On the impact of the diabatic component in the Forecast Sensitivity Observation Impact diagnostics

Over the years, a comprehensive set of the linearized physical parametrization schemes has been developed at ECMWF. These linearized schemes, operationally used in data assimilation, parametrize both the dry physical processes (vertical diffusion, gravity wave drag, shortwave and longwave radiation) and the moist processes (convection, large-scale condensation and clouds) consistently with the physical parametrization of the nonlinear model (though some simplifications are applied).

In this work, the representation of the moist physical processes in the adjoint assimilation model is compared with the representation of humidity in the energy norm used to compute the forecast sensitivity to observations in the short-range forecasts. Forecast Sensitivity Observation Impact using the adjoint model with only dry processes (dry adjoint) but moist energy norm in the sensitivity gradient calculation is examined in contrast with the observation impact obtained when moist processes (moist adjoint) and dry energy norm are used. The performed study indicates that the use of the humidity term in the norm produces unrealistic humidity and temperature sensitivity gradients, which largely affect the observation forecast impact results.