Informing National Statistics and Managing Disaster Risks
Samoa, a beautiful island nation in the South Pacific is a tropical paradise — quiet, serene and traditional with its Fa’a Samoa (the Samoan Way) culture firmly intact. But Samoa also lies on the Pacific Ring of Fire and is subject to the treacherous earthquakes, tsunamis and cyclones that frequent this part of the world not to mention the very real impact of climate change that is being felt in Samoa as well as across the Pacific islands.
Pulse Lab Jakarta, through its ongoing partnership with mobile network operator, Digicel, were interested in a proof of concept project designed to test whether artificial intelligence, machine learning or predictive analytics can be useful in public sector decision making.
Specifically, we were interested in operationalising mobile phone network and financial transaction data for the benefit of the Sustainable Development Goals and disaster risk reduction as well as implementing the Sendai Framework for DRR (particularly “Priority 1 Understanding disaster risk” and “Priority 4 Enhancing disaster preparedness for effective response and to “Build Back Better” in recovery, rehabilitation and reconstruction.
Managing these projects remotely is often a challenge but the key to success for Pulse Lab Jakarta is working with our UN colleagues on the ground who can provide us with the domain expertise and the contextual knowledge that enriches our research. During one of our missions to Samoa and in discussions with stakeholders it became evident that the disaster management office quickly gets overwhelmed during a disaster and has no way of accessing real or near real-time information in order to be able to make evidence-based decisions. The hypothesis for this research was whether the use of ‘new’ or non-conventional data sources can be harnessed for better decision making and so that’s what project team set out to discover.
But first things first:
The hardest part of developing any data innovation project is access to data and this project was no different. A new server had to be installed in the mobile network operator’s data warehouse which took time to purchase and install so the project was invariably delayed due to this constraint. Since preserving data privacy is always our main concern we were adamant that responsible data access would prevail even if that meant waiting for a server to be shipped to the data warehouse and the subsequent delays that followed. However, once installed, it meant that our analysis could be done within the mobile network operator’s data warehouse and only aggregated insights from the research would be extracted and made public.
Timelines are a factor.
We wanted to study the movement patterns of the Samoan population during Cyclone Gita as measured through mobile data usage but given the constraints we faced above with logistics, we missed the cyclone season and had to instead prepare a baseline analysis. This is an important point to reiterate. The legal agreements and bureaucracy often don’t coincide with humanitarian needs so its important to prepare well in advance for this research and always worthwhile keeping this in mind when it comes to testing a new methodology during an upcoming natural disaster. Do you have data access? If not, how long before the legal agreements and the infrastructure are in place in order to move ahead with data access and analysis?
In our original proposal, we had suggested using new forms of data for operational use in national statistics production and in disaster risk reduction in Samoa. However, we were overly optimistic with getting access to various data sets. We were able to test the potential of using mobile phone data as proxies for education, household characteristics, expenditure and income diversity given our unique and collaborative partnership with Digicel. But our research really focused in on what our stakeholders were requesting — better understanding of human mobility in times of disasters. And we generated evidence-based estimates on population vulnerability in order to fill in one of the knowledge gaps faced by government decision-makers in the region.
The risks associated with climate change to communities in the Pacific are increasing. The negative impact from cyclones need to be offset by a better-targeted preparation and response by communities, governments and other partners. Limited secondary data and remaining data gaps in the Pacific region restricts efficient humanitarian decision-making. Also given the high programmatic delivery costs due to underdeveloped infrastructure across the web of islands in the Pacific region, the potential gains in terms of efficiency and effectiveness of using new digital data sources were compelling factors in pursuing this analysis.
Nodes and edges strengthen community structures
By analyzing pseudonymised people’s movements through mobile network data and correlating it with evacuation points, we were able to identify that churches are important storm shelters in Samoa. It is also known that churches are places that build and maintain the community structure and in this case the proxy for community structure was social networks derived from pseudonymised call detail records (CDRs). Social networks were calculated by identifying each caller as a node, each receiver is another node and the connection between them is an edge and the number of times a person calls a different receiver is captured by degree — so the higher degree value reflects more inter-connectedness in the community as many people make multiple calls to different recipients while lower degree values indicate many people call a limited number of people.
These proxies of community structures were calculated for each antenna and each antenna is co-located or mapped with a church. The analysis was revealing in that social networks exist amongst communities surrounding the mapped churches. This analysis assumes that if the churches are indeed important community structures, those churches with weak social networks would need entirely different humanitarian or development assistance in the event of a disaster than those churches with strong networked communities. We can also glean numbers of people affected by disasters through pseudonymised mobile network data which can help build up the evidence base for decision-makers. This analysis has contributed to the growing field of the use of big data for disaster statistics. Observations in big data patterns is beneficial for a wide range of efficiencies in planning and understanding disaster risks and mapping out quicker and more effective responses
So what?
Insights on short and medium-term behavioural changes associated with cyclones as derived from mobile network data give a different needs assessment to disaster management officials and can help build effective models of disaster impact.
This project was done in parallel with several other research projects in other Pacific island countries and the methods for modelling education, household characteristics, expenditure and income diversity from mobile network data was published in an academic paper.
What next?
During our visit to Samoa to meet and engage with stakeholders and understand the gaps which we felt could be filled through the use of non-conventional data sources, interest in data innovation was heightened. Through our colleagues in the UN office in Samoa, we had great access to government officials who were impressed with the work of the UN Global Pulse lab network and are keen to leverage the opportunities for data innovation not only in Samoa but also throughout the Pacific islands. This has since led to official communication from the Prime Minister of Samoa to establish a Lab in Samoa to serve the Pacific. Pulse Lab Jakarta is also getting Cyclomon ready again for the coming cyclone season in the South Pacific.
Pulse Lab Jakarta is grateful to UNDOCO’s Innovation Facility which funded this preliminary research.
Pulse Lab Jakarta is grateful for the generous support from the Government of Australia
