Resumen
Large-scale humanitarian disasters often disproportionately damage poor communities. This effect is compounded when communities are remote with limited connectivity and response is slow. While humanitarian response organizations are increasingly using a wide range of satellites to detect damaged areas, these images can be delayed days or weeks and may not tell the story of how many or where people are affected. In order to address the need of identifying severely damaged communities due to humanitarian disasters, we present an algorithmic approach to leverage pseudonymization locational data collected from personal cell phones to detect the depopulation of localities severely affected by the 2017 Puebla earthquake in Mexico. This algorithm capitalizes on building a pattern of life for these localities, first establishing which pseudonymous IDs are a resident of the locality and then establishing what percent of those residents leave those localities after the earthquake. Using a study of 15 localities severely damaged and 15 control localities unaffected by the earthquake, this approach successfully identified 73% of severely damaged localities. This individual-focused system provides a promising approach for organizations to understand the size and severity of a humanitarian disaster, detect which localities are most severely damaged, and aid them in prioritizing response and reconstruction efforts.