Accelerated and accurate Radiative Transfer simulations in OMI Ozone Profile retrievals: combining PCA-based RT model and Look-up table correction

Due to the limitation of the computational capability, only 20 % of Ozone Monitoring Instrument (OMI) measurements are processed to produce operational Ozone Profile data (OMO3PR), and spatial resolutions are degraded by 4 times to produce our research ozone profile product (PROFOZ). The PROFOZ retrieval is still slow despite of manly approximations made in Radiative Transfer (RT) calculations including the interpolation of RT calculations from ~ 80 wavelengths into 0.05nm resolution, on-line polarization correction using differences between vector and scalar calculation at ~10 wavelengths, the use of 4 half streams, and a vertical grid of ~26 layers. This study accelerate the RT calculations and improve the accuracy through combining a fast RT model based on the Principal Component Analysis (PCA) technique and correction with pre-calculated look-up tables (LUTs). In the PCA-RT approach, single and multiple scattering terms are separately calculated from three RT models. Line-by-line calculations are efficiently performed for accurate single scattering and approximate multiple scattering (MS) radiances using "First-Order" and 2Stream (2S) RTs, respectively. The time-consuming VLIDORT model is employed to correct 2S results, but limited for a few representative optical profiles which are expressed by mean optical state and its perturbation in terms of the first few EOFs in a single bin: all spectral points are previously grouped into "bins" according to the optical similarity of the absorption optical depth. We optimized the binning and the number of EOFs to simulate high-resolution radiance (270-330 nm) using 54 VLIDORT calls with errors of ~ 0.03 %. Using PCA-RT alone does not speed up the RT calculation over our current retrievals. In on-line application to OMI ozone profile retrievals, the PCA-RT is performed for lower accuracy configuration such as 4-stream, scalar mode, and coarser vertical layering. Simulation errors arising from vector vs. scalar, low vs. high stream, and coarse vs. fine vertical grids are corrected with LUTs. To reduce the space of LUTs, RT calculations are parameterized with respect to azimuth angle and surface albedo and hence correction spectrum is interpolated as functions of solar and viewing angles, and surface pressure from LUTs created using 22 total ozone dependent profiles.