Title: Estimation of displacement error in a variational framework

Author: R. Legrand (CNRM)
Y. Michel (CNRM)
T. Montmerle (CNRM)

With a probabilistic approach of the data assimilation problem in numerical weather prediction, distributions of error contained in observations and background state are specified through Probability Density Functions (PDFs). A largely adopted hypothesis is the Gaussianity of such PDFs. Hence models of errors are completely specified with a mean and a covariance operator. With forecast errors being strongly driven by non-linear physical processes and by displacement errors, such representation has shown to be questionable for some variables including humidity and hydrometeors.

First this work aims in diagnosing the non-Gaussianity of forecast errors in a convective scale model. The non-Gaussianity is documented using a statistical test based on univariate skewness and kurtosis of the sample background errors. Then as a mean to model a part of this non-Gaussianity, a correction of displacement errors is added to the standard correction of amplitude errors. The chosen methodology is to estimate displacement error in a Gaussian framework following a variational formalism.

For this study the dataset is a large ensemble of forecasts performed by the convective scale model AROME. Presented results are of two kinds. i) First an overview of the main non- Gaussian features is presented. We describe some quantitative aspects of Non-Gaussianity such as time evolution during forecasts and analyses steps, and sensitivity to model level or variables including control variables of the assimilation scheme. ii) We then describe first results about the objective estimation of displacement error, and their statistical structures.

GMAO Head: Steven Pawson Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: Feb 9 2015