Principal component analysis of remote sensing of aerosols over oceans

We apply principal component analysis (PCA) to estimate how much information about atmospheric aerosols could be retrieved from solar-reflected radiances observed over oceans by a satellite sensor as a function of the number of wavelength bands, viewing angles, and Stokes parameters. It is assumed that our virtual satellite sensor can simultaneously perform multispectral, multiangle, and linear polarization measurements of the radiances, and the following quantities are used to vary: aerosol optical thickness, single-scattering albedo (SSA) of aerosol particles, height of the aerosol layer, aerosol model (includes size distribution parameters and optical properties), and wind speed. The real refractive index was kept constant and, therefore, is not part of the analysis. To calculate the number of significant principal components (SPCs), the cumulative percent variance rule is used, which takes into account anticipated errors of measurements. The reported results predict how much additional information can be retrieved from observations by adding more wavelength, angle, and polarization channels. For example, for the Moderate Resolution Imaging Spectroradiometer instruments (lambda > 0. 5 mu m), the number of SPCs is two to three; for Multiangle Imaging SpectroRadiometer, three to five; for Polarization and Directionality of the Earth Reflectances, five to ten; while for the future Glory/Aerosol Polarization Sensor instrument, it is 6-11 (when using eight out of its > 100 view angles). The ranges reflect view conditions, and analysis method. Our calculations show that the observations should be most sensitive to the aerosol model followed in decreasing order by optical thickness, SSA, and aerosol height. We found that there is no systematic increase in the information about aerosol starting from 10-15 view angles for unpolarized observations and 30 view angles for those with linear polarization. It is achievable with modern detectors to retrieve up to 10 and 16 SPCs from unpolarized and polarized observations, respectively. The methodology and results of our PCA can be useful for estimating the reliability of aerosol parameters retrieved from existing and future satellite observations.