DSSG-Supported Project Uses AI to Reveal How Climate Events Drive Migration
A research project supported by DDDI's Data Science for Social Good (DSSG) program is featured in Penn Today for its work on one of the hardest questions in climate research: how environmental disasters change what people do, including whether and when they migrate.
Machine Learning for Environmental Event Detection (MLEED) is led by Erik Wibbels, Presidential Penn Compact Professor of Political Science and co-director of PDRI-DevLab, in collaboration with Irina Marinov, associate professor of Earth and Environmental Science, and Anna Cabré, a researcher in Marinov's Oceans & Climate Group. The project grew out of a gap the team identified together: there was little good data connecting climate events to social outcomes such as protest, violence, legal change, and migration.
To fill that gap, the team adapted the tools behind Machine Learning for Peace, an AI model built at PDRI-DevLab with chief data scientist Zung-Ru Lin. MLEED processes about 130 million locally sourced news articles from 71 countries. It tracks three kinds of environmental events (sudden-onset disasters like floods, slow-onset disasters like droughts, and human-induced disasters like oil spills) along with 13 social responses, over time and at the subnational level.
What the data show
In a report published by the Kleinman Center for Energy Policy, Wibbels, Marinov, Cabré, and predoctoral fellow Iman Dahr examined El Salvador, Guatemala, Honduras, and Nicaragua. They found that sudden-onset disasters are associated with higher international migration in the months that follow, while slow-onset disasters are associated with lower migration. Drought gives farmers a reason to leave but also wipes out the resources they would need to do so. The team also found that major disasters in Central America were followed by increased encounters at the U.S. southwest border six to nine months later, which challenges the common view that disaster displacement is mostly local and short-lived.
The researchers hope these patterns can help policymakers anticipate migration trends months ahead and support communities' ability to adapt in place.
Students at work in the data
MLEED is also a teaching resource. Marinov has built the dataset into her Climate and Big Data course, where students this past spring explored links among environmental disasters, food prices, and displacement in West Africa and South Asia. Undergraduates Julian Huang and Luca Zhang continued that work over the summer with Cabré and Marinov, and the Kleinman Center will publish two policy digests based on the group's findings. Wibbels told Penn Today that having students "mucking around in the data is actually very helpful" for making sense of a dataset this large.
About the support
This research was supported by DDDI's Data Science for Social Good program and the Kleinman Center for Energy Policy. DSSG supports projects that apply data science to pressing societal challenges.