Uncertainties of SeaWiFS and MODIS remote sensing reflectance: Implications from clear water measurements

A fundamental parameter derived from satellite ocean color measurements is the spectral remote sensing reflectance, R-rs(lambda) (sr(-1)), which is used as the input to all inversion algorithms to derive bio-optical properties (e.g., chlorophyll-a concentration or Chl in mg m(-3)) and water's inherent optical properties (IOPs). The accuracy and uncertainties of the satellite-derived R-rs have only been assessed through comparisons with in situ measurements that were often limited in both space and time. Here, a novel approach was developed and used to estimate R-rs uncertainties from SeaWiFS and MODIS/Aqua (MODISA) measurements over clear waters. The study focused on two oligotrophic ocean gyres in the North Atlantic and South Pacific, and used a recently developed new Chl algorithm to provide a constraint to determine the highest-quality R-rs data with minimal errors. These data were used as surrogates of ground truth or references (termed as R-rs,R-true) to estimate the R-rs error in each data point, with uncertainty estimates (in both relative and absolute forms) generated from statistical analyses. The study led to several findings: One, both SeaWiFS and MODISA have met their mission goals of achieving R-rs uncertainties and absolute accuracy (assuming that the R-rs,R-true values can represent the truth) to within 5% for blue bands and blue waters. As a comparison, nearly all previous in situ-based validation efforts reported mean (or median) percentage differences exceeding 10% between in situ and satellite R-rs in the blue bands. Two, for the green bands, R-rs uncertainties are significantly higher, often in the range of 10-15% for oligotrophic waters. Three, SeaWiFS R-rs uncertainties are generally higher than those of MODISA, possibly due to its lower signal-to-noise ratio (SNR). Four, all R-rs errors are spectrally related in a monotonous way from the blue to the red wavelengths, suggesting that these errors are resulted primarily from the imperfect atmospheric correction algorithms as opposed to sensor noise or vicarious calibration. Such empirical relationships are shown to be useful in reducing the R-rs uncertainties for the North Atlantic Gyre and may also be useful for most of the ocean waters. Finally, the tabulated results provide lower bounds of R-rs(lambda) uncertainties for more productive waters. The findings may serve as references for future ocean color missions, and they have also significant implications for uncertainty estimates of other ocean color data products. (c) 2013 Elsevier Inc. All rights reserved.