A new method of assessing filtering schemes in data assimilation systems

Abstract

AbstractA new statistical diagnostic is proposed to assess the impact of externally applied filters in data assimilation systems with high‐lid atmospheric models. The diagnostic involves comparing the variances of 6 h differences of 6 h forecasts with those of a free model simulation. It was applied to the Canadian Middle Atmosphere Model Data Assimilation System (CMAM‐DAS) to choose among various filtering options. The variances of 6 h difference fields are shown to suppress long time‐scales and highlight short ones. This explains their sensitivity to a variety of filters considered and their relative insensitivity to the choice of initial conditions used for the time series. The method was used to quantify the extent to which a digital filter applied to the full state reduced the desired level of variability, as well as to determine objectively the most appropriate filter for our system from among several incremental analysis updating schemes. The method should be especially useful for models extending to the stratosphere and mesosphere, where short time‐scales represent significant contributions to the energy spectrum. Copyright © 2009 Royal Meteorological Society and Crown in the Right of Canada