The bigger picture

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

The Rohingya crisis is one of the world’s fastest-growing refugee emergencies, with hundreds of thousands of people having fled from Myanmar to Bangladesh over the past two years to escape extreme violence. Most of the refugees have settled in the Cox’s Bazar district of Bangladesh, living in very congested conditions, and increasing the pressure on existing infrastructure and facilities in an area that is also under threat from extreme weather conditions.

In February 2018, the World Bank asked Ipsos’s risk analytics team, based in Washington, DC, to carry out an assessment of Cox’s Bazar to locate the refugee population and help identify their needs. At the time, the population of Rohingya refugees in the area was around 650,000.

Taking place over a three-week period, the assessment project required data to be integrated from multiple origins, including drone imagery and a ground survey. “The question the Bangladesh government and other organisations, such as the World Bank, were trying to figure out was where these refugees are, what their needs are, and how they can help them and make the integration process a little easier,” says lead researcher Mark Polyak, who runs the risk analytics division, part of Ipsos Public Affairs.

To understand rapidly where people were, how many of them were there and what potential problems they were facing, Ipsos used satellite imagery – acquired from drones run by the UN – to understand more about the scale of the displacement.

Using machine learning, Ipsos trained the system to understand what percentage of buildings in informal refugee settlements have unsafe conditions and estimate the number of people allocated to each tent. It then ran machine-learning algorithms against all the drone imagery to estimate the maximum number of settlements and to work out the maximum capacity of those settlements.

Hyperspectral imagery was also used by the team, to understand how much light was in a particular area and to identify which settlements had issues with unsafe lighting.

“The areas of unsafe lighting are the areas that have a significantly higher percentage of gender-based violence – sexual assault is much higher in those areas,” says Polyak.

Drone imagery also allowed Ipsos to assess the conditions of roads and calculate the distance that refugees would have to travel from their settlements to access water, health and education facilities, and other services. This is helping the government and the World Bank to decide what areas to focus on and what type of investments are needed.

At the same time, a ground survey in the area spoke to the refugees and the host population to understand issues including access to jobs and wood for fuel. The result of refugees having no access to wood was highlighted by the analysis of the satellite imagery, says Polyak.

“A lot of the time, they would resort to using local forest wood to keep themselves warm, and that resulted in major deforestation in the area,” says Polyak. “That was one of the unforeseen impacts of the situation, and something that we could draw to the attention of the government and World Bank, because this was a tertiary impact that was not anticipated at all.”

While on-the-ground researchers faced some obstacles in terms of difficult terrain (in much of the area there are no roads, so the only access is via footpaths), the main challenges of the project lay in cross-checking information based on when it was received, and layering data from multiple sources to create a comprehensive picture, Polyak explains.

“If a survey of an area was done by the UN and the drone flight flew over the area three weeks later, how do you cross-check what people have said about housing conditions in this area against what you see from the drone flight?”

He adds: “The second challenge is having comprehensive access to the same data sources during the same or comparative timeframe. For example, flight conditions may have been better in one area.

“Dealing with some of the gaps and layering sources on top of each other to create a comprehensive picture is what’s important. Every source has value. Every source has useful information, but all of it is a mosaic – no source of information has a monopoly on truth or accuracy. Combining all of them is what gives
this strength.”

Advances in machine-learning mean that this type of detailed analysis can be done in a much shorter timeframe than previously, according to Polyak.

“In the past, to know the issues with housing conditions, you needed somebody to look at satellite imagery – and over areas as big as Cox’s Bazar, it would take weeks to months for even two or three people to look and identify structures. By running machine algorithms and AI over large groups of images, whether satellite or drone images, I can do in hours what used to take weeks.”

The results of this work and subsequent analysis helped to inform the World Bank’s strategic and operational planning in its response to the Rohingya crisis.

In March 2019, it approved a $165m grant – the third in a series of planned financings – to help the Bangladesh government provide services for Rohingya refugees, and to build and rehabilitate basic infrastructure. This will include constructing a water supply system, building and improving multipurpose cyclone shelters, roads, footpaths, drains, culverts and bridges, and installing solar street lights inside refugee camps.

STATS

Since August 2017, an estimated 740,000 people from Myanmar have fled to Cox’s Bazar, Bangladesh, because of extreme violence in Rakhine state:

• Persecution has forced Rohingya people into Bangladesh for many years, with spikes in 1978, 1991-1992 and 2016
• As of March 2019, more than 900,000 stateless Rohingya refugees live in the Ukhiya and Teknaf Upazilas areas of Cox’s Bazar.

Figures: United Nations Office for the Coordination of Humanitarian Affairs

This article was first published in the October issue of Impact.