One of the country's leading commercial forecasters, AccuWeather, said earlier this month that it could predict the weather conditions and temperature three months ahead of time. It's a bold claim that has been refuted by some meteorologists who say such a 90-day forecast will only be as good as historical averages and not much use to someone planning a hike or outdoor wedding this summer.
That's because the Earth's atmosphere is a chaotic system that doesn't follow an easily predictable path, according to Keith Seitter, executive director of the American Meteorological Society in Boston.
"If anybody kept track about how (AccuWeather) did, they would find it's a pretty horrible forecast," Seitter said.
The best weather forecasters can do now is seven to 10 days. After that, accuracy drops off quickly.
The good news is that forecasting has gotten better over the years. Improvements in computer technology, data collection and weather models have improved this forecasting number about one day each decade.
One of the biggest advancements has come in boosting computer power. The National Oceanic and Atmospheric Administration, the parent agency of the National Weather Service, operates supercomputers in Reston, Va., ("Luna") and Orlando, Fla., ("Surge") to come up with weather forecasts.
After a $44 million upgrade in January, each one has the capacity of 2.89 petaflops, or 2.89 quadrillion calculations per second, according to Richard Michaud, director of NOAA's office of central processing. That's up from 778 teraflops (1 petaflop equals 1,000 teraflops) of computing power last year.
In simpler terms, it means that the faster, bigger computers will allow the agency's scientists to:
Better predict the amount, timing and type of precipitation in both winter storms and thunderstorms Create "water forecasts" and more accurately predict drought and floods Connect the air, ocean and waves to track eight hurricanes at once These supercomputers are the brains behind the weather forecasts you see on TV each night or your smartphone when you wake up.
"We collect trillions of observations per day, then we create a state of the atmosphere," Michaud said. "Then we run weather models to forecast into the future."
But Michaud cautions that the upgraded supercomputers still won't be able to push out an accurate local forecast months ahead.
Forecasts are getting better, just not longer.
"Its not just about adding more computing time," he said. "You have to have the appropriate way to assimilate the data into the models."
Building better models is the work of computer scientists, climate scientists and meteorologists.
In this instance, the European forecast model (which includes forecasts for both the United States and Europe) has a slight advantage over the U.S. weather model. That's because the European weather agency has fewer weather-related forecasts to produce, according to Seitter, so they can focus more effort on one product.
"Every time the U.S. model gets better and is running close to the European model, they get an infusion of computing power," Seitter said. "We have always been lagging a tiny bit."
In the future, NOAA's Michaud says that advances like quantum computing and new kinds of processors could make a difference extending long-range forecasting. But there will always be limits that can't be overcome, said AMS's Seitter.
"The atmosphere itself doesn't know what it's going to do a month from now," Seitter said. "It does have predictability in a probabilistic sense. You can say is it wetter or dryer than normal in the next three months. What you can't do is say today that it is going to rain in 42 days and then be sunny after that."