アンサンブル手法に基づく簡易予報感度解析

Abstract

Simplified methods are developed to identify initial perturbations that have largest influence on a given norm in a user-defined verification region. Our methods do not require a data assimilation system such as an adjoint model or an ensemble Kalman filter but an ensemble forecast with simple matrix operations. Two ensemble-based methods are formulated based on the linear perturbation theory: one that approximates the singular vector method and the other the adjoint method. The sensitive initial perturbations are obtained by finding the coefficients of the linear combination of ensemble members. Our formulations have been applied to the weekly ensemble forecasts of the Japan Meteorological Agency with the verification region over Japan. Our methods successfully identify sensitive perturbations associated with mid-latitude and tropical disturbances.