A house submerged in flood waters following Hurricane Irene in New Jersey

Hurricane Irene devastated parts of New Jersey in August 2011. Severe flooding and power outages forced thousands of residents into shelters. 

A Sharper Focus on Catastrophe Modeling

Paolo Bocchini, Daniel Conus, Brian Davison and their colleagues leverage their collaborative experience in probabilistic modeling to sharpen their focus on catastrophe modeling, a discipline not traditionally explored in academia.

Story by

Kelly Hochbein

Photography by

Spencer Platt and Bryan Anselm

Early on the morning of January 17, 1994, a magnitude 6.7 earthquake rocked California’s San Fernando Valley. The Northridge Earthquake killed at least 57 people, injured thousands and resulted in tens of billions of dollars in damage. The cause of this particular quake—one crustal block moving over a second crustal block—produced extremely powerful ground shaking, making it even more destructive. 

To recover from the devastation of a rare disaster like the Northridge Earthquake, individuals and communities need to be able to rebuild quickly while navigating tremendous loss. This comes at a cost, most often covered by insurance, and the speed with which insurers make payments can impact the long-term recovery of a region. To operate effectively, insurance companies must prepare for the unknown, predicting the future without much information from events of the past. They do this with catastrophe modeling. 

In cases of standard coverage, insurers look for patterns in their extensive claims data to calculate an individual customer’s likelihood of having to make a claim. This straightforward process works well in determining a customer’s auto insurance premium. It is not, however, effective in the case of rare events such as earthquakes, for which insurers cannot conduct statistical analyses of historical claims and losses. With a catastrophe, “what if?” becomes the critical question.

“The necessary premise—that we have a lot of past claims data that we can use to inform our models—basically disappears. Because these are, by definition, rare events, there is not going to be a large amount of past data. [Insurers] cannot use their normal way of operation,” says Paolo Bocchini, associate professor of civil and environmental engineering. 

“If we don't have much data, we can't do traditional modeling based on lots of data to figure out what the underlying relationships are among the events and their impacts and so on,” says Brian Davison, professor of computer science and engineering.  

A rigorous probabilistic approach to the study of disasters and their consequences, catastrophe modeling, or CatModeling, attempts to estimate potential events and their associated risks, including financial losses. Researchers take what they know about a particular scenario and incorporate methods from a variety of disciplines to make predictions about the likelihood of certain outcomes. In the case of a hurricane, for example, researchers might gather data about past hurricane activity and details about the infrastructure in a particular region. They then build a model to predict the likelihood of the region experiencing losses in a similar storm in the future, as well as the potential cost of recovery from those losses. Beyond natural disasters, decision-makers can use information from CatModeling to plan for rare events such as pandemics, financial crises and political unrest.  

Unlike most disciplines, which are born in academia and then applied in industry, says Bocchini, catastrophe modeling was born in the insurance sector as “a sister to actuarial science.” The information it generates can help insurance companies estimate losses and calculate premiums.  Its use is not limited to the insurance industry, and it can help leaders at all levels make the best possible decisions about how to protect and ensure the recovery of their communities.

A young boy stands in front of debris following a tornado

Tornadoes tore a path of destruction through Tuscaloosa, Al. on April 27, 2011, killing more than 200 people.

“Natural disasters, political unrest, pandemics—these are all kinds of events that could benefit from planning for the future and trying to anticipate what damage they can do to society for multiple factors: from a financial perspective, from a social perspective on the well-being of people, from physical damage to infrastructure if we talk about hurricanes or earthquakes. These can have an immediate impact on what functions in our communities,” says Daniel Conus, associate professor of mathematics. 

Bocchini, Davison, Conus and their colleagues at Lehigh are leveraging a wealth of experience and expertise in their respective fields and their collaborative experience in probabilistic modeling to sharpen their focus on CatModeling, a discipline traditionally not explored in academia. 

Building on Existing Expertise

Lehigh has had teams of researchers studying disaster resilience for years. Led by Bocchini, the university’s Probabilistic Modeling Group has built momentum in its work, developed relationships with collaborators in industry and academia, and secured funding for a variety of projects. Eventually, the group began receiving grants for applications of probabilistic modeling to catastrophe modeling. This prompted Bocchini and his colleagues to narrow their focus and assemble a catastrophe modeling team based on ongoing Lehigh efforts. 

Among the probabilistic modeling efforts is “Probabilistic Resilience Assessment of Interdependent Systems (PRAISys),” a five-year project that used a probabilistic approach to examine how interdependent infrastructure systems work together during and after an extreme event, such as a natural disaster. The PRAISys platform combines models of individual infrastructure systems—such as power distribution systems, transportation networks and communications systems—with models of their interdependencies to allow researchers to assess the resilience of the systems under uncertainty. 

“We essentially built a simulator for how communities will act and respond to catastrophic events and how they would repair their infrastructure and how long it might take,” explains Davison, whose own research focuses on machine learning and its applications in a variety of domains. “Given that it is a simulator, you can simulate anything: You can change the environment, you can change your investments or the way things are configured and see how that would impact your resilience, your ability to adapt and respond to the failures that result from a flood or from a tornado or earthquake.”

The project, an interdisciplinary collaboration between Lehigh researchers and researchers from Florida Atlantic University and Georgia State University, involved 58 scholars and was funded by the National Science Foundation. It resulted in a number of publications, conference proceedings and book chapters, including papers co-authored by former postdoctoral fellow Wenjuan Sun, Bocchini and Davison in Structure and Infrastructure Engineering, Journal of Infrastructure Systems and Sustainable and Resilient Infrastructure, among others.  

Sun coordinated the project, communicating the research progress between different groups and merging them into a final deliverable: the PRAISys platform. 

“The work itself is very interesting,” says Sun. “I hope what we did here can be helpful. … At the very least it built the foundation for the improvement of CatModeling at Lehigh.”

Although the financial support for the project has ended, the work continues, Bocchini says. “The idea was to develop a platform for interconnected infrastructure systems. We were looking at power, telecommunication and electricity and studying how they are affected by disaster, how they can be damaged, how the decision-making takes place, and how they are recovered—with all the uncertainties that this process obviously involves. The idea was to be able to combine and predict the impact of different choices.”

For example, a power company must decide which transmission towers to retrofit to withstand a severe storm. Is this the right investment? Or is it better to invest in more cranes to repair failed equipment more quickly? Should they focus instead on mobile generators to power critical facilities? PRAISys can help company leadership determine how to best allocate resources today to improve the system’s resilience tomorrow. 

Three people survey the extreme damage caused by a tornado

Tuscaloosa, Al., April 28, 2011

 

“These are very, very different types of investments, and normally they are not compared because there were no models that could put all of them together. So we tried to create a platform that had all these features, so analysts can introduce different types of changes to the model and see which one is most effective. I think this was really a first, and the platform is available for download,” Bocchini explains. “This is just the beginning.” 

Although CatModeling started with natural disasters, it can also be applied to other rare events, including infectious disease outbreaks. Bocchini and his colleagues, including former Lehigh researcher Javier Buceta, currently at the Institute for Integrative Systems Biology (I2SysBio) at the University of Valencia in Spain, are applying the traditional steps of CatModeling to frame Ebola virus outbreaks in Africa and attempt to forecast their impacts. In particular, they try to predict the spread of Ebola in various cities within a broad geographic region. Focusing on bat-transmitted Ebola spillover to humans, they combine a variety of methodologies, including probabilistic regional hazard modeling and big-data analysis, to determine the cities most at risk and what officials in those areas might do to minimize that risk. Their model utilizes bat birth and death rates, the rate of Ebola infection in bats and their recovery rates, bat mobility, seasonal changes and data about the availability of food and shelter for bats in a particular location. 

This project is supported by the National Institutes of Health and was initially funded by a Lehigh Collaborative Research (CORE) grant. It has resulted in several publications, including a 2018 paper in Scientific Reports that proposes the team’s predictive spatial distribution framework and demonstrates its ability to predict where and when an Ebola outbreak is likely to appear. A preprint on arXiv outlines the team’s use of regression and machine learning techniques to analyze survey data that they personally collected in Africa with a team of undergraduate students; identify the features that best predict an individual’s tendency toward behaviors that expose them to Ebola infection; and develop a predictive model about spillover risk statistics that researchers can calibrate for different regions, including Sierra Leone. 

The Best Choices for Tomorrow, Today

CatModeling is an application of probabilistic modeling, which incorporates random variables and probability distributions into models of events, taking into account uncertainties. Researchers “try to look at the future and determine what is the likelihood of something happening [and] at what is the choice that we make today that will have the highest likelihood to be a good choice tomorrow,” explains Bocchini.

The approach is rooted in mathematics but has applications across many disciplines, including engineering, science and the social sciences. 

Conus brings theoretical expertise to the team’s work. Unlike statistics, which involves collecting data from the past and trying to explain what happens out of the data, “probability takes sort of the reverse approach,” he explains. “We’re actually trying to come up with a mathematical model that can explain from a theoretical basis what we expect would happen even before we collect the data.” 

Prior to his work in CatModeling, Conus worked on mathematical finance and models for particle physics, which, he says, sound drastically different, but they rely on the same mathematical tools. “The approach we take is that what happens in a catastrophe is that you have a lot of small entities that play a role and impact what happens in the big picture. Essentially, the heart of the mathematical techniques that we use is to take all those small effects, and instead of modeling every single one of them, to actually put it together and try to come up with models and equations that describe the big picture without having to look down on every single little entity that makes up the model,” he says.  

For any kind of disaster, CatModeling includes the probabilities of different event scenarios, levels of intensity, the region’s vulnerability to damage and the potential financial impact. 

Liyang Ma ’20 Ph.D. used intensity measure maps to determine the performance of structures under the conditions of particular catastrophes while working on his doctorate with Bocchini as his advisor. Now a Lehigh postdoctoral fellow working specifically on catastrophe modeling, he develops these maps for different hazards, such as the wind speed, storm-surge level or earthquake magnitude of a region. 

“How do I produce very good maps that can represent these catastrophes more precisely [and] more accurately so that when people do[work related to] how structures behave in a hazard, they can get better or more accurate results?” he asks. 

For these maps to be representative, he explains, researchers use Monte Carlo simulations to generate a large set of maps given the uncertainty of an event like an earthquake. His goal is to select a smaller set of the most representative maps for that hazard.

Postdoctoral fellow Haifeng Wang, who received his Ph.D. from the University of Buffalo, focuses on wind field simulation. “From a single wind speed, I can generate a large number of wind fields, stochastic pressure fields for a building,” he says. Wang collaborates with Ma by providing the wind load field of an entire building for use in structural response analysis. 

“Hazards nowadays become more intense and severe, and human societies are more vulnerable,” says Ma. “In the past, people didn't need electricity or the internet. Nowadays they need those things. They need hot water. When a hazard strikes, like a hurricane, people will be without power for days. So the goal is to make society better prepared for those hazards. ... Catastrophe modeling is a new area that has a bright future for civil engineers.”

People survey the damage following a tornado

Tuscaloosa, Al., April 28, 2011

CatModeling invites collaboration across disciplines, challenging researchers to determine the most relevant questions to ask, explains Conus.  

“That brings in all sorts of other disciplines that are not necessarily hard science and engineering, but disciplines in the humanities and the social sciences, to decide what is the relevant question if a major hurricane came on the East Coast,” he says. “Are we really worried about the solvency of the insurance company? Maybe that's what an economist would answer, but as a society is that the very first question we should answer? Or should we focus on the remaining buildings? How many people can they host to make sure everyone has a shelter? … What are the right questions to ask and to model? And then, once you have answers, you can start leading to policy creation. What's the policy that we should develop to try and diminish the impact or make it better for people?” 

Several Lehigh faculty in the social sciences have joined the effort, forming a parallel group to collaborate on projects related to community resilience. In March 2020, members of this group, including Jessecae Marsh, associate professor of psychology; Dominic Packer, professor of psychology; and David Casagrande, professor of anthropology; in collaboration with Bocchini, Davison, and others within Lehigh’s Institute for Cyber Physical Infrastructure and Energy (I-CPIE), shifted the focus of their planned project on community resilience and recovery in the areas of North Carolina affected in 2018 by Hurricane Florence to the COVID-19 pandemic unfolding in their midst. The team conducted a real-time longitudinal investigation of how individuals in the United States perceive, respond to, and recover from the impact of COVID-19, with the aim of understanding whether traditional metrics align with the lived experience of individuals living in communities recovering from a catastrophe. 

Embracing Opportunity

In recent years, major insurance companies have started to build their own catastrophe modeling departments to better prepare for extreme events. This has increased demand for professionals in the field. As a significant next step, the team has received a Lehigh Research Futures grant to tackle this next challenge: the launch of the Catastrophe Modeling Center and the development of an academic program in catastrophe modeling at Lehigh. 

“The world has people who are interested in catastrophe modeling in industry, in consulting and occasional academics,” says Davison. “But what was interesting was that there was, in the U.S., at least, no focus area for that. And it seemed like it was a good match for what our experiences said we could do at Lehigh. [Establishing] an academic program in that area would be novel.” 

In addition to postdoctoral fellows Ma and Wang, four Lehigh doctoral students are currently focused on CatModeling. Even without a formal program, the CatModeling team already has had several students graduate and take positions in or related to this field, including at the two largest CatModeling firms in the world. The establishment of an educational component to the team’s work would strengthen that pipeline. 

“The intent would be that the students that are in such a program would be prepared for careers in catastrophe modeling. Today they sort of come at it from the side—they might be a statistician or [work in] financial modeling or insurance, but they don't have all these things come together. You need to have an interdisciplinary education to really prepare you for catastrophe modeling,” says Davison. 

Lehigh faculty have partnered with colleagues at Rice University, Stanford University, and Florida Atlantic University, as well as AIG General Insurance, on this work. The process of applying for grants with researchers from other institutions has been beneficial, Bocchini says. 

“The opportunity for us to create this network with the other universities [is valuable in and of itself],” says Bocchini. “Writing a proposal together provided the opportunity to clarify ideas, roles [and] participants.” 

This collaboration extends to advising the team’s postdoctoral fellows. Ma is co-supervised by Bocchini and Conus, Wang by Bocchini and faculty at Rice University. 

“We’re trying to make these connections by using people to bridge gaps,” Bocchini says. 

Since 2020, members of the team have published seven papers and a book chapter on CatModeling, and the group is planning a workshop for Spring 2022 that will include colleagues and collaborators in academia and the insurance industry. The team also has begun collaborating with faculty from Lehigh’s College of Health, making CatModeling an even broader initiative at the university. 

Says Bocchini: “There is this entire field, this entire discipline that we need to be a part of. I think academia has built all the building blocks and then we have completely delegated to the industry the job of assembling the pieces and making it a science. I think we can play an important role in this.” 

Story by

Kelly Hochbein

Photography by

Spencer Platt and Bryan Anselm

Related Stories

researchers in sierra leone

Individual Risk-Factor Data Could Help Predict the Next Ebola Outbreak, New Study Shows

Researchers confirm a relationship between social, economic and demographic factors and the propensity for individuals to engage in behaviors that expose them to Ebola spillover. 

Ron Nersesian ’82 and Clay Boice '22

New Keysight Technologies Lab Added to Packard Lab

Made possible by a gift from Keysight Technologies Inc., the lab will provide experiential learning opportunities for those entering the fields of electrical engineering and computer engineering.

Aurelia Honerkamp-Smith

NIH Grant Funds Lehigh Researchers' Exploration of Lipid Membranes

Aurelia Honerkamp-Smith and Damien Thévenin envision that the models they produce will apply to multiple cell lines and flow conditions, and will lay the groundwork for future research directions.