In the aftermath of Hurricane Sandy, President Obama made clear that the debate about climate change was over. As the leader of the Federal Emergency Management Agency (FEMA) at that time, I was part of the senior team tasked with ensuring that the federal government was taking the necessary steps to begin adapting to a future of extreme weather and record-breaking storms.
Despite some meaningful progress, we were handicapped by tools that lacked clarity about what that future would mean in terms of specific impacts on individual communities and their citizens.
Since leaving office, our society has been under constant attack from record-setting disasters that are occurring monthly, if not weekly, and this threat is only expected to grow, alongside the urgency we face to address it.
The terrifying wildfires in Southern California and the massive Sonoma County blaze in Northern California are our most recent life-threatening reminders, not to mention last summer’s 7.1 earthquake north of Los Angeles.
The magnitude of these disasters is testing the critical informational tools available to emergency managers, who rely on traditional disaster assessment and prediction modeling, as well as human experience from past events.
The unprecedented nature of current hazards is challenging these tools and experiences. In addition to the raging wildfires in California, the deadly Typhoon Hagibis in Japan and the devastating impact of Hurricane Dorian in the Bahamas are examples of the growing magnitude of disasters in the 21st century.
At FEMA, I saw first-hand the heroic work of emergency managers, who are in dire need of additional help to meet the enormous challenges of the coming disasters. They need new tools that can visualize events and outcomes that are unimaginable today but could be hard realities tomorrow. Knowing what could happen and understanding future worst-case scenarios will help emergency managers explore how they can better protect their communities.
The good news is that pioneering data scientists and engineers are developing new software tools that augment existing disaster assessment and prediction methods. These tools are based on machine learning technology that enables computers to analyze vast amounts of data to make predictions about future events.
These predictions can show emergency managers where there may be unanticipated weaknesses and gaps in their plans so they can make adjustments. Put another way, machine-learning tools can show emergency managers at what point their response program fails to be effective.
Machine-learning tools could have helped emergency managers with Typhoon Hagibis or Hurricane Dorian response efforts. These tools could have analyzed data from past hurricanes, as well as dynamic data on weather, storm surge, infrastructure, and population, and predicted the scale of impact that these record-breaking storms would have on individual communities and regions.
With more dynamic and precise information, emergency managers would have had actionable information to change outcomes and get help where it is needed most, as quickly as possible.
Machine-learning tools are also faster at predicting damages from future events than other technologies that are currently used. Take the Cascadia Subduction Zone, which is a mega-fault line that stretches from Vancouver, Canada, to Northern California. A significant earthquake here could result in one of the worst disasters the U.S. has ever seen. To prepare for a potential catastrophe, FEMA used what was then state-of-the-art technology to run worst-case scenarios in the subduction zone.
But this system took two years to develop, and the technology used is cumbersome and expensive to run, and scenarios can be run only once a year. In contrast, machine-learning tools could allow local stakeholders to test scenarios at any time. These tools can quickly analyze data to show what could happen at multiple points where an earthquake might originate, allowing for more informed and robust preparation.
Machine learning-based predictions are similar to the National Hurricane Center’s forecasts in that both guide emergency managers and are meant to be used as decision support tools.
However, neither machine learning nor the National Hurricane Center can provide 100 percent accuracy in their predictions and forecasts, and they cannot save lives directly. What they can do is provide valuable information to emergency managers, who can use the information to make different or better decisions in crises.
Emergency managers cannot be expected to adequately protect us from the increasing threat of disasters if they are forced to rely on historical models and experience to do their essential work.
They must be armed with new tools, such as machine learning-based decision support systems, help them prepare for scenarios we can’t even imagine yet. In a changing world, relying on the past for answers is like driving 100 miles per hour while looking through the rearview mirror. It’s not going to end well. We can and must do better for our emergency managers and the safety of our society.
Craig Fugate is the chief emergency management officer at One Concern Inc. He also served as head of FEMA during the Obama administration and director of Florida Division of Emergency Management.