Articles | Volume 8, issue 1
https://doi.org/10.5194/amt-8-281-2015
https://doi.org/10.5194/amt-8-281-2015
Research article
 | 
14 Jan 2015
Research article |  | 14 Jan 2015

Use of neural networks in ground-based aerosol retrievals from multi-angle spectropolarimetric observations

A. Di Noia, O. P. Hasekamp, G. van Harten, J. H. H. Rietjens, J. M. Smit, F. Snik, J. S. Henzing, J. de Boer, C. U. Keller, and H. Volten

Abstract. In this paper, the use of a neural network algorithm for the retrieval of the aerosol properties from ground-based spectropolarimetric measurements is discussed. The neural network is able to retrieve the aerosol properties with an accuracy that is almost comparable to that of an iterative retrieval. By using the outcome of the neural network as first guess in the iterative retrieval scheme, the accuracy of the retrieved fine- and coarse-mode optical thickness is further improved, while for the other parameters the improvement is small or absent. The resulting scheme (neural network + iterative retrieval) is compared to the original one (look-up table + iterative retrieval) on a set of simulated ground-based measurements, and on a small set of real observations carried out by an accurate ground-based spectropolarimeter. The results show that the use of a neural-network-based first guess leads to an increase in the number of converging retrievals, and possibly to more accurate estimates of the aerosol effective radius and complex refractive index.

Download
Short summary
A neural network algorithm has been developed to retrieve aerosol microphysical parameters from ground-based measurements of skylight intensity and polarization. The neural network is capable of producing accurate estimates of aerosol optical thicknesses, effective radii and refractive index. In addition, it is shown that the use of the neural retrievals as initial guess for an iterative retrieval algorithm results in improved convergence and retrieval accuracy.