Acceleration of radiative transfer model calculations for the retrieval of trace gases under cloudy conditions

In the independent pixel approximation (IPA), radiative transfer computations involving cloudy scenes require two separate calls to the radiative transfer model (RTM), one call for a clear sky scenario, the other for an atmosphere containing clouds. In this paper, clouds are considered as an optically homogeneous layer. We present two novel methods for RTM performance enhancement with particular application to trace gas retrievals under cloudy conditions. Both methods are based on reusing results from clear-sky RTM calculations to speed up corresponding calculations for the cloud-filled scenario. The first approach is numerically exact, and has been applied to the discrete-ordinate with matrix exponential (DOME) RTM. Results from the original clear sky computation can be saved in the memory and reused for the non-cloudy layers in the second computation. In addition, for the whole-atmosphere boundary-value approach to the determination of the intensity field, we can exploit a 'telescoping technique' to reduce the dimensionality (and hence the computational effort for the solution) of the boundary value problem in the absence of Rayleigh scattering contributions for higher azimuthal components of the radiation field. The second approach is (for the cloudy scenario) to generate a spectral correction applied to the radiation field from a fast two-stream RTM. This correction is based on the use of principal-component analysis (PCA) applied to a given window of spectral optical property data, in order to exploit redundancy in the data and confine the number of full-stream multiple scatter computations to the first few EOFs (Empirical Orthogonal Functions) arising from the PCA. This method has been applied to the LIDORT RTM; although the method involves some approximation, it provides accuracy better than 0.2%, and a speed-up factor of approximately 2 compared with two calls of RTM. -- Highlights: • Reusing results from clear-sky computations for a model with a cloud layer. • A spectral correction from a fast two-stream model based on PCA. • A speed-up factor of approximately two and accuracy better than 0.2%