IBM leverages machine learning for hyper-local weather

New precision forecasts will help businesses in industries ranging from aviation and agriculture to energy generation and retail better respond to their environment.

Changeable weather
Credit: flickr/Nicholas A. Tonelli

It's been just about six months since IBM closed its acquisition of The Weather Company, but it's not resting on its laurels. This week Big Blue moved to leverage The Weather Company's go-to-market strength to launch Deep Thunder, a machine learning-driven weather model developed by IBM Research to help industries ranging from aviation and agriculture to retail better predict the business impact of weather.

"One of the greatest things about being part of IBM is having a relationship with IBM's Research arm," says Mary Glackin, head of Science & Forecast Operations for The Weather Company.

The Weather Company is actually merging its existing Rapid Precision Mesoscale (RPM) model — a numerical weather prediction system based on the Advanced Research Weather Research and Forecast System (WRF-ARW) — with Deep Thunder. RPM generates forecasts up to 24 hours ahead, with updates every three hours in the U.S. and every six hours outside the U.S. Precipitation forecasts are calculated from half-hourly instantaneous precipitation forecasts provided by RPM.

Beyond local forecasts

Deep Thunder brings hyper-local short-term forecasts to the table. Glackin says the combination of Deep Thunder's hyper-local forecasts with The Weather Company's existing global forecast capabilities represents a game-changer.

[ Related: Weather Company forecasts more big data for IBM Watson Analytics ]

"What we're announcing is the merger of these two models," she says. "We think that in this case, one plus one really is going to equal three."

Deep Thunder is tuned for forecasts at a 0.2 to 1.2 mile resolution, allowing businesses to understand the exact weather conditions in their location. For example, in aviation, local inclement weather contributes in a direct and measurable way to congestion at airports. With accurate insight into local weather, airlines can better predict congestion to make more precise decisions about exactly how much fuel to put on any given plane.

"So many of their decisions are in the very near term," Glackin says. "Any improvement we can give them in terms of understanding what the convection will be this afternoon is dollars in their pockets. Knowing what fuel load they should put in the airplane because of what the congestion is going to be at LaGuardia at four in the afternoon goes directly to their bottom line."

Getting hyperlocal down on the farm

Another case in point is precision agriculture. Pesticides and fertilizers are sensitive to environmental factors like rain — some need periods of clear weather while others require rain immediately after application. In addition, Glackin says, farmers now want to apply pesticides and fertilizers at very particular points crop lifecycles to maximize yields. Better, more accurate forecasts allow them to hit those windows.

[ Related: With a torrent of weather data, IBM hopes to know the world ]

Energy, particularly renewables, are another area where precision weather forecasts can translate directly to the bottom line.

Deep Thunder can forecast weather for targeted areas retrospectively, and use machine learning-based weather impact models to help businesses more precisely predict how even modest variations in temperature could affect their business. This ranges from consumer buying behavior to how retailers should manage their supply chains and stock shelves, to how insurance companies analyze the impact of past weather events to assess the validity of insurance claims related to weather damage. Utility companies could mine and model historical data of damage to power lines or telephone poles and then couple that information with hyper-local forecasts to better plan for how many repair crews would be needed and where.

And this is just the beginning, Glackin says. For now, The Weather Company is helping airlines decide how much fuel to put on any particular plane, but the opportunity is much bigger than that.

"Ultimately, this is not just about helping with the decision about how much fuel to put on the airplane, it's about how does an airport manage its overall operations," she says. "That's the kind of scale of things IBM can bring."

Glackin notes that The Weather Company is now aggressively seeking out new data sources to feed its analytics. Already, its models use more than 100 terabytes of third-party data daily, including one of the largest troves of location data available anywhere and Weather Underground's network of more than 195,000 personal weather stations that report in minute-by-minute. But it's also starting to work on ingesting air pressure data gathered from cell phones. Instrumentation data from aircraft can help it analyze wind speed and turbulence data, which can then be deployed to other aircraft.

[ Related: When storms hit, the Weather Company needs the cloud ]

Last week, The Weather Company announced a deal with Gogo Business Aviation, under which Gogo will implement Weather's patented Turbulence Auto PIREP System (TAPS), a turbulence detection algorithm that will provide access to required avionics inputs to calculate and report turbulence intensity and transmit to the ground via Gogo's U.S.-based air-to-ground and global satellite communication network.

"Leveraging Gogo's expanded fleet of aircrafts, The Weather Company can quickly share real-time turbulence data directly with pilots and dispatchers, thereby improving crew and passenger safety," Mark Gildersleeve, president of business solutions at The Weather Company, said in a statement last week. "It is a great example of the internet of things in action, where we are collecting massive amounts of data very quickly and then using that insight to provide guidance to all flights that will be traveling through impacted airspace."

"We're also on the path to taking data from connected vehicles," Glackin adds. "We could take data from windshield wipers on a fleet of cars, or data from thermostats in houses that are wired — that allows us to make inferences about what the outside temperature is. When I think about really moving the dial on accuracy, especially in the short term, it's this proliferation of observations from non-traditional sources that I think is going to really make a difference."

Next watch this:

To comment on this article and other CIO content, visit us on Facebook, LinkedIn or Twitter.
Download the CIO Nov/Dec 2016 Digital Magazine
Notice to our Readers
We're now using social media to take your comments and feedback. Learn more about this here.