Ten Years of Regional Climate Outlook Forums

The African Centre of Meteorological Application for Development (ACMAD) and the IRI have analyzed the accuracy of a decade's worth of forecasts issued by Regional Climate Outlook Forums (RCOFs) in Africa. The findings, available in a new report, indicate the forecasts have performed moderately well in each of the three regions where RCOFs have been held regularly - Southern Africa, the Greater Horn, and West Africa. However, they also show clear tendencies of over-predicting normal rainfall, and under-predicting dry conditions.

"The RCOFs have had numerous benefits, such as the development of capacity within the climate services of the national meteorological agencies in most of Africa. The fact that the forecasts themselves have been skillful, despite us having to learn to make them on the job, provides a powerful endorsement to the whole process," says Simon Mason, the head of IRI's Climate Program.

Since 1997, the RCOFs have regularly convened members of the climate and meteorological communities to share, analyze and discuss upcoming seasonal forecasts. Today, the forecasts are routinely used to make planning decisions for agriculture, water and other sectors.

The IRI has played a key role in the RCOFs since their inception, and has continued to participate in most of the forums around the globe. It regularly provides forecast products, participates in capacity building activities and develops software such as the Climate Predictability Tool specifically for use in the forums.

The forums were started as a way to address concerns by nontechnical users over conflicting seasonal climate forecasts issued by different agencies.

"People didn't know which forecast to believe," says Mason.

Each RCOF produces a consensus forecast endorsed by all the participating forecasters, who, together with sectoral experts, discuss the implications of any predicted deviations from normal rainfall or temperatures in the upcoming season. Some RCOFs have started to focus on problems related to global warming as well.

The first RCOF, held in Kadoma, Zimbabwe in 1997, provided forecasts for most of southern Africa. Forums for the other regions in Africa quickly followed. Since then, similar meetings have been held in South America, Southeast Asia, the Caribbean, the Pacific, and parts of Europe. Others are being planned for the remaining parts of the globe. For a complete list, visit the World Meteorological Organization web site.

Seasonal climate forecasts are typically issued in probabilistic terms, meaning that a probability is provided for each of three categories: below-normal, normal and above-normal. For example, in the three-month forecast for the Greater Horn of Africa shown on this page, green areas have a 35% chance of above-normal rainfall, a 45% chance that rainfall will be near-normal, and a 20% chance that it will be below-normal (a tutorial on forecasts is available here).

After many years of issuing such forecasts, climatologists can calculate how good, or skillfull, the forecasts have been by comparing WHAT ACTUALLY HAPPENED TO WHAT WAS FORECAST.

Although the forecasts for all three African regions show some skill, there are systematic problems that could be addressed to improve overall quality.

"The most common ‘error' we found was that forecasters tended toward high probabilities of near-normal rainfall and avoided indicating high probabilities of below-normal rainfall," says Mason. " In all three regions, the forecasters frequently suggested that the normal category was the most likely to occur, whereas the below-normal category was observed most frequently."

THERE IS A SUBJECTIVE COMPONENT TO GENERATING FORECASTS, SAYS MASON. SOPHISTICATED MATHEMATICAL MODELS GENERATE A ROUGH DRAFT OF A FORECAST, WHICH IS THEN ADJUSTED BY THE FORECASTERS, WHO DRAW ON THEIR OWN EXPERTISE TO RECONCILE OUTPUTS FROM DIFFERENT MODELS AND FOR NEIGHBORING AREAS.

The consistently high probabilities on the normal category are an indication that the forecasters are hedging, Mason explains. "They are avoiding being seen as producing a potentially bad forecast because if they issue high probabilities for above-normal rainfall and it is unusually dry, the forecast will look much worse than if they had issued high probabilities on the normal category. The forecasters are therefore playing safe."

For similar reasons, the forecasters avoid putting high probabilities on the below-normal category because they do not want to cause undue alarm.

"While it may at first sight seem irresponsible for the forecasters to hedge their forecasts, they do have to worry about over-reactions to forecasts that call for possible drought conditions, for example. So they tend to weaken their warnings of dry conditions in the hope of realizing an appropriate level of response," says Mason.

Mason says these biases do not mean the forecasts were poor at forecasting below-normal rainfall. "When the forecasters indicated A GREATER CHANCE of below-normal rainfall, dry conditions did in fact occur more frequently than when they indicated A LESSER CHANCE," he says. "The error lies in the fact that the probabilities for below-normal rainfall were consistently too low."

"IRI AND ACMAD HAVE TO WORK TO HELP ADD MORE OBJECTIVE ASPECTS IN THE WAY FORECASTS ARE ISSUED IN ORDER TO SUBSTRACT THE SUBJECTIVE ASPECT, SINCE THIS IS DEPENDENT ON EACH INDIVIDUAL," SAYS ALHASSANE DIALLO, THE NEW DIRECTOR OF ACMAD.

The report also recommends that forecasters rely more heavily on the global-forecast products from various official WMO-designated centers.

The full report is available on the IRI's web site: http://here

RELEVANT LINKS:
WMO's Regional Climate Outlook Forums pamphlet