network structure, asking (2) "how centralized or diffuse is the structure of the research network?"For this, we used the one-mode network and calculated network density and centralization.Density measures the proportion of actual to potential ties in a network and serves as a proxy for network cohesion, with values closer to one indicating greater coherence (Zare-Farashbandi et al. 2014).Centralization, on the other hand, is the level to which a few nodes have the greatest number of connections in the network and can indicate the degree of core-periphery structure or inequality in the distribution of influence (Bordons et al. 2015).Values closer to one indicate a greater network centralization and less even distribution of centrality.We also looked at clustering within the network, comparing two different community detection algorithms that have different strengths in detecting clusters.The Louvain Community Detection Algorithm (LCDA) maximizes connections within communities and minimizes connections between, using a "greedy" optimization function to increase modularity.Conversely, the label propagation algorithm (LPA) is a diffusion method that converges when each node has the same label as the majority of its neighbors.The latter method is better at detecting large clusters and the presence of a core group of actors.Therefore, for the whole network, we present the only LPA results in the main text but include results for both algorithms in the supplemental materials.Only the LCDA was used for the time series analysis, as LPA over-clusters small networks.We summarized the geographic and institutional composition of the largest clusters (>5 members) and calculated the Krackhardt e/i ratio as a measure of homophily in the networks-the tendency to connect to nodes of similar attribute (Everett and Borgatti 2012).
Finally, the third question asks (3) "how has the composition of the network evolved over time?"To address this question, we split the publications into four periods of time: (A) 2010-2013, (B) 2014-2016, (C) 2017-2019, and (D) 2020-2022.Three-year increments were chosen to ensure adequate studies to create a network, as well as account for the annual variation in publications due to the length of the publication process and the delay between the time of carrying out a study and actual publication.A separate one-mode network was constructed for each period, we calculated betweenness centrality scores for individual nodes to compare across years and documented which ties repeated over multiple time periods.We also ran a Louvain clustering algorithm for each period, to see if cluster membership and community structure shifted over time.Finally, we summarized how the representation of organization types was tracked over time.
Workshop and focus group outcomes were transcribed, and responses to the two questions on evidence gaps and research roles were considered in the interpretation and discussion of the implications of the network analysis.We note that the qualitative data is limited by the perspectives available and is only included in this study to nuance the findings of the network analysis.

The seascape of organizations involved in research
A total of 158 unique institutions (not distinguishing departments or sub-divisions within an organization) were engaged in the research reviewed (Table 1).Academic institutions were the most prevalent author institutions in the reviewed studies (n=104) on climate change impacts on food systems, making up half (50%, n=79) of all actors and represented in over half of studies (60%, n=62).Regional and multilateral organizations were similarly highly represented in the studies reviewed (51%, n=53); however, there were far fewer unique actors (see Table 1), suggesting that only a few organizations were involved in a large number of studies.For example, the SPC-the principal scientific and technical organization, governed by its 27 country and territory members, that supports development in the Pacific region-was involved in 24 of the 53 studies with regional or multilateral authorship.Similarly, non-academic research institutions were well represented in studies (28%, n=29) but made up only 6% (n=10) of the unique actors.
In terms of geography, individual countries of the "Global North" are best represented by institutions based in Australia (20%, n=32), the USA (15%, n=23), and France (9%, n=14).However, upon aggregating institutions across the Pacific region, the actors based in PICTs were the most common (22%, n=34).Breaking geography down by institution type, academic institutions were mostly based in Australia (25%, n=20), the USA (15%, n=12), or the UK (11%, n=9).The only bilateral donors (e.g., aid agencies) represented in this study were from Australia, the USA, and New Zealand.Government authors were primarily based on the PICTs (72%, n=18), while NGOs were mainly from the USA (42%, n=5).See Supplemental Materials for a list of all the organizations included in the study (Table S2).
It is clear there was a subset of highly prolific organizations in the co-authorship network.Figure 1 shows several prominent and highly influential actors in the research landscape, connected to other actors through direct collaboration on a publication or on the shortest path between unconnected actors (betweenness).For example, SPC, the Commonwealth Scientific and Industrial Research Organisation (CSIRO), The Australian National University, the University of the Sunshine Coast, and the Secretariat of the Pacific Regional Environment Programme (SPREP) had the five highest weighted betweenness centrality scores and were represented on publications that link them to the most other actors in the network.The network map also illustrated the occurrence of disconnected pieces of research, primarily focused on food production.Isolation does not necessarily mean an organization produced few studies.For example, the World Bank had a high degree centrality score (0.13) due to its 13 studies; however, the organization did not collaborate with others on the production of this research (so has a normalized betweenness centrality score of 0 in the one-mode network).See Supplemental Material for the full table of centrality scores (Table S3).

Network structure and community detection
The one-mode actor network depicted the semi-centralized nature of the research collaborations in the region (betweenness centralization = 0.23), highlighting a core-periphery structure of one connected and 16 disconnected network components.Network edge density was low (0.06), likely as a product of the high number of unconnected actors.The clustering algorithms corroborated the visual observation of a centralized core and disconnected peripheral actors.
The label propagation algorithm (LPA) detected one primary community of research collaborations, consisting of 79 (50%) actors (Fig. 2).An additional 25 communities were identified, including 16 unattached to the main network.
These were mainly single publications with one or multiple authors.
In the primary community, SPC could be considered the most influential actor due to its high betweenness centrality.This community had a moderate level of heterogeneity of organization types (e/i = 0.22)-still dominated by academic institutions-and regions covered (e/i = 0.36).The next largest communities had only 9 members each, one focused around the University of Newcastle and the other with the World Health Organization and the University of Auckland.These were more homogeneous in the type of organization (e/i = −0.5 and −0.4,respectively), with the former dominated by national meteorological organizations, and the latter primarily composed of universities.One community centered on the University of Oxford had high heterogeneity for organization type (e/i = 0.78) but lower heterogeneity for regional representation (e/i = 0.22), with actors mainly from the Global North.Finally, the most diverse community centered on The Nature Conservancy and had representation by non-governmental, academic, and government actors (e/i = 0.64) and across five regions (e/i = 0.82).The composition of the top clusters can be seen in Fig. 3, and Supplemental Material (Tables S4 and S5).Bridging ties between communities are also evident, connecting the core network community to other smaller communities.In these instances, there is only one tie linking a more peripheral community to the core community, such as between the International Food Policy Research Institute (IFPRI) and the University of Oxford, The Australian National University and Fiji National University, and SMEC Australia Pty Ltd and Griffith University.While most of the actors serving these roles are academic institutions (University of Queensland, University of Exeter, University of the South Pacific, The Australian National University, University of Washington), there are also a couple of prominent research organizations (WorldFish and CSIRO) and the SPC.

Network evolution over time
The number of studies published per year has generally increased in the last decade and notably since 2016.In the final period under study, from 2020 to 2022, the number of studies more than tripled from the first period (2010)(2011)(2012)(2013).Studies that took system approaches grew over this time (Fig. 4).The regional representation of organizations in studies has also shifted slightly over time.While Australia (Oceania) has been a dominant force for the period of this study, other regions grew in representation since 2016.European and American actors, primarily academic institutions, became more involved, and actors based in PICTs-mostly government and regional institutions-have taken more ownership over research in the region (Fig. 5).Although academic institutions were still the most prominent, there has been a rise in non-governmental, governmental, and regional or multilateral organization engagement since 2016 as well (Supplemental Materials, Figures S4 and S5).
Key influential actors fluctuated over the time periods under study.For instance, from 2010 to 2013, CSIRO had the highest degree of centrality, which only rebounded again in the 2020-2022 period (Supplemental Materials, Figure S6).Conversely, SPC retained a relatively consistent and comparatively high degree of centrality during all four periods.In terms of connectivity, SPC retained its position of influence with the highest betweenness Fig. 2 One-mode network with actors colored based on their community, identified using the label propagation algorithm.Green denotes the core cluster.Nodes are scaled by their normalized betweenness centrality scores 153 Page 8 of 14 centrality for the first three time periods, with the rise in importance of SPREP for the final period (Figure S7), after a new partnership to improve Pacific data management, and sharing3 was announced in 2019.The heterophily of organization type decreased over time-meaning actors of the same organization type published together more often-while the regional heterophily increased over time, moving from Oceania-based institutions dominating publication to more trans-national and inter-regional partnerships (Supplemental Material Table S6, Figure S8).
Finally, there were some shifts over time in the size and composition of communities detected in each of the four networks (Supplemental Materials).In the first time period, two main communities emerged (Figure S9): one primarily with collaborations between French and Australian academic institutions and the second between Fig. 3 a The organizational composition of communities with more than five members.b Regional representation of actors in communities with more than five members SPC and a diversity of actors.There were also several disconnected communities.In the second time period, the network became more dispersed (aligning with the drop in network density), with three primary communities emerging (Figure S10).SPC served as a central hub connecting European and Australian actors.Another community was mainly composed of Australian academic and research institutions.The Australian National University acted as an anchor for a community of mainly Australian, New Zealand, and Japanese universities.The third time period exhibited even greater modularity, although SPC remained a central actor in the largest community (Figure S11).McGill University served as a link to a community of diverse actor types and geographies.University of the Sunshine Coast and University of the South Pacific served as central nodes in two bridging communities in the core network, composed of a mix of universities and local Pacific institutions.In the final time period, network density increased again, returning to a more tightly-knit core network (Figure S12).Although SPC was still prominent in one of the main identified communities, SPREP was positioned as a prominent bridge between a community of Fig. 4 Studies included in the review, organized by the primary food system components assessed and the year of publication.The dot size corresponds to the number of individual studies.The total number of studies has grown since 2016, though the relative focus does not seem to have altered substantially a majority Francophone institutions and a community of universities and Pacific organizations.The last major cluster comprised geographically diverse universities, with the Sorbonne, Wollongong, and Conservation International as the most central actors.The prominence of universities corroborates the decreased heterophily of organization type, yet the rich transnational collaborations support the rise in heterophily of region (Figure S8).
While no ties repeated themselves across all four time periods, there were 35 interactions that repeated at least once.SPC and CSIRO, SPC and University of Wollongong, SPC and University of New South Wales, SPC and Collecte Localisation Satellites, and CSIRO and University of Wollongong co-authored publications in three of the four time periods.SPC had another seven co-authorships that repeated in two time periods, including one with the highly central Australian National University.Other repeat co-authorships between the top influential actors included CSIRO and the University of the Sunshine Coast (periods 2 and 4), the Australian National University and the University of the Sunshine Coast (periods 2 and 4), and CSIRO and SPREP (periods 3 and 4).The greatest number of repeat co-authorships were between periods 3 and 4, as the network expanded substantially.

Expert insight on research collaboration in Pacific Island Countries and Territories
The expert workshops highlighted a number of gaps in the research on climate change and food systems in the region and reflections on who is and should be involved in moving this forward.Generally, workshop participants agreed that in the region there is a lack of basic baseline and regular collection of data on food security, related to both production and access, which makes it difficult to accurately study the impacts that can be attributed to climate change.Overall, the level of research collaboration was perceived as inadequate.Several participants recognized the need for more participatory and applied research, which could entail more collaboration between government or research institutions and non-governmental or community-based organizations.Other participants suggested more interaction between academics and government or the private sector, as well as across food system sectors.What such interaction should look like in terms of planning, execution, and sustainability needs unpacking and is potentially a subject for further conversations among the stakeholders.This theme reflects the high levels of homophily for organization type found in the network analysis, though the decrease in homophily over time also aligns with a recognition that more collaboration across organization types is needed.Finally, there were some comments about who sets research priorities and a need to consult stakeholders rather than base these solely on government and academic interests.Farmer-or user-led research could improve applications of key findings and innovations.One participant from Vanuatu argued for more research into the impacts of climate change on traditional farming practices-such as utilizing local and wild food crops-which would require the consultation and active engagement of producers who employ these practices.

Discussion
This study set out to understand the composition, structure, and evolution of the research collaboration network on climate change and food systems in PICTs.Overall, the results highlight a moderately centralized research collaboration network with a few influential actors, particularly from regional and academic institutions.The growth and diversification of the network over time speak to the importance of the food security issues under study, while strengthening of Pacific engagement in the network suggests growing consideration for regional interests and needs.The ensuing discussion elaborates on the composition, structure, and evolution of the research collaboration network and contemplates the implications for future research on food systems and climate change in the region.We conclude this section by addressing the limitations of using co-authorship to examine collaborative research on an applied topic like climate change and food systems and recommend complementary research to develop a more complete picture of the knowledge-generating landscape in PICTs.

Universities abound, but regional organizations knit the network together
Existing literature alludes to the power dynamics in research collaboration networks, with universities shaping the design and management of research projects (Bukvova 2010), such as through the Australian Research Council Linkage scheme (McCabe et al. 2021).Our study, which drew on both peerreviewed and gray literature due to the applied nature of food security research, reaffirmed the prevalence of academic institutions across the network.However, in terms of power, the five most influential actors (highest betweenness centrality scores) comprised two Pacific regional organizations-SPC and SPREP-and three Australian research bodies-The Australian National University (ANU), CSIRO, and University of the Sunshine Coast.While the governance architecture of the Pacific Islands region continues to shift and adapt, the prominence of external actors reflects the historical role of Australia in Pacific governance and its continued involvement in regional bodies like the Pacific Island Forum (Tarte 2014).
The prominence of SPC in the network is important because of the Secretariat's mandate as a technical body that empowers national governments to direct the focal points of any research.As one SPC participant in the workshops discussed, the Secretariat houses technical networks wherein country researchers promote their research priorities.While this does not ensure that local research needs are necessarily covered, it does at least place the direction of those research priorities in the hands of the PICTs themselves.Local, regional, and traditional ownership over research is considered particularly important for decolonizing research processes and outputs (Smith 2012).The workshops also highlighted that there are some important sub-national variations in climate impacts on food systems.As such, grassroots voices and local researchers have a role to play in providing insights into these changes at the sub-national level.However, national and sub-national level studies and actors were conspicuously underrepresented in the literature reviewed.
The limited role of Pacific academic institutions in the research collaboration network has implications both for decolonizing research and for filling important knowledge gaps at sub-national levels.Yet there are a number of barriers to overcome in order to improve this regional representation.Funding is one of the main constraints, with no Pacific regional or national research funding bodies (comparable to the Australian Research Council or the National Science Foundation in the USA), and an associated lack of public investment in higher education (Healey 2022).In response to a dearth of qualified teaching staff, academic researchers face a heavy teaching load that undermines the production of research outputs.Many Pacific Islander researchers also end up working in institutions outside of the region, leading to a "brain drain" from PICTs (ibid).Finally, while COVID-19 might have provided a wake-up call to emphasize the importance of food system resilience beyond the farm (Davila et al. 2021), there are also persistent local stigmas against agriculture and food production as career paths, further impeding the study of food system topics in the region.As such, the need to change young people's negative perceptions about agriculture as a non-career and to broaden awareness about the core role of food production in strengthening food security, health, business, and research has never been more critical.Integrating agriculture into both primary and secondary school curricula could be part of the solution, but it is also critical for key actors in PICT governments and regional bodies like SPC and the University of the South Pacific to make concerted efforts to increase awareness and meaningful nurturing of potential agricultural specialists.

Centralized core with isolated periphery
The structure of a research network has a bearing on the flows and integration of different forms of knowledge.This analysis depicted a clear structure of a "global" core with peripheral local cores in the research collaboration network on Pacific Island food systems and climate change.This was a similar structure to that identified in a study on climate-based communication between professionals in the Pacific (Corlew et al. 2015).However, the clustering and characteristics of actors within this core-periphery structure have implications for the inclusion of valuable perspectives in research formulation or integration of existing knowledge and understanding.
Within the main core network, five larger clusters of actors emerged, varying in their levels of organization-type heterogeneity.The largest cluster, with SPC as its most influential node, had only modest levels of heterophily for organization type, suggesting a slight tendency toward co-authoring with different types of organizations.Two other clusters with universities as their most central actors demonstrated homophily of organization type, tending to co-author with other academic actors.The final two smaller clusters showed high levels of heterophily for organization type and included representatives from NGOs, universities, government, and research organizations.These latter clusters reflect the findings of other studies that demonstrated how heterophily can breed creativity (Huang 2014) and that collaborations with different types of organizations can yield more innovative or even transformative research (Faems et al. 2005;Tether and Tajar 2008).For a complex topic such as food security under climate change, the infusion of novel or out-of-the-box ideas from new sources can be particularly beneficial for identifying potential transformative adaptation pathways (Dowd et al. 2014).
Workshop participants corroborated the value of multiactor collaborations, emphasizing the need for government institutions and NGOs to work together to formulate research projects.For instance, one participant from a farmer support organization in Vanuatu noted that "most of the time, government is doing their own thing and NGOs are doing their own thing."Similarly, a university participant stressed the need for more interaction between academics and policy actors, stressing the importance of "multiple touch points."This indicates a recognized opportunity for the development of climate and food boundary organizations with the explicit purpose to bridge and translate across science, policy, and implementation.While there were also calls for participatory and farmerdriven research to ensure studies connect to on-the-ground priorities (university participant), the extent to which this is pursued is difficult to glean from a co-authorship network.

Network has become more complex, but also more diverse, over time
The collaboration network on climate change and food system research in PICTs has become larger, more complex, and more diverse over time.Notably for regional representation, the SPC was consistently central throughout, while the prominence of SPREP emerged over the four time periods.The sheer number 153 Page 12 of 14 of studies also increased substantially since 2016.This reflects the increasing urgency of the issue over the last half a decade (Schipper et al. 2021), as well as the strengthening leadership of the Pacific Islands in climate change efforts (Carter 2020).Major research and development initiatives aiming to enhance research, science and technology innovation, and collaboration coincided with this uptick.For instance, the Pacific Island Universities Research Network was founded in 2013, and PACE-NET-a cooperation between PICTs and the European Union-concluded in 2016.Besides more published studies over the study period, there were also more non-academic actors represented and a greater tendency for cross-regional collaborations.Hearkening back to the previous section, this could also reflect a growing tendency for interdisciplinary, collaborative, and grounded research on climate change (Cundill et al. 2019;Currie-Alder et al. 2020;Schipper et al. 2021).
Although the expansion of the network is visible, it is also critical to note that key influential collaborations have persisted over the study time period.Repeated partnerships have been shown to build trust and can enhance the value and relevance of collaborative projects (Fritsch and Kauffeld-Monz 2010;Young and Freytag 2021).All of the collaborations that repeated for three of the four periods were between either SPC or CSIRO and a university or research institution (not based on PICTs).SPC held the most repeat co-authorships, reinforcing the role of the regional body as pivotal in shaping research, and doing so with the support of these longer-term relationships.Finally, the greatest number of repeat co-authorships were between time periods 3 and 4, as the network grew substantially.This indicates the potential for further strengthening of the research collaboration network over time.That said, one workshop participant from a national research institution indicated active efforts to form new partnerships, suggesting not only the prospect of further expansion of the network, but also relationships that will require close cultivation in years to come.

Limitations and future research
While the network of co-authorship on food systems and climate change research provides a useful starting point for understanding the actors shaping knowledge generation in the Pacific region, there are a number of limitations in the methods and data that merit further examination.Importantly, not all research collaborations lead to available publications, and not all collaborators are included as co-authors (Bukvova 2010).As noted earlier, traditional knowledge is likely not well captured in the published literature, and so relevant actors might have been excluded, yet it can play a critical role in understanding and adapting food systems in the region to climate change (Chand et al. 2014;Mcleod et al. 2019).Integrating traditional knowledge-which could include the use of natural cues to modify farming practices or cultivating crop varieties passed down over generations-with the existing Western scientific understanding could be a fruitful direction for future applied or participatory research collaborations.Furthermore, for organizations that are more involved in implementing projects or activities on the ground, such as private entities/consultancies and NGOs, collaboration may focus on rural development, social cohesion, capacity building, etc. and may not result in published literature.While these organizations often do produce reports and other gray literature, they still may be underrepresented in this study due to their emphasis on other types of knowledge production outputs and outcomes.As such, further research could expand the network to include partnerships on projects and funded proposals in the region.
This analysis of research collaboration could be expanded to other regions, to see if there are comparable patterns in other networks.While Australian and Pacific Regional government actors were prominent in this network, government actors from other geographies (e.g., North America) were conspicuously absent.Conducting similar analyses in other contexts could help capture where the research efforts of these institutions are concentrated, for a more global snapshot of research collaborations on climate change and food systems.Food systems and climate change work is also a relatively new field of inquiry, so some actors may become more prominent as research programs come online and become established, such as those at the University of Hawai'i and the University of Guam.Finally, many of the local Pacific government agencies were not represented in this network.This could stem from the tendency for much of the sub-national research to be housed as hard copies in local research institutions.Without digital copies, more investigative work would be needed to uncover the extent of knowledge currently stored.However, due to the nature of this research, it likely has not been widely shared and would not be well integrated into the regional networks.
The gaps in co-authorship connections within the network could also arise due to the challenges associated with working collaboratively.Research collaborations can be difficult, often requiring mediating multiple perspectives or shifting whole philosophies of research management (Currie-Alder et al. 2020;Huang 2014).Therefore, it can be difficult to determine to what extent co-authorship equates to meaningful co-production and the reciprocal participation of all collaborators.Co-authorship does not provide a benchmark for power dynamics within a collaboration nor how closely the knowledge co-production meets a theoretical ideal (McCabe et al. 2021).Teasing apart the quality of collaboration would require additional methods, such as interviews with project partners.

Conclusion
This study examined the co-authorship network for research on climate change and food systems in the Pacific Island Countries and Territories.On the one hand, it confirmed trends that Page 13 of 14 153 have been previously documented in the literature, such as the dominance of academic institutions from the Global North and the tendency for a primary core of collaboration to emerge.On the other hand, our analysis showed sustained engagement by Pacific regional organizations-namely SPC and SPREP-and the increasing focus of research on the nexus of climate change and food security.The growing interest in Pacific regional visibility and participation (such as in efforts like the Intergovernmental Panel on Climate Change reports) is a potential positive step toward building resilience in food systems.However, the paucity of Pacific national and sub-national actors in the coauthorship network, and the related comments raised during the workshops, suggest there are lingering capacity constraints to meeting this demand.Capturing this research collaboration network in the midst of a substantial expansion and a transition to even stronger regional leadership does stir up questions about the future evolution of the network.Will calls for greater ownership of research by Pacific actors result in more prominent authorship?Will this drive resources toward building and retaining the necessary expertise in local academic and research organizations?And what role do the historical collaborators, such as those in Australian institutions, play in supporting future research development and capacity building in the region?While only time will tell how these transformations unfold, the products of this network have never been so critical to ensuring PICTs have the knowledge and tools needed to safeguard the food and nutrition security of their populations as the climate continues to change.

Fig. 1
Fig. 1 Whole network with author institutions in blue and publications in red.Ties represent authorship.The size of the node reflects the normalized betweenness centrality for the actors

Fig. 5
Fig. 5 Representation of different regions in reviewed studies over time.Dot size corresponds to the number of individual studies in which institutions from the region are represented

Table 1
Actor types represented in the review.Percent representation was based on a total of 104 studies and 158 unique actors