The Optimal Spectral Sampling (OSS) method is a new rapid and accurate technique developed at AER for the numerical modeling of narrow band transmittances in media with non-homogeneous thermodynamic properties containing a mixture of absorbing gases with variable concentrations. The method has been specifically designed for the modeling of radiances measured by Earth-orbiting down-looking microwave and infrared radiometers, but can be applied to any spectral domain and instrument viewing geometry. This technique can also be applied to the general problem of flux or radiance computation in emitting and scattering atmospheres. The OSS method is particularly well suited for remote sensing applications and for the assimilation of satellite observations in Numerical Weather Prediction models. Both applications require fast and accurate models to produce radiances and Jacobians used for the real-time inversion of multi-spectral satellite observations. The OSS method is a generalized form of the exponential sum fit technique, ESFT, and is equivalent to the ESFT technique in the special case of a homogeneous atmosphere and a single constituent. The OSS method approximates the atmospheric transmittance (or radiance) in a given channel as a weighted sum of monochromatic transmittances (or radiances) evaluated at selected wavenumbers (nodes) within the interval spanned by the instrument function:

In the OSS approach, the wavelengths () and weights ()are obtained by minimizing the RMS difference between the exact and estimated transmittance profiles (or radiances) for a set of training profiles chosen to span the range of conditions to which the model will be applied. The optimal selection of the s is done using a Monte-Carlo search method applied to a set of uniformly spaced monochromatic transmittances,  (or radiances) produced by a reference line-by-line model (e.g., LBLRTM) output.

OSS attributes:

• OSS model parameters are derived for any instrument using a fully automated and unsupervised search procedure (no manual tuning is required).
• RT calculations are performed monochromatically from tabulated optical depths for fixed and variable constituents, which provides an accurate treatment of the surface reflection effects and significant computational advantages over other existent RT parameterization techniques used in remote sensing applications for the derivation of the analytical Jacobians.
• The model accuracy can be improved by simply increasing the number of nodes