Comments on: Machine learning shines in weather-forecasting study – Innovate Long Island

  • Lauren
  • June 1, 2021
  • Comments Off on Comments on: Machine learning shines in weather-forecasting study – Innovate Long Island

Jokes about the weatherman’s predictive powers will be harder to land, if artificial intelligence takes over the long-range forecasting.
That’s according to a new scientific study published in May by Nature Communications, an open-access journal sharing high-quality research from across the natural sciences. The study – led be Hyemi Kim, an associate professor in Stony Brook University’s School of Marine and Atmospheric Sciences – says highly accurate weather forecasts extending beyond two weeks are possible with the use of machine-learning technologies.
Machine learning, technically a branch of AI, involves computer algorithms that improve automatically through experience and the addition of new data. Plug it into weather-forecasting models, according to the SoMAS scientist (and collaborators at Seoul National University and Chonnam National University, both in South Korea), and spot-on weather forecasting – beyond the current limitations of a week or two – becomes possible.

From the weather desk: Hyemi Kim, looking ahead.

There’s more at stake here than planning a backyard barbecue or a trip to the beach. Accurate weather forecasting has immense socioeconomic value, affecting international trade, travel industries, military maneuvers and policymakers across government and commerce.
And as extreme weather events proliferate, so do floods, windstorms, heat waves and other potentially devastating side effects – making reliable forecasts in the subseasonal range, defined roughly as two weeks to two months out, even more important.
Enter Kim and her colleagues, who focused their study – funded in part by a $749,232 National Science Foundation grant awarded in 2017 – on the Madden-Julian Oscillation, a thunderstorm belt that “pulses” east, from the Indian Ocean toward the Pacific Ocean, every 30 to 60 days.
Scientists have long used the MJO as a key tool for three-to-four-week weather forecasting, but computer models have proven somewhat unreliable – current data cannot completely simulate the MJO, and long-range predictions based on MJO information are historically hit-or-miss.

Storm front: The MJO, a key tool for long-range weather forecasts around the globe.

By combining state-of-the-art weather-forecasting models with deep-learning protocols (a machine-learning subset where machines learn on their own, basically unsupervised), Kim and friends reduced errors in four-week prediction models by nearly 90 percent.
“Our study demonstrates that machine learning substantially reduces the MJO forecast errors from models,” the professor noted. “This will help improve global extended-range forecasts.”
For their next trick, the researchers hope to improve predictions of extreme weather events, including hurricanes that could potentially strike Long Island and other coastal regions.
But they may have already improved the accuracy of the seven-day forecast on your local news program – meaning you can plan next week’s barbecue with more confidence.
“We created a simple approach with machine learning,” Kim said. “This method can be implemented into operational forecasts that are currently used for two-week weather forecasting and longer.”