Climate-related disasters and temperature extremes are not associated with conflict risk in Pacific Island countries

Abstract

Concerns about climate change and conflict are ubiquitous. Systematic research on the climate-conflict nexus in the Pacific is lacking, despite the region’s high vulnerability to climate change as well as several past episodes of political instability. Using a new dataset, we analyze the impact of climate-related disasters and temperature extremes on social conflict events in Fiji, Solomon Islands, and Vanuatu. According to theoretical expectations, climate change should engender more social conflicts, for instance due to intensified grievances, resource competition (especially about land), and rural-to-urban migration. The results indicate that in the three countries studied, disaster shocks and temperature extremes do not noticeably affect conflict risks, even after controlling political, economic, and environmental context factors. Major cities, commonly considered most prone to climate-related conflicts in the region, do also not experience higher conflict risks following disasters. These findings indicate that Pacific Island countries may be more resilient to climate-related conflict risks than commonly assumed.

Keywords

climate change Pacific resilience security war violence

1. Introduction

Climate change is increasingly considered a security concern by politicians, militaries, peacebuilders, and development practitioners (McDonald, 2021; Vogler, 2023). There is an emerging scholarly consensus that climate change increases the risk of violent conflict within states if certain scope conditions like ethnic exclusion or poor adaptive capacity are present (Buhaug et al., 2023; Ide et al., 2020; Mach et al., 2019; Sharifi et al., 2021).

So far, research has mostly focused on Africa, Asia, and the Middle East. A systematic literature review did not find a single empirical study on the climate-conflict nexus in the Pacific Island region (Adams et al., 2018). This lack of research impedes our ability to analyze whether theories of the climate-conflict nexus are valid across different geographical settings and world regions. Furthermore, portraying regions like the Pacific as being at high risk of climate-related conflict without empirical evidence risks a stigmatization of these regions. Considering the Pacific Island countries as unable to deal with environmental stress and destined to experience violence in the future could, for instance, precludes much-needed investments in adaptation and livelihood diversity (Barnett, 2017; Telford, 2023).

The lack of studies on climate change and conflict in Pacific Island countries is especially problematic because the region is widely considered to have low levels of human development and to be highly vulnerable to climate change¹ (ND-GAIN, 2025; Weir & Virani, 2011). The region’s geopolitical relevance has also been growing with renewed competition between China and the USA (and their respective allies; Tarte, 2022). Regional experts argue that livelihood insecurity, migration, and land conflicts resulting from climate change can increase conflict risks in the Pacific (Boege, 2022; Campbell, 2022). The Boe Declaration signed by the leaders of the Pacific Island states also argues “that climate change remains the single greatest threat to the livelihoods, security and wellbeing of the peoples of the Pacific” (Pacific Islands Forum, 2018).

Yet comprehensive cross-case evidence on a climate-conflict nexus in the region is still missing. We address this knowledge gap by analyzing the relationship between climate impacts and conflict risks in three Pacific Island countries: Fiji, Solomon Islands, and Vanuatu. We selected those three because they are among the largest four Pacific Island countries in terms of population and land area, making them crucial cases to study. They also display significant variation, for instance in terms of human development and political system (Boege, 2022), which increases the external validity of our results (we nevertheless encourage an extension of our study to the other Pacific Island countries). Yet, by focusing on countries with a history of political instability and that are highly prone to be adversely affected by climate change, our study follows a most-likely case design for the climate-conflict nexus (Gerring & Seawright, 2007).

A key reason for why climate-conflict research has so far largely ignored the Pacific Island countries is the lack of quantitative data for the region. To address this gap, we create the first geocoded dataset on social conflict events in Fiji, Solomon Islands, and Vanuatu for the period 2012 to 2020 (see Section 3).

In the next sections, we provide a theoretical background (Section 2) and explain our data and methods (Section 3). Afterwards, we present (Section 4) and discuss (Section 5) our results. We find that neither temperature extremes nor climate-related disasters have a statistically significant effect on increasing the risks of social conflicts in the Pacific Island countries under study, even when accounting for socioeconomic and political context factors. This indicates that the Pacific could be rather resilient to the climate-conflict nexus, hence casting doubt on the emerging consensus about climate change indirectly increasing conflict risks.

2. Theoretical background

Our study builds on previous work on the climate-conflict nexus. We focus on three (overlapping and interacting) pathways, namely climate change (a) contributing to disaster-induced grievances against state institutions (e.g., due to insufficient preparation or relief) and (b) intensifying resource competition between social groups (e.g., over increasingly scarce land and water resources), both of which might be accelerated by (c) climate-related migration. For instance, if extreme temperatures and disasters adversely impact rural livelihoods, people affected may migrate to urban areas (permanently or temporarily). These migrants can hold grievances against the government or compete about resources with existing residents. For those residents, immigration could induce grievances for political, economic, or cultural reasons (Brzoska & Fröhlich, 2016; Koubi, 2019; Petrova, 2021; Vesco et al., 2021).

Existing qualitative research on the Pacific Island countries shows that the current and projected impacts of climate change may already contribute to disaster-related grievances. In recent years, popular discontent has emerged over state-induced (planned) and unplanned resettlements to increase disaster resilience, lack of proper state support after disasters, and economic turbulences induced by extreme weather events (Boege, 2016; Weir & Virani, 2011). Such discontent is most likely to transform into manifest conflict events like protests in urban areas, where high population density facilitates mobilization and collective action (Staniland, 2010; Toft, 2002).

Furthermore, several recent qualitative studies indicate that climate change affects land and water availability in the Pacific. This can, in turn, contribute to competition over these resources, and particularly regarding land. Livelihoods in the region are often based on agriculture, and land has a high spiritual value to many Pacific Islanders. Again, such conflicts may be more likely in urban areas of the Pacific Island countries (even though they will also occur in rural regions), not just because of higher mobilization potential, but also because land is more scarce and valuable there (Shibata, 2022; Weir & Virani, 2011). For instance, a descriptive analysis of our new dataset reveals that 18.6% of all social conflict events are directly related to three natural resources: forests (4.1%), water (0.3%), and especially land (14.1%).²

Finally, qualitative evidence suggests that climate-induced environmental stress (and particularly disasters like floods and storms) may induce migration from rural to urban areas in Pacific Island countries. Capital cities are the most likely destination due to more social services and jobs being available there. Such migration is expected to happen in the first months after a disaster when resources for in situ adaptation and recovery are most limited. In turn, migration increases competition over land, which is already scarce in urban areas, and can also contribute to general grievances about a lack of government services. This is particularly so because many displaced persons move into informal settlements (Boege, 2022; Campbell, 2022; Shibata et al., 2023).

Our new datasets contain several events explicitly in line with this logic. In April 2014, for instance, a series of intense floods struck Solomon Islands. On May 16 of the same year, a group of people displaced by the floods to an informal settlement in the capital city Honiara started a protest-turned-riot because they considered the government rehabilitation package insufficient to cover their disaster-induced losses (event ID 28).

As discussed above, testing these three pathways for Fiji, Solomon Islands, and Vanuatu is currently not possible for statistical analyses due to insufficient data availability. Future research can build on our data and results and test pathways more specifically, for example by connecting more fine-grained quantitative data (e.g., in the form of surveys; see Koubi et al., 2021) or in the form of additional qualitative case studies (see Shibata et al., 2023).

Additionally, our study focuses on social conflicts and does not consider linkages between climate change and conflicts involving organized armed groups like rebels, militias, or warlords, which are not present in the countries we analyze over out period of interest. Such groups might require attention to other pathways, such as the effects of climate change on weakening the state or enhancing the recruitment opportunities for armed groups (Buhaug et al., 2023; Ide, 2023; Koubi, 2019).

3. Methods

The unit of analysis is the 0.5-decimal degree (∼55 × 55 km cells at the equator) year. Due to data availability, we focus on the period 2012 to 2020. Summary statistics for all variables are provided by Supplemental Table S1.

To operationalize the dependent variable (social conflict), we built a unique geo-referenced, time-coded dataset on social conflict events in Fiji, Solomon Islands, and Vanuatu for the period 2012 to 2020. In line with the literature on contentious politics (Tilly & Tarrow, 2015), we define social conflict events as situations where at least two social groups perceive their interests to be incompatible with each other and take contentious actions to enforce these interests, resulting in events like protests, riots, assaults, and violence against civilians. To identify social groups, we rely on claims made by the event participants (describing themselves as “local residents,” etc., while also ascribing a collective identity to their opponents) or on attributions made by local media (referring to the conflict parties as “police,”“disaster victims,” etc.). Each action carried out by a designated actor (or group of actors) at a specific place during a limited period was considered a conflict event. Data was collected from local media and news sources, which increases the reliability of the data (see Supplemental Material for a list of sources and a codebook).

In constructing this dataset, we are motivated by the fact that standard conflict datasets are generally unsuitable to analyze conflict risks in the Pacific. Existing data either only include high-intensity violent events (which are rare in the region) in selected countries (UCDP), do not cover the Pacific Island countries at all (SCAD; see Hendrix & Salehyan, 2012), or only cover the region since 2022 and record few events due to their strong reliance on international news media (ACLED; Öberg & Yilmaz, 2025).

Our two main independent variables are climate-related disasters and temperature extremes. To identify storms, floods, and droughts in the three Pacific Island states, we used the Global Dataset on Geocoded Disaster Locations (GDIS) to identify climate-related disasters in the three countries under study (Rosvold & Buhaug, 2021). As the dataset only covers the years up to 2018, we coded the remaining years following the codebook compiled by the dataset’s creators. For descriptive statistics, Supplemental Figure S1 visualizes the geographical distribution of climate-related disasters and social conflict events in Fiji (panel a), Solomon Islands (panel b), and Vanuatu (panel c). Supplemental Figure S2 plots annual correlations and Supplemental Figure S3 plots a breakdown of social conflict rates by type across the three countries. Furthermore, in line with previous work (Ge et al., 2022; Helman & Zaitchik, 2020), we also test the effects of temperature. To this end, we relied on average annual country-level temperature data from the World Bank’s Climate Change Knowledge Portal (World Bank, 2024a).

To account for alternative confounders, our analyses also employ a set of control variables and potentially relevant context factors from both the literature on violent conflict and climate security (Ge et al., 2022; Koren & Schon, 2023). These controls include first variables to account for the effect of development and state capacity on social conflict, for instance, due to greater ability of the state to prevent conflict and lower opportunity for contention. This variable is operationalized as average levels of pixels with illumination based on the defense meteorological satellite program (DMSP) operational linescan system (OLS) system within each 0.5-degree cell, calibrated for time series analysis (2012 estimates). Another possibility is that cropland and local agricultural production pressures might shape social conflict (Koren & Schon, 2023). To account for this, we include an indicator for percentage of the 0.5-degree grid cell that was farmland (based on year 2000 estimates).

Next, we account for proximity to the equator (which can affect the frequency and intensity of storms) by accounting for 0.5-degree cell area, seeing that due to the Mercator projection, cells located closer to the equator are slightly smaller. Another possibility is that both social conflict events and storm damage might be more likely in rural areas. To account for this risk, we also included a control estimating the travel time from the center of the 0.5-degree grid cell to the nearest major city (50,000 or more residents). These controls all came from the PRIO-GRID dataset (Tollefsen et al., 2012). Additionally, in some of our robustness models, we use capital-city locations. This variable was coded manually based on each city’s coordinates.

Although nighttime lights and travel distance may overlap, we decided to still include them in some of the models since they also capture distinct dimensions of development, accessibility, and political centrality. To illustrate our results are not driven by this decision, our baseline model is estimated without controls, and subsequent specifications add them sequentially; results remain stable across models. Moreover, our variance inflation factor (VIF) tests and ridge regression models (described below) address the risk that any findings (or lack thereof) are driven by multicollinearity concerns. Finally, capital-city location may also overlap with these indicators, but since this variable is only used in one robustness check, not as a main control, this is less of a concern.

In addition to 0.5-degree cell-level indicators, our models also include country-level variables to account for the potential political and national confounders. Here, we first account for (log) country population size annually, considering bigger countries might be at a higher risk, for example, due to having more social groups that perceive their interests as incompatible. We also account for the size of the economy and its potential impact on both conflict and disaster preparedness by including the (log) gross domestic product per capita (GDP, in 2015 U.S.$). These two indicators were obtained from the World Bank (2024b). Finally, conflict frequency might be shaped by the existence of democratic institutions to mitigate tensions. To this end, we include a control for democracy operationalized base on the democracy score from the updated Democracy-Dictatorship data (Bjørnskov & Rode, 2020).

Our modeling strategy (designed to identify the relationship between climate-related disasters and temperature on conflict) proceeds in four steps, adjusting the sets of controls to account for specific concerns. The baseline specification (Equation 1) estimates the relationship between disasters and social conflict while adjusting only for grid-cell fixed effects, lagged conflict, and a linear time trend, thus ensuring that no potential multicollinearity across the independent variables is driving the results. In the second step (Equation 2), we add country-level controls for temperature, population size, GDP per capita, and life expectancy to capture national context while accounting for all constant (non-time-varying) cell-level confounders. The third specification (Equation 3) instead includes country fixed effects to account for within-country heterogeneity, while adding both time-varying and constant 0.5-degree cell-level covariates for nighttime lights, farmland, and travel distance to the nearest city, and omitting country-level controls to ensure these are not impacting the observed relationships. The final specification (Equation 4) then combines both sets of controls in addition to using country fixed effects.

This sequential approach ensures that any relationship between climate disasters, temperature, and conflict is not driven by the specific concerns each model seeks to address. To additionally account for temporal and spatial dependence, all models include a lagged dependent variable and a linear time trend. Standard errors are clustered by grid cell to adjust for local autocorrelation and spatial clustering. This modeling strategy balances parsimony with rigor and helps ensure that null findings are not artifacts of model misspecification. Moreover, it helps to effectively minimize concerns about serial autocorrelation and spatial clustering. The inclusion of a lagged dependent variable and a linear time trend captures persistence in conflict dynamics over time, while clustering standard errors by grid cell adjusts for within-cell correlation across years and local spatial dependence. This approach provides a conservative test of null effects, as failure to account for such dependence would typically inflate significance rather than produce null results. Accordingly, the stability of our findings across all specifications reinforces that they are not artifacts of temporal or spatial autocorrelation.

Formally, the models are identified using the following equations:

y_{i t} = β_{1} s_{i t} + β_{2} d_{i t} + β_{3} y_{i t - 1} + β_{4} t_{t} + ω_{i} + ϵ_{i}

(1)

y_{it} = β_{1} s_{it} + β_{2} d_{it} + β_{3} y_{it - 1} + β_{4} t_{t} + β_{5 - 8} X_{jt} + ω_{i} + ϵ_{i}

(2)

y_{it} = β_{1} s_{it} + β_{2} d_{it} + β_{3} y_{it - 1} + β_{4} t_{t} + β_{5 - 7} Z_{it} + ψ_{i} + ϵ_{i}

(3)

y_{i t} = β_{1} s_{i t} + β_{2} d_{i t} + β_{3} y_{i t - 1} + β_{4} t_{t} + β_{5 - 8} X_{j t} + β_{9 - 11} Z_{i t} + ψ_{i} + ϵ_{i}

(4)

Here, the subscripts i, j, and t, correspond to the 0.5-degree grid cell, country, and year, respectively, and β are each independent variable’s respective coefficient; y_it is a variable measuring social conflict events and y_it₋₁ its lag; s_it are the number of storms and d_it are the number of non-storm disasters in a given cell during a given year; τ_t is the time trend; X_jt are country-level controls for temperature, (log) life expectancy at birth, (log) population, and (log) GDP per capita; Z_it are 0.5-degree cell-level controls for nighttime light, presence of farmland, and travel time to the nearest city; ω_i and ψ_j are fixed effects by 0.5-grid cell and country, respectively; and ε_i are standard errors clustered by grid cell to adjust for local heterogeneities. As we discuss below, we conducted a wide set of robustness checks—including varying disaster measures, disaggregating conflict types, adding political controls, standardizing disaster indicators, estimating system generalized method of moments (GMM) models, and distinguishing the effects of disasters in the capital from other areas—to test the sensitivity of our OLS findings. In some of these models (reported in Supplemental Table S14), we adjusted Equations 3 and 4 as follows:

y_{i t} = β_{1} s d_{i t} + β_{2} u_{i t} + β_{3} s d_{i t} X u_{i t} + β_{4} y_{i t - 1} + β_{5} t_{t} + β_{6 - 8} Z_{i t} + ψ_{i} + ϵ_{i}

(5)

y_{it} = β_{1} s d_{it} + β_{2} u_{it} + β_{3} s d_{it} + β_{4} y_{it - 1} + β_{5} t_{t} + β_{6 - 9} X_{jt} + β_{10 - 12} Z_{it} + ψ_{i} + ϵ_{i}

(6)

Here, sd_it is the aggregate indicator of all climate-induced disasters, u_it is an indicator of whether a city was a capital, and sd_it × u_it is their interaction. Otherwise, Equations 5 and 6 are identical to Equations 3 and 4, respectively. All resource-driven subsample models similarly use Equations 1 to 6.

Additionally, as discussed above, we are aware that using fixed effects at different geospatial levels, as well as the country-level variables and cell-level variables that measure similar features, can all introduce multicollinearity and serial autocorrelation in 0.5 grid-level analysis that may lead to falsely observing null results; we also conduct two sets of tests specifically designed to assess the impact of collinearity on our estimates. First, we report models where the 0.5-grid cell or country fixed effects (which induce multicollinearity) are removed (Supplemental Table S15). Second, we conduct VIF tests, finding that multicollinearity is not a concern in all 0.5-degree grid-level variables (VIF <5), including in our two disasters variables (VIF = ∼1; Supplemental Table S16). However, we also find that our country-level variables do exhibit, unsurprisingly, noticeable (VIF >5) multicollinearity.

Therefore, to systematically assess the impact of multicollinearity, we additionally estimate a set of ridge regressions corresponding to each equation. Ridge regressions seek to minimize the residual errors caused by multicollinearity by “penalizing” (i.e., reducing or fully omitting the effect of) specific independent variables (Hoerl & Kennard, 1970). To this end, it introduces a “shrinkage penalty” term, λ, into the residuals. When λ = 0, this penalty term has no effect and ridge regression produces the same coefficient estimates as the OLS models; as λ approaches infinity, the shrinkage penalty becomes more influential, and the ridge regression coefficient estimates approach zero.

The estimation process takes place in two steps. In the first step, we use k-fold cross-validation, where k = 10, to estimate the λ value that minimizes the test mean squared errors. For this purpose, we standardize all our predictors to ensure no single predictor variable is overly influential when performing ridge regression. In the second step, we take the best λ values, and then estimate the coefficients of each of our (standardized) predictors using this value, to identify each variable’s “multicollinearity penalized” coefficient, and compare if it statistically different from zero using standard errors from each corresponding OLS model, which would suggest multicollinearity is driving our key results.

4. Results

Figure 1 visualizes the OLS coefficients of climate-related disasters for a change in the expected number of social conflicts, after (1) adjusting for 0.5 grid fixed effects, conflict lags, and time trends (Equation 1); (2) then adding country controls for temperature, life expectancy, population, and GDP per capita (Equation 2); (3) adding country fixed effects with local-level controls for nighttime light, farmland, and distance from big cities (Equation 3); and then (4) adding the country controls from (2) (Equation 4; Supplemental Table S2 reports estimates for all models).

Figure 1.

The effect of climate-related disasters on social conflict in three Pacific Islands. Coefficient estimates for storm (left) and non-storm (right) disasters are reported with 95% confidence intervals for each respective model. Numbering corresponds with specifications discussed above.

Even after adjusting for socioeconomic, political, and other environmental confounders, the results demonstrate that climate-related disasters have no statistically significant impact on conflict risks in Fiji, Solomon Islands, and Vanuatu. Equality of coefficient tests suggest there is no statistical difference in effect size between storms and other types of disasters (with χ² test p-values ranging from .2 to .99). Adjusted R² values across the models range from .45 (country fixed effects, full specification) to .64 (grid cell fixed effects, full specification), suggesting the models provide overall good fit for these observational data. Moreover, while some studies highlight temperature as a key driver of conflict risks (Helman & Zaitchik, 2020), we find it has no noticeable effect on social conflict in our sample (Figure 2).

Figure 2.

T effect of temperature on social conflict in three Pacific Island countries. Coefficient estimates for temperature are reported with 95% confidence intervals for each respective model. Numbering corresponds to the specifications discussed above.

To test the sensitivity of our OLS results, we conduct a range of robustness checks (Supplemental Tables S3–S14). We begin by varying the disaster indicators. Because storms are the most frequent disasters in our sample (28 compared to 11 non-storm events), we re-estimate the models (1) using only storms, (2) using only non-storm disasters, and (3) using an aggregated disaster indicator. In addition, we estimate models with (4) country-by-year fixed effects, which absorb all confounders that vary at the country-year level. These checks confirm that our findings are not sensitive to how disasters are measured or to unobserved national shocks (Supplemental Tables S3–S6).

Next, we assess whether the null findings are driven by the aggregation of different types of social conflict. We re-estimate the models disaggregating the dependent variable into (5) protests, (6) riots, (7) assaults, and (8) violence against civilians. In all cases, the results remain substantively unchanged (Supplemental Tables S7–S10), indicating that our conclusions are not an artifact of outcome aggregation.

We also consider institutional and political context. Adding an indicator for democracy and removing fixed effects does not alter the results, suggesting they are not contingent on a particular modeling choice for regime type or on the inclusion of fixed effects (Supplemental Table S11). There is also a possibility that a better way to account for the impact of disaster is by standardizing them based on anomaly in disaster frequency. To this end, we estimate models where each disaster indicator is constructed by standardizing their effect time, subtracting the mean of storm and non-storm disaster for each 0.5-degree cell from the annual indicator, and then dividing it by the standard deviation (Supplemental Table S12).

A further concern is potential endogeneity between disaster onset and conflict, which could bias OLS estimates toward zero. To address this, we re-estimate Equations (1) and (2) using GMM, specifically system GMM models (Blundell & Bond, 1998). These models generate internal instruments that help “exogenize” the effect of climate-related disasters on conflict. They also incorporate two-way effects that adjust for both year and grid-cell heterogeneity, obviating the need for fixed effects or linear time trends, and employ two-step robust standard errors for additional reliability. The system GMM estimates (Supplemental Table S13) corroborate the OLS results, indicating that the null findings are not artifacts of simultaneity between disasters and conflict.

Finally, we explore whether disasters affect capital cities differently from other areas, given potential dynamics of migration, land pressures, mobilization potential, or institutional capacity in urban areas. We interact the disaster indicator with a capital-city dummy (Suva, Honiara, Port Vila). Results (Supplemental Table S14) show no significant interaction effect: disasters do not increase or decrease conflict risk in capitals relative to other areas. However, the constitutive capital term is positive and significant, indicating that capitals were generally more conflict-prone during disaster periods, regardless of disaster shocks themselves.

Turning to our collinearity effect tests, we first find that removing 0.5-degree grid (Equations 1 and 2) or country (Equations 3 and 4) fixed effects have no impact on our disaster coefficients’ size and significance (Supplemental Table S15). Moving on to our ridge regressions, designed to specifically account for the risk that some highly autocorrelated country-level variables affect the results (Supplemental Table S16), we find that in all four ride specifications, the most penalizing value of λ is practically = 0, meaning that the OLS estimates are unaffected by multicollinearity (Supplemental Figure S4).

Moreover, to test whether the coefficients of both our storm and non-disaster coefficient become statistically significant after adjusting for multicollinearity in these ridge regression models, we first estimate our four OLS models using the same standardized data used in the ridge regression to create a standardized version of the OLS coefficients. We then use the formula β_{ridge regression} − 1.97 × SE_{OLS regression} to test whether the penalized coefficients are now statistically significant. As long as this formula is <0, which happens in every case, the penalized/adjusted coefficient fails to reach statistical significance. This means that the (lack of) effect of each storm and non-storm disaster coefficient is practically the same across the ridge regression and OLS models, and hence multicollinearity is not inducing lack of significance in these models.

5. Discussion

Using a new database on localized social conflict events—and accounting for key confounders, endogeneity, and multicollinearity—our analysis finds no statistically significant effect of climate-related disasters and temperature extremes on social conflict risks in three Pacific Island countries: Fiji, Solomon Islands, and Vanuatu. If these results are illustrative of a broader trend, they have several relevant implications.

Many experts argue that climate change tends to increase conflict risks in settings characterized by high climate vulnerability and pre-existing tensions, with low-intensity (community) conflicts being particularly sensitive to climate-related disasters (e.g., Birch, 2025; Delina et al., 2024; Koubi, 2019; Mach et al., 2019). Our study illustrates these claims might be limited. Even when focusing on most-likely cases (Pacific Island countries), potential hotspot areas within such contexts (urban centers), and conflict types particularly sensitive to climate change (low-intensity conflicts), we do not detect a significant relationship between climate extremes and conflict risks.

Granted, while theory emphasizes mechanisms such as disaster-related grievances, land competition, and migration, our empirical leverage is necessarily limited to identifying aggregate associations rather than directly observing these causal pathways. We are also aware that our time series is relatively short and that the number of disasters is limited, which could raise concerns about statistical power explaining the null results. Nevertheless, the dataset spans 13 years across 72 distinct grid cells (n = 936), providing substantial variation over time and space that allows us to detect more than just very large effects. Therefore, despite these concerns, we believe our findings at least indicate that claims about a climate-conflict nexus in the Pacific should be moderated. This is in line with arguments made by Daoust and Selby (2023) for Western Africa, Koren and Schon (2023) for the Sahel, Dinc and Eklund (2023) for the Middle East, and Suza et al. (2024) for South Asia, among others.

Moreover, this does not mean that climate change will have no conflict implications for the Pacific in the future. It is important to stress that the lack of ability to reject a null effect should not be interpreted as conclusive evidence that no effects exist, only that statistically we were unable to establish such an impact. Precipitation extremes (which are not completely covered by our analysis due to data availability issues) might drive higher conflict risks. Likewise, climate-related inter-island migration and slow livelihood deterioration could increase conflict risks over large distances or long time periods, going beyond the spatial and temporal lags employed here. Future sea-level rise will further aggravate the situation. Finally, the climatic changes projected for the next decades will very likely be more severe than what the region experienced in recent years. We also did not study all Pacific Island countries.

Nevertheless, our results caution against calls that climate change significantly increases conflict risks in contexts of high climate vulnerability and pre-existing conflict risks: the Pacific Island countries may not be destined to face a protracted conflict crisis in a climate-changed world. This is in line with other evidence that communities in the Pacific region are remarkably resilient and can successfully adapt to climate change in situ, suggesting they may be more resilient than expected (Farbotko & Campbell, 2022; Keen et al., 2022).

In Vanuatu, for instance, Community Climate Change Committees were effective in preparing locals for cyclone Pam in 2015 and played a key role in delivering relief and recovery support. Such community cooperation often cut across local cleavages and therefore likely even lowered conflict risks (this is in line with other success stories of disaster diplomacy in the Pacific; see Ide et al., 2025). The government’s response to the cyclone was also swift and well-coordinated, hence boosting citizen’s support of state institutions (Keen et al., 2022). Understanding the drivers of such peaceful responses despite major climate impacts will be a key task for future research that could focus on, for instance, local social capital and the response strategies by governments and donors. This could also tie into the growing research field of environmental peacebuilding (Ide et al., 2021).

In these regards, future qualitative case studies or mixed-methods designs could complement our aggregate quantitative analysis by illuminating specific micro-level mechanisms—such as community cooperation, institutional trust, and donor engagement—that can explain when and where societal impacts of climate disasters are mitigated. These mechanisms currently remain hidden in large-N data. Another approach would be to improve large-N data coverage and resolution. Finer-grained measures—such as monthly climate anomalies (building on our robustness model in Supplemental Table S12 that uses annual local anomalies), cyclone track data, and displacement records—would sharpen estimates and allow for an empirical testing of specific pathways. Together, these approaches would provide a fuller picture of why disasters sometimes mitigate rather than exacerbate conflict risks in the Pacific Island countries.

If climate-change-related livelihood loss, social disintegration, and conflict in the Pacific are assumed to be inevitable, incentives to support local adaptation efforts and peacebuilding processes are likely to decline (Barnett, 2017). Our findings instead provide room for cautious optimism: If even highly vulnerable countries with a history of political instability do not yet experience climate-related conflicts, there is still room to build climate-resilient peace (Barnett, 2019).

Supplemental Material

sj-docx-1-eas-10.1177_27538796251408642 – Supplemental material for Climate-related disasters and temperature extremes are not associated with conflict risk in Pacific Island countries

Supplemental material, sj-docx-1-eas-10.1177_27538796251408642 for Climate-related disasters and temperature extremes are not associated with conflict risk in Pacific Island countries by Tobias Ide, Ore Koren and Luke Derrick in Environment and Security

Footnotes

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Tobias Ide’s research was funded by the Australian Research Council (ARC) under the DECRA scheme (DE190101268). Ore Koren’s work was supported by XCEPT Research Fund, the Harry Frank Guggenheim Foundation and NSF Grant No. SAP 2149053. Koren’s opinions do not reflect those of the Harry Frank Guggenheim Foundation or the NSF.

ORCID iDs

Tobias Ide

Ore Koren

Data availability statement

Replication data for this article are available in the online appendix.

Supplemental material

Supplemental material for this article is available online.

Notes

Author biographies

Tobias Ide is an Associate Professor of Politics and International Relations at Murdoch University in Perth, Australia. He has published widely on environment, peace and conflict, including in Journal of Peace Research, International Affairs, International Security, Nature Climate Change, and World Development.

Ore Koren (PhD 2018, University of Minnesota) is an Associate Professor of Political Science at Indiana University Bloomington. His research examines how environmental shocks, infectious disease, and development shape conflict and political violence trends. He has published in leading interdisciplinary and disciplinary journals.

Luke Derrick holds a Master in Development Studies from Murdoch University and worked there as a research assistant with a focus on social conflict events in the Pacific. He now is a Policy Officer at the Northern Territory Government.

References

Adams

Ide

Barnett

Detges

(2018). Sampling bias in climate-conflict research. Nature Climate Change, 8(3), 200–203.

Barnett

(2017). The dilemmas of normalising losses from climate change: Towards hope for Pacific atoll countries. Asia Pacific Viewpoint, 58(1), 3–13.

Barnett

(2019). Global environmental change I: Climate resilient peace? Progress in Human Geography, 43(5), 927–936.

Birch

(2025). Extreme weather and contentious elections. Journal of Peace Research, 62(4), 1144–1158.

Bjørnskov

Rode

(2020). Regime types and regime change: A new dataset on democracy, coups, and political institutions. The Review of International Organizations, 15(2), 531–551.

Blundell

Bond

(1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), 115–143.

Boege

(2016). Climate change and planned relocation in Oceania. Security and Peace, 34(1), 60–65.

Boege

(2022). Climate change, conflict and peace in the Pacific: Challenges and a Pacific way forward. Peace Review, 34(1), 1–10.

Brzoska

Fröhlich

(2016). Climate change, migration and violent conflict: Vulnerabilities, pathways and adaptation strategies. Migration and Development, 5(2), 190–210.

10.

Buhaug

Benjaminsen

T. A.

Gilmore

E. A.

Hendrix

C. S.

(2023). Climate-driven risks to peace over the 21st century. Climate Risk Management, 39(1), 100471.

11.

Campbell

J. R.

(2022). From the frying pan into the fire? Climate change, urbanization and (in)security in Pacific Island countries and territories. Peace Review, 34(1), 11–21.

12.

Daoust

Selby

(2023). Understanding the politics of climate security policy discourse: The case of the Lake Chad Basin. Geopolitics, 28(3), 1285–1322.

13.

Delina

L. L.

Gaviola

Pagkalinawan

(2024). Unveiling the climate–conflict nexus: Examining the influence of weather extremes on conflict emergence in the Philippine Bangsamoro. Environment and Security, 2(2), 235–258.

14.

Dinc

Eklund

(2023). Syrian farmers in the midst of drought and conflict: The causes, patterns, and aftermath of land abandonment and migration. Climate and Development, 16, 349–362.

15.

Farbotko

Campbell

(2022). Who defines atoll ‘uninhabitability’? Environmental Science & Policy, 138(1), 182–190.

16.

Hao

Ding

Jiang

Scheffran

Helman

Ide

(2022). Modelling armed conflict risk under climate change with machine learning and time-series data. Nature Communications, 13(1), 2839.

17.

Gerring

Seawright

(2007). Techniques for choosing cases. In: Gerring

(Ed.), Case study research: Principles and practices (pp. 86–150). Cambridge University Press.

18.

Helman

Zaitchik

B. F.

(2020). Temperature anomalies affect violent conflicts in African and Middle Eastern warm regions. Global Environmental Change, 63(1), 102118.

19.

Hendrix

C. S.

Salehyan

(2012). Climate change, rainfall, and social conflict in Africa. Journal of Peace Research, 49(1), 35–50.

20.

Hoerl

A. E.

Kennard

R. W.

(1970). Ridge regression: Biased estimation for nonorthogonal problems. Technometrics, 12(1), 55–67.

21.

Ide

(2023). Rise or recede? How climate disasters affect armed conflict intensity. International Security, 47(4), 50–78.

22.

Ide

Bruch

Carius

Conca

Dabelko

G. D.

Matthew

Weinthal

(2021). The past and future(s) of environmental peacebuilding. International Affairs, 97(1), 1–16.

23.

Ide

Brzoska

Donges

J. F.

Schleussner

C.-F.

(2020). Multi-method evidence for when and how climate-related disasters contribute to armed conflict risk. Global Environmental Change, 62(1), 102063.

24.

Ide

Hayes

Larasati

Canavan

(2025). De-escalation and diplomacy: Disasters as drivers of reduced conflict risks in the Indo-Pacific. Australian Journal of International Affairs, 79(2), 276–294.

25.

Keen

Sanderson

Osborne

Deo

Faith

Ride

(2022). Area-based approaches and urban recovery in the Pacific: Lessons from Fiji, Solomon Islands and Vanuatu. Environment and Urbanization, 34(1), 151–169.

26.

Koren

Schon

(2023). Climate change, cash crops, and violence against civilians in the Sahel. Regional Environmental Change, 23(3), 112.

27.

Koubi

(2019). Climate change and conflict. Annual Review of Political Science, 22(1), 18.11–18.18.

28.

Koubi

Nguyen

Spilker

Böhmelt

(2021). Environmental migrants and social-movement participation. Journal of Peace Research, 58(1), 18–32.

29.

Mach

K. J.

Kraan

C. M.

Adger

W. N.

Buhaug

Burke

Fearon

J. D.

Field

C. B.

Hendrix

C. S.

Maystadt

J.-F.

O’Loughlin

Roessler

Scheffran

Schultz

K. A.

von Uexkull

(2019). Climate as a risk factor for armed conflict. Nature, 571(7764), 193–197.

30.

McDonald

(2021). Ecological security: Climate change and the construction of security. Cambridge University Press.

31.

ND-GAIN. (2025). ND-GAIN country index. https://gain.nd.edu/our-work/country-index/.

32.

Öberg

Yilmaz

M. C.

(2025). Measurement issues in conflict event data: Addressing some misconceptions about what drives differences between human-coded event datasets. Research & Politics, 12(3), 20531680251362440.

33.

Pacific Islands Forum. (2018). Boe declaration on regional security. https://www.forumsec.org/2018/09/05/boe-declaration-on-regional-security/

34.

Petrova

(2021). Natural hazards, internal migration and protests in Bangladesh. Journal of Peace Research, 59(1), 33–49.

35.

Rosvold

E. L.

Buhaug

(2021). GDIS: A global dataset of geocoded disaster locations. Scientific Data, 8(1), 61.

36.

Sharifi

Simangan

Lee

Reyes

S. R.

Katramiz

Josol

J. C.

Dos Muchangos

Virji

Kaneko

Tandog

T. K.

Tandog

Islam

(2021). Climate-induced stressors to peace: a review of recent literature. Environmental Research Letters, 16(7), 073006.

37.

Shibata

(2022). Climate change in the Pacific: Land, identity, and security. Peace Review, 34(1), 22–30.

38.

Shibata

Seforosa

Boege

(2023). Climate change and conflict in the Pacific: Challenges and responses. Taylor and Francis.

39.

Staniland

(2010). Cities on fire: Social mobilization, state policy, and urban insurgency. Comparative Political Studies, 43(12), 1623–1649.

40.

Suza

Warner

van Dijk

(2024). Climate, food (in)security and conflict in the Bangladesh delta: Myths and puzzles. Environment and Security, 2(4), 525–547.

41.

Tarte

(2022). Reconciling regional security narratives in the Pacific. East Asia, 39(1), 29–43.

42.

Telford

(2023). Where to draw the line? Climate change-conflict-migration-terrorism causal relations and a contested politics of implication. Environmental Science & Policy, 141(1), 138–145.

43.

Tilly

Tarrow

(2015). Contentious politics. Oxford University Press.

44.

Toft

M. D.

(2002). Indivisible territory, geographic concentration, and ethnic war. Security Studies, 12(2), 82–119.

45.

Tollefsen

A. F.

Strand

Buhaug

(2012). PRIO-GRID: A unified spatial data structure. Journal of Peace Research, 49(2), 363–374.

46.

Vesco

Kovacic

Mistry

Croicu

(2021). Climate variability, crop and conflict: Exploring the impacts of spatial concentration in agricultural production. Journal of Peace Research, 58(1), 98–113.

47.

Vogler

(2023). Tracking climate securitization: Framings of climate security by civil and defense ministries. International Studies Review, 25(2), viad010.

48.

Weir

Virani

(2011). Three linked risks for development in the Pacific Islands: Climate change, disasters and conflict. Climate and Development, 3(3), 193–208.

49.

World Bank. (2024a). Climate change knowledge portal. https://climateknowledgeportal.worldbank.org/

50.

World Bank. (2024b). World Bank open data. https://data.worldbank.org/

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

3.39 MB