Conflict Risk Analysis in Pakistan Based on Multi-source Geographic Data

Objectives: As the Belt and Road Initiative (BRI) continues to advance, China has made significant progress in cooperation with countries along the BRI route. However, frequent conflicts in these countries pose risks and challenges for Chinese enterprises investing in BRI projects. We focus on Pakistan, one of the initiative's earliest partners, which experienced an increase in conflict events in 2023, accompanied by heightened military involvement, as reported by the ACLED dataset. Methods: First, a binary conflict occurrence variable was constructed from the ACLED dataset, serving as the explained variable. Explanatory variables were derived from a variety of multi-source datasets, including nighttime light (NTL) imagery, population data, and other geographic data, to investigate the underlying factors behind conflict events. Second, the spatial and temporal characteristics of conflicts in Pakistan in 2023 were analyzed using hotspot method. Then, the XGBoost model was applied to explore the relationship between conflict occurrences and various multidimensional factors, such as nighttime light intensity, population density, and transportation infrastructure. The model was further interpreted using the SHAP method from game theory, providing granular insights into the contribution of each variable to conflict occurrences. Results: The results demonstrated that conflicts in Pakistan during 2023 were concentrated in seven distinct hotspot regions, mainly located in Khyber Pakhtunkhwa Province, Sindh Province, and the capital Islamabad. In the temporal dimension, there were more conflict hotspot regions in July, September, and November compared to other months in 2023. Among all variables, social development factors had a greater impact on conflict occurrence than natural geographic factors. The variable "Population" emerged as the most significant contributor to conflict occurrence (15.9%), followed by "Transport" (15.0%), "Traffic" (14.0%), "Gini" (11.3%), "Road" (11.1%), and "NTL" (9.8%). Furthermore, there was a positive relationship between "Population" and conflict occurrence, which means regions with dense populations were more prone to conflicts. Similarly, economic inequality, represented by the Gini coefficient, showed a positive correlation with conflict occurrence, suggesting that areas with uneven economic development were more susceptible to conflicts. In contrast, transportation-related variables (Transport, Traffic, and Road) exhibited a negative relationship with conflict occurrence, meaning that regions well-developed transportation networks, especially transportation hubs, were more prone to conflicts. Conclusions: By incorporating nighttime light imagery and other geographic data, we construct a conflict model for Pakistan at a geospatial grid scale, providing a more nuanced understanding of the key factors influencing conflict occurrence and expanding the existing research on conflict risk analysis in countries along the BRI route. All findings aim to support insurance companies in designing more precise investment insurance policies and to assist Chinese enterprises in successfully investing in and implementing BRI projects.