How Data and a Good Algorithm Can Help Predict Where Fires Will Start

The New York City Fire Department is using a tool called FireCast to predict which buildings are most likely to have fires

FireCast 2.0 targets the most fire-prone buildings, many of which haven't been inspected in years. © Paul A. Souders/CORBIS

There may be no job more reactive than firefighting. You wait for the alarm to sound; when it does, you go fight fires.

But what if there was an algorithm that could take the guesswork out of fire prevention? What if enough data could be analyzed that fire departments would be able to identify where fires are most likely to happen?

For more than a year now, the New York City Fire Department (FDNY) has been doing just that. Using a data tool called FireCast 2.0, it has been prioritizing which of the hundreds of thousands of buildings in the city are at the highest risk of having a fire. The software applies an algorithm from five city agencies, taking into account as many as 60 different risk factors—not only obvious ones, such as a building’s age, but also whether it was in the middle of foreclosure proceedings or had active tax liens. It’s not a big leap to see why a property’s financial plight could make it a bigger fire risk, but until now, fire departments had no formal way of knowing such things.

It wasn’t that long ago, in fact, that even a fire department as sophisticated as FDNY was keeping track of buildings in card catalogues in local fire houses. Each structure would have its own card with basic information—when it was built, square footage, construction materials—and from that, company commanders were expected to determine which buildings were to be inspected how often. 

Building inspections are a key part of fire prevention in cities like New York, and that, as you might suspect, wasn’t a very efficient way to handle them. Usually, the FDNY struggled to meet its annual goal of inspecting 10 percent of the 330,000 buildings in the city for which it is responsible. It is a massive job when you consider that one of those edifices is the Empire State Building.

But FireCast 2.0 has already simplified that process, allowing the department to more precisely target the most fire-prone buildings, many of which hadn’t been inspected in years. Of course, inspections can’t always prevent fires.  But FDNY officials point out that since FireCast 2.0 was deployed in 2013, more than 16 percent of the city’s fires were in buildings that had been inspected in the past 90 days, suggesting that not only had the right structures been moved to the top of the list, but also when the firemen returned to fight the fires, they had up-to-date information on the layout of the buildings.  

Getting smarter

The FDNY is pleased with the big stride it’s taken into what’s known as “smart firefighting,” but it’s only a first step. Later this year, the department is expected to upgrade to FireCast 3.0, an even more powerful tool that will analyze three years of data from 17 different city agencies for every one of the 330,000 buildings. Each will be given a fire risk score. But that list will be updated daily—if a building receives a trash violation, for instance, its score may rise on the next day’s list. Compiling the data from all those buildings will take only 90 minutes, according to a report from the National Fire Protection Association.

The information processed by FireCast 3.0 will also be much more refined. FireCast 2.0 bunched the whole city into a one big data set. The upgraded tool will instead separately analyze each of the city’s 49 battalion districts, basing fire risk scores on the fire history and characteristics of individual neighborhoods. It will incorporate data every day from the city’s 311 non-emergency phone reporting system. That may not seem that helpful to identifying fire hazards, but more than half of the calls that come in through that system are complaints or reports about buildings.

The idea is to keep a steady stream of fresh data coming in to sharpen the algorithm, with the hope that firefighting can become more of a science. As Ryan Zirngibl, the lead data scientist for FireCast, told the National Fire Protection Association Journal, the goal is to identify as many of the characteristics of buildings that have had fires and compare them to the characteristics of buildings that haven’t.

“What’s the difference between two buildings that look exactly the same, except one building had a fire,” he said.  “What is it we’re not seeing about these buildings?”

Robots at sea

A very different approach to the future of firefighting was unveiled recently by the U.S. Office of Naval Research. It’s a 5’10”, 143-pound robot named SAFFiR, short for Shipboard Autonomous Firefighting Robot, and it was designed by engineers at Virginia Tech to put out fires where they’re most dangerous—at sea.

During a recent test, SAFFiR was able to use its infrared stereovision to find a fire through thick smoke and handle a hose with its hands well enough to extinguish the flames. Maybe more impressively, it displayed its sea legs, able to stay upright on a rolling ship. That, according to SAFFiR’s designers, may have been their biggest challenge.

Shipboard Autonomous Firefighting Robot - SAFFiR

SAFFiR still has a ways to go before its ready to head out to sea. It still struggles to navigate doorways and stairwells. For the test, in fact, its movements were controlled by a human. While it will likely be paired with a human for some time, SAFFiR may eventually be able to move and make decisions on its own. In time, when a fire starts on a ship, it will be the machine, not the human, that faces the flames.

Get the latest stories in your inbox every weekday.