Ensemble approach to weather forecasting

Abstract

Numerical simulations of the Earth System evolution are the basis of weather forecasts. Due to chaotic nature of the atmosphere, an inevitable inaccuracy of initial state information, and models imperfection their results are always uncertain. To be correctly interpreted by a user, each weather forecast should be supplemented by an estimate of its uncertainty. An ensemble approach is the only practically affordable method to solve the problem. With this approach, it is possible not only to get an a priori estimate of forecast uncertainty, but also to improve a mean forecast skill and to present alternative scenarios of future weather evolution. Ensemble prediction systems are highly time-consuming and can be implemented only on powerful supercomputers. A global ensemble medium-range prediction system has been running operationally at the Hydrometcenter of Russia since 2015. A short-range limited-area convection-permitting ensemble prediction system was developed for meteorological support of the Sochi Olympic Games 2014 and is now applied for research. Characteristics of the systems, peculiarities of their computer implementation and verification results are discussed.