Evaluation of the jacobians of infrared radiation models for variational data assimilation

In this paper, linearized versions of fast infrared radiative transfer sche mes for variational data assimilation are studied. A neural network-based i nfrared broadband radiation model (NeuroFlux) is compared with the European Centre for Medium-Range Weather Forecasts operational radiation model. Als o, the Radiative Transfer for Television and Infrared Observation Satellite Operational Vertical Sounder (RTTOV) scheme for satellite brightness tempe rature computation is compared with a more physically based scheme: the nar rowband Synsatrad model developed at the European Organization for the Expl oitation of Meteorological Satellites. The Jacobians are examined. They are converted into flux perturbations with the tangent-linear approximation an d into atmospheric variable increments with a one-dimensional variational a ssimilation system. For NeuroFlux and RTTOV, despite accurate flux and radi ance computation, the sensitivity with respect to water vapor needs to be i mproved. However, the random structure of the neural network derivative err or allows the use of NeuroFlux with a single mean Jacobian in the variation al context. Also, further improvements to RTTOV are expected from ongoing w ork on the regression dataset and on the choice of the regression predictor s.