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

Armed conflict significantly hinders sustainable development, making its reduction a crucial goal of the 2030 Sustainable Development Goals (SDGs). This study focuses on identifying the drivers of armed conflict, particularly the role of climatic conditions, across Sub-Saharan Africa, the Middle East, and South Asia. These regions experience a high incidence of armed conflict, accounting for over 80% of global conflicts since World War II. While some studies link climate change to conflict, others find limited evidence, often due to variations in methodologies and data. This study aims to address this knowledge gap by using advanced machine learning techniques to analyze the relative impact of climatic factors on conflict risk, considering a broader range of social, economic, and geographical factors. The study hypothesizes that climatic conditions, while important, interact with social and geographical factors in complex ways to shape regional conflict patterns. The three regions were selected due to their high conflict rates and diverse geographical and climatic conditions, allowing for an investigation into regional variability in conflict drivers.

Existing research on the relationship between climate change and armed conflict presents conflicting results. While studies like Burke et al. (2009) and Hsiang et al. (2013) suggest a link between climate change and increased conflict risk, others, such as Raleigh and Urdal (2007) and Buhaug (2010), find limited evidence. This inconsistency stems from variations in methodologies, data selection, and the complex interplay of social, economic, and environmental factors. The consensus from expert elicitation (Mach et al., 2019) indicates that climate change will likely increase future conflict risk, but the magnitude of this impact remains uncertain. This study draws on previous research highlighting the importance of social vulnerability (Minhas and Radford, 2017; Rozmarin, 2017; Bell and Keys, 2018; Elbadawi and Hegre, 2008), resource scarcity (Koubi et al., 2014), and weak governance (Elbadawi and Hegre, 2008) as conflict drivers. This study seeks to build on this foundation by employing advanced machine learning to quantify the relative contributions of climatic and socio-geographic factors to conflict risk.

This study employs a machine learning approach using Boosted Regression Trees (BRT) to model the relationship between armed conflict risk and a range of climatic, social, and geographical factors. Data on armed conflict events are drawn from the UCDP Georeferenced Event Dataset (GED) for the period 1989-2018, aggregated at a 0.1° × 0.1° grid-year level. The dependent variable is a binary indicator of armed conflict incidence (1 for conflict, 0 otherwise). The independent variables, representing social vulnerability, include population density, urban accessibility, nighttime lights, and ethnic diversity. Geographical conditions are represented by land use, elevation, and natural disaster hotspots. Climatic conditions are divided into climate average conditions (average maximum temperature, minimum temperature, and precipitation from 1960-1988) and climate variation conditions (annual deviations from the average conditions from 1989-2018). The BRT model, implemented using the 'dismo' package in R, is trained on a balanced dataset of high-risk (conflict occurrence) and low-risk (no conflict) grid-years. To reduce randomness, 20 ensemble BRT models were constructed. The analysis is performed separately for three decades (1989-1998, 1999-2008, 2009-2018) to examine temporal changes in the influence of various factors on conflict risk. The relative contributions of climatic, social, and geographical conditions to conflict risk are assessed and visualized to identify regional C-S-G patterns.

The study finds a significant increase in the contribution of climatic conditions to armed conflict risk over the past 30 years across all three regions. The overall percentage increase in the contribution of climatic conditions to conflict risks over the last 30 years in Sub-Saharan Africa, the Middle East, and South Asia are 4.25, 4.76, and 10.65 percentage points, respectively. However, the relative importance of climatic conditions varies regionally. In South Asia, climatic conditions contributed 56.96–72.64% to the risk of armed conflict, significantly higher than in Sub-Saharan Africa (37.09–41.77%) and the Middle East (24.16–28.92%). Socio-economic factors remain the dominant drivers of conflict risk in the Middle East. The impact of climate variation conditions also varies: Sub-Saharan Africa shows the strongest relationship (8.80–11.27%), followed by South Asia (2.67–8.64%) and the Middle East (0.94–3.37%). Marginal effect plots illustrate complex relationships between specific climate anomalies (temperature and precipitation) and conflict risk, varying across regions. The analysis reveals distinct C-S-G patterns: social conditions dominate in the Middle East, climatic conditions are most important in South Asia, and all three conditions contribute significantly to conflict risk in Sub-Saharan Africa. These patterns remain relatively stable over time. The findings suggest that while climate change contributes to conflict risk, its impact is mediated by other social and geographic factors; the climate-conflict linkage is not deterministic.

The findings highlight the complex interplay of climatic, social, and geographic factors in shaping armed conflict risk. The study's high-resolution spatial analysis reveals regional variations in the relative importance of these factors, rejecting a simplistic, universal climate-conflict relationship. The distinct C-S-G patterns identified underscore the need for tailored conflict prevention strategies for each region. The results align with some existing research demonstrating climate's role in conflict, particularly in Sub-Saharan Africa, but also highlight the significance of social and economic conditions. The surprisingly high contribution of climate to conflict risk in South Asia warrants further investigation into region-specific vulnerabilities. The study's approach of considering multiple factors and using advanced modeling techniques offers a more nuanced understanding of the climate-conflict nexus compared to previous research that offered inconclusive results.

This study makes several important contributions to the climate-conflict literature. It uses high-resolution spatial data and a sophisticated machine learning model to identify regional variations in the drivers of armed conflict risk. The findings highlight the significant and growing contribution of climate change to conflict risk, but emphasize the crucial role of social and geographic context. Region-specific C-S-G patterns provide strong justification for regionally-tailored conflict prevention and mitigation strategies, reflecting the complex and multifaceted nature of conflict. Future research could focus on investigating the specific pathways through which climatic conditions influence conflict, incorporating additional social and economic variables, and delving into sub-regional variations in conflict dynamics.

The study's limitations primarily relate to data availability and model assumptions. The inclusion of certain variables may be constrained by the availability of high-resolution spatial data across all three regions. Some important factors, such as the historical legacy of colonialism, may not be fully captured due to data limitations. The BRT model, while powerful, relies on assumptions about the relationships between variables that may not perfectly reflect the real-world complexity of the phenomenon. Future research could address these limitations by exploring alternative modeling approaches and incorporating additional data sources.