Characterizing the uncertainties in spectral remote sensing reflectance for SeaWiFS and MODIS-Aqua based on global in situ matchup data sets

- A global matchup data set is used to assess Rrs accuracy of SeaWiFS and Modis-Aqua.
- Uncertainty metrics are assessed based on optical water type.
- Rrs RMS differences vary with wavelength and optical water types for both sensors.
- The standard deviation contributes more than the bias to RMS.
- The bias, while small, shows a zigzag pattern in both sensors across all water types.

Uncertainties in the spectral remote-sensing reflectance for SeaWiFS and MODIS-Aqua are characterized based on globally distributed data sets of in situ/satellite matchups. The in situ data sets were derived from fixed mooring sites and ship data using a variety of field radiometers. An optical classification procedure was applied to the satellite reflectance from each matchup pair to derive its fuzzy membership to eight previously characterized, optical water types. The memberships were then used as weights to derive uncertainty measures for the optical water types. The resulting uncertainty measures are shown to vary by optical water type for both sensors. Overall, the root-mean-square difference (RMSD) for open ocean waters is low and diminishes with increasing wavelength. The RMSD increases with optical complexity and is the highest in coastal waters associated with highly absorbing or highly scattering conditions. Although it plays a small role in RMSD, a spectral bias with a distinct zigzag pattern was detected across all water types. The most sensitive measure associated with optical water types is the relative uncertainty, which is shown to be near the desired 5% threshold for much of the open ocean for both SeaWiFS and MODIS-Aqua, but exceeds 50% in waters where the reflectance signal is low. This approach provides an objective method for characterizing uncertainty in water environments with distinct optical properties, and the results can be used in the processing of future satellite data to derive their associated uncertainty fields.