UW News

April 4, 2019

April ‘Weather Madness’: UW wins top team, individual prizes in national forecasting contest, now enters tournament round

UW News

group of people against cloudy sky

Faculty lead Lynn McMurdie, center, and some of the students from the winning UW team pose on the roof of the Atmospheric Sciences-Geophysics Building.Dennis Wise/University of Washington

 

The University of Washington has won a national competition in which colleges vie to deliver the most accurate daily forecast for cities across the country. A UW student also developed a machine-learning model that for the first time delivered a more accurate forecast than any human competitor.

In results announced this week, the UW team placed first among 36 teams in the annual WxChallenge operated by the University of Oklahoma. Lynn McMurdie, a research associate professor in atmospheric sciences, has led the UW team since its inception in 2011.

“We’ve been close many times, but this is our first win,” McMurdie said. “We were in first place for the last couple of weeks. It’s very exciting to bring the trophy to Seattle.”

Jonathan Weyn, a doctoral student in atmospheric sciences, claimed the top individual prize. Several other UW contestants — including McMurdie and team captain Joe Zagrodnik —  placed in the top 10. Undergraduate Jamin Rader placed third among juniors and seniors, and undergraduates Mason Friedman and Peter Brechner placed well and contributed to the team’s success.

During the challenge, participants make a forecast for the next day’s weather in a U.S. city. The contest moves to a new location every two weeks: This year’s cities included Omaha, Phoenix and Anchorage.

All students and current employees at higher education institutions in the U.S. and Canada can participate. At the UW, undergraduates can enroll in a one-credit pass/fail class, “The Weather Challenge,” that meets to discuss forecasting techniques.

“Our graduate students don’t do it for credit, they just do it for fun, but they often show up to the weekly meetings just to talk about the weather,” McMurdie said.

three people in front of gray sky

Graduate student Jonathan Weyn, who took the top individual prize; faculty lead Lynn McMurdie, who also placed second in the contest’s faculty/staff category; and team captain Joe Zagrodnik, who is also a first seed going into the tournament round.Dennis Wise/University of Washington

Zagrodnik, who recently earned a doctorate for his study of precipitation over coastal mountain ranges, has been the team’s student leader since joining the UW in 2014. As captain he helps with the team’s recruiting, logistics and coaching. He does it for fun, and enjoys sharing his forecasting knowledge.

“The real challenge in building a competitive team is you have new people every year and you have to quickly coach them up to learn the ropes. There’s always a scramble in September to get the newbies ready. But I enjoy it,” Zagrodnik said.

Teammates are allowed to discuss available forecasts and assess the conditions together, but they cannot directly influence each other’s forecasts. Each Monday through Thursday, contestants submit a forecast by 5 p.m. Seattle time for the next 24 hours of maximum and minimum temperatures, precipitation and sustained winds in the given contest city. Then they wait to see if the weather cooperates.

“They update the weather on the challenge site every hour, so you can watch the next day and see how you’re doing,” McMurdie said.

This year the UW team included more than 30 members. Of those, 14 submitted forecasts from September through March so they could be included in the team score. Team scores are a combination of the individual scores.

In true Seattle style, the UW team credits its win partly to custom software. Zagrodnik and others built a dashboard that brings together more than a dozen forecast models.

“There’s weather data everywhere on the internet,” Zagrodnik said. “We do a lot of work to pull that data into one location so our team has a dashboard they can look at to make a fast decision, without having to hunt all over.”

The dashboard also includes a computer model that uses machine learning to improve cities’ forecasts. The tool, developed by Weyn, compares historical forecasts for a given city with the actual weather to learn the models’ biases in different scenarios, and correct for them.

After the announcement of a new location, Weyn would train the tool with six or seven years of past forecasts and weather records for that city, and then make those results visible to the rest of the team.

Weyn’s machine-learning tool, MOS-X, also entered in the “model” category along with many government models. Over two weeks in Pueblo, Colorado — the most unpredictable of this year’s cities — his model delivered the best forecasts.

It was the first time in the competition’s history that a computer beat all the human forecasters, and the organizers awarded Weyn a special category trophy.

The competition is not yet over. A playoff round, in March Madness bracket style, began April 1. The top 32 finishers from the regular season will sit out the first week.

Sixty-four contestants will then compete in two-day, head-to-head contests to forecast the weather in Columbia, South Carolina, with winners advancing to the next round. Weyn is the top seed for Region 1, and Zagrodnik is the top seed for Region 2. You can follow the UW team members’ progress here through April 26.

Weyn, who will be the student lead for the team next year, said he’s most proud of the team’s first-place trophy.

“The overall team win is a big deal. The tournament is like a playoff season — random things happen,” Weyn said. “But the top standing for the entire regular season is quite an accomplishment.”

###

For more information, or to join next year’s team, contact McMurdie at lynnm@uw.edu or 206-685-9405.

Tag(s):