Scientists Seek Insights into Outlier Drought Projections

By Andrew Freedman

Although official drought outlooks failed to provide Americans with advanced notice of one of the worst droughts to strike the U.S. since the Dust Bowl-era — a drought that is still ongoing — there were some computer models that got the forecast right. Viewed as outliers at the time by climate forecasters tasked with making seasonal forecasts, such models look downright prescient with hindsight.

In the wake of the flawed forecasts, climate researchers are seeking to understand what enabled certain computer models to anticipate the drought and intense heat that affected much of the U.S. beginning in March, in order to recognize the early warning signs the next time around. Their task is a complex one, since models show varying levels of skill depending on the initial climate conditions and time of year when a forecast is made.

The GFDL's experimental climate forecast for temperatures during June that was initialized on March 1. Actual June temperature anomalies are shown in the bottom left of the image. Click on the image for a larger version.
Credit: Gabe Vecchi/GFDL.

For example, a computer model ensemble developed at NOAA's Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, N.J., correctly anticipated the heat and dryness in projections made as early as January 2012. Later projections in early March, with the model fed with, or “initialized,” with weather and climate data from February, also showed a likelihood of much above-average temperatures and below-average precipitation across the Central U.S. during the spring and summer.

Importantly, the March 1 projections were issued before a rare heat wave gripped the country later that month. The extreme heat event tied or shattered more than 2,000 temperature records, melted snow cover, and dried soils, putting the U.S. on a path toward even drier and hotter conditions in the months to follow.

Gabriel Vecchi, a researcher at GFDL, is working with colleagues at other U.S. and Canadian institutions to develop improved computer models for making skillful seasonal weather forecasts. Known as the “multi-model ensemble for seasonal prediction,” this effort includes agencies such as NASA and NOAA, as well as the National Center for Atmospheric Research.

Vecchi said GFDL’s model correctly anticipated the warmer-than-average temperatures that were prevalent from March through August, as well as the drought, although its depiction of the hardest hit areas did not match up exactly with the areas that turned out to be most severely affected. The U.S. experienced its third-warmest summer on record, while July was the warmest month of any month on record. The drought is still underway, with about 65 percent of the contiguous U.S. suffering from at least moderate drought as of September 18.

The GFDL's experimental climate forecast for temperatures during Julythat was initialized on March 1. Actual July temperature anomalies are shown in the bottom left of the image. Click on the image for a larger version.
Credit: Gabe Vecchi/GFDL.

“Our model was convinced that this year there should have been a heat wave,” Vecchi said.

Within the interagency modeling group, GFDL’s projections were an outlier result, and therefore they were discounted when it came time to make the official climate outlooks. This was the right thing to do at the time, Vecchi said, since forecasters had no reason to believe that the GFDL model was going to outperform other models.

“It’s really easy looking backwards to say what we should have done,” he said. “Even if we find that our forecast was, in fact, right for the right reasons — and not luck/chance — we need to work to understand how to use that information going forward. How do we distinguish when an outlier forecast is a 'Galileo' from that unfortunate fellow with the tinfoil hat?”

The GFDL forecast illustrates a dilemma that climate forecasters face, since models show varying levels of skill depending on the conditions at the time they are run. When an El Niño event is underway, for example, one model might become more reliable than another, just because of the ways it incorporates ocean temperatures into its complex series of calculations.

According to Randall Dole, the deputy director of research at NOAA’s Earth Systems Research Lab in Boulder, Colo., one factor that complicated the drought forecast for the summer is that models typically have less skill in predicting summer conditions compared to other seasons. But under certain conditions, he said, some models may be able to produce more skillful forecasts even at times when general forecast skill is thought to be low. In an email message, he called this “a kind of conditional probability that in retrospect may well have been the case for summer 2012.”

Dole cited a European climate forecasting system, known as EUROSIP, that also showed some skill in predicting the drought and heat.

Vecchi is working on efforts to conduct a drought and heat wave post-mortem to try and identify the predictability of the unusual conditions so far in 2012, a process that he hopes will identify “skillfull outliers” among the many computer models that seasonal climate forecasters use.

Related Content
Lack of Warning on Drought Reflects Forecasting Flaws
Global Warming May Have Fueled March Heat Wave Odds
Ongoing Heat Wave Rivals Events of Dust Bowl
Coverage of Summer 2012 Heat Waves