Mobile Phone Data as Alternative Source of Information for Affected Disaster Areas Detection after a Severe Cyclone

Natural disasters cause devastating losses in human life, environmental assets and personal, and economic impacts in many countries every year. Tropical cyclones and derived floods are one of the most frequent natural disasters, mainly in those countries located in the coastal areas across the world. Rapid and effective response to these natural disasters depend on timely available information about affected disaster areas. Traditional methods of affected disaster areas extraction are based on the field visit, questionnaire survey, official documents such as available publications and newspapers information, which some of them are time-consuming and costly, mainly in developing countries with limited financial resources. In the last decades, satellite imagery have been used for this purpose. However, assessing the impacted areas usually, require high-to-moderate spatial resolutions satellite imagery which are sometime costly. Recently, social media posts and mobile phone Global Positioning System (GPS) traces have been used for this purpose. However, they fail to provide accurate results due to the limited sample of data that are usually available. To overcome such limitations, in this research, anonymized mobile Call Detail Records (CDRs) data are proposed as alternative source of information to infer affected disaster areas by analyzing mobile phone use patterns in each mobile phone cell tower before and after a disaster. If the mobile CDRs data are available right after a disaster, the proposed method can provide the results on affected disaster areas in a timely manner and hence support the disaster response activities, which among other measures, aims to provide for the basic humanitarian needs of affected people. The effectiveness of the proposed method is evaluated using damage assessment results from remote sensing data of a real world severe cyclone in central Mozambique, in March 2019. The results show an encouraging overall accuracy over 70%, which prove that CDRs have potential use as a complementary, if not an alternative, source of information for affected disaster areas detection after a cyclone.