A new study published in the journal Nature examines how combining different types of data can lead to a more comprehensive understanding of armed conflicts. Lead author Valerie Sticher uses the examples of Ukraine and Myanmar to highlight both the benefits and limitations of using automatically analyzed satellite data. The study suggests that the number of armed conflicts worldwide has reached its highest level since the end of the Cold War in 1989, and violence against civilians is on the rise. While casualty figures are often used as the main measure of conflict severity, the study argues that they are not sufficient, as conflicts also cause displacement, loss of livelihoods, and other long-term effects. Casualty data is typically gathered from media reports and other textual sources, which can miss important aspects of violence, especially those that don't result in immediate deaths. Satellite imagery, on the other hand, can reveal physical damage, such as burned-out homes or destroyed infrastructure. However, it cannot determine who caused the damage or whether people had already fled the area. In Myanmar, for example, text-based sources indicated that the worst violence against the Rohingya occurred in the first week of the conflict, but satellite data showed that destruction continued for months. Similarly, in Ukraine, damage from the war is more common during Russian territorial advances than during Ukrainian counterattacks. Combining satellite data with traditional text-based sources provides a more accurate picture of a conflict. In two Ukrainian cities that experienced similar levels of destruction, casualty figures varied significantly because many residents had already fled before the Russian offensive began. The effectiveness of automated satellite analysis depends heavily on the datasets used to train the models. If these datasets are too limited or focused on a few high-profile conflicts—like the one in Ukraine—the models may not perform as well in other regions with different types of infrastructure or conflict patterns. For example, some conflicts involve burning homes rather than heavy weapons, and more reference data from these situations is needed to improve analysis accuracy. The study also notes a significant gap in data on conflict-related sexual violence, which is underrepresented in textual sources and cannot be detected through satellite imagery. This means such violence is likely being underestimated in global conflict assessments. The research could help humanitarian organizations like the ICRC and the UN by offering a better understanding of conflict dynamics and more precise targeting of aid. The researchers hope to encourage greater collaboration between computer scientists and humanitarian workers to improve the protection of civilians in conflict zones.