Expected Improvements in Atmospheric Correction of Ocean Color Imagery Over Case 1 Waters

Atmospheric correction is a crucial and difficult step in the processing of ocean color satellite imagery because of the high variability in space and time of the concentrations in small particles in the atmosphere. Interestingly, the basic principle for atmospheric correction has not changed since the proof-of-concept CZCS in the late 70's. Even for the most recent and high-performance ocean color such as MODIS and MERIS, the aerosol contribution is estimated from the near-infrared (NIR) and is then extrapolated toward the visible to correct the measured radiances. This technique has two major drawbacks: (1) Errors in aerosol characterization in the NIR are amplified once propagated in the blue-green; (2) This technique is not capable of distinguishing between weakly and strongly absorbing aerosols because aerosol absorption is much more efficient in the blue-green than in the NIR. An alternative approach for atmospheric correction is to use a model of the water- leaving radiance together with a model of the aerosol radiance to determine which set of aerosol (optical depth, single scattering albedo...) and marine (chlorophyll, CDOM...) parameters leads to the best match between computed and measured radiances over the whole (visible + NIR) spectrum. Two different techniques were developed: the Spectral Optimization Algorithm (SOA) and the Spectral Matching Algorithm (SMA). Both algorithms were tested with SEAWIFS imagery and were shown to strongly improve ocean color estimates for case 1 waters in the presence of pollution aerosols (SOA) and Saharan dust (SMA). Such algorithms should also be applicable to case 2 waters provided that accurate models of the water-leaving radiance are available. These algorithms however require a good knowledge on the aerosol properties. It is likely that future atmospheric corrections will not be "universal" anymore and will use different aerosol data set depending on the oceanic region observed by the sensor. Another difficulty in using both SOA and SMA is that they require much more computer time than the standard atmospheric correction. The use of advanced programming techniques based, for example, on neural networks is now attempted to speed up the processing. Improved aerosol models and more efficient computations are certainly the key parameters toward a new generation of operational atmospheric correction algorithms.