Skip to main content

Why Machine Learning Is Critical for Disaster Response

It can help decision makers answer questions such as “When?” and “How bad?”—and “How many people are in harm’s way?”

This article was published in Scientific American’s former blog network and reflects the views of the author, not necessarily those of Scientific American


Hurricane Dorian wreaked havoc in the Bahamas. Massive fires raged through the Amazon forest. A 7.1-magnitude earthquake and aftershocks rocked Southern California this summer. Kerala, India, suffered the biggest flood in nearly a century.

It is painfully obvious that natural disasters all over the world are inflicting increasing amounts of damage—and it is likely that even more destructive events will occur in the future. But how can we defend and protect ourselves against the inevitable disasters to come?

The answer lies in our ability to better forecast, plan for and respond to natural disasters. New technologies that can analyze massive amounts of data are very promising tools that can help community leaders and emergency managers make more informed decisions. These technologies, developed from the field of machine learning, can supplement and enhance existing disaster response programs.


On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


Machine-learning technologies can help decision makers more accurately answer urgent questions such as: When will the disaster hit? How destructive will it be? What areas will be hit hardest, and how many people live and work in these areas? What buildings will be most vulnerable? Will there be power outages, and if so, where? What equipment and resources will be needed and for how long? How much will the disaster response effort cost? And so forth.

In a nutshell, machine learning lets computers mimic human learning to analyze large amounts of data from past disasters to generate new insights about current and future similar events. The computer is trained to "think," processing information and developing insights that far exceed what a human brain can compute.

There's a wealth of data available on past disasters that machine learning can leverage. In fact, it's already being used to improve disaster response. For example, some utilities are using machine-learning tools developed in my research group, in collaboration with Steven Quiring of Ohio State University, to predict power outages from hurricanes and other severe weather events. The utilities report that machine learning has provided critical information to help them improve decision making.

In another example, a startup has developed a cross-hazard platform using both engineering-based and machine-learning models to provide information to community leaders and emergency managers that enhance both long-term disaster resilience and short-term disaster response. Another nonprofit startup is using data analytics and mapping to connect disaster victims to first responders and volunteer groups.

Moreover, machine-learning technologies do have limitations. They can only process and analyze information that has been input into the computer. For example, if data for an extremely large disaster is not part of the data set, machine-learning technologies likely cannot make accurate predictions for a comparable event in the future. Machine-learning predictions come with uncertainty that can be difficult for decision makers to fully understand.

It is important to emphasize that machine learning in no way replaces human decision making; it only supplements expert judgement and traditional disaster response methods. This is a critical difference from how machine learning is used in other areas, such as self-driving vehicles, where the technology seeks to at least partially replace human decision making. Machine learning cannot and should not replace traditional methods of disaster response. Expert human judgement is absolutely critical, given the complexity and magnitude of the situations.

I know many people are skeptical about machine learning. They worry that the science is unproven and there isn't enough data to predict future events. But these are just myths. Machine learning, when used correctly and based on solid data relevant to future situations, has been proven in many industries. Enormous amounts of data do exist that can be leveraged for different events and situations, even in the realm of natural disasters.

As floods, earthquakes and wildfires inflict increasing amounts of damage in the future, machine learning should be an essential part of disaster response programs. If we don't use it, we are depriving community leaders and emergency managers of an important tool that can improve their decision making in critical times.

Seth Guikema, PhD, is a professor in the Department of Industrial and Operations Engineering and the Department of Civil and Environmental Engineering at the University of Michigan. He is also an adjunct professor in the Department of Safety, Economics and Planning at the University of Stavanger in Norway, and a Data Science Research Fellow at One Concern, Inc.

More by Seth Guikema