Modelling armed conflict risk under climate change with machine learning and time-series data

Area of Science:

• Environmental Science
• Political Science
• Data Science

Background:

• Decades of research have explored climate variability and armed conflict, yet global causal links remain unclear.
• Existing studies use diverse methods across various scales, highlighting the need for a unified global perspective.
• Understanding climate-conflict dynamics is vital for effective peacebuilding and conflict prevention strategies.

Purpose of the Study:

• To quantitatively model and infer potential causal linkages between climate variability and armed conflict at a global scale.
• To simulate the worldwide risk of armed conflict from 2000-2015 using a machine learning framework.
• To identify key climatic and contextual factors influencing the risk of armed conflict.

Main Methods:

• Employed a quantitative modeling framework utilizing machine learning algorithms.
• Analyzed high-frequency time-series data to infer causal relationships.
• Simulated global armed conflict risk for the period 2000-2015.

Main Results:

• The risk of armed conflict is predominantly shaped by complex, stable background contexts.
• Climate deviations, specifically positive temperature anomalies and precipitation extremes, are significant covariates.
• Inferred patterns demonstrate a clear association between these climate factors and increased global armed conflict risk.

Conclusions:

• Stable background conditions and climate deviations are key drivers of armed conflict risk.
• Positive temperature deviations and extreme precipitation events correlate with heightened conflict risk worldwide.
• Enhanced understanding of climate-conflict linkages improves global spatiotemporal conflict risk modeling capabilities.