Development of Precipitation Postprocessing at the Weather Company

The Weather Company’s (TWC’s) global precipitation amount and precipitation probability forecasts are based upon a multi-model synthesis of numerical guidance, with some adjustments for common biases. Human edits are also made in situations where the forecast guidance is judged to be of sub-standard quality. With a suitable time series of past forecasts (f) and observations (o), it is possible to improve TWC’s real-time forecast guidance using statistical corrections based on the discrepancies between the past forecasts and observations. This talk will describe the postprocessing method that has been developed at TWC. Some aspects are relatively well understood, including the postprocessing of 6-hourly amounts using the extended logistic regression of Wilks (2009). There are novel elements, though, including data handling in areas with sparse surface observations, and methods of aggregating the post-processed guidance to longer periods (day/night, or calendar day) or disaggregating to hourly. These are based on a method of reconstructing 6-hourly ensemble forecast members, using a scaling factor for each member based on the ratio of the post-processed and raw values, and scaling the hourly amounts. From the scaled hourly amounts, accumulations over arbitrarily longer periods can be performed on each scaled, calibrated member, and probabilities derived from the calibrated ensemble relative frequency . A more full description of the methodology along with comprehensive results will be provided in the talk.