Understanding Population Movement After the 2018 Central Sulawesi Natural Disasters
Through a shared value partnership with mobile telecom Digicel, Pulse Lab Jakarta (PLJ) over the past few years has investigated how pseudonymised mobile network data from subscribers in the Pacific region can be used to support evidence-based decision making. From modelling population displacement to understanding changes in citizens behaviour after natural disasters, the research has yielded actionable insights for policymakers and a wealth of experiential knowledge. So much so that when a massive earthquake struck Central Sulawesi, Indonesia in September 2018, our team was well placed to apply our learnings from the research to better understand the impact on local communities.
How much do we know about the scale of population displacement? Where have residents travelled to? What about the people who have remained in the impacted areas? How might we harness mobile network data to help governments and humanitarian agencies mobilise aid and resources to those most in need? In collaboration with the International Organisation for Migration (IOM), we set out to explore whether we could answer these questions by analysing mobile positioning data from one of Indonesia’s mobile network providers.
The Existing Need
Natural disasters tend to disrupt everyday life and people are sometimes forced to leave their local communities. Within the United Nations system, IOM has been leading efforts to assist various displaced cohorts, including in scenarios where displacement has been induced by natural disasters. The Displacement Tracking Matrix (DTM) has functioned as the agency’s primary tool for monitoring population displacement and mobility since 2004, and it has been used to capture, process and disseminate relevant information pertaining to affected individuals. While the DTM methodology has been refined over the years to meet emerging challenges, time remains a crucial factor when it comes to the collection, processing and analysis of the data. We thus expanded our foray into exploring mobile positioning data as an alternative data source from which faster and more detailed information on human mobility during times of disaster can be gleaned.
Following the earthquake, tsunami and subsequent liquefaction in Central Sulawesi, IOM sought to provide evidence to the Indonesian Government on how Call Detail Records (CDRs) might be utilised to better understand human mobility during and after a natural disaster. Leveraging Pulse Lab Jakarta’s advanced data analytics capacity and experience in developing disaster risk reduction data analytics tools gained from prior modelling with Digicel data, IOM joined forces with the Lab to undertake research using data from an Indonesian telecom. In particular, the research aimed to gather insights on internal displacement throughout the most affected districts in Central Sulawesi with a view to developing a proof-of-concept on the feasibility of using mobile network data for effective displacement assessment.
Piloting the Approach
Throughout the three most affected districts (Palu, Sigi and Donggala), there were approximately 600,000 active mobile subscribers on this particular network. Considering that the series of disasters began with the massive earthquake on 28th September 2018, our team decided to establish a baseline for resident subscribers (based on subscribers’ mobile positioning data from 1–27 September 2018) and then compared changes in the number of subscribers who travelled outside the districts following the disasters (based on data from 1 October 2018 to 31 January 2019). An origin-destination matrix was also built to determine whether popular destinations could be identified. The analysis was done on weekly aggregates, covering 7-day periods.
To make sense of the results in a way that could be communicated to a broad audience, an interactive visualisation dashboard prototype was designed. This visualisation communicates the distribution of subscribers in Palu, Sigi and Donggala, highlights the most popular destinations where people travelled to after the disasters, and provides an estimate of the number of displaced subscribers based on analysis of subscribers’ movements before and after the disasters.
Engaging with Stakeholders to Identify Potential Uses
PLJ and IOM invited representatives from the Indonesian Government, United Nations, and mobile telecoms to participate in a training workshop to discuss how mobile network data can be better harnessed to support government-led disaster resilience efforts. The workshop was also intended to gather feedback to inform further development of the visualisation dashboard and explore adoption and scaling of the tool to aid in future disaster response. These were the takeaways:
Data: The discussion primarily focused on the advantages and limitations of using mobile phone data, data management, as well as data protection and privacy. There were thoughtful questions regarding the weekly aggregates the platform currently uses, and whether there’s a possibility to also enable daily measures to conduct a more granular displacement assessment. To validate the findings of the analysis, one of the suggestions was to overlay additional data sets such as from social media and data collected from disaster management authorities that might provide deeper insights. The team was commended for having a set of data protection and privacy principles in place to guide the research, and was also encouraged to do regular assessment of risks, harms and benefits as the nature of the research may evolve.
Platform: The usability testing conducted during the workshop was a unique way to gather first-hand feedback from potential end-users. To synthesise the comments offered by the stakeholders, we created the following clusters: not-so-useful features, difficult-to-comprehend features, most-useful features and general ideas. One of the not-so-useful features was the use of coloured dots on the map to represent the number of people in an area, because they appeared similar to the radiating coloured dots indicating subscribers’ movement. This feedback is also linked to the absence of a map legend and information boxes to explain various icons and features. Having an interactive timeline was highly appreciated as it allows users to glide between different weeks and compare aggregate changes. Other general suggestions included: adjusting the default display to focus on Palu as the main site; using a line instead of a dot to depict subscribers’ movements; adding relevant information pop-up boxes; and incorporating a translation feature as end users might speak different local languages.
Ecosystem: Scaling up the tool and nurturing an ecosystem in which it can be maintained was a critical part of the feedback discussion. This included inputs on relevant policies that should be put in place to support such an ecosystem, and the best positioned stakeholders to become custodians and ensure its sustainability. The importance of setting up a working group within the Indonesian Government to facilitate access to relevant data sets and encourage regular training on the use of mobile network data was stressed as one of the initial steps needed. This would then be followed up with regular high-level discussions between the established working group, related ministries, as well as mobile telecoms operating in the country. However, for these discussions to bear fruit, existing disaster management regulations in Indonesia need to be revised to create room for mobile network data research for disaster preparedness and response, as well as the integration of emerging insights.
Refining the Prototype for Future Use
The feedback we gathered during the workshop raised a critical question for our team: In what ways should the prototype be refined to ensure it is practical and sustainable for end-users within the Government? Mapping potential stakeholders as well as their capacity to utilise the tool is paramount. Allocating resources to assist citizens who travel outside of their affected communities during times of disaster can often be overlooked, as in most cases aid and resource mobilisation are prioritised for citizens remaining in affected areas. Our analysis demonstrates that by identifying the movement of affected cohorts of a population, we can also ensure that they are not left behind in the Government’s disaster response and resilience strategy. The case of the 2018 Central Sulawesi disaster episodes revealed that data on displaced citizens is scarce, hampering the speed and effectiveness of humanitarian agencies in distributing vital resources to those most in need. There’s still more research to be done on how mobile network data can be harnessed to inform the Government’s disaster response effort, and this proof-of-concept demonstrates the value of the approach and how it can be applied in different disaster scenarios to reduce the scope of impact.
Improving on the prototype of the visualisation dashboard and incorporating feedback from the workshop, we built a platform that could better serve the government and the humanitarian community, as well as serve as a tangible conversation opener when the Government is approaching mobile network providers.
Here are some of the key features of the platform:
Distribution of Subscribers in Palu, Sigi, and Donggala
The line graphs below provide an intuitive representation of changes in the number of subscribers before and after the disaster event throughout the three most affected districts. They highlight in particular how many subscribers remained in the affected districts in the immediate days following the disasters, and how the numbers fluctuated in the weeks later. One can observe that the number of subscribers in Palu and Sigi decreased the first week after the disasters, while the weekly aggregate for Donggala saw an increase. This increase then noticeably reduced in the following weeks, while the numbers of subscribers in Palu and Sigi steadily increased (an indication that people might have begun returning to their local communities).
Popular Destinations
Our team inferred the popular destinations by examining where subscribers travelled to (outside of the three most affected districts) after the disasters struck. The chart below shows the popular destinations in the first week after the disaster event, with the top five destinations being nearby Parigi Moutong, Makassar, Poso, Toli-toli and Banggai. The platform allows users to interact with the timeline from September 2018 to January 2019, thus showing changes in population flows over time. Having a sense of where subscribers are moving to (instead of only focusing on their places of origin) is particularly useful in helping disaster authorities to allocate additional resources to areas with more than usual overflows. These insights can complement our automated Managing Information for Natural Disasters (MIND) platform, which is to provide the locations of public infrastructure in these destinations (for instance to identify potential sites for temporary shelters) and geotagged social media data (to understand what people in the top destinations are saying and what they might be in need of).
Population Displacement Proxy Indicator
The table below presents some useful statistics from Palu, the provincial capital which was the most affected district. Again, observing subscribers’ movements during the week of the disasters, one can see that of the near 200,000 subscribers who were in the location before the disasters hit, only about 50,000 subscribers left. It’s interesting to highlight that our analysis showed that around 30,000 subscribers moved into the region during the week of the disasters, which might be a signal of disaster recovery efforts. The number of subscribers totalled in ‘balance’ on the other hand is an inference of the approximate number of subscribers who might be displaced.
Creating An Enabling Environment
We used the platform we developed as a basis to introduce training engagements for government officials from the National Disaster Management Authority (BNPB), the Ministry of Social Affairs (Kemensos) and Statistics Indonesia (BPS), and received positive feedback on how it can be further contextualised to inform government decision-making, in addition to functioning as a complementary information tool to help shape disaster analysis and response strategies.
While there remains a decision to be made on which institution should become the future custodian of the platform, the hands-on training provided will allow government officials to have a better understanding of the potential use of mobile network data for humanitarian response, and at the same time enhance their ability to leverage the platform in future crises.
Pulse Lab Jakarta is grateful for the generous support from the Government of Australia
