Detecting village burnings with high-cadence smallsats: A case-study in the Rakhine State of Myanmar

Mass atrocities continue to occur in areas away from international observers and with poor information outflow, with the international community often learning days or weeks after the onset of violence. While organizations concerned with human rights are increasingly using remote sensing, high-resolution commercial satellites remain an expensive option for monitoring large areas at risk of human rights violations.In order to address the need for rapid alerting of possible human rights violations in remote areas, we present an algorithmic approach to leverage affordable, high-cadence, smallsat satellite imagery to detect the burning of villages within as little as two days of occurrence. This algorithm capitalizes on the constellation's systematic observations by detecting a potentially destroyed village if its near-infrared reflectance is less than 20% of its control pair for two sequential observations. The algorithm is based on Planet's Dove satellite's near-infrared band, which is a persistent indicator of the destruction of plant cell structure and is less affected by atmospheric scattering than a Dove satellite's visible bands.Comparison of this product with a database constructed from manual analysis of high-resolution satellite imagery shows high levels of accuracy among villages in the study area. We evaluated algorithm performance in the Republic of the Union of Myanmar, a sovereign state in Southeast Asia, from 1 August 2017 until 31 December 2017. In the study approximately 7000 images were downloaded and analyzed, producing algorithmic estimates that were within 82.5% of the ground-reference database, demonstrating a functioning system to alert human rights practitioners to a potential destruction of villages and to corroborate refugee eyewitness accounts of destruction. This information can also be used to validate refugee asylum claims or for the prosecution of the perpetrators in international courts.