Detecting population transitions after Cyclone Mahasen in Bangladesh (Lu et al., 2016)

Background

Climate resilience is a serious concern for countries like Bangladesh where cyclone vulnerability is increasing with sea-level rise. It is occurring faster than global averages and exposes millions of people to inundation risks. Climate change is one of drivers of migration from environmentally stressed areas. However, quantifying short and long-term movements across large areas is challenging due to difficulties in the collection of highly spatially and temporally resolved human mobility data.

This use case highlights how mobile network data can be used to generate statistics on population transitions based on changes in number of active SIMs during and after weather events across a wide range of temporal scales.

Data

This project uses the following two datasets:

  • First dataset (D1) covers 1 April–30 June 2013, the period before and after Cyclone Mahasen, which struck Bangladesh on 16 May 2013. The data includes, for each call, the position of the mobile phone tower closest to the caller for all 5.1 million GP phones in Barisal division and Chittagong district, the primary impact zones of Cyclone Mahasen.
  • The second dataset (D2) covers a simple random sample of 1 million mobile phones drawn from the entire national set of mobile phones in the GP network. This dataset spans almost two years (1 January 2012–30 November 2013) and includes, for each calendar month, the location of each mobile phone's most frequently used tower that month.

Insights on this approach

  • Small movements across vast areas are extremely difficult or impossible to measure using traditional survey-based approaches, highlighting the benefit of temporally resolved individual-level data that can be collected on a national level.

Key steps taken for producing statistics

  • Filtering: Subscribers who were not active in the study area before the cyclone and those who were not active in the last ten days of the data collection period (20–30 June, 2013) were filtered away. This filtering excludes phones which were destroyed due to the cyclone or which belonged to incoming relief workers. This process made the final dataset for D1 include 2.95 million users and for D2 include 64% of users.
  • Correspondence with known population statistics: The spatial distribution of users in D2 was compared to the spatial distribution of the population from the Bangladesh 2011 census, resulting in a correlation of r = 0.948 (p < 0.001).
  • Examining changes in the number of SIM cards: Absolute changes in SIM cards are computed for examining the weekly number of unique subscribers in Chittagong City. There is a clear increase in the number of unique subscribers (SIM cards) in Chittagong City after the cyclone (16th May). The increase was generated by adding a large number of small, highly distributed mobility streams from across the whole of Bangladesh together

Areas for improvement and challenges

The key contribution of mobile data could come from combining the vast spatial, temporal and population coverage of mobile network data with targeted phone-based and household-based panel surveys. This is crucial to characterise how especially vulnerable groups such as women, children and the poorest are represented in the mobile phone data. With further methodological development and continued increases in mobile penetration rates, stratified large samples based on country specific mobile usage patterns will likely provide the most accurate results.


  • No labels