In the context of variational data assimilation for Numerical Weather Prediction (NWP), adjoint operators of suitable linearization of observation operators is needed. In a state of the art NWP model where convective parameterization is used, some form of linearized convection scheme is desirable for such purposes. Unfortunately, constructing a linearized (tangent-linear) version of a complex convective parameterization scheme (such as the Kain-Fritsch scheme)can be a rather demanding task. In addition, these linearized schemes can be unstable to small perturbations. Identifying, understanding their nature and removing these instabilities can be complex. We propose an approach that stably approximate the linear part of the full nonlinear convective scheme to a degree generally higher than the classical tangent-linear approach.

Accuracy results in vertical column mode with this new approach will be presented using convective regimes extracted from state of the art Canadian Global NWP forecasts at 35 Km horizontal resolution with the Kain-Fritsch moist convective parameterization scheme.