Natural disasters such as wildfires, hurricanes and floods are increasing in frequency and intensity, and devastating our environment and local communities. In January 2025, the Los Angeles wildfires caused 31 deaths, damaged over 16,000 homes and businesses, are estimated to have contributed to an additional 409 deaths, while exposing thousands to immediate and long-term physical and mental health risks. Yet, beyond infrastructure and emergency services, it is often community response in the face of natural disasters — such as mutual aid and collective action — that mitigate the most immediate impacts of environmental-related crises: displacement, food insecurity and traumatic stress. These community resilience behaviors play a vital, yet understudied, role in reducing and protecting against harm to health during and post-crisis.

Two new interconnected research collaborations that span four USC schools are seeking to address this often-overlooked factor by combining social science with machine learning to study the positive effects of community resilience at the geographic level. The goal is to create novel tools that measure the current social-psychological resilience of communities while using predictive modeling to enable early detection of potentially vulnerable populations, combining expertise from the USC Suzanne Dworak-Peck School of Social Work, USC Viterbi School of Engineering, and USC Marshall School of Business, with datasets from an ongoing study at the USC Dana and David Dornsife College of Letters, Arts and Sciences. 

“What we see in the immediate aftermath of disasters such as the L.A. wildfires is that there is confusion, and state and federal systems are overwhelmed and not able to react fast enough,” said Michàlle Mor Barak, Dean Endowed Professor of Social Work and Business and principal investigator on one of the studies. “Community resilience steps in to fill that gap. We want to understand how to identify, measure and predict this resilience so that we can better prepare communities.” 

Understanding from a social-psychological perspective

The first study, “Care Amidst Crisis: Applying AI to Understanding Community Resilience Responses,” funded through a grant from the USC Office of Research and Innovation Zumberge Preliminary Studies Research program, is examining the key indicators for resilience and developing a measurement tool to identify communities at greatest risk for poor outcomes to crises versus those likely to show high resilience behaviors. Key indicators being measured and analyzed include the capacity of the community for collaboration, communication and belief in its ability to resolve challenges. 

This study builds upon the Understanding America Study (UAS) dataset established by USC Dornsife, a nationally representative Internet panel of 14,700 adults, including the LABarometer Panel, a probability-based sub-panel of 2,000 adults randomly selected from LA County households, which is led by USC Dornsife sociologist Kyla Thomas. The researchers are using the existing infrastructure of this study to conduct additional data collection from Los Angeles–area residents impacted by the recent wildfires, then incorporate machine learning to better understand the data and construct a model that helps to identify the key factors driving community resilience at the geographic level in the immediate aftermath of a disaster. 

“Sometimes you build machine learning models to forecast,” said Bistra Dilkina, the Dr. Allen and Charlotte Ginsburg Early Career Chair in Computer Science and associate professor at USC Viterbi, who serves as principal investigator of the “Care Amidst Crisis” study. “This is more like an introspective way of using machine learning, where we are building these models in order to understand something about the driving factors.”

Dilkina and Mor Barak, co-principal investigator on “Care Amidst Crisis,” have harnessed an international, interdisciplinary team of experts for the project, and applying social science principles to study affected communities combined with machine learning analysis. Dilkina points out that machine learning is particularly efficient and applicable in the world of social science research where financial resources are often quite limited. 

“AI has many well-established applications in finance, social media, energy and transportation, but there are already a lot of people doing this,” Dilkina said. “I've always been passionate about the opportunity to do something related to public good where there is no obvious financial interest and not very much work or funding already. We’re trying to see where we can use AI to help vulnerable communities.”

Applying the model to predict future vulnerabilities

The second study, “From Exposure to Action: Applying AI to Understand and Strengthen Community Resilience During Environmental Crises,” builds upon the work of the “Care Amidst Crisis” study with the design of a measurement tool and machine learning to apply the tool to specific geographic communities to assess their existing resources, identify predictors of community crisis resilience (CCR), and establish risk profiles in the event of a natural disaster. Funded by the Southern California Environmental Health Sciences Center (SCEHSC), this study uses the same UAS LABarometer dataset as the foundation and applies the novel CCR predictive modeling measurement tool to enhance the understanding of community resilience and the mitigation of health impacts following environmental exposure. Findings will inform targeted policy and preparedness strategies. 

“We feel that an area that has been neglected is how communities as a unit prepare for disasters,” said Mor Barak, principal investigator on the “From Exposure to Action” study. “The community is not often a unit of attention for research and policy. But in a natural disaster, it is definitely geographic community that is the focus. Neighbors helping neighbors, particularly in the immediate aftermath before first responders get to the area, is also critical in the long-term rehabilitation process.  That is why we are shining a spotlight on community crisis resilience.” 

Co-principal investigator Shinyi Wu, associate professor at USC Social Work with a joint  appointment as associate professor of industrial and systems engineering at USC Viterbi, brings an engineering approach to research that spans many social work applications to develop and analyze real-world approaches and technology applications.

"It’s deeply fulfilling to weave my passion for data-driven modeling with a compassionate commitment to human connections,” Wu said. “We’re crafting a tool not only to measure community resilience but to spark a long-term vision of empowering communities to prepare and support one another through environmental crises."

Both studies demonstrate the power of interdisciplinary and international collaboration. The ideas around the impact of community resilience on environmental justice that led to these studies were born out of an international symposium on environmental issues led by Mor Barak in France last summer, funded by the Borchard Foundation, with researchers from around the globe in a variety of disciplines. The “From Exposure to Action” study also collaborates with researchers from the University of California, Irvine, environmental experts in France, Canada and Israel, and several graduate students from USC Social Work and USC Viterbi, including social work doctoral student Meghana Nallajerla taking a leading role. 

“Together, we are uniting our unique strengths to develop a validated measure to deepen our understanding of community resilience and its impact on health outcomes after environmental crises,” Wu said. 

Dilkina points out that USC is an ideal epicenter for interdisciplinary projects such as these to thrive. 

“Part of the beauty of USC is that we have the strength in so many different domains that we are actually very well positioned to bring together these interdisciplinary teams,” Dilkina said.