New famine modeling studies offer clues on how extreme starvation develops and how earlier intervention could help prevent catastrophic crises

Famines were once thought to be fading into history. But after devastating crises in places such as Somalia and Sudan—and with levels of extreme food insecurity rising worldwide—researchers are asking whether a better understanding of famines could lead to improved forecasting and help prevent catastrophic hunger.

A collection of new studies from more than 30 researchers at Tufts University and institutions spanning academia, humanitarian organizations, global development, and data science suggest that famine is not a sudden event, but rather a complex system that evolves—and may be interrupted.

The papers, published in the Journal of Public Health Policy, demonstrate how various types of models can provide insight into when famines begin and end, where they occur, how and why they form, and who is involved.

“We have learned a lot from infectious disease forecasting about how to identify early signals and understand how many interacting factors shape outcomes. And the COVID pandemic showed us the power of combining complex data from many sources to better understand crises as they unfold,” said Elena Naumova, the journal’s editor-in-chief and Barry J. Rosenbaum Professor of Data Science at the Gerald J. and Dorothy R. Friedman School of Nutrition Science and Policy at Tufts University. “We think famine research can benefit from similar tools and interdisciplinary collaboration.”

Naumova and Paul Howe, Irwin H. Rosenberg Professor of Nutrition and Human Security and director of the Feinstein International Center, co-editedthe special issue of the journal. Below, they share their insights on how these approaches could improve famine prevention and response.

What causes famine?

Paul Howe: People sometimes think famine is simply caused by drought or conflict, but recent research has given us clearer insight into the way famines form. Usually, some set of factors—such as conflict, drought, or an economic shock—disrupts people’s normal ability to get food. This combines with their underlying vulnerability to create pressure.

But this pressure alone doesn’t create famine. There also needs to be something that prevents release from this pressure, which I describe as a “hold.” For example, a conflict may destroy crops and livelihoods, but the hold may be the lack of safety created by the conflict or a deliberate policy restricting access that prevents humanitarian assistance from reaching people in need. When pressure continues unchecked because of the hold, the crisis can escalate, and the situation in communities can tip into famine.

What is different about this approach to fighting famine?

Paul Howe: Historically, famine has often been viewed either as a sudden event where people rapidly starve or as a longer, socioeconomic process. This new approach tries to understand famine as a system—something that is dynamic and forms under certain conditions. It considers how many different interacting factors come together, how crises escalate, and when they shift into a state with very high mortality. As it turns out, modeling is particularly well suited to capturing these kinds of dynamics.

In some ways, you can compare famine to a hurricane. If you understand better the dynamics of how the “perfect storm” develops, you can improve your forecasting because you know which signs to look for. But in other ways, the analogy is misleading since famines almost always involve human action or inaction. Modeling also allows us to capture these political dimensions and to determine how different decisions affect the evolution of crises.

Can famine prediction actually help prevent people from starving?

Elena Naumova: From a modeling perspective, famine is a cascade of events manifesting interconnected processes. Different forces amplify one another over time, and sometimes they counteract one another. The challenge is understanding those feedback loops and identifying the moments when conditions begin to intensify.

However, it’s not enough to build the models or to collect data. Preventing the enormous suffering and loss caused by famine means not just recognizing the warning signs early but acting on that information before conditions turn deadly.

Why is famine so difficult to predict?

Elena Naumova: Famine often happens in places where collecting reliable data is incredibly difficult. During a crisis, the focus is rightly on saving lives, not building data systems. Infrastructure may be damaged, data may be incomplete, and there often are serious ethical and political concerns around collecting and sharing sensitive information.

At the same time, we know our prediction abilities can improve. People once said it was almost impossible to predict influenza outbreaks or the path of major storms. But with sustained investment in data collection and analysis, those forecasts have steadily improved. If we want to trust predictive models—including AI tools in the future—we have to start with trustworthy data.

Can different kinds of models help answer different questions about famine?

Howe: Yes, that’s really one of the central ideas behind this work.

Some models help us think about when and where famine risks may increase. For example, a new study led by Erin Coughlan de Perez, an associate professor at the Friedman School, used climate and statistical models to examine how drought, extreme weather, and food insecurity may raise the risk of famine in the future, and how adaptation by planting alternative crops may help prevent conditions from worsening over time.

Other models help us understand how famines form. A new study led by Bingjie Zhou, who earned her Ph.D.at the Friedman School, traced the sequence of events in Yemen—from climate shocks to food insecurity to rising malnutrition—to better understand how crises escalate over time and whether there are warning signs along the way.

Another team explored how and why a devastating famine occurred in southern Sudan in 1998. The model suggested that it was largely generated by counterinsurgency warfare and drought (the pressure) and a ban on humanitarian flights (one of the holds); the analysis helped identify key points where political action and technical interventions might help prevent future crises.

Naumova: Models also can operate at very different scales. Some examine what happens at the household level, including how families make difficult decisions during crises, such as selling assets or changing how they eat.

Udita Sanga, an assistant professor at Friedman School, meanwhile used qualitative systems modeling and archetype analysis to investigate the devastating 2011–12 famine in Somalia at the community and national levels, identifying how interacting environmental, economic, social, and political pressures shaped the crisis.

Other models dive into the biology of starvation itself or expand our perspective to see how global supply chains may affect food security.

One of the exciting parts of this work is that we are bringing together many disciplines and ways of thinking. Some approaches help identify patterns and probabilities. Others help us understand feedback loops.

In public health, we often use different models to answer different questions, and famine should be no different. The goal is not to choose one “best” model. The goal is to create a space where different methods can contribute to a deeper understanding of famine and, ultimately, help us act earlier to save more lives.