Evaluation of the Jacobians of Infrared Radiation Models for Variational Data Assimilation

Abstract

In this paper, linearized versions of fast infrared radiative transfer schemes for variational data assimilation are studied. A neural network?based infrared broadband radiation model (NeuroFlux) is compared with the European Centre for Medium-Range Weather Forecasts operational radiation model. Also, the Radiative Transfer for Television and Infrared Observation Satellite Operational Vertical Sounder (RTTOV) scheme for satellite brightness temperature computation is compared with a more physically based scheme: the narrowband Synsatrad model developed at the European Organization for the Exploitation of Meteorological Satellites. The Jacobians are examined. They are converted into flux perturbations with the tangent-linear approximation and into atmospheric variable increments with a one-dimensional variational assimilation system. For NeuroFlux and RTTOV, despite accurate flux and radiance computation, the sensitivity with respect to water vapor needs to be improved. However, the random structure of the neural network derivative error allows the use of NeuroFlux with a single mean Jacobian in the variational context. Also, further improvements to RTTOV are expected from ongoing work on the regression dataset and on the choice of the regression predictors.