Modern seasonal forecast systems employ powerful computers to calculate a number of forecasts simultaneously and thereby produce estimates of the probability distributions of various atmospheric variables for lead times of months, seasons, and one year in advance. These probabilistic forecasts indicate uncertainty and likelihood and can be adapted to provide information critical to decision making in the energy industry.

But their usefulness in decision support depends on whether the predicted probabilities are reliable. Thus climate variability forecasts are assessed and improved by applying the seasonal forecast system to several decades of historical cases and then comparing these retrospective forecasts to the actual observations. The observed historical errors are summarized statistically to serve as corrections for future forecasts, with differences in mean values and various measures of spread being the main focus of the calibration process. The foundation for all seasonal forecast calibration is the assumption that past errors are a prolog to future errors and can be used to make meaningful refinements.

The key issues are whether the verification observation is likely to fall within the ensemble of forecasts, whether it is likely to be near the ensemble average, and whether the distribution of ensemble members correctly represents the actual probabilities of occurrence. Thus the aim is to ensure that forecasts are centered properly and then to adjust the spread of the ensemble so that the probabilities, and thus the uncertainty, are portrayed reliably.

For more than eight years, the World Climate Service (WCS), a joint venture of Prescient Weather Ltd and MeteoGroup, has provided energy industry clients in the U.S. and Europe with calibrated seasonal forecasts of the European Centre for Medium Range Weather Forecasts (ECMWF) and the U.S. National Weather Service and makes them available to clients along with other climate information on an interactive website. The WCS website allows clients to examine the forecasts in probabilistic form as calibrated with 10-year and 30-year training periods.

The efficacy of two methods of calibration will be examined as applied to the new ECMWF Seasonal Forecast System 4 and the new NWS Climate Forecast System V2. One calibration method is Bayesian and attempts to align the forecast ensemble with the observed climatology. The second and newer method is a “climate-conserving” calibration that centers the forecasts on averages and then scales both the variance of the ensemble averages and the members of the ensemble to match observations.

The presentation will show how the meteorological variables of the calibrated seasonal forecasts can be converted into variables that are directly relevant in energy decisions. Examples of decision-oriented predictands are probabilistic estimates of degree-day totals or wind power potential, which depend nonlinearly on daily or hourly variation that is not usually depicted in monthly average seasonal forecasts.. Finally, the prospects for reliable prediction of extreme events in the tails of the distributions and for meaningful estimates of intra-seasonal variability will be examined.