Noise Suppression in AEOLUS Optical Properties Retrieval by Maximum Likelihood Estimation

The Aladin instrument on-board the ESA Earth Explorer satellite Aeolus is a UV high spectral resolution Doppler Wind Lidar. The main mission product is profiles of horizontally projected line-of-sight winds, and the instrument design is therefore optimized to measure Doppler shifts of the atmospheric backscatter signals compared to the UV light emitted at ~355 nm (ESA, 2008; Stoffelen, 2005). Since the lidar backscatter contains information on the location of optically thin aerosol and cloud layers and cloud tops, spin-off products have been developed to retrieve aerosol and cloud backscatter and extinction coefficient and lidar ratio profile products (ESA, 2008; Flamant, 2008; Flamant, 2017). The advantage of a high spectral resolution lidar is that it measures molecular and particle backscatter separately in two dedicated channels. Still, some contributions from molecular backscatter exists in the measurements from the Fizeau channel and vice versa. This channel cross-talk requires correction during the product retrieval.

The Aeolus L2A operational aerosol and cloud retrieval algorithm is applying the so-called high spectral resolution retrieval method for the calculation of the particle and extinction backscatter coefficient products. The algorithm, developed at IPSL and Météo-France, is called the Standard Correct Algorithm (SCA) (Flamant, 2008; Flamant, 2017). High signal noise is obtained due to ever-decreasing laser energies and instrument receive path transmission. As a result, the Aeolus SCA optical properties retrieval is hampered. Particularly the ill-posed particle extinction coefficient retrieval is severely affected. In the past, attempts were made to mitigate nonphysical optical properties by measures like zero-flooring or signal accumulation in even coarser range gates (Flamant, 2017). Their success was limited.

An alternative noise suppression approach by Maximum Likelihood Estimation has therefore been prototyped that permits the retrieval of extinction coefficients and lidar ratios solely within pre-defined physical bounds. The optical properties are fitted to the 24 Aeolus atmospheric range gates within single atmospheric columns, minimizing the corresponding distance to the observed L1B useful signals measured by both spectrometers. This up to 48-dimensional non-linear regression problem is solved by means of the L-BFGS-B algorithm (Zhu, 1997). The method has proven its usefulness in noise suppression with astonishing efficiency. Particularly, the retrieved extinction coefficient profiles are less noisy, clearly revealing atmospheric layers also visible in the L1B useful signal profiles. The method is validated on end-to-end simulations and in-orbit observations.

References

ESA, ADM-Aeolus Science Report. ESA SP-1311, ESA Communication Production Office, 121 pp., 2008, available on http://www.esa.int/aeolus.

Flamant, P. H., Cuesta, J., Denneulin, M.-L., Dabas, A., Huber, D. ADM-Aeolus retrieval algorithms for aerosol and cloud products, Tellus, 60A, 273-286, 2008, https://doi.org/10.1111/j.1600-0870.2007.00287.x.

Flamant, P. et al. ADM-Aeolus L2A Algorithm Theoretical Baseline Document, 2017, available on https://earth.esa.int/aos/AeolusCalVal.

Stoffelen, A. et al. The atmospheric dynamics mission for global wind field measurement, Bulletin of the American Meteorological Society, 86, 73-88, 2005, https://doi.org/10.1175/BAMS-86-1-73.

Zhu, C., Byrd R. H. and Nocedal, J. L-BFGS-B: Algorithm 778: L-BFGS-B, FORTRAN routines for large scale bound constrained optimization, ACM Transactions on Mathematical Software, 23 (4), 550-560, 1997, https://doi.org/10.1145/279232.279236.