MAXWELL AIR FORCE BASE, Ala. -- U.S. Air Force Capt. P. Trent Vonich, Secretary of the Air Force PhD Fellow studying weather prediction, recently published a paper with his advisor Dr. Greg Hakim showing how using AI to adjust initial atmospheric data allows a nearly perfect prediction of a 10-day weather forecast.
Every year, the Department of the Air and Space Force selects high-performing officers and civilians to immerse with distinguished civilian institutions to earn post-graduate degrees. The Secretary of the Air Force PhD is the Technologist pillar of the Air Force’s Strategic PhD programs, alongside the Chief of Staff of the Air Force PhDs which focus on Social Sciences and History. Both are designed to explore broad-ranging topics seeking high-impact, operationally meaningful outcomes.
“For the last, 60-70 years we have used physics models to produce your 10-day forecast,” said Vonich. “In the last three years, AI weather models have come online and they work differently. They may understand physics in a sense, but nobody is putting explicit physical equations into them. It is just trained on weather data from the past. That allows them to be a lot faster. For reference, it is now possible to generate a 10-day forecast on your home computer for the whole world with an AI weather model in about 20 minutes, but with the leading physics models, it would take you an hour on one of the most powerful supercomputers in the world.”
Because of how quickly and efficiently they run, Vonich decided to use an AI model to analyze previous forecasts and let it figure out what initial conditions could have been adjusted to give a more accurate forecast.
“When you initialize a weather model, you give it a starting point. Let’s say you plug in 80 degrees as your starting point, but in reality, it’s not really 80 degrees, it’s 80.1, and that tenth of a degree though small, ten days later can make a really big difference,” Vonich explained. “That is where this whole idea of improving your starting point comes in, and the AI models can potentially show ‘Hey, you should have increased your temperature by a tenth of a degree here,’ and then you expand that to the whole globe.”
Vonich’s AI model was able to review the forecast of the 2021 Pacific Northwest heatwave, adjust the initial atmospheric data hundreds of times according to how the heatwave played out, and reduced the initial forecasting errors by 90%.
Although he is currently a PhD candidate at the University of Washington, Vonich is also an active-duty combat rescue officer in the U.S. Air Force. Like many field grade officers, he paused his typical active-duty service commitment to earn a PhD through the Dr. Heather Wilson STEM PhD fellowship program.
“I am a year into it right now, and you are given a maximum of three years to get it done,” Vonich explained. “It’s important to know that officers from any AFSC can compete for a slot in this PhD program. Ultimately, the goal is to get you in a position within your career field or perhaps working on a key defense problem set that draws on your scientific background.”
If you are an active-duty U.S. Air and Space Force Captain who would like to pursue a STEM PhD, you can learn more about the annual application process for the Dr. Heather Wilson STEM PhD Program here: https://myfss.us.af.mil/USAFCommunity/s/knowledge-detail?pid=kA08300000001sDCAQ