Decision‐making factor interactions influencing climate migration: A systems‐based systematic review

Global migration and mobility dynamics are expected to shift in the coming decades as a result of climate change. However, the extent to which migration is caused by climate hazards, in contrast or addition to other intervening factors, is a point of debate in literature. In this study, we conducted a systematic literature review to identify and consolidate factors which directly and indirectly influence climate change migration. In our review of the literature, we found a total of 21 economic, environmental, demographic, political, social, and personal intervening decision‐making factors which affect climate migration. Causal interactions between these factors were identified using an axial qualitative coding technique called purposive text analysis. By combining causal links, a semi‐quantitative causal loop diagram was created that represented factor interaction and feedback within the “climate migration system.” Using this model, we highlight influential feedback loops and point to how intervention strategies may cause downstream effects. This research helps to address calls for a better understanding of the complex decision‐making dynamics in climate migration. In particular, results from our causal feedback loops show that intervention strategies targeted toward economic factors such as financial capital and livelihoods, as well as food security, would have the greatest impact in assisting climate‐affected communities. These results help inform climate migration policy and aid planners in the future to better understand the interconnected system of factors that lead to the emergent outcome of climate migration.


| INTRODUCTION
In 2020, more than 30 million new people became displaced as a result of weather-related disasters (IDMC, 2021). As the impacts of climate change increase, global population movement will also continue to experience larger shifts in people on the move (IPCC, 2019). Although there is a relationship between climate change and mobility, it does not necessarily follow that those affected by climate hazards will move (Duijndam et al., 2022;Stojanov et al., 2014;Wisner et al., 2004;Zander et al., 2016). Migration relating to climate change results from multi-faceted decision-making and the confluence of multiple factors (Black, 2001;Oliver-Smith, 2012;Xu et al., 2020;Zander et al., 2016). Although past studies have contributed significantly to the identification of decision-making factors that drive migration (Parrish et al., 2020;Piguet, 2013), a gap remains in understanding exactly how these factors interact to lead to migration (Bates-Eamer, 2019;Piguet, 2022;Willett & Sears, 2020).
In this study, we address these gaps and advance understanding of migration processes through a comprehensive systematic literature review of the decision-making factors of climate migration and their interactions, asking the questions: RQ1. Within the existing literature, what are the decision-making factors that influence migration for households affected by climate change?
RQ2. How do these decision-making factors interact to create climate (non)migration?
To answer these questions, we begin by presenting a brief background on migration relating to climate change, relevant terminology, and existing climate mobility frameworks in the literature. We then detail our systematic literature review which identified 21 factors influencing migration decision-making, and the interactions between these factors, from 206 studies. We then used causal loop diagramming to map and analyze the interaction of these factors. Through the identification of top feedback loops in the climate migration literature, this study lays the groundwork for policy makers and future researchers to understand the systemic and dynamic drivers of climate migration so that the consequences of climate change may be better anticipated.

| BACKGROUND
Population movement due to environmental challenges is not a new phenomenon (Stojanov et al., 2014). Historical migration practices are deeply embedded into many cultures and act as risk management strategies to cope with periods of environmental hardship (Farbotko et al., 2018;Opeskin & MacDermott, 2009). Residents of seasonally dry environments, for example, employ circular migration in order to diversify livelihood opportunities during nongrowing periods (Milan & Ho, 2014;Shi et al., 2019;Willett & Sears, 2020). Recently we have seen examples of climate migration in hazard-prone locations such as California in the United States (Casey, 2022) as residents looked to escape future wildfires. Although most of the global population remains relatively sedentary (McAuliffe et al., 2020), mobile and mixmobility lifestyles are also a significant aspect of global population dynamics (Kelman, 2019) leading many scholars to argue that migration relating to climate change does not imply instability, but is a recognized resiliency strategy .
However, while environmental and climate changes are known to catalyze migration, identifying and classifying instances of climate migration can be unclear. Due to inconsistent migration movement, including episodes of movement and nonmovement, as well compounded migration stimuli, difficulties arise in delineating classifications of people on the move. Especially when coupled with other drivers, many scholars argue that environmental or climate migrants are difficult to separate from other migrants, such as those moving on economic grounds Oliver-Smith, 2012). The classification and terminology used for those whose movement is related to climate change is a debated subject in which there is a marked lack of consensus. Renaud et al. (2011) offers typologies such as environmentally motivated, environmentally forced, and environmental emergency migrants to classify such populations, while the International Organization for Migration (IOM, 2019) simply uses environmental or climate migrants. In contrast, Boas et al. (2019) recommends moving away from the term "migration" altogether in favor of climate mobilities to capture more diversity in the duration, direction, and multicausality of movement. The terms "environmental or climate refugee" have also quickly fallen out of favor. While used in the past, these terms are now widely agreed to be technically incorrect as environmental justification is not given within the 1951 Convention Relating to the Status of Refugee's definition of refugees. Many also argue that "refugee" deceptively implies mono-causal movement and denies agency of those affected by climate change (Bettini, 2013;Constable, 2017;Olwig & Gough, 2013). Using the knowledge and learning from literature, we provide a summary of operationalized definitions used in this study to describe mobility, migration, factors, and migrants in Box 1.
Along with the evolution of terminology, conceptual frameworks of migration have also evolved to better include the agency and autonomy for those affected by climate change. McLeman and Smit (2006) developed an algorithmic flowchart style model to determine migration response to climate change, accounting for a community's capacity for adaptation. Later, Black et al. (2011) proposed a conceptual framework identifying five main families of drivers which affect migration decision-making: economic, political, social, demographic, and environmental. As the impacts of climate change become more apparent, existing vulnerabilities within these five families are multiplied and contribute to how mobility decisions unfold (Harper, 2021). Later work, such as that of Parrish et al. (2020), has since expanded Black et al.'s (2011) framework into new conceptual models showing driver dynamics over time, space, and society. Subsequent case studies have applied these frameworks to specific hazards and cultural contexts to help explain migration decision-making (Ayeb-Karlsson et al., 2016;Hauer, 2017;Khavarian-Garmsir et al., 2019;Parnell & Walawege, 2011).
In this work, we are closely aligned with Beine and Jeusette (2019) who used statistical analyses to identify positive and negative relationships between factors influencing climate migration. Similarly, we used qualitative coding to highlight positive and negative linkages in the literature while taking a new and progressive step of analyzing these linkages through causal loop diagramming. Although interactions between influencing factors and migration are well acknowledged in past reports and literature (Foresight, 2011), we present explicit interactions not only between identified factors and migration but also between the factors themselves. Through this, we can analyze both direct causal relationships and indirect relationships between factors may contribute to non(migration).

Mobility
The ability to move freely and easily. This is the umbrella term for movement and includes a spectrum of varying degrees of agency (e.g., forced displacement, migration, voluntary tourism) (Thornton et al., 2019;Warner, 2012).
Migration The process of moving within or across borders, temporarily, seasonally, or permanently. Migration falls along the continuum between forced and voluntary mobility (Piguet, 2018). Migration in this text is shorthanded for what is more rightfully termed "migration influenced by climate change" (Foresight, 2011).

Factors
We use "factors" synonymously with drivers. Similar to how climate is not a direct linkage to migration, the presence of the factors identified does not necessarily imply movement and may only influence the decisionmaking of climate affected households or individuals.

Migrants
The term, "migrants" in this text refers specifically to those whose movement is related to climate change. As the articles in this review focus on slow-onset climate hazards, we highlight the agency of individuals in responding to these gradual changes as opposed to individuals forcibly displaced by rapid-onset hazards. For more reading on migrant typologies refer to Renaud (2011).

| METHODS
We conducted a systematic literature review to identify the interaction of factors influencing climate migration. We first discuss our process for the selection of relevant literature and the identification of migration factors and their interactions. Then, we describe how we mapped and analyzed migration factor interactions using purposive text analysis (PTA) and causal loop diagramming.

| Systematic literature review
A systematic literature review was employed to document the current state of knowledge on migration factors relating to climate change and how these interactions lead to migration decision-making. While long used in health research, systematic reviews are gaining attention within the field of climate change in order to summarize and aggregate existing knowledge on specific topics (Berrang-Ford et al., 2015;Ford et al., 2011). Our systematic literature review was adapted from Preferred Reporting Items for Systematic Reviews and Meta-Analyses and consisted of identification, screening, and eligibility review of articles (Page et al., 2021). A keyword search was used within Scopus, following the example of the Intergovernmental Panel on Climate Change (IPCC) which used this database in its review of scientific literature (IPCC, 2014). In comparison to other often used databases, Scopus was selected to capture the multidisciplinary nature of literature where migration scholarship is situated (Gusenbauer & Haddaway, 2020), offering higher coverage than Web of Science and a higher result consistency than Google Scholar (Falagas et al., 2008;Mongeon & Paul-Hus, 2016). Using a Population, Exposure, Outcome (PEO) format, modified from the Population, Intervention, Comparison, Outcome search strategy, keywords were selected for search terms ( Table 1). The PEO search strategy allowed us to clarify and define our search terms based on population, exposure, and outcomes set in our research question. The search covered the use of these terms in document titles, abstracts, and keywords. A Boolean "or" operator was used for any of the terms within the population, exposure, and outcome categories, while "and" operators were used across these groups. Wild card search strings (*) were added to terms that had several expected terms, such as "climate change" or "climatic changes." The search was limited to peer-reviewed journal articles written in English, with all dates included up until May 2021. As primary empirical data were the focus of the search, review and meta-analysis articles were excluded. The initial search yielded 665 articles. Article abstracts were then preliminary screened using the following exclusion criteria: • Noncontemporary climate migration (migration before 1970).
• Short-term mobility (i.e., daily mobility, movement in which dwelling remains consistent per week).
• Mobility not related to climate change (e.g., displacement from infrastructure projects).
Screening consistency was validated by the third author through an independent review of inclusion for 50 randomly selected abstracts. Following the abstract screening, 240 articles remained. An additional 34 articles were excluded after full text review showed them to be not relevant using the same exclusion criteria above, leaving a total of 206 articles used for analysis. Documents were then imported to Nvivo software for analysis of causality between factors influencing climate change migration to model factor interactions and dynamics.
T A B L E 1 Search terms for climate change migration

| Purposive text analysis
Casual factor interaction themes were deductively coded based on Black et al.'s (2011) conceptual framework with factors categorized according to "Economic," "Political," "Social," "Demographic," and "Environmental" families. Each causal factor interaction took the form of a pairwise connection; one factor influencing the other. In this analysis, only pairwise connections identified directly in each article's results were coded to avoid duplicate coding and skews to frequency data. This means that discussions of others' work which identified pairwise connections were not included in the coding. This coding methodology called PTA has been used in other system dynamics studies such as in Kim and Anderson (2012) for transcript data and Valcourt et al. (2020) for workshop interviews to build models of factor interactions. Examples are provided in Table 2 to show how causal statements in the literature were coded to form pairwise interactions. In the first PTA example, Opeskin & MacDermott (2009) identify that population pressures have led to depletion natural resources. As such, the entire phrase was highlighted and added to the coding theme of "population growth ! resource depletion." The quote also mentions population growth's effect on environmental degradation and was likewise added to the coding theme of "population growth ! environmental degradation." In the second quote, we provide an example of how in situ adaptation was recognized to affect migration. Although the relationship is negative, the quote was still coded to the relevant factors, while the polarity of the connection was assessed later using the polarity of the majority of codes for each pairwise connection. For each of the factors identified, a coding dictionary was maintained to ensure consistent identification of themes across the included studies.
The process of systematically coding pairwise factor interactions allowed for the development of an aggregated model of factor interactions, described in the next section. Although article excerpts were coded into pairwise factors to show relationships ("population growth ! resource depletion"), each individual factor was categorized as "Economic," "Political," "Social," "Demographic," or "Environmental." A full list of factors identified in literature as well as how they were defined for the coding is shown in Table 3.

| Modeling factor interactions with causal loop diagramming
Following the coding of pairwise factor interactions found in the literature, the identified connections were then analyzed using a form of qualitative system dynamics modeling known as causal loop diagramming. CLDs are an approach to study complex systems that have interdependence, mutual interaction, and circular causality known as feedback loops, with a goal of building theoretical understanding on systems leverage points for policy improvements (Richardson, 2020). In this study, we aggregate pairwise interactions between factors identified using PTA (addressing RQ1 and RQ2) to develop a CLD to analyze the feedback loops driving climate migration (addressing RQ2).
The systems insights gained from CLDs are derived from the combination of simple model components-pairwise interaction between factors (presented as arrows), interaction polarity (presented as + or À), and feedback loops. Interaction polarity allows for the characterization of dynamics within each pairwise connection, where a positive polarity (+) implies a direct influence (i.e., if Factor A increases, Factor B increases, or vice versa), and negative polarity (À) implies an inverse influence (i.e., if Factor A increases, Factor B decreases, or vice versa). The combination of pairwise interactions with interaction polarity (positive or negative) in a CLD enables the identification and characterization of T A B L E 2 Example of purposive text analysis (PTA) "There has been significant depletion and degradation of natural resources in some Pacific countries because of population pressures and over-exploitation." (Opeskin & MacDermott, 2009) Cause factor Effect factor Population growth ! Resource depletion Population growth ! Environmental degradation "Autonomous adaptation measures in flood-prone areas in the city of Belem have become a good instrument to enable low-income populations to stay in central parts of the city." (Szlafsztein & de Araújo, 2021) Cause factor Effect factor In situ adaptation ! Migration Note: This example shows a negative causal relationship where in situ adaptation leads to a decrease in migration.
feedback loops representing circular causality between two or more factors. A feedback loop can be characterized as reinforcing, causing an exponential increase in system behavior, or balancing, causing a goal-seeking or stabilizing system behavior (Sterman, 2000). Reinforcing loops are characterized within a CLD by an even or zero sum of negative polarities, while balancing loops are characterized by an odd sum of negative polarities. The combined output of pairwise interactions, polarities, and characterized loops allows for the hypothesis on the dynamic drivers of a complex and nonlinear problem, emerging from reinforcing and balancing loops interacting in concert. Refer to Box 2 for more information on understanding a CLD. Although CLDs can identify and characterize feedback loops driving a complex problem, they cannot ascertain which feedback loop dominates system outcomes, as all pairwise interactions are assumed to have the same strength (Luna-Reyes & Andersen, 2003;Richardson, 1995). In order to identify the dominant loops within the CLD emerging from our analyses, we added interaction strength based on the frequency a pairwise interaction was mentioned in the literature and used these strengths to evaluate factor influence. It is important to note that though the frequency of factor interaction occurrence in literature does highlight strength of evidence for connections, it does not necessarily point to importance in the overall climate-migration system. For this reason, we classified the strength of factor links into four categories: weak, moderate, strong, and very strong (corresponding to values of 1-4). The middle 50% of factor links, represented as interquartile range (IQR), were defined as having moderate strength as the value difference within this range was minimal. The lowest quartile of factor interactions was assigned as weak connections, while the highest quartile was defined as strong connections. Any factor pair that was more than 1.5 times the IQR above the upper quartile was classified as very strong. Vensim, a system dynamics modeling software, was used to consolidate the pairwise interactions identified with PTA into a single CLD used to identify feedback loops. As with all complex systems, factors within the climate migration system are deeply intertwined and well understood through a systems feedback loop approach. In our analysis, we focused on the feedback loops that "passed through" the factor migration. This enabled us to directly evaluate the dynamic processes within the diagram that influenced migration. These loops ranged from two to seven factors in length.

| Scoring and ranking feedback loops
Eigenvector centrality values were computed for each factor within the CLD, and loop scores were calculated as the normalized summation of eigenvector values for each factor within the loop. Eigenvector centrality is a measure of a BOX 2 : Reading a climate migration causal loop diagram

Symbol
Meaning Example

Pairwise interactions
Denoted by an arrow pointed from the causing factor to the effected factor. The thickness of the arrow connection is based on the factor pair's frequency of occurrence in literature, on a 1-4 scale.
The top left link shows that political stability affects migration.
The factors' circle sizes are directly proportional to their eigenvector centrality value (i.e., its relative importance in the system).
Eigenvector centrality values for migration, financial capital, and political stability are 1.00, 0.42, and 0.26, respectively.
In the presence of political stability, households are less likely to migrate (i.e., there will be a decrease in migration).

Reinforcing
Reinforcing loops can be identified as having an even or zero sum of negative polarities. These loops lead to exponential increase or decrease, continually compounding change in the system.
In the presence of political stability, households are less like to migrate. The less incoming migrants in a community, the likelihood for political stability increases.

Balancing
Balancing loops can be identified as having an odd sum of negative polarities. These loops cause a goal-seeking or stabilizing system behavior.
As a household's financial capital increases, their likelihood of migration also increases. However, through migration, relocation costs and higher costs of living in destination locations leads to decreases in financial capital.
factor's influence in a system or "network," where influence is calculated based on its connection to other influential factors (Borgatti, 2005), scored on a range of 0 (extremely low) to 1 (extremely high). This metric was used because it highlights not only a factor's individual connections to adjacent factors, but its relative importance in the whole system. The calculation of an eigenvector centrality scores requires multiple calculation iterations to arrive at a dominant eigenvector score for factors in the network. We calculated eigenvector scores based on the weighted and directed network (the CLD) using the "igraph" package in R (Csardi & Nepusz, 2006). Although a metric typically used in social network analysis, Murphy and Jones (2021) propose the integration of eigenvector centrality into systemic design as an indicator of potential leverage points in the system. In this study, we used eigenvector centrality values to score loops in order to evaluate loop dominance within the context of the whole system of factor interactions. Loop scores were based on the average eigenvector centrality values for factors included in the loop. For example, a loop with three factors having eigenvector centrality scores of 0.8, 0.5, and 0.7, respectively, would result in a loop score of 0.666. Loops were ranked by their scores to determine the dominant loop for each length (number of factors in the loop). By highlighting the top loop for each length, we show a progression of how factors are folded into varying dynamic layers as analysis of the system becomes more complex.

| RESULTS AND DISCUSSION
We first provide a condensed summary of main themes found through the systematic literature review. Next, we present a comprehensive list of migration decision-making factors from the literature (addressing RQ1). Through our use of systems analysis, we then show the interactions between migration factors and identify the dominant feedback loops within the CLD (addressing RQ2) driving climate migration systems.

| Emerging themes in climate migration literature
There has been rapid growth of the climate migration literature over the past three decades, as shown in Figure 1 with tallies of the annual publication counts on literature focused on the climate-migration nexus. Prior to 2010, we found few studies that explicitly examined migration decision-making processes to climate change and a 10-fold increase in articles published in the decade between 2010 and 2020. This evolution is well detailed by Piguet (2013) who posits a number of reasons for the noted lack of environmental mention in migration studies in the 1900s and its current resurgence. In his work, Piguet suggests the current uptick may relate to the politicization and securitization of climate migration (Boas, 2015), increasing natural hazards driving displacement, and growing climate change anxiety, among others. Fussell et al. (2014) also suggest that improved data collection methods and analysis tools may have contributed to this increase. Of the 206 total articles examined, 34% (n = 70) took a global approach using aggregated global data or a theoretical framing, 15% (n = 30) had focuses on Bangladesh, and 3% (n = 6) in both India and Mexico. In Figure 2, we provide a heat map showing darker colors for countries with high references in the climate migration literature. The focus of environmental migration literature on low-and middle-income countries has been previously documented in other reviews (Obokata et al., 2014;Piguet et al., 2018). In their review, Piguet et al. (2018) suggest that this uneven geography may be explained by an attitude of immunity held by high-income countries, leading to research and funding focused more on countries perceived to have less adaptive capacity. Other suggested reasonings for the research geography distribution include the attraction to high-risk regions, stereotypical racialization of "climate refugees" to low-income country populations , and a built-up fear that incoming migrants from low-income countries will lead to security threats in high-income countries (Boas, 2015).

| Identifying leading migration decision factors
We found 21 leading factors discussed in scholarship that influence migration relating to climate change. These factors were categorized into 'Personal and Household Characteristics/Intervening, Facilitators and Barriers," "Demographic," "Economic," "Environmental," "Political," and "Social" based on Black et al.'s (2011) classification of migration factors.

| Climate migration factor interactions
Using the interactions identified through literature, the next step in our study was to consolidate the relationships into a climate migration system. The 21 identified factors, as well as their relationships between each other, are presented as a CLD in Figure 3. As shown in Box 2, the CLD presents the interaction of factors within the climate migration system where link size is based on the the strength of relationship (1-4 quartile value), the factor and font size is based on the on the eigenvector centrality of the factor, and the factor color is based on the factor category.
Analysis of the climate migration CLD in Vensim revealed 155 unique feedback loops which included the migration factor. Within these loops, the top 10 factors with the highest factor occurrences are shown in Table 4 along with their respective eigenvector centrality score. Financial capital, food security, and livelihood were the most frequently occurring factors within these feedback loops.
Although the frequency of a factor's occurrence in the 155 feedback loops shows that it is a core factor in the climate migration system, the eigenvector centrality score shows its importance relative to other high influential factors in the system. For example, food security has a more frequent occurrence in the feedback loops than livelihood but we can see that livelihood has a higher eigenvector centrality score implying that while livelihood is more connected throughout the system, food security is a more acute driver of migration.

| Climate migration system feedback loops
In addition to ranking the most commonly occuring factors in the migration-focused loops we also ranked loops based on their overall loop score. Loop score was based on the average of eivenvector centrality values for factors included in each loop. In order to not overvalue loops with smaller lengths, and as a way to let the top feedback loops capture the nuance of climate migration, we show the top feedback loop(s) of each loop length in Figure 4. Some loop lengths had multiple feedback loops with the same eigenvector loop score, but only one was selected as a visual in Figure 4 for clarity. In these instances, each variation shared the same factor combinations in varying sequential order.
Through examining the growth of the top feedback loops, from two factor loops to eight factor loops, a pattern emerges showing migration and financial capital as the epicenter of dominant loop behavior, adding one new factor as the loop length increase. Starting with a feedback loop of migration and financial capital, the factors added in order of increasing loop length were livelihood, food security, health, political stability, environmental degradation, and resource security. These loops illustrate the most significant interactions and dynamics leading to climate migration, as identified in the climate migration literature.
A key finding is the interaction between migration and financial capital as the anchor for subsequent series of influential loops. Although some studies share cases of movement spurred by lack of financial capital (Bernzen et al., 2019;Dreier & Sow, 2015;Shi et al., 2019), more often studies highlight that those with stable finances are the ones who will move (Constable, 2017;Kartiki, 2011;Logan et al., 2016). This result was also found in Borderon et al.'s (2019, p. 528) systematic review of empirical evidence in Africa which concluded, "…even though climate change will increase population exposure to environmental hazards, high levels of poverty mean that a large part of African populations do not have sufficient resources to be mobile." A notable aspect of our coding scheme was the inclusion of remittances under financial capital. As one member of a household moves away and sends back money, the liquidity constraints of other household members are relieved, making them more likely to have the financial capacity to move as well (Bernzen et al., 2019). This point is, however, contested in literature where some studies have also found that instead of encouraging future migration (Kleemans, 2015;Tan, 2017;Tiwari & Winters, 2019), remittances may also be used for strengthening a family to remain in their home F I G U R E 3 Climate change migration system causal loop diagram location (Calero et al., 2009;Das Sharma et al., 2020;Shi et al., 2019). Indeed, many families choose to have a member migrate in order to diversify income and stabilize their finances against future climate shocks (Sakdapolrak et al., 2016;Scheffran et al., 2012). Postmigration, transition costs as well as higher costs of living in destination locations were reasons that literature pointed to decreases in financial capital after migration (Barua et al., 2017;Haque et al., 2020;Mallick & Sultana, 2017). The story that this loop (Figure 4a) implies is that as a household gains enough financial capital to choose to move, their capacity to move in the future is lessened because of the costs of their initial movement. In summary, this loop ( Figure 4a) supports arguments for liquidity constraint effects on migration, showing that individuals will migrate as this constraint is relieved and will have at least an initial negative financial shock after migrating.
The next factor added into the loops is livelihood. In the literature, opportunities for more secure livelihoods were a strong motivating factor for migration (Chen & Caldeira, 2020;Davis et al., 2018;Kolmannskog, 2010). As traditional livelihoods are strained under climate change, many people will venture into urban areas looking for work (Abah & Petja, 2016;Schwerdtle et al., 2021). Both failed livelihoods in origin locations as well as opportunities for more stable livelihoods in destination locations were strongly cited as motivating factors for migration. A significant criticism towards the idea of "climate migrants" is that it is difficult to separate climate-influenced migrants from those who are economically motivated (Kartiki, 2011;Lonergan, 1998). We, however, point to the clear link that as climate change increases, communities will experience shifts in their natural environments and major disruptions to traditional livelihoods leading to increases in migration. The latest IPCC report also confirms with high confidence that deteriorating economic conditions and livelihoods are main pathways toward climate-induced migration (IPCC, 2022).
Food security was the next factor to appear in four-length loops. For this loop length there were three loops that had the same loop score, each showing various sequences of factors in relation to migration. All three of these loops tell variations of the same theme that as livelihoods suffer, particularly those based in agriculture, a household or individual's food security decreases (Kartiki, 2011) and their likelihood of migration increases. For example, Milan and Ho (2014) show how changes in rainfall patterns effect livelihoods in Peru, leading to food insecure households to employ migration as a coping strategy. Similarly, health is introduced as an intermediary factor in Figure 4d. Food Insecurity is directly related to poor health outcomes (Meze-Hausken, 2000;Rakib et al., 2019), while illness decreases the ability to migrate (Morrissey, 2013;Parrish et al., 2020).
The next factor added in loops of lengths six, seven, and eight are political stability, environmental degradation, and resource security, respectively. There is abundant literature documenting how political instability and conflict lead to migration (Abah & Petja, 2016;Atapattu, 2020;Owain & Maslin, 2018;Smith, 2007). In the literature, cases show how increasing intensity and frequency of hazards, such as drought and desertification, lead to environmental degradation (Missirian & Schlenker, 2017). As the environment degrades, food insecurity rises and conflicts over resources such as water are likely (Levy & Sidel, 2014;Reuveny & Moore, 2009). The combination of these and other factors contribute to decisions to migrate. Likewise, as new populations enter destination communities there is a likelihood for ethnic and resource tensions as pressures increase on infrastructure (Kamta et al., 2021;Kartiki, 2011;Nunn & Campbell, 2020).
Out of these top loops, only two of the seven (Figure 4b,c) exhibit reinforcing behavior while the rest are balancing loops. This means that the feedback between migration, financial capital, livelihood, and food security will lead to perpetuating behavior in the system. Practically, this shows that communities whose livelihoods are negatively affected may have difficulties settling in a new location if they employ migration as an adaptation strategy. A case study in Bangladesh supports this with evidence of migrants with limited skill sets experiencing difficulty findings jobs at urban centers, as well as women losing access to their traditional livelihood options (Kartiki, 2011). This should be a point of focus for policy planners in aiding climate-resilient livelihoods to stem migration from the onset or supporting livelihood training programs for incoming migrants. The remaining balancing loops (Figure 4a,d-g) show a stabilization of the climate migration system. This means that the combination of factors within these loops combine to limit the "growth" of migration. Alternatively, if climaterelated migration continues to increase this could point to other mechanisms (factors) at play that were not included in this analysis. A second alternative would be that financial capital, livelihood, and food security are indeed the dominant factors influencing migration, while the others are more incidental.
Although this study successfully identifies factors and interactions influencing climate change from within the climate migration literature, more work is needed to understand these interactions in a fuller context of migration literature. Although some factor pairs may have had low-frequency counts in this study, we do not suggest that there are F I G U R E 4 Top feedback loops in the climate migration system (per feedback loop length) weak or nonexistent relationships, but rather that there was a lack of explicit mention of these relationships in the climate migration literature. In the future, we recommend researchers dive into these relationships with expanded scope to develop a clearer understanding of how factors interact. This study highlights gaps in climate migration literature where no linkages were coded such how "Social" and "Environmental" factors interact. The suitability of migration as a climate change adaptation strategy is highly context specific and will be unique to each individual and household. Although this study presents a general, global view of the climate migration system, geographically specific case studies are also needed to understand the nuance of particular societies and cultures. In addition, this research develops a map of relationships within climate migration systems that can be furthered with future research. In particular, the results of this research lend well to future system dynamics modeling, where the addition of unique demographic data can help to determine context specific consequences of factor build-up or reduction over time (Fussell et al., 2014).

| Limitations
We reiterate that the results of this research reflect the findings from the state of current literature on the topic of climate migration. Although these results are useful for gaining an idea of the interconnection and complexities within the climate migration system, they are also subject to the inherent biases found within scientific research at large. In our work, we acknowledge the English language bias of the articles selected for review and the disproportionate authorship from high-income, western countries (see Piguet et al. 2018, for a discussion on the uneven geographies of environmental migration research). As such, we lack important and relevant insights from experts originating from the lower income countries which are the most vulnerable to impacts of climate change.

| CONCLUSION
In this study, we set out to identify the factors that influence climate migration in literature (RQ1) and how these factors interact to lead to migration or non-migration (RQ2). To date, there has yet to be a thorough, comprehensive review of climate migration literature to specifically identify key factors and interactions that lead to climate migration decision-making. Through this study, we provide not only an updated view of the current state of literature in the field but also a novel systems-based approach for the synthesis of interactions and feedback loops.
In our review of literature, we presented a breakdown of publications by year and country setting and identified 21 factors that contribute to migration decision-making under climate change. Our work extends previous metaanalyses and syntheses which have also sought to identify factors affecting climate and environmental migration (Hoffmann et al., 2020;Parrish et al., 2020;Sedova & Kalkuhl, 2020), emphasizing a new and more dynamic view of migration systems. For instance, in their meta-analysis, Hoffmann et al. (2020, p. 910) concluded by stating, "environmental change can influence migration through several other channels" and identified urbanization, labor market conditions, conflicts, and health, among the factors playing a role in triggering migration. Although this study agrees with the intermediary relationships between environmental change and migration, the conceptualization of "channels" of influence still denote a linear causal thinking. We, meanwhile, expand upon previous studies such as these to emphasize feedback loops within systems, rather than chains of causality. Using PTA and CLDs as a means to map of the pairwise interactions in literature allowed us to visually explore the interconnection of the identified factors. A major output of this research is the development of a climate migration CLD which can be used to visualize the multitude of interactions with system ( Figure 3).
The use of systems thinking, and causal loop diagramming, is a new approach to analyzing and displaying the known interconnection and complexities in the field of climate migration. Notably, in this study we not only include direct causal relationships between specific factors and migration but also relationships between factors. This is an important aspect of systems thinking that allows us to better understand how tweaks and perturbations to any part of the system causes repercussions directly and indirectly to migration decision-making. Using this visualization aid, cascading effects can be more easily understood and predicted by policy makers and planners, thereby minimizing future unintended consequences. We specifically recommend policy makers to reference these causal connections when determining development strategies in climate-affected areas so that down-stream effects on migration and other key areas can be better understood. In this way, we compliment meta-syntheses, such as Nayna Schwerdtle et al. (2020), in helping to better understand ensuing effects and possibly prioritize policy recommendations for aiding health in human mobility. In our analysis of the top feedback loops within the climate migration system CLD, we highlight that particular emphasis for policy and practice needs to be made on "Economic" factors, such as financial capital and livelihood to assist climate-affected communities. Likewise, the relationship of how livelihood struggles lead to food insecurity is another point where planners should direct their attention.
In addition to its practical use, the results of this study are also useful to further understandings within the literature of climate migration. Obokata et al. (2014) call researchers to include greater context in key areas (economic, demographic, social, political, and environmental) to increase our understanding of the interplay between these factors and migration. We likewise recommend the continued and increased collection of empirical data to contextualize these identified relationships to specific geographic or hazard settings. In this way, future studies may address the limitations associated with the global nature of this study.
In the future, the results of this study may also be useful in the development of more empirically or quantitatively derived systems models. From here, we recommend the exploration of temporal dynamics between these factors to determine how factor importance and relationship shift over time. Hoffman et al. (2020) likewise points to the need to consider time by noting that certain factors may only affect migration after specific threshold points have been reached or passed.
As the impacts of climate change are expected to increase in the coming decades, so too can we expect the mechanisms of the climate migration system to be amplified. In the future, studies may be needed to validate and revise these relationships at a localized level. While the results are generalized across geographic locations and hazard types, we still offer valuable insight into factor interactions that are useful for policy and aid planners.