Google's AI Weather Forecasting Model Might Improve Forecasts

Old fart that I am, I tend toward skepticism with AI, or to spell it out, Artificial Intelligence.

Google's GraphCast AI weather forecasting model
performs better than current weather models,
according to studies. 

I know that's where we're going and there's certain benefits to the technology, but don't get me started on the downsides. 

However, if I must focus on the positive,  Google has developed some AI technology consisting of a weather forecasting model that experts say could vastly improve predictions. 

For years now,  it seems like every time there's a storm, you hear about the American model and the European model disagreeing on the track and intensity of expected bad weather. Both are computer models of course, with the European model arguably the better of the two.

According to studies, the AI weather forecasting model from Google is a big improvement over these old school models. I guess all roads lead to Google these days. 

According to the Washington Post: 

"Google DeepMind's AI model, named "GraphCast," was trained on nearly 40 years of historical data and can make a 10-day forecast at six-hour intervals for locations spread around the globe in less than a minute on a computer the size of a small box. 

It takes a traditional model an hour or more on a supercomputer the size of a school bus to accomplish the same feat. GraphCast as more accurate than the European model on more than 90 percent of the weather variables evaluated."

GraphCast is actually one of several AI weather forecasting models under development. Government agencies world wide are looking into AI models because they are fast, efficient and can save money. 

As the Washington Post explains, current models like the American and European make forecasts based on really complex mathematical equations that would make anybody's head spin. These models are expensive because they require tons and TONS of computing power.

The AI models work differently. They suck up pretty much all the historical weather data available, then generate forecasts by looking at current conditions and figuring out what will happen next based on that historical data.

In a sense, AI weather forecasting models are old school.  In addition to relying on computer models, many human meteorologists often compare historical weather events to what's going on now to help them predict what the weather will be like tomorrow. 

One question about AI and weather forecasting is its ability to forecast extreme events.  There are not many super wild weather events from the past from which AI can "learn" about and then project into the future.  

This is especially true in our age of climate change, considering that we're seeing more and more record storms, record temperatures and weird storm behavior. 

However, GraphCast performed well in these event. Says the Washington Post: 

"GraphCast reduced cyclone forecast track errors by around 10 to 15 miles at lead time of two to four days, improved forecasts of water vapor associated with atmospheric rivers by 10 to 25 percent, and provided more precise forecasts of extreme heat and cold five to 10 days ahead of time."

However, questions remain as to whether GraphCast and other AI models can accurately forecast small scale weather trouble, like severe thunderstorms, tornadoes and localized flash floods. We also don't know how well AI does with larger storms that produce heavy precipitation in one area, but light rain or snowfall just miles away. 

That's particularly important here in Vermont. The most dangerous weather in the Green Mountain State include flash floods roaring out of the mountains;  downslope winds during winter storms that can be destructive in one town and no big deal the next; and nor'easters that can dump a couple feet of snow in say, Ludlow, but just a few inches up the road in Rutland.  

The Washington Post also cautions that AI isn't going to replace traditional computer models, or human meteorologists for that matter. At least not right away. Traditional models offer the initial conditions that AI uses to project the future. Also, the old computer models are still necessary to "teach" AI how to interpret historical weather. 

Also, it'll take real-life meteorologists to transmit what AI is telling us what will happen into the Plain English we need to know whether to take an umbrella with us tomorrow.  

 Some of the headlines are already overblown on the idea of AI "replacing" meteorologists. The headline about GraphCast Tuesday in Gizmodo was "Google's New AI Weatherman Will Leave Forecasters In The Dust."

Read the actual Gizmodo article and you discover, not quite. As that article states, "The study authors say their work is meant to work alongside the standard systems meteorologists rely on."

So AI isn't going to chase away your favorite meteorologists like Al Roker, Jim Cantore, Ginger Zee or local Vermont favorites like Gary Sadowski, Tyler Jankowski or Mark Breen.   

If it all works out, AI will be another tool these meteorologists use to give us their most accurate forecasts.  I'd hate to see our forecasts just coming from some sterile machine.  I need the human touch when I'm being told how much snow I'm going to have to shovel from my driveway. 

I just hope the bean counters at major media outlets don't think they can fully replace humans. Because they can't.