A new era in operational numerical weather prediction(NWP) began in recent years when the ensemble predictions were performed in some meteorological centers in the world.Since the actual state of the atmosphere at any time is known only approximately,a complete description of the weather-prediction problem should be formulated in terms of the time evolution of an appropriate probability density function in the atmosphere′s phase space.Ensemble forecasting appears to be the only feasible method to predict the evolution of the atmospheric probability density function beyond the range in which error growth can be described by linearized dynamics.The generation of effective perturbation is one of the major problems in ensemble forecasting.Perturbing the initial conditions along the most unstable directions of the phase space of the system is a technique which may be most effective for determining the principal weather types consistent with given initial data.The singular vectors and Chinese ensemble prediction systems were described briefly.