Using ECMWF's Forecasts (UEF2019)

MeteoSwiss as a national met service commits itself to provide weather forecasts for arbitrary locations in Switzerland and also to adequately communicate forecast uncertainty. While in the past such information has been communicated mostly in text form, MeteoSwiss tries to increasingly use such uncertainty information in all products. We rely on ensemble predictions as a data source, both on the IFS-ENS from ECMWF and on our local area ensemble model COSMO-E. While both of these numerical weather prediction (NWP) models are run at increasingly high resolutions, raw ensemble forecasts still tend to be biased and underdispersed. Hence, statistical postprocessing is expected to improve forecast skill and may help to condense the forecast information. At MeteoSwiss, a project on postprocessing of relevant variables for public weather forecasts, i.e. near surface temperature, precipitation, 10 meters wind speed, and cloud cover, has recently been launched. The project’s goal is the introduction of an operational suite providing spatial fields of calibrated probabilistic multi-model consensus predictions integrating the available NWPs. The focus on introduction into operations implies that performance in terms of skill improvement is not the only requirement for the postprocessing suite, but also its data demand and ability to cope with model updates or the flexibility to incorporate additional NWP models are equally relevant. This poster will present an overview of the project, summarize the challenges and show our approach and first results towards seamless probabilistic forecasts from hours to weeks ahead.