AI to help improve weather forecasting

In recent decades, there has been a lot of progress with weather forecasting based on numerical predictions. According to NOAA, a five-day forecast has an accuracy rating of approximately 90%. For a seven-day outlook, the accurate rate is about 80%. Many weather apps and websites have recently expanded their outlooks to 14-days, or even higher in some cases. Beyond 10 days, the outlooks are about 50% accurate, but certainly better than what it used to be. But, as the technology improves, so will the long-term outlooks.

Forecasting the weather based upon a numerical process was introduced by an English mathematician named Lewis Fry Richardson. He worked relentlessly to create mathematical equations to enhance weather prediction, but his calculations did not work out very well. In his book published in 1922, “Weather Prediction by Numerical Process,” he wanted to highlight this process of forecasting, but there was not enough scientific data about our atmosphere to make accurate forecasts. It wasn’t until 1955, when numerical weather prediction started to gain more traction with the development of programmable computers.

Fast forward to 2023 where modern forecasts are primarily based upon numerical computer models. I do believe that Richardson would be impressed with today’s weather technology. Meteorologists are now able to predict hurricane tracks, severe weather, storm movement and other phenomenon with very good accuracy, especially within a one-to-three-day window. Farther out past seven days, I believe that many of the computer forecast models are doing a good job in catching overall weather trends, but there is still a long way to go. For example, most long-range models did very well in predicting our massive cold waves here in North Idaho in December and in late January.

Some of the major issues of forecasting are the isolated thunderstorms and microclimates. For example, I have noticed on many occasions where one location will receive heavy rain and hail, but several miles away, there was no rain at all. Scenarios like this are very difficult to predict, especially beyond several days.

Data in remote locations has been limited as much of the historical weather data has been from major cities and airport locations. However, thanks to services like Weather Underground and other services, electronic weather stations have been popping up in major metropolitan regions as well as remote locations. As more weather stations are added, this will help with forecasting in areas that have their own microclimates.

According to a recent study published in the Journal of Advances in Modeling Earth Systems, there is new weather forecasting research that is using AI (Artificial Intelligence) to help with extreme weather forecasting as much as six weeks in advance. Forecasting with the current system can take hours to analyze, in addition to large computer resources, as many computer forecast models are accessed to make the best weather prediction as possible. With the new advanced AI computers, models can run over 300 versions of a forecast, collect local, satellite and other critical data, and create a forecast out to six weeks in a matter of minutes. They can also learn and adapt from using observed data to help predict future weather. A one-week forecast could be computed as less than a second with the help of AI.

With the frequency of extreme weather increasing each year, the importance of more precise long-term and seasonal forecasting is much higher. According to NOAA, there were 18-billion-dollar weather and climate disasters in the U.S. in 2022 that were over $1 billion each. Better and more precise long-term predictions could help communities prepare for potential future extreme weather events that could lead to widespread power outages and other infrastructures.

Artificial Intelligence for weather prediction has already seen some limited use. According to an article by TechTarget.com, an AI-powered weather system was used for the farming community in parts of India. The purpose was to optimate farming practices for crops and soil managements. The new technology did help to increase revenues by 37%.

The technology of weather forecasting continues to improve. With better satellite data, increased local observations and faster super-computing, perhaps there will be a day where forecasting will be close to 100% accurate. However, Mother Nature will always have the final say.

In terms of our local weather, after a strong start to the snowfall season, only 7.7 inches has fallen since Jan. 1. Weather patterns are starting to look more favorable for snow across the Inland Northwest around the middle of the month, but if this pattern takes too long to develop, then our chances for the 80-inch-plus snowfall season in Coeur d’Alene will likely go down.

The normal snowfall in our region from Feb. 1 through the end of April is around 19 inches. A normal snowfall for this timeframe is 18.9 inches, which would take our seasonal total for the 2022-23 season to approximately 75 inches. There’s still a long way to go, so we’ll have to see what happens.

• • •

Contact Randy Mann at randy@longrangeweather.com.