(Inside Science) — Today, if you're wondering whether rain will wash out your upcoming weekend camping trip, you can whip out your smartphone and check a 10- or even 15-day forecast. And if you're hoping for a white winter, you can turn to a seasonal outlook to judge the chances for extra snowfall. But what if you want to know if the average weather will be out-of-the-ordinary in the second half of the coming month?
Up until just recently you'd be more or less out of luck. The period between when the day-by-day weather forecasts end and the seasonal forecasts begin used to be considered a "predictability desert," said Andrew Robertson, a senior research scientist at the International Research Institute for Climate and Society at Columbia University in New York.
But now scientists are cultivating growth in this formerly barren terrain. A new paper in the Journal of Climate demonstrates that a model that meteorologists primarily use to make three-month seasonal forecasts can reveal temperature and precipitation trends in North America three to four weeks in advance.
Weather forecasts get less accurate the further out in time they go. But even after they lose the ability to predict the arrival time of a storm, for example, they still contain information about the average temperature and precipitation. To extract this information, seasonal forecasts lump long stretches of days together to make predictions.
The so-called week 3-4 forecasts look similar to seasonal forecasts — they predict if broad areas will have higher or lower than normal temperature and precipitation, averaged over a two-week period starting 15 days in the future. But researchers weren't sure if the seasonal model would work at this range, because two weeks is significantly shorter than the typical three-month averaging period for seasonal forecasts.
"One of the reasons we can predict averages, like three-month averages, is that when you compute a three-month average you are smoothing out a lot of weather variability that you know you can't predict," said Timothy DelSole, a climate scientist at George Mason University in Fairfax, Virginia and first author on the new study. "But when you come to two weeks, it's not so clear that that's long enough to average out all of the effects of weather."
To test if the week 3-4 forecasts would hold up to scrutiny, DelSole and his colleagues looked at past seasonal forecasts generated between 1999 and 2010 by the Climate Prediction Center, a part of the National Weather Service. The models automatically generate data for the 15-28 day period as part of a forecast running many months into the future.
Get the mach newsletter.
The researchers compared the model's week 3-4 predictions for January and July to the actual temperature and precipitation measured during those months. Their analysis showed that the forecasts did indeed have "skill," a technical measurement of how well the forecast matches the actual weather. This is one of the first demonstrations of forecasting skill in the three- to four-week range.
"This paper is exciting to me," said Robertson, who was not involved in the study. He serves as the co-chair of an international effort called the Subseasonal-to-Seasonal Prediction Project, or S2S, which is working to improve forecast skill in this challenging range. "There are not many papers that have looked at this before, so it's something new," he said.
DelSole and his colleagues found that winter trends were more predictable than summer trends, and temperature was more predictable than precipitation. For example, 59 percent of the land area in North America showed forecast skill for January temperature, but only 9 percent showed skill for July precipitation. For those land areas that do show skill, the model can give a forecaster at best a slight edge in predicting above or below normal trends, equivalent to predicting a coin flip correctly 67 percent of the time, DelSole said.
"There are not many papers that have looked at this before, so it's something new."
One reason why the week 3-4 range is tricky is scientists are less sure which atmospheric, oceanic and terrestrial phenomena leave an imprint on the weather at this time scale — and whether the current weather models can capture the effects of those phenomena accurately enough to make useful predictions.
DelSole and his colleagues tested whether the week 3-4 forecast skill they found could be linked to two longer-term weather phenomena: the El Nino-Southern Oscillation, which involves fluctuating water temperatures in the Pacific Ocean and is often tied to seasonal forecast variation, and the Madden-Julian Oscillation, which is a regular buildup of clouds, wind and rain that moves eastward above the Indian and Pacific Oceans, at tropical latitudes.
The researchers found that the most predictable components of the winter forecast are related to El Nino. They also linked the Madden-Julian Oscillation to other predictable components of winter precipitation.
DelSole emphasized that there could be additional phenomena that help explain the forecasts' skill. For example, moisture absorbed by the soil after a rain could change how the air and ground exchange heat, and could leave an imprint on the weather three or four weeks out.
The researchers' efforts are part of a broader push by climate and weather scientists to advance forecasts in the three-to-four-week range. In fiscal year 2016, the National Oceanic and Atmospheric Administration launched an initiative to improve predictions on this timescale. On April 18, President Donald Trump signed the Weather Research and Forecasting Innovation Act of 2017, which supports efforts to improve subseasonal forecasts.
Forecasts in this range could be used by dam operators deciding when to release water, utility providers considering future power needs and farmers planning when to harvest their crops.
Bob Rose is a meteorologist for the Lower Colorado River Authority, a public utility that manages the lower Colorado River in Texas, among other services. He said knowing up to three weeks out about potential shifts in weather patterns could help the agency, for example by pushing them to complete maintenance on power units before a big heat wave or cold snap.
Rose said he currently looks at the experimental week 3-4 outlooks and hopes the Climate Prediction Center further develops the product, for example by releasing the forecasts more often and by assigning probabilities to above and below normal predictions.
“I do think there is value looking out to three weeks,” he said. “We have our hands in a lot of different things, and all of it has to do with what’s going on in the weather, both in the short term and in the long term.”