Twenty-five years of ensemble prediction at ECMWF

Twenty-five years ago, in November 1992, two of the major meteorological centers, the European-Centre for Medium-Range Weather Forecasts (ECMWF) and the US National Centers for Environmental Prediction (NCEP), started producing global, medium-range ensemble forecasts as part of their operational suites. The implementation of these ensembles followed years of research in predictability, which saw many scientists both in academia and in operational numerical weather prediction (NWP) centers investigating how best to deal with the sources of forecast uncertainties. The ECMWF and NCEP decisions induced a paradigm shift in NWP from providing a single forecast, to issuing a range of forecasts that can be used to identify possible future scenarios, compute the probability of events of interest, and in general to estimate forecast confidence levels.

At ECMWF, the first operational run of the ECMWF global, medium-range ensemble (ENS) was completed on the 24th of November 1992. At that time, ENS included only a representation of initial uncertainties based on initial-time singular vectors and no model uncertainty scheme, had 33 members and was run three-times a week. It had a resolution of about 200 km and 19 vertical levels, and produced forecasts up to 10 days.

Today (August 2016), twnety-five years after its implementation, the ECMWF medium-range/monthly ENS includes more sophisticated initial perturbations constructed using singular vectors and an ensemble of data assimilations. Model uncertainties are also simulated by two stochastic schemes, and the ensemble size has been enlarged to 51 members. ENS is run four times a day: at 06 and 18 UTC up to 6.5 days as part of an ECMWF Special Boundary Condition Project, and at 00 and 12 UTC up to 15 days. Furthermore, twice a week ENS is extended to 46 days with a lower land/atmosphere resolution of 36 km. ENS has an 18 km resolution in the land, ocean waves and atmosphere grid, and it is coupled to a dynamical ocean model with a horizontal resolution of 1 degree and 42 levels. In October 2016, the resolution of the dynamical ocean will be increased to 0.25 degrees and 75 levels with an upper ocean vertical resolution of 1m, and an interactive sea-ice model will be introduced. To compute the model climate and to generate calibrated forecast products such as the extreme forecast index, ENS also includes a re-forecast suite, with an 11-member ensemble run twice every week for the past 20 years.

During this talk, the main upgrades of the ENS configurations throughout its first 25 years of life will be reviewed, and our plans for future improvements will be discussed.