News at Mason
Breaking new ground in extended weather predictions
July 7, 2017 / by John Hollis
Figuring out the weather in store for an upcoming weekend camping trip is easy these days with three- and five-day forecasts so readily available. Planning for a frigid or mild winter is possible to some extent now, too.
But the intermediary period of several weeks from the current date has previously been anybody’s guess.
Until now. A team headed by George Mason University’s Timothy Delsole has shown that prediction models currently used in operational forecasting can skillfully predict weather outcomes three to four weeks in advance. Working with Delsole were Mason’s Laurie Trenary and Kathleen Pegion, as well as Columbia University’s Michael K. Tippett.
In a paper recently published in the American Meteorological Society’s Journal of Climate, Delsole unveiled a sophisticated statistical analysis that found the National Oceanic and Atmospheric Administration’s Climate Forecast System, which is currently used for seasonal forecasts, can also skillfully predict precipitation trends three to four weeks in advance for most of the United States. It is believed to be the first scientific look at forecasts of this particular type.
“There’s been a well-known gap in forecasting,” said Delsole, a professor in Mason’s Department of Atmospheric, Oceanic and Earth Sciences and a senior research scientist at the Center for Ocean-Land-Atmosphere Studies. “Predicting a two-week average—less than a month—has been a challenge that’s just been out there for a long time.”
Weather forecasts typically become less accurate over time, but often contain useful information about the average temperature and precipitation, Delsole said.
They look similar to seasonal forecasts, but the predictions of three to four weeks out have long proven tricky to navigate as scientists are less certain about the effects of atmospheric, oceanic and terrestrial phenomena on the weather during this time frame. Researchers were unclear whether the seasonal model would work with a significantly shorter time period.
“The goal of the study was to take the statistical tools and filter out enough noise to see what’s predictable,” Delsole said.
Delsole insisted that it wasn’t enough to make just one forecast and come close, because those results could easily be attributed to chance. Rigorous statistical analysis and repeated tests proved winter trends more predictable than summer ones, and that temperature was more predictable than precipitation.
Delsole and his team found that 59 percent of the land area in North America showed forecasts matched the actual weather better than random chance for January temperature, but that number dipped to just nine percent for July precipitation. Delsole likened the improved prognostication to being able to correctly predict a coin flip 67 percent of the time.
Their findings could be significant as reliable extended weather forecasts “would have significant social and economic value because many management decisions in agriculture, food security, water resources, and disaster risk are made on this time scale,” Delsole said in the paper.
Utility companies and their customers would immediately benefit from the improved prediction of the weather, Delsole said. Cognizant of impending bad weather several weeks out, power companies would be better able to anticipate their needs in advance, gather the necessary resources at a considerably lower price to consumers and allocate them more efficiently.
“It could save a lot of money,” Delsole said.