Impact of sea ice on the retrieval of water-leaving reflectance, chlorophyll a concentration and inherent optical properties from satellite ocean color data

Two physical phenomena by which satellite remotely sensed ocean color data are contaminated by sea ice at high latitudes are described through simulations and observations: (1) the adjacency effect that occurs along sea ice margins and (2) the sub-pixel contamination by a small amount of sea ice within an ocean pixel. The signal at the top of the atmosphere (TOA) was simulated using the 6S radiative transfer code that allows modeling of the adjacency effect for various types of sea ice surrounding an open water area. In situ sea ice reflectance spectra used in the simulations were measured prior to and during the melt period as part of the 2004 Canadian Arctic Shelf Exchange Study (CASES). For sub-pixel contamination, the TOA signal was simulated for various surface reflectances obtained by linear mixture of both sea ice and water-leaving reflectances (rho(w)). The standard atmospheric correction algorithm was then applied to the simulated TOA spectra to retrieve rho(w) spectra from which chlorophyll a concentrations (CHL) and inherent optical properties (IOps) were derived. The adjacency effect was associated with large errors (> 0.002) in the retrieval of rho(w) as far as 24 km from an ice edge in the blue part of the spectrum (443 nm). Therefore, for moderate to high CHL (> 0.5 mg m(-3)), any pixel located within a distance of similar to 10-20 km from the ice edge were unreliable. It was also found necessary to consider the adjacency effect when the total absorption coefficient (a(t)) was to be retrieved using a semi-analytical algorithm. a(t)(443) was underestimated by more than 35% at a distance of 20 km from an ice edge for CHL > 0.5 mg m(-3). The effect on the retrieval of the particle backscattering coefficient (b(bp)) was important only for clear waters (CHL similar to 0.05 mg m(-3)). In contrast, sub-pixel contamination by a small amount of sea ice produced systematic underestimation of rho(w) in the blue because of incorrect interpretation of enhanced reflectance in the near infrared that is attributed to higher concentrations of atmospheric aerosols. In general, sub-pixel contamination was found to result in overestimations of CHL and at, and underestimations of bbp. A simple method was proposed to flag pixels contaminated by adjacency effect. (c) 2007 Elsevier Inc. All rights reserved.