Infrared Radiance Modeling by Optimal Spectral Sampling

Abstract This paper describes a rapid and accurate technique for the numerical modeling of band transmittances and radiances in media with nonhomogeneous thermodynamic properties (i.e., temperature and pressure), containing a mixture of absorbing gases with variable concentrations. The optimal spectral sampling (OSS) method has been designed specifically for the modeling of radiances measured by sounding radiometers in the infrared and has been extended to the microwave; it is applicable also through the visible and ultraviolet spectrum. OSS is particularly well suited for remote sensing applications and for the assimilation of satellite observations in numerical weather prediction models. The novel OSS approach is an extension of the exponential sum fitting of transmittances technique in that channel-average radiative transfer is obtained from a weighted sum of monochromatic calculations. The fact that OSS is fundamentally a monochromatic method provides the ability to accurately treat surface reflectance and spectral variations of the Planck function and surface emissivity within the channel passband, given that the proper training is applied. In addition, the method is readily coupled to multiple scattering calculations, an important factor for treating cloudy radiances. The OSS method is directly applicable to nonpositive instrument line shapes such as unapodized or weakly apodized interferometric measurements. Among the advantages of the OSS method is that its numerical accuracy, with respect to a reference line-by-line model, is selectable, allowing the model to provide whatever balance of accuracy and computational speed is optimal for a particular application. Generally only a few monochromatic points are required to model channel radiances with a brightness temperature accuracy of 0.05 K, and computation of Jacobians in a monochromatic radiative transfer scheme is straightforward. These efficiencies yield execution speeds that compare favorably to those achieved with other existing, less accurate parameterizations.