A filtering technique was developed to reduce the noise in the NIR band ratio, and thereby reduce the small-scale variability in aerosol model selection. The smoothing filter adjusts the band 7 radiance to minimize local variability in the band 7 to band 8 aerosol ratio (i.e., the multi-scattering equivalent of epsilon), while leaving the band 8 radiance (i.e., the aerosol concentration) unchanged..
The effect of this smoothing is to reduce pixel-to-pixel variability in the retrieved aerosol type, and thereby reduce atmospheric correction noise in the retrieved water-leaving radiances. The impact of this filtering can easily be seen in Level-2 images of epsilon, Angstrom coefficient, and to a lesser extent AOT(865). The value of this smoothing will obviously diminish with increasing spatial and temporal averaging, and is therefore more readily seen as reduced speckling in Level-2 oceanic and atmospheric optical property retrievals. The smoothing is not expected to induce any bias-change in either the aerosol optical thickness or the water-leaving radiances. The aerosol model smoothing resulted in a decrease in the frequency where epsilon was outside the acceptable range, resulting in the reduction of the atmospheric warning flag (ATMWARN).
As the NIR filtering technique showed no significant improvement to the GAC data products, and a potential increase in variance in regions where, due to the GAC subsampling, stray light information is incomplete, it is recommended that the filtering not be applied to GAC resolution data. There is a clear benefit to applying the filter to the LAC resolution data, and it's use is highly recommended for these data.
Fixed aerosol model for very low or negative B7/B8 rho-a
Under very clear atmospheric conditions, the Rayleigh-subtracted radiance in the NIR approaches zero. When other uncertainties are factored-in, the retrieved aerosol path radiances in the NIR may even go slightly negative. The impact of this is that aerosol model selection becomes highly uncertain or the atmospheric correction fails altogether in clear atrmospheric conditions. As a result, the current atmospheric correction algorithm often fails to obtain ocean-color retrievals in the best of atmospheric conditions. A simple solution to this problem is to fix the aerosol type/model when the aerosol path radiance in one or both of the NIR bands approaches zero, and limit the aerosol radiance at 865nm to be greater than or equal to zero. With these two changes, it is possible for the atmospheric correction algorithm to procede when the retrieved aerosol concenteration is effectively zero.
Ignore (Lt - Lr < 0):
In conjunction with the above enhancement, we will no longer consider the occurrence of [(Lt-tLf)/toz - Lr) < 0] in band 2-8 to be grounds for atmospheric correction failure.
The nLw's are computed as band-averages values, and have an out-of-band correction applied prior to use in downstream computations (e.g., chlorophyll retrieval). The current correction, described in Wang et. al (2001 Applied Optics Vol 40, No. 3, pp 343 - 348), is computed from the Band 3/Band 5 ratio, and is based on a chlorophyll-dominated nLw spectrum which uses the Gordon 1988 model. A switch to the Morel and Maritorena (2001) model has been implemented. The figure below shows the comparison between the Gordon(1988)-based and Morel(2001)-based correction factors. The solid lines are for Morel(2001) model results.
The solar irradiance values used through-out the atmospheric correction processing are band-pass averaged quantities. In reprocessing #3, an algorithm was introduced to correct the nLw retrievals to a nominal wavelength. Unfortunately, the out-of-band corrected nLw was still being divided by the band-averaged F0 when computing Rrs. This may be introducing a slight bias in the computation of OC4 chlorophyll. We have corrected this error.