Accurate retrieval of water leaving radiances from ocean color satellite observations, such as SeaWIFS or MODIS, requires accurate correction algorithms to compensate for atmospheric impacts. For open ocean conditions, an atmospheric correction scheme was developed where the aerosol contribution was estimated using top of atmosphere (TOA) reflectances obtained from the SeaWIFS 765-865 (or MODIS 748-869) NIR channels under the reasonable assumption that the water leaving radiance at these wavelengths is negligible (black pixel approximation) due to strong water absorption. This atmospheric correction algorithm works well in open ocean waters, but tends to over correct for the atmosphere in coastal waters where the black pixel assumption is no longer valid due to the increased reflection from hydrosol scattering. Thus, if the water leaving radiance is not negligible in the NIR bands, the retrieved aerosol loading will be overestimated resulting in underestimated or even negative water leaving radiances for the 412nm channel

Fig. 1  Comparison of  insitu measurements of normalized water leaving reflectance of SWIR retrieval using standard processing and regional model.

An algorithm which uses the 412nm water reflectance obtained statistically in the Chesapeake Bay as a constraint for aerosol correction of coastal waters has been developed and preliminary matchup of retrievals with insitu remote sensing reflectance and aerosol optical depth have been performed. In particular, we compare in figure 1, the remote sensing reflectance at 443nm is obtained from our algorithm with the standard SWIR model as well as the NIR algorithm (when possible). Although neither retrieval approach seems completely satisfactory, we do note the removal of negative retrievals in our approach which occur in the standard processing due to the presence of a moderately absorbing aerosol modes. This undesirable feature is also seen when the NIR algorithm is used. On the other hand, we also see significant improvement using our constrained regional algorithm retrieval to insitu data for matchup sites 11-17 in comparison to the SWIR algorithm. These matchups occur in the coastal ocean outside the Chesapeake Bay area where the NIR algorithm is expected to be superior to the SWIR which is indeed the case. However, even under these difficult retrieval conditions, the 412nm constraints seem to be sufficient to improve the retrieval to the level of the NIR algorithm. More matchups over different regions are needed to understand the performance improvement and much work is needed to attempt to study this algorithm within the operational SeaDAS environment.

Fig. 2. Rrs reflectance comparisons a) constrained retrieval at 443nm b) SeaDAS retrieval at 443nm c) constrained retrieval at 551nm d) SeaDAS retrieval at 551nm 

The most interesting comparisons between SeaDAS SWIR retrievals of the remote sensing reflectance for 443nm and 551 nm and our regionally constrained approach are made in fig 2. We first note (panel b) the dramatic anomalous negative Rrs at 443 nm from SeaDAS together with the fact that that the effect is dramatically stratified between the west (negative) and east (positive) shorelines. The constrained regional retrieval on the other hand removes the anomalous negative Rrs values and the unphysical stratification seems to be significantly removed. The results at 551nm are less dramatic since the effects of aerosol diversity is less pronounced and the overall water leaving signal in most cases is significantly larger than at 443nm.

However, it is important to point out that our constrained regional retrieval does not find a suitable solution in all cases where SeaDAS retrieval was successful. In particular, the narrow western tributaries as well as a gap near the center of the bay were unsuitable for retrieval. In examining these regions, looking at the TOA reflectances (figure 16), it is clear that retrievals would be grossly inaccurate since these regions have anomalous bright reflectance patches due to either inadequate cloud mask or in the case of the tributaries, to possible ground reflectance contamination from the surrounding shore. The fact that these questionable regions are being processed by SeaDAS is evident in dramatic anomalies in the retrieved Rrs values as well as the aerosol optical depth as seen in fig 18. Interestingly, we note the same east-west stratification in the retrieved AOD retrieval which is again eliminated in our approach. In fact, the overestimation of AOD for the western bank seems to be highly correlated to the underestimation of the Rrs. 

Acknowledgements

This work is partially supported under grants  NOAA #NA17AE1625 and ONR N00014-08-1-0325