Varying climatic-social-geographical patterns shape the conflict risk at regional and global scales

The study investigates how climatic, social, and geographical factors jointly shape the risk of intrastate armed conflict, focusing on Sub-Saharan Africa, the Middle East, and South Asia—regions with a high share of global conflicts since World War II. Motivated by Sustainable Development Goal 16 and ongoing debates about climate–conflict links, the authors aim to quantify the relative importance of climate conditions (both long-term averages and interannual anomalies) vis-à-vis social and geographic drivers, and to examine how these contributions vary across regions and over time (1989–2018). The central question is: to what extent do climatic conditions contribute to armed conflict risk relative to social and geographic conditions, and how do these contributions differ by region and decade?

Prior research emphasizes social vulnerability as a core driver of conflict, including resource scarcity, weak economies, governance deficits, and ethnic discrimination. The climate–conflict nexus remains contested: some studies find significant associations between climatic variability/change and conflict (e.g., Zhang et al., Burke et al., O’Loughlin et al., Hsiang et al., Ide et al.), while others find limited or no effects (e.g., Raleigh & Urdal; Buhaug; Owain & Maslin; De Juan & Hänze; van Weezel). Methodological choices in definitions, spatial/temporal coverage, and model specification can sway results. Expert elicitation suggests climate change is likely to increase future conflict risk, but the magnitude remains debated (Mach et al., 2019). The study builds on calls to assess the relative importance of climate as a driver using flexible data-science methods capable of capturing heterogeneous, conditional relationships.

Study regions: Sub-Saharan Africa, the Middle East, and South Asia. Period: 1989–2018, analyzed in three decades: 1989–1998, 1999–2008, and 2009–2018. Outcome: Armed conflict incidence at 0.1° × 0.1° grid-year resolution, derived from UCDP Georeferenced Event Dataset (GED) Global v20.1. A binary indicator equals 1 if any armed conflict event occurs in a grid-year, else 0. Sampling: For each year, high-risk samples (grids with conflict) are paired with an equal number of low-risk samples (no conflict). To reduce randomness, 20 ensemble Boosted Regression Trees (BRT) models are fitted per analysis. Covariates (all harmonized to 0.1° resolution):

• Social conditions: population density; urban accessibility; nighttime lights; ethnic diversity—capturing demographic distribution, transport access, economic activity, and ethnic differences.
• Geographical conditions: land use/land cover; elevation; natural disaster hotspots—capturing natural resources, topography, and disaster risk.
• Climatic conditions, split into:
  • Climate average conditions: long-term means (1960–1988) of maximum temperature, minimum temperature, and precipitation.
  • Climate variation conditions: annual anomalies (1989–2018) for each climate variable relative to its 1960–1988 mean. Modeling approach: Boosted Regression Trees (ensemble gradient boosting) implemented with the 'dismo' package in R 3.3.3. Models are trained per region and time slice to quantify the relative contribution (RC) of each covariate and aggregated into category-level RCs (climatic, social, geographical). Performance diagnostics are reported in supplementary materials (Table S4, Fig. S4). Response curves (marginal effects) from ensemble BRTs illustrate relationships between climate anomalies and conflict risk.

• Climatic conditions are significant drivers of armed conflict risk across all three regions, with regionally varying importance.
• Relative contribution (RC) of climatic conditions (total = climate average + climate variation):
  • Sub-Saharan Africa (SSA): 37.09% (1989–1998), 41.77% (1999–2008), 41.34% (2009–2018). Overall increase ≈ +4.25 percentage points over 30 years.
  • Middle East (ME): 24.16%, 24.53%, 28.92%. Overall increase ≈ +4.76 pp. Social conditions remain dominant in ME (63.28%, 67.82%, 52.29% across decades).
  • South Asia (SA): 61.99%, 56.96%, 72.64%. Overall increase ≈ +10.65 pp, indicating a strong and rising climatic influence.
• Climate variation (anomaly) contributions specifically:
  • SSA: 8.80–11.27% of RC.
  • SA: 2.67–8.64%.
  • ME: 0.94–3.37%.
• Marginal effects of climate anomalies:
  • SSA: Minimum temperature anomaly positively associated with conflict risk; relationships for maximum temperature anomaly and precipitation anomaly are complex/nonlinear.
  • ME: Minimum temperature anomaly negatively associated; precipitation anomaly positively associated; maximum temperature anomaly shows complex effects.
  • SA: Minimum temperature and precipitation anomalies tend to have positive associations; maximum temperature anomaly effects are complex.
• C-S-G patterns:
  • ME: Social conditions dominate conflict risk relative to climatic and geographical conditions.
  • SA: Climatic conditions dominate; social and geographical contributions are smaller.
  • SSA: More balanced contributions across climatic, social, and geographical conditions.
  • Patterns remain relatively stable over the three decades in each region.
• Implication: The rising and regionally heterogeneous climatic influence underscores the need for region-specific conflict risk mitigation strategies aligned with SDGs.

The findings address the central question by quantifying how much climatic conditions contribute to armed conflict risk relative to social and geographical factors and how these roles differ by region and time. Climatic conditions are important drivers in all regions, with the strongest and increasing influence in South Asia, substantial and growing influence in Sub-Saharan Africa, and a smaller but rising role in the Middle East where social drivers are paramount. The region-specific C-S-G patterns suggest climate–conflict linkages are contingent rather than deterministic and operate in interaction with socio-economic vulnerability, resource dependence (e.g., rainfed agriculture in SSA), and regional particularities (e.g., oil, water scarcity, identity politics in ME; high-altitude, hazard-prone zones and transboundary water issues in SA). The anomaly effects (e.g., minimum temperature and precipitation anomalies) further highlight non-linear, context-dependent relationships. These results align with parts of the literature that identify climatic impacts on conflict but reconcile prior disagreements by demonstrating substantial spatial heterogeneity and the necessity of comprehensive controls. Policy-wise, interventions should be tailored: strengthen socio-economic resilience and climate adaptation in SSA; address social-political drivers in ME while accounting for climate stress; and prioritize climate risk management (e.g., flood, water scarcity, migration pressures) alongside governance in SA.

This study contributes by (1) using high-resolution (0.1°) geospatial data to capture local heterogeneity in conflict risk; (2) integrating a broad set of social and geographical controls to isolate climatic effects; and (3) quantifying the relative contributions of climatic, social, and geographical conditions while identifying stable, region-specific C-S-G patterns. Climatic influence on conflict risk has increased over the past three decades, most notably in South Asia. The results support designing region-specific conflict prevention and resilience strategies that consider local climatic, social, and geographic contexts. Future research should incorporate coarser but relevant historical and institutional variables where possible, disentangle economic channels (prices, productivity) linking climate to conflict, and identify finer sub-regional patterns and differential vulnerability/resilience to better explain spatial heterogeneity and improve forecasting.

Key limitations include: (1) inability to incorporate important drivers available only at coarser spatial/temporal resolutions (e.g., legacies of slave trade, colonization, institutional capacity, interpersonal trust); (2) limited identification of mechanisms (e.g., via prices, productivity) through which climate affects conflict risk; (3) potential unobserved confounding despite extensive controls; (4) need for sub-regional analyses since patterns can vary substantially within regions and across neighboring countries; and (5) model dependence on available proxies (e.g., nighttime lights) that may imperfectly capture socio-economic conditions.