Calibrated Probabilistic Seasonal Forecasts at IBM/The Weather Company: Business Applications

Seasonal forecasts have been created and disseminated to IBM/The Weather Company clients for two decades, with the primary focus being products supporting clients involved in trading the commodity markets. Given the lack of appetite for probabilistic information in the trading world until recently, these forecasts have been primarily expressed in deterministic terms, with the goal being to correctly capture the skew relative to “normal”.

However, there appears to be a (slowly) increasing appetite for probabilistic information at all forecast time scales, including the seasonal window. Further, there are many different industries with the desire to improve the quality of their long-range weather assumptions, as many companies have been relying on climatology instead of using seasonal forecasts.

Over the past year, scientists at The Weather Company/IBM have developed a fully calibrated probabilistic seasonal forecasting system that uses the new ECMWF S5 seasonal forecasting model. Forecasts for precipitation and maximum/minimum/mean temperature are produced at daily resolution out to 6 months on a global 0.4 degree grid. For each day, a 50-member (prototype) calibrated ensemble and 11 percentiles are provided for each of the four variables.

Rather than providing one deterministic forecast, we are now providing fifty equally likely forecasts that span the calibrated distribution. This paradigm fundamentally changes how seasonal forecasting information is used, as businesses can now run fifty different scenarios through their decision models to produce a distribution of possible business outcomes. The availability of a set of fully calibrated hindcasts back to 1981, in the same format as the live forecasts, allows businesses to do a thorough evaluation of the efficacy of the forecasting system for their particular use case.

This talk will discuss the scientific development of the forecasting system and will also focus on a variety of business applications and use cases that have been encountered to date.