A Place to Go: Locating Damaged Regions After Natural Disasters Through Mobile Phone Data

Abstract

Large scale natural disasters involve budgetary problems for governments even when local and foreign humanitarian aid is available. Prioritizing investment requires near real time information about the impact of the hazard in different locations. However, such information is not available through sensors or other devices specially in developing countries that do not have such infrastructure. A rich source of information is the data resulting from mobile phones activity that citizens in affected areas start using as soon as it becomes available post-disaster. In this work, we exploit such source of information to conduct different analyses in order to infer the affected zones in the Ecuadorian province of Manabí, after the 2016 earthquake, with epicenter in the same province. We propose a series of features to characterize a geographic area, as granular as a canton, after a natural disaster and label its level of damage using mobile phone data. Our methods result in a classifier based on the K-Nearest Neighbors algorithm to detect affected zones with a 75% of accuracy. We compared our results with official data published two months after the disaster.