Evaluation of the Naval Research Laboratory algorithm for identifying clouds in ocean color data

The performance of an algorithm which identifies and removes clouds from ocean color satellite data over the ocean is evaluated. The evaluations were accomplished with archived sets of Coastal Zone Color Scanner (CZCS) and simulated SeaWiFS data. The simulated SeaWiFS sets were produced by spectrally and spatially convoluting the data of the Airborne Visible-Infrared Imaging Spectrometer to fit the spectral responses of the SeaWiFS channels and achieve the same spatial resolution of the SeaWiFS instrument. This study also included evaluations of the effect that changes in spatial resolution have on the texture computations of the algorithm. Convolution procedures were used to generate images at various spatial resolutions: 20 m, 50 m, 100 m, 500 m and 1 km. Results from the investigations involving CZCS Channel 5 (720 - 800 nm) data indicate that the algorithm performs well in all the oceanic data sets tested but not in the coastal imagery. Cloud masking in the coastal sets was clearly affected by bottom reflectances and river run off. In the evaluations involving simulated SeaWiFS Channel 8 (845 - 885 nm) data, the algorithm performed accurately every time and was not affected by coastal or oceanic features. It began to confuse cloud shadow edges with cloud edges at resolutions of 30 m or less. From the results obtained, it appears that this algorithm has the potential to identify clouds not only in SeaWiFS data, but also in data from high resolution color sensors.