Statistical post-processing or calibration can significantly improve the value of forecasts for the end user.
An expert review of the state-of-the-art of calibration methods for medium-range weather forecasts was carried out for ECMWF by Professor Tilmann Gneiting of the University of Heidelberg. The report showed that well-established, state-of-the-art post-processing techniques can significantly improve the skill of these forecasts. The report also highlighted important areas where further research is required, including the requirements for reforecast datasets. The report has been issued as an ECMWF Technical Memorandum (Calibration of medium-range weather forecasts, TM 719, March 2014).
Following the report, ECMWF initiated a study on the calibration of surface weather forecasts, using the meteogram parameters, verified against SYNOP observations as an example. The aim of the work was to demonstrate the benefits of using state-of-the-art calibration for the ECMWF forecasts, including an objective approach to combine the ENS and HRES components of ECMWF’s forecast. It was found that calibration can provide substantial additional skill compared to the raw ECMWF forecasts. The study was carried out in collaboration with Prof Gneiting and members of his Group on Computational Statistics at the Heidelberg Institute for Theoretical Studies (HITS). The results of this study are reported in an article in the ECMWF Newsletter (Calibration of ECMWF forecasts, ECMWF Newsletter, 142, Winter 2014-15).
A number of relevant aspects were not addressed in the above study, which focused on individual locations and on overall performance as measured by the CRPS. Important areas for further study include the spatial and temporal structure of calibrated products and the impact of calibration on the forecasting of extreme events. The enhanced re-forecast dataset to be introduced in 2015 will allow ECMWF to begin investigating these topics. ECMWF will explore the potential for calibration of gridded fields (against analyses). This work will allow the development of ‘seamless’ forecast products that cover all time-ranges from the medium-range to seasonal.
ECMWF will continue to work together with Prof Gneiting and his group at HITS, and will also develop collaboration with experts in statistical post-processing in the Member States and Co-operating States
Calibration Meeting 12 February 2015