Is it possible to predict future forced displacement? | Fixing Aid

Fixing Aid | Episode 5 | Foresight

Alae Ismail:

Hey listeners, welcome to Fixing Aid!

I’m Alae Ismail, and in this podcast series by The New Humanitarian, we take a look at innovations meant to improve the lives of people in humanitarian crises – from those fleeing conflict to communities facing flooding, prolonged drought, and hunger.

Today, we take a look at the future: Is it possible to predict which people and communities will be displaced in the next year, and the years after? And if it is, does that mean people who must leave their homes because of conflict and disaster can get the help they need more quickly?

There is a prediction model that claims to know where people might be uprooted this year, next year, and the year after.

But before we dive into just how this model works, if it’s accurate, and how it can be used, let’s look at a region with high rates of displacement: the Sahel. Conflict and the impacts of climate change are threatening the lives of people raising livestock and other communities in the Sahel. In a scenario like this, how useful is it for humanitarians to have advance warning of likely displacements?

Leila Adamo Armand:

If the model is accurate enough in its predictions, it could help guide actions in anticipated emergency situations, to understand on which elements to act, which actors should intervene, and in which areas, to avoid or minimise the negative effects on pastoral populations and save time in mobilising the necessary resources.

Ismail:

That’s Leila Adamou Arouna. She is part of the RBM — Billital Maroobé Network — a group of herders and pastoralists in the Sahel, focusing on the well-being of two and a half million people.

Leila frequently sees the impacts of competition over natural resources. Climate change means less and less land is available to be cultivated, and the Covid-19 pandemic further intensified the negative impact on farmers. Leila says getting information from the future would help her make better decisions on behalf of the pastoral communities and their interest:

Adamo Armand:

It is necessary to share the information with the partners so that together we can find the best strategy of action and respond to people’s needs as soon as possible.

Ismail:

RBM plans and creates programs for pastoralists with input from an already existing early warning system in the region, but this one cannot make future predictions. Every day, Leila’s organisation and its members share alerts on security, food, drought, and other factors that are based on information from the previous day. It’s their attempt to mitigate negative impacts. But having information on the future instead of what happened yesterday could be really important for pastoralists in the Sahel, Leila says:

Adamo Armand:

It is important for the RBM to be able to look into the future and to make the best decisions that respect the pastoral communities, whose interests it defends.

Ismail:

Trying to predict the future by estimating the number of people who may be displaced in a region — and when that might happen — is Foresight. It’s a prediction model developed in a pro-bono partnership between tech corporation IBM and the Danish Refugee Council, an international humanitarian organisation. The partnership started in 2017, following the so-called migration crisis in Europe, and ended in 2020. The online tool uses 120 indicators and open-source data to anticipate where people might be on the move this year, next year, and the year after.

Alexander Kjærum is a global advisor and senior analyst for the DRC in Denmark. By managing the Foresight prediction model, he has access to lots of data on how many people will be displaced in different countries around the world. The data shows that displacement will be a serious issue for millions of people in the coming years:

Alexander Kjærum:

For example, for Burkina Faso, it’s approximately around 1.6 million people that were displaced by the end of 2021. The model is predicting that that number will increase to more than 2 million this year, and then increase further to 2.3 million in 2023. In the case of Cameroon, we have around 1.1 million people displaced at the moment. The model is predicting that will increase to almost 1.5 million this year and 1.6 million by the end of 2023. For Ethiopia, we have around 4.5 million displaced people by the end of 2021. And we expect that number to increase to almost 5.5 million by the end of 2023. So almost an increase of 1 million over the course of two years.

Ismail:

Other countries that might see significant increases in displacement in coming years, according to the Foresight model, include Afghanistan, Yemen, South Sudan, and Somalia. Hearing about these countries might not be that surprising. DRC admits that the model is better at predicting stable trends and patterns that could help with better planning when it has more accurate numbers.

To estimate how many people will be displaced in different countries or regions around the world, the model relies on open source data, typically from organisations such as the World Bank or United Nations institutions. That data is then processed and analysed with the help of algorithms. These algorithms identify patterns and relationships in the data, informed by 25 years of historical data. All of that together results in future forecasting.

Kjærum:

The type of data that we focus on are sort of the root causes of displacement. What are the different factors that drive displacement? So we look at conflict data, violence data, we look at economic data, aspects related to governance, environmental climate data, as well as different social demographic data that’s available. Most of the data that we collect is available at a national level and typically sort of updated on an annual basis. We try to avoid collecting personal, individual-level data. First of all, it's also difficult to get. But of course, also due to the, to the sensitivities around it. We are also not, for example, using social media data both related to the capacity and resources to actually extract that data and make something meaningful out of it, but again, also because in a lot of the countries where we work, social media penetration isn't very high. So we would get a very biased data set if we were to base our models on such data.

Ismail:

So let me take a look at this magical prediction model. I’m on the DRC login page, and I add my email and password — given to me by Alexander — which takes me directly to a dashboard with a world map and graphs with forecasts and data trends. The first thing that draws my attention is a map of countries covered in red, amber, and green circles, which indicate the total number of people forced into displacement.

Countries highlighted in yellow mean that the DRC has forecast figures. And if I hover over a country with a red circle, it means I can get more insights automatically on their forecast and trend. Red countries include Afghanistan, with 6.39 million people forced into displacement; Syria with 13.3 million; and Yemen with 3.69 million — according to the prediction tool.

I can click onto the indicator timeline, and I see a timeline of total forced displacement from 2000 to 2020. And what's pretty impressive is, by viewing the visual insights tab, it looks like a network diagram with keywords like Afghanistan, refugee, and Ukraine, and I see articles related to the topic. All of this information can be exported to create a customised report.

This all looks super impressive. But to go back to Leila, who works with pastoralists in the Sahel — a region faced with a lot of displacement — what can the Foresight model do for those communities?

Kjærum:

So what we hope to achieve is to enhance their early warning system that they are providing to their members, so that they can better support, essentially, their members, and the pastoralists on the ground in the Sahel by being able to anticipate when a drought might occur, and what the impact of that drought might be. And also, how can different interventions, how can we support these pastoralists communities better before a drought actually happens in these areas.

Ismail:

Understanding how many people will be impacted could mean, in an ideal world, that humanitarians prepare and deliver food, shelter, and other things that people lack very quickly after the need arises. And it can also help in other ways: when knowing more about a potential disaster, people could be warned and protected before things go bad. In recent years, humanitarians have started to shift their focus to be more anticipatory, by handing out funding based on certain risk triggers. But this approach is still not widely used.

The model covers 26 countries at the moment. And as the algorithm continues to learn, the accuracy is improving, according to the DRC. They’ve seen the margin of error on average decline from 22 percent to 15 percent. And in some countries, it’s even less: In Afghanistan, the margin of error is now down to 8 percent; for Guatemala, the margin of error on the forecast is now at 6 percent. This is more accurate than the 26-percent margin of error among the planning figures used by the aid sector, says Alexander.

But the model has limited success when it comes to predicting sudden onset crises — there was no heads up that a war would break out in Ukraine, or that millions of Rohingya refugees would flee Myanmar in 2017. And the shorter a crisis, the less historical data there is to rely on to find patterns or trends, which results in less accurate data.

So how can those within the aid sector work with the prediction model?

Jade Kahhaleh is the coordinator of WeExist, an advocacy alliance of 28 Syrian civil society organisations. They try to raise awareness on what is happening within Syria with European policy makers, but also in countries around Syria. For Syrians, displacement has been a major issue in the past decade:

Jade Kahhaleh:

Being able to expect displacement movements is crucial, for the organisations that are working in humanitarian aid this is really a key asset because when people are being displaced, programs are also being displaced, organisations are being displaced. So you then have to adapt some of your programs and some of your material needs and staff to new places, new amounts of people coming all of a sudden. So, this is from an operational perspective and programming perspective, and also funds-related perspective, it's really drastically changing everything.

Ismail:

Jade conducted research in 2021 on forced displacement within Syria. Her qualitative research — based on interviews with civil society, experts, and so on — more or less confirmed what the Foresight tool had predicted when it comes to how many Syrians are likely to be displaced in the coming years.

But one thing you need to know about the Foresight tool is that even though it can push out displacement numbers for next year or the year after, the tool doesn't tell you why it’s predicting what it’s predicting.

Jade’s additional research helps put those predictions into context and come up with concrete recommendations to organisations and donors:

Kahhaleh:

If we were to ask a donor to launch funds to support women who have been displaced for psycho-social support, well, if we have some numbers that can actually push and show that we expect more women to be displaced, or we expect more women to suffer from these specific issues because their children are not accessing education, for instance, then I think this can only give more weight to the need and the legitimacy of the demands that we have.

Ismail:

The prediction by Foresight on Syria is that there will be less displacement in real numbers than in previous years, but it will be driven more by economic factors than violence, which was the case before. Another thing Jade noticed: the prediction tool doesn’t take any external factors and data from neighbouring countries into account. But as millions of Syrians have fled abroad, Jade says it’s important to include a more regional context to the predictions, as displacement figures are not only impacted by what happens inside the country:

Kahhaleh:

I think the main limitation is that it's built around national level indicators. So it doesn't take into account any factor that takes place outside of the analysed country. When you are being displaced, it means you're going somewhere. So there is not only just push factors, there are also pull factors. And sometimes you want to go back because the situation is also really dire. Like, for instance, Syrians in Lebanon right now, are really facing some huge difficulties, and in Turkey as well. If there would be any way to improve it, I think it would be to try to include indicators that are in the region, not just in the country.

Ismail:

Hazem Rihawi is a Syrian humanitarian worker. He used to coordinate the health clusters for various aid organisations and UN institutions, in Syria starting 2011, and then in Somalia in 2017.

After three years of developing the tool, IBM and the DRC asked some people for input and feedback. Hazem was one of them.

Hazem says that when looking back at his time coordinating health responses, insight into the future would have significantly informed the way he planned the programs for the people affected by conflict and drought:

Rihawi:

Prediction models will allow us to look at any crisis from the macro and micro levels. From the macro level, what we mean is that we need to start, especially when reaching out to donors and preparing for such responses, to have an idea of what we will be anticipating for the next year, to be able to put a plan that is useful, that is applicable. But also at the same time, it's important to put [in place a] preparedness plan for any shocks that can happen. If we anticipate that there will be certain displacement, forced displacement, that we are not we are not seeing right now but is anticipated, we would be able to put preposition supplies in anticipation of such movement.

Ismail:

Hazem says, the aid program he had to oversee for displaced people could have provided them maybe with better aid, as he might have had better understanding of what he was planning for.

But planning three years in advance is not always easy, as Alexander learned from colleagues working in humanitarian responses across the world. So to make it more user-friendly for aid workers, the model is adjusting the type of predictions it can generate.

Kjærum:

The feedback we got when we presented it to our colleagues was ‘this is super useful, we can use it for strategic planning purposes and so on, but it's not really operational so, can you give us something shorter term — three to four months — and also, the national level, if you can give us more numbers on the sub-national levels, that will help us actually know where should we respond and when exactly should we respond.’

Ismail:

But even after adding shorter-term predictions, requests keep coming in for even more localised knowledge to help make better decisions when it comes to offering aid to people in need.

Alexander says that when IBM and DRC started developing the tool, they looked at predicting mass migration instead of forced displacement.

Mass migration involves large numbers of people moving across international borders for various reasons, that can include better economic opportunities. Forced displacement means people are uprooted from their homes because of conflict or climate disasters or human rights violations, they could be forced to find a safer place either within their home country, or across the border.

First, they looked at Ethiopia, and whether they could predict where in Europe Ethiopians would migrate.

While working towards a model that is as accurate as it can be, can there be a downside to knowing what the future might hold when it comes to displacement? Not all countries are willing to take in refugees and migrants, even when they are fleeing war, conflict, or other types of disaster. So after some internal discussions, the DRC decided to shift the focus on what they would predict:

Kjærum:

We were working with an ethos of being transparent and open about it, so if this information would end up in the wrong hands, could we risk fueling an anti-migration narrative and policies? Would these forecasts be used to, for example, close borders and so on? We took the decision to then say, we'll move away from that, and instead focus our efforts on predicting forced displacement, which is closer to our mandate as an organisation. But we also decided then to move away from predicting where people are going, and just predicting how many will be displaced.

Ismail:

And to make it even less sensitive, the prediction tool now doesn’t even distinguish whether people will be internally displaced or cross a border and become refugees and asylum seekers.

The model can provide better information on preparing pastoralists in the Sahel, as we heard Leila tell us earlier on. It can help inform advocacy around the resources needed for Syrian refugees, as Jade explained. I wonder, can knowing the future prevent people from being displaced altogether?

If I know something bad will happen to me tomorrow, can’t I take action today to prevent that bad thing from happening to me? And by adjusting the course of the future — avoiding or stopping things that are bad and negative and force people into displacement — will that stop the prediction from becoming reality?

Kjærum:

As a single actor, as the Danish Refugee Council, it may be difficult for us to really prevent some of those larger root causes of conflict and so on from occurring. So that may not be a realistic objective. If we as a concerted effort by humanitarian actors, and of course, with donors and diplomacy and other actors step in, then we may have an opportunity. But it is maybe a bit far-fetched that we will be able to prevent some of this displacement from happening. So the more sort of immediate objective is really to be able to respond better to some of these things that we are forecasting.

Ismail:

I want to share one final prediction on displacement with you. For the 26 countries, the Danish Refugee Council tracks through its Foresight tool, the future in terms of displacement is bleak. It’s expected that another 6.8 million people in those 26 countries will be displaced in the coming years. And when including new crises — such as Ukraine — we’re looking at an increase of 17 million displaced between the end of 2021 and the end of 2023.

If we can look into the future with new tools such as the one we discussed today, can the humanitarian sector go beyond predictions and actually be a tool to change the prediction of our future?

That’s it for this week’s episode of The New Humanitarian’s podcast series: Fixing Aid.

We have one more episode coming up, and instead of looking at a tech solution, we’ll look at the use of technology — such as drones, artificial intelligence, and algorithms — to stop people fleeing conflict and disaster from crossing borders, and why this sets a dangerous precedent for the future.

Send us your feedback. Find us on Instagram or Twitter: @NewHumanitarian, or send us an email: [email protected].

Or, visit thenewhumanitarian.org/podcast where you can find a form to share your thoughts on the Fixing Aid podcast.

Don’t forget to subscribe and leave a review.

This podcast is a production by The New Humanitarian.

This episode was produced and edited by Marthe van der Wolf.

And I’m Alae Ismail. See you next time!