| description abstract | A continuous assimilation of high-density global satellite observations is required in order to improve numerical weather prediction analyses used to start forecasts. Until now, it was assumed that efficiency requirements imposed the use of regression-based models of atmospheric transmittance (typically on fixed pressure layers with coefficients varying for each layer) and prohibited the use of physically based models. Here, it is demonstrated that an explicit calculation of infrared transmittances for each absorbing gas (H2O, CO2, O3, CH4, N2O, and O2) can be done efficiently, provided that a monochromatic approach is followed as in a regression model such as Radiative Transfer for TOVS (the TIROS Operational Vertical Sounder) (operational in most weather centers). The classical Goody random model is chosen as a physical formulation for spectral line absorption along with established water vapor and oxygen continua parameterizations. Line-by-line transmittance calculations for 189 atmospheric profiles are used as reference in the evaluation. By adjusting the individual gas optical depths by a constant multiplicative factor (typically near unity), it is shown that an accuracy better than 0.3 K in brightness temperature can be obtained for most satellite infrared sounding channels. Jacobians defining the adjoint of the model are readily obtained by analytical differentiation of the radiance with respect to level temperature and humidity. The new model was introduced into the Canadian Meteorological Center 3D variational data assimilation system, and a comparison was carried out between the regression and physical models for a 2-week period for the first 12 sounding channels of the NOAA-12 satellite. The numerous advantages of the physical model over the regression model are emphasized. Biases introduced by the use of fixed or outdated mixing ratio estimates for CO2, O3, and CH4 are largely reduced using current and location dependent concentrations. Global statistics and maps of observed minus calculated radiances reveal the general superiority of the physical model. The proposed model is efficient and is well suited for the massive assimilation of satellite radiances and other remote sensing applications. | |
