Title: Ensemble methods for weather prediction
Authors: Roberto Buizza - Scuola Superiore Sant'Anna (SSSA) Pisa (Italy) [presenting]
Abstract: Ensemble methods based on a limited number of numerical integrations, have so far proven to be the only feasible way to estimate the probability distribution function (PDF) of forecast states. Probabilistic forecast products are generated by computing statistics based on a finite number of members, which are generated to produce accurate and reliable forecasts. The implementation of operational ensembles in 1992 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 operational implementations at the European Center for Medium-Range Weather Forecasts (ECMWF, Europe) and at the National Centers for Environmental Prediction (NCEP, US) 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. Ensembles helped the development of subseasonal and seasonal prediction systems. In climate studies, ensembles are used to estimate the range of possible future scenario. In NWP, today, ensembles are used also to estimate the PDF of initial states. A brief overview will be provided about the key characteristics of ensemble methods used in weather prediction, to estimate the initial and forecast PDF.