Strengthening Humanitarian Action with Insights from Mobile Network Data

Insights on the evacuation destinations of anonymous mobile phone subscribers, submitted during the August 2018 state of emergency on Ambae island.

The volcanic eruptions over the past couple years on the island of Ambae in Vanuatu have forced hundreds of residents to evacuate. Our partnership with Digicel (one of the local mobile network operators) has enabled us to analyse anonymised mobile network data to generate timely insights for the Government as well as humanitarian agencies. Below we outline some of our research in Vanuatu using mobile network data and how the research approach and findings may be relevant to Indonesia and other countries responding to similar crises.

Privacy Up Front

During and after natural disasters, humanitarian response needs timely information on the distribution of affected people and communities, in particular the geographic locations of displaced populations, to be effective. Throughout the past decade, many research papers have underlined the potential of using mobile network data to strengthen public policy and humanitarian action, yet the full operational potential that exists with the data source is still to be unlocked.

In a recent post, we discussed some of the political-economy challenges associated with operationalising mobile network data for sustainable development and humanitarian action. Alongside these challenges, privacy protection is a really crucial aspect of working with this data set. There is not enough space in this blog to dive into the intricacies of the data mining behind our Vanuatu research, we will save that for an academic paper; it is important however to highlight that all our mobile network data analysis is in compliance with the GSM Association privacy guidelines for the use of mobile phone data. Within our Global Pulse network, we also have risk assessment tools for undertaking analysis with new digital data sets as well as privacy and data protection principles that are at the heart of our research.

Among other principles, the GSMA guidelines dictate that subscriber data be anonymised and that all analysis take place on the mobile network operator’s systems and under the operator’s supervision. Similar to how a statistics office processes census data to produce national statistics, our data science team has analysed anonymised mobile network information to produce statistics relevant to areas of humanitarian action.

Modelling the Effects of Two Natural Disasters in 2017

Before the eruption of Monaro Voui Volcano on Ambae in August 2018, our team had already been working to model the effects of two natural disasters in Vanuatu: 2017 Tropical Cyclone Donna and an earlier eruption of the Monaro Voui Volcano on Ambae island in 2017.

We started out by examining network signals at the cell tower level, conducting wavelet and multifractal analysis. Results of which indicated that regular patterns exist in the data set, confirming its value as a general source of indicators and insights on outliers that are of use for real-time sensing.

From there, we looked at Tropical Cyclone Donna, specifically at changes to aggregate behavioural indicators and human mobility networks, finding that the cyclone had clear effects that could be detected in the mobile network data. In the four northern provinces, for instance, aggregate subscriber movements between areas were diminished, and on the strongest day of the cyclone, negative subscriber movement anomalies characterised the areas closest to the eye of the cyclone.

The figures below provide province level and tower level insights on the effects of Tropical Cyclone Donna on human mobility networks:

Subscriber flow by province: aggregate distance travelled (8th May in red, Cyclone Donna’s most intense day)
Active user flow anomalies pre-, during and post-Tropical Cyclone Donna (Red = negative anomaly in active user flow or an anomalous decrease. Blue = positive anomaly in active user flow or an anomalous increase) clockwise from the top left, 1st, 3rd, 8th and 12th May 2017.

Next we investigated the 2017 eruption of Monaro Voui Volcano on Ambae island, which revealed that mobile network data was useful for understanding the effectiveness of the state of emergency issued by the Government. The figure below represents the proportion of Digicel subscribers who are normally resident on Ambae (seen in patterns from week 34 and prior) and whether they remained on the island or evacuated during the state of emergency.

Subscribers who remained on or evacuated from Ambae island over several weeks in 2017

The Government made the first evacuation announcement for Ambae residents in week 35, and as seen in the figure above the proportion of mobile network subscribers evacuating from Ambae increased significantly in week 40, compared to previous weeks, which correlates with the Government’s off-island evacuation order, issued at the end of week 39, on 29th September 2017.

Also, Government reports implied that all residents of Ambae were evacuated; but, the mobile network data suggested that a not inconsequential number of residents remained on the island throughout the eruptions and the state of emergency. This is a useful insight for evaluating the effectiveness of public emergency policies.

Timely Operational Insights in 2018

Our visit to Vanuatu in mid 2018 to present the research findings to the Government coincided with another episode of active volcanic eruption on Ambae. Since we already had all the infrastructure, data and code in place from the previous research, we were able to quickly apply it and respond to a request from the Office of the Prime Minister for insights on population displacement from the island. We provided information on the status of Digicel subscriber evacuations from Ambae as the volcano erupted, as well as aggregate information on evacuation destinations which, alongside data from Government officials, can be leveraged to improve better targeting of assistance and enhance recovery efforts.

We have also taken our learning of how to mine and model mobile network data to understand the impact of natural hazards into the response to the Central Sulawesi Earthquake & Tsunami in Indonesia via that Humanitarian Country Team.

We are grateful to Digicel for their continued partnership and we look forward to sharing more aspects and findings from this research partnership over the coming months.

This research was partially funded by the UN Delivering Together Facility, in partnership with WFP, UNCDF, UNDP and the UN Resident Coordinator’s Office of the UN Multi-Country Office for Fiji, Tonga and Vanuatu.


Pulse Lab Jakarta is grateful for the generous support from the Government of Australia.