In April 2018, a major storm hit Ontario, bringing torrential rain, an inch of ice and wind gusts up to 60 miles an hour. More than half a million people lost power. Within four days, Hydro One -- Ontario's largest distribution utility -- restored power to its customers' homes and businesses. By contrast, after a major storm in 2016, it took six days to restore power.
Give artificial intelligence some of the credit. Hydro One used an electrical-outage prediction tool developed by International Business Machines Corp. that combines AI technology and the resources of IBM's Weather Co. subsidiary. The tool helped predict the severity of the storm and the locations that would be hardest hit, so Hydro One knew where to position 1,400 front-line staff who were needed to restore power and to handle the nearly 130,000 customer calls that came in during the outage.
Improving accuracy
IBM's outage-prediction tool is also being used, with 70% accuracy, by other cities throughout North America to predict power outages as far in advance as 72 hours before storms are expected. The outage-prediction tool gets more accurate in the final hours before a storm as it incorporates real-time updates. "During severe weather events, every hour of advance notice counts and helps minimize the impacts," says Mary Glackin, head of weather business solutions at IBM.
Many companies and universities are developing AI tools to help cities better predict and prepare for weather events and natural disasters. In the U.S., weather and climate events caused $1.5 trillion in damage from 1980 to 2017, according to the National Oceanic and Atmospheric Administration. When cities can predict more accurately the severity of weather, natural disasters and which areas will be affected most, they can better allocate resources to prepare for relief efforts such as restoring power or evacuating residents at risk.
Weather forecasting has improved dramatically since the early 2000s, thanks to the dramatic increase in cheap sensors that track weather data and the increased capacity of computers to process the plethora of data from sensors, radar, satellite and other sources. New AI systems can comb through years of historical data from storms, hurricanes and earthquakes to detect patterns and better predict new events and their impact.
Forecasting becomes even more accurate -- and more useful to teams that respond as natural disasters unfold -- when it can incorporate real-time information. A startup in Palo Alto, Calif., One Concern, has developed an AI tool used by some emergency-response centers and other government agencies in California to plan for the impact of earthquakes on a block-by-block basis. Emergency-response centers in Los Angeles, San Francisco and Woodside all confirmed they use the tool.
The company's founder, Ahmad Wani, says he started to think about developing such a tool after a severe flood left him stranded on the roof of his home in Kashmir for a week. The rescue effort was so slow and poorly organized, "I realized there was a lack of science in disaster response," Mr. Wani says. Similarly, Mr. Wani saw a basic flaw in relying on 911 calls after an earthquake struck in Napa, Calif., near where he now lives. Many of the people in the worst-hit areas had lost cell coverage and weren't able to call 911, he says. These experiences, Mr. Wani says, led him to work on a tool for getting help quickly to where it is needed most urgently.
First 15 minutes
Now, within the first 15 minutes of an earthquake, the One Concern platform makes calculations that try to predict how bad the damage is in specific blocks of a city. The AI tool gets its real-time information on the strength of the earthquake and its location from sensors and damage reports. The system also is trained on data from hundreds of past earthquakes throughout the U.S., and each new quake gives it more data to help it become even more accurate in its predictions.
One Concern's platform is meant as a tool for emergency responders who ultimately decide how to best respond in light of limited resources and competing priorities, says Mr. Wani. The tool is hosted in the cloud so disaster-response teams can access it with any device, whether a desktop projecting it onto a large screen during a training session, or on smartphones if crews are out in the field during a natural disaster.
'Vulnerable people'
"We make it easier to understand where vulnerable people are likely to have been adversely affected, so rescuers can know who is likely to need saving and what unique conditions they might face," Mr. Wani says.
Michael Dayton, a deputy director in San Francisco's Department of Emergency Management, says his department has used the tool for two years now and says it has improved their "ability to respond quickly and understand where precious resources are needed most."
This summer, One Concern plans to roll out a flood-damage prediction tool that attempts to show how much water will accumulate and where it will flow up to five days in advance of potential storms.
Researchers at universities throughout the world, meanwhile, are developing other AI systems that can better predict dangerous weather, including tornadoes, hail, lightning and severe wind. Such projects leverage the increased number of sensors available, including those on airplanes that provide turbulence data, which can provide helpful hyperlocal information.
There are experiments using social media and crowdsourced data to supplement data from sensors as well. In a 2018 paper published in the journal Computers and Geosciences, a team at the University of Dundee in Scotland reported developing a flooding monitoring tool using information from Twitter and the crowdsourced app MyCoast, which asks people to submit flood photos.
Using hundreds of photos from regular citizens, the AI system could potentially determine flooding in a given area much more quickly than relying on human staff to monitor all at-risk areas at once. The researchers say their AI system was 70% accurate in recognizing flooding.
Some roadblocks still exist before wide-scale adoption of AI tools to predict and prepare for weather events and natural disasters will occur, says Seth Cutler, an environment and water program manager at research firm Frost & Sullivan. For one, greater standardization is needed between technologies and platforms so that sensors and systems can communicate effectively with one another, says Mr. Cutler.
Comment list ( 0 )
