How meteorologists are saving lives with smarter forecasting
Weather researchers decipher mother nature’s next moves with artificial intelligence
With the annual number of natural disasters steadily growing, weather forecasts are becoming more vital. Lives are at stake and businesses can be severely impacted by unpredictable storm patterns. Forward thinking meteorologists now use big data, complex algorithms and “extreme learning machines” to scope out much more detailed information than if tomorrow is going to be a rainy day.
Such cutting-edge information is extremely useful in everyday life. For example, advanced weather modeling helps electricity companies prepare for outages and governments notify citizens of life-threatening conditions in hopes of mitigating potential damage.
At the University of British Columbia (UBC), the Weather Forecast Research Team regularly tests hypothetical models against observed weather to improve predictions. While satellites, beacons and sensors have been capturing data for decades, advances in “artificial neural networks” and machine learning make it possible to compare many more stats.
Most meteorologists combine data collected by governments and then use formulas to correct for unknown factors. How complex are these formulas? The amount of computing power traditionally required to sift through the numbers costs hundreds of thousands of dollars, explains Tim Chui, a doctoral candidate at UBC.
Many of the early weather forecasting formulas were developed in the midwestern United States, where flat terrain leads to relatively predictable weather patterns. British Columbia, on the other hand, has a lot of mountains and microclimates where weather patterns can come and go in a matter of hours. Dr. Greg West, a research associate at UBC, says that prediction methods supported with machine learning are 10 to 20 percent more accurate than traditional models.
The improved reliability also influences your local weather report. Dan DePodwin, Director of Core Weather Content at AccuWeather, explains that artificial intelligence and supercomputers have had a particularly big impact on meteorology in the past decade.
Twenty years ago, forecasters could predict about three days of weather in advance with a good level of accuracy, he says. Various scientific and technological improvements have helped extend that timespan to five days and, in some cases, a week.
That progress also helps warn people about natural disasters and save countless lives. “Now you have decision makers who are calling evacuations three or four days in advance because of the confidence of a forecast,” says DePodwin. “You would not have seen that two decades ago. You'd be waiting until a day or two days ahead of time and you still might not be confident.”
In the case of Hurricane Florence, projections like these helped to evacuate the right areas days in advance.
The UBC Weather Forecast Research Team works with local power company BC Hydro to help to mitigate damage from potential disasters, but also to optimize the electricity generated from wind and hydro systems.
The west coast of British Columbia is subject to wind storms, which are similar to other high-velocity squalls without the rainfall. These intense gusts often leave thousands of people without electricity. The nearly 800,000 residents of Vancouver Island, which is just off the coast of Canada, can be particularly hard hit by these blasts.
“During a windstorm, power crews can’t get over to the island because all of the ferries get shut down,” explains West. “If the power company has a windstorm forecast one day in advance, they’re able to send crews over to the ground ahead of time so they can restore the power faster.”
The same is true for other companies that have important infrastructure to protect and support. The more advanced and accurate the predictions, the more proportional and accurate the preparation for response can be in events like Hurricane Florence or Hurricane Irma in the United States.
In the future, detailed weather forecasts driven by artificial intelligence will not only be more common, but also cheaper. The UBC team has been testing how to decrease the cost of in-depth modeling using cloud computing in place of the multi-server clusters utilized by universities and other large organizations. The approach adjusts the upfront costs of processing these algorithms from six figures to four.
“This cost reduction is extremely important,” explains West. “It makes modeling accessible for smaller research teams, companies with smaller budgets and countries with smaller budgets. All of a sudden it makes research and real time forecasting possible to a lot of groups that wouldn’t otherwise have been able to do it.”
For additional resources, please visit the following: