“Climate change is ‘out of control’ according to the UN secretary-general. Last week, the globe experienced the hottest weather on record, and many environmentalists expect this record to be broken several times in the near future.
Heat waves and extreme weather events are predicted to become more commonplace in many parts of the world, making it more critical to produce accurate weather forecasts.
Introducing AI into forecasting models will help with this.
The Push is On
Google & DeepMind (GraphCast), Nvidia (FourcastNet), and Huawei (Pangu-Weather) have published papers in recent weeks that introduce machine-learning methods which predict the weather as well as conventional methods but more quickly.
This AI is compelling meteorologists to “reconsider how we use machine learning” for weather forecasting,” says Peter Dueben, head of Earth system modeling at the European Center for Medium-Range Weather Forecasts (ECMWF). ECMWF’s weather forecasting is the gold standard for 15-day term weather forecasting, yet in test studies, AI weather predictions have matched and sometimes exceeded the accuracy of ECMWF’s models.
Timing is Everything
The biggest advantage to AI weather modeling is timing. Traditionally, forecasting models have used complex computer algorithms that take hours to run. AI can run similar models in a matter of seconds. Once fully incorporated into weather forecasting, these time differences may save thousands of lives.
Though much faster, AI weather models are based on historical data and work best on predicting the patterns of similar weather events. AI, or Machine learning, is “essentially doing a glorified version of pattern recognition,” says Jonathan Weyn, a researcher at the University of Washington. “It sees a typical pattern, recognizes how it usually evolves, and decides what to do based on the examples it has seen in the past 40 years of data.” It is also much more efficient, using 7,000 times less computing power to create forecasts.
The speed at which AI can create models allows forecasting centers to quickly run many models with slight variations, which is called “ensemble forecasting.” This creates a broader range of possible outcomes, such as where a hurricane might strike. Another speed advantage is if a storm makes an unpredictable change. AI can quickly calculate a new pattern that will give residents in the path more time to take measures to evacuate or hunker down.
An example of where AI might have helped was last year’s Hurricane Ian. Traditional models had the storm striking higher up the gulf coast of Florida. By the time they were able to see that the storm had shifted (using traditional methods), the residents in the southern part of the state were left scrambling to prepare. AI modeling could have been used to calculate the sudden shift and its impact more quickly.
It’s because current weather patterns are unpredictable and change more frequently, however, using AI in partnership with traditional methods may be the better choice than relying on AI solely. Over time, AI will have the historical data that will make its predictions even more reliable, but until we have more than 40 years of data on which to rely, experience and the speed of AI will produce more solid results.
Economics Will Influence Adoption Rates
Global economics is impacted by weather, particularly with the rise of renewable energy. Retailers and tech companies are even impacted based on search queries for things like generators, gas prices, and even ice cream.
Many businesses will benefit from faster weather forecasting. Transportation businesses, in particular, will be able to remove a ship or plane from dangerous weather, for example, while manufacturing and construction businesses can streamline productivity, manage product distribution, so it is where it is most needed, reduce insurance needs, and more effectively manage personnel assignments.
The pressure to improve weather predictions in an increasingly volatile environment will come from all across the board so that businesses, people, crops, infrastructure, and property are better protected.
Sources: MIT Technology Review, World Economic Forum, SmartAsset.com