A decade of multimodel seasonal climate forecasts
One of IRI's first net assessment forecasts
October marks the tenth anniversary of IRI's monthly net assessment forecasts, which are a series of freely available maps that show temperature and rainfall predictions for every region of the world. The IRI Forecast group looks back on the evolution of this popular service, and shares a few of its favorite anecdotes from the past ten years. But first, a bit about what the IRI forecasts are all about. On the third Thursday of each month,IRI releases a series of regional and global maps that show the probability of any particular area having above-normal, near-normal or below-normal temperatures and rainfall in a three-month period. The forecasts extend six months into the future. In October, for example, users can view the forecasts for the November-December-January, December-January-February, January-February-March, and February-March-April periods. [click here to see the current suite of forecasts]. Of course, the further away from the current month one gets, the less confident the forecast. Still, it's important to go out six months because many of IRI's forecast users need a long lead time to take action, says Tony Barnston, who runs theforecast operations. "During La Nina, for example, there is a greater likelihood of above average rainfall in Botswana, which could pose an increased threat to epidemic malaria. Several months of lead time provide a needed margin of safety in planning mitigative action such as spraying or immunization," he says. Moreover, there are some situations where forecasts can still have fairly good confidence at longer lead times. "For example, an El Nino-Southern Oscillation episode that has developed by August is very likely to still be occurring in December and January," says Barnston. The IRI produces its net assessments by using current and forecast sea-surface temperature data to drive model-based predictions of how the atmosphere will change during the next several months. But the atmosphere is a chaotic system, and small changes in current conditions can result in big differences in the seasonal patterns that models predict, so a single prediction doesn't tell the whole story. Instead, the IRI runs multiple models numerous times to get its best estimate of the upcoming season's mean climate. These days, the forecast maps are computer generated, dividing up the globe into grid cells that are approximately 250km x 250km in size. Eleven colors are used to show the probability of below-normal, normal or above-normal conditions for each cell. "Ten years ago, however, we would draw in the probabilities by hand," says Simon Mason [see the map at the top of this page for an example]. Mason and his colleagues, Lisa Goddard and Liqiang Sun, would gather in a "war room" to pore over and make sense of the model printouts. Mason, who leads IRI's Climate Program, recalls that the whole endeavor would take a full week of the group's time and effort because it had to "go over everything by hand and eye and discuss it." The process was also more subjective than it is today, says Goddard, because back then the forecasters didn't have an algorithm to combine the different models. "Our heads would be swimming by the end of the day!" she says. For the first few years, the group issued its forecasts on a quarterly basis. They switched to the current monthly format in 2001. "Updating each month allows new and possibly unexpected climate developments to be taken into account in the new forecast," says Barnston. Goddard adds that quarterly forecasts would miss key target seasons for some users. "For example, in Northeast Brazil, the main rainy season occurs in February, March and April. None of our quarterly forecasts would have hit that directly. By updating every month we're hopefully providing more accurate information to our users," she says. Approximately 6,000 people visit the net assessment web pages each month, and 1,500 people subscribe to get email alerts when new forecasts are posted. In the pre-web-site years, the group would have to email or transfer the raw data files to specific users who requested the forecasts. "For example, we were collaborating with rangeland decision makers in Queensland, Australia, so we'd send them the actual numerical values coming out of the models. And for our colleagues in Brazil, we'd print out and fax simplified contour maps," Goddard says. The kinds of users who download the forecasts have varied over the years, says Mason. "The interest in our earliest net assessments was very much tied to the strong El Nino that developed in 1997. We lost count of the number of phone calls we had from the United States Agency for International Development (USAID) and the State Department." Mason says that numerous country-level forecasts created by national meteorological services currently use IRI data as a starting point. "Also, we know that our forecasts are monitored regularly by the International Federation of Red Cross and Red Crescent Societies, Famine Early Warning Systems, World Vision and other humanitarian organizations." The quirkiest request Mason remembers came from an ice-cream manufacturer in South Africa, who contacted him almost every month, hoping to figure out how much ice cream to produce for the upcoming season. "I also used to get an hour-long phone call every month from a man who lived in Namibia. He would report to me exactly how much rainfall had been recorded in every station in the country. He had an encyclopedic knowledge of the rainfall history there," Mason says. "I tried to get him to do some analyses, but it never worked out!"
Relevant Links About the IRI
|