What Drives Conflict in the Lake Chad Basin? - Assessing the Impact of Environmental and Socioeconomic Factors Using Earth Observation and Machine Learning

Abstract

Since the early 2000s, the Lake Chad Basin (LCB) has witnessed a rising number of violent attacks from insurgent groups, as well as confrontations among armed militias. Often, civilians and their means of subsistence are the primary targets. While various geographical factors are suspected to influence the timing and location of conflicts, there remains a lack of consensus on what predictors must be considered for conflict modeling efforts. This research explores the importance of socioeconomic and environmental predictors for conflict in the LCB. We present a quantitative assessment of how these variables inform a machine learning model aimed at predicting conflict events in the region. We utilize documented conflicts in the LCB, as recorded in the Armed Conflict Location & Event Data, for both training and testing the model. The model is based on Earth observation-derived environmental and socioeconomic features from time series data spanning the last two decades. We analyze means, anomalies, and trends for each month and across the entire time series of environmental factors, which include air temperature, precipitation, potential and total evapotranspiration, soil moisture, surface water extent, and gross primary productivity in both irrigated and unirrigated areas. Additionally, we incorporate means, anomalies, and trends of socioeconomic factors such as population density, the Subnational Human Development Index, and the number of ethnic claims in specific areas. We also consider the means, anomalies, and trends of prior conflicts as indicators of a region's general instability. All these parameters are used in a random forest regression model to forecast conflict occurrence. We identify which features are significant to the model for each experiment using Shapley Additive Explanations for individual features. Our results indicate that it is crucial to consider both socioeconomic and environmental variables when discussing potential future conflicts. The quantitative insights highlighting the relative importance of factors across various domains can serve as a foundation for developing integrated approaches in future conflict modeling research. Therefore, we believe this information is valuable for researchers and stakeholders in sustainable development.