Armed conflicts are often defined at too broad a geographic scale. This study introduces a machine learning approach to map conflict zones more precisely based on event data, creating a new dataset of geospatial conflict zones.

Why Current Methods Fall Short

Existing approaches rely heavily on coarse administrative units or arbitrarily label entire provinces as "conflict areas," leading to inaccurate representations and loss of nuance. This creates challenges for both political analysis and ecological impact studies.

Methodology: Geospatial Learning from Events

We developed a machine learning technique that directly models conflict intensity across geographic space, minimizing the need for predefined administrative boundaries or arbitrary cutoffs.

New Dataset & Key Findings

Applying this method to event data produced a refined map of armed conflicts. This allowed us to replicate Daskin and Pringle's (2018) analysis with greater precision: the ecological impact of civil war on mammal populations was significantly lower than previously estimated when zones were properly defined by machine learning.

Implications for Political Science Research

This approach offers a more nuanced understanding of armed conflict geography. It helps political scientists better isolate causal effects in research and improves our ability to target interventions effectively.