Implementing and Testing of a Lognormal-Based Humidity Control Variable for the NCEP GSI Hybrid EnVAR Assimilation Scheme

As the resolution of operational global numerical weather prediction system approach the meso-scale, then the assumption of Gaussianity for the errors at these scales may not valid. However, it is also true that synoptic variables that are positive definite in behavior, for example humidity, cannot be optimally analyzed with a Gaussian error structure, where the increment could force the full field to go negative.

In this presentation we present the initial work of implementing lognormal incremental approximation for the moisture variable in both the ensemble and variational component of the NCEP hybrid GSI EnVAR, as well as present the initial work in implementing a mixed Gaussian-lognormal incremental approach for temperature and moisture to capture a more consistent covariance model between the two. As part of this work we shall also present the foundations for implementing a mutlivariate lognormal based approach for cloud-related control variables that should allow for the more consistent assimilation of cloudy radiances.