Estimating the resilience to natural disasters by using call detail records to analyse the mobility of internally-displaced persons

This paper presents work that was performed as part of the project ‘Contributing to a better understanding of human mobility in crisis and enhancing linkages with citizen-driven assistance’, a collaborative project between the International Organisation for Migration (IOM), the Internal Displacement Monitoring Centre (IDMC), the Humanitarian Data Exchange (HDX, part of the United Nations Office for the Coordination of Humanitarian Affairs (OCHA)), and Flowminder, funded by the European Civil Protection and Humanitarian Aid Operations (ECHO).

We use mobile phone call detail records to estimate the resettlement times of a subset of individuals that have been previously identified to be internally displaced persons (IDPs) following a sudden-onset disaster [1]. Four different mobility metrics - two versions of radius of gyration and two versions of entropy - are used to study the behaviour of populations during three disasters - the 2010 earthquake in Haiti, the 2015 Gorkha earthquake in Nepal, and Hurricane Matthew in Haiti in 2016. We characterise the rate at which a disrupted population resettles by the fraction of individuals who remain disrupted each week after the disaster. We find that this rate can be modelled very well as the sum of two exponential decays and observe that the resettling rate for all three disasters is similar, with half the original number of displaced persons having resettled within four to five weeks of the disaster. If the study of further disasters leads to the observation of similar exponential decay rates, then it would imply that the number of IDPs at any time can be inferred from an estimate of the initial number of IDPs immediately following the disaster. Alternatively, the method provides a way to monitor disaster resilience and compare recovery rates across disasters. The method has the advantage that no assumptions need to be made regarding the location or time of resettlement. We also find that metrics that are skewed by long distances, such as the radius of gyration, may not be ideal in this specific context because post-disaster disruption manifests in many individuals as a decrease in long-distance travel and an increase in short-distance travel. A metric that is skewed by long distances can therefore make it appear as though these individuals have recovered earlier than they really have, resulting in an underestimate of the number of people requiring support at a given time. Our results indicate that CDRs can significantly contribute to measuring and predicting displacement durations, distances, and locations of IDPs in post-disaster scenarios. We believe that information and estimates provided by specifically developed CDR analytics, coupled with field data collection and traditional survey methods, can assist the humanitarian response to natural disasters and the subsequent resettlement efforts.