Global Statistical Postprocessing of Precipitation Amount at the IBM/Weather Company

IBM has a range of precipitation-related weather forecast products, including hourly probability of precipitation and expected amount, as well as forecasts for periods such as morning and afternoon and for the calendar day. These show up in your mobile phone application as well as through APIs used by many other businesses. Past practice has been to generate these with weighted combinations of data from deterministic and ensemble forecast guidance. Given that precipitation forecasts are subject to bias such as the common over-forecast of light precipitation in the midlatitudes, these multi-system products are typically biased.

IBM is in the process of a once-generational upgrade of its precipitation product generation. The synthesis of multi-system data is preserved, but products will be generated from the postprocessing of each system’s numerical guidance against a combination of surface observations and satellite-based precipitation reanalysis products. This talk will provide a background on the arrangement of the data used in the process, the algorithm for producing the calibrated forecast guidance, and results of how much the precipitation forecast products have been improved through application of these technologies.