Real-time estimates of sea surface temperature reliability in an operational production environment

The Naval Oceanographic Office (NAVOCEANO) recently started to operationally provide quantitative reliability estimates with all sea surface temperature (SST) values derived from NOAA-16 data. Reliability estimates, in the form of root-mean-square (RMS) error estimates, are assigned by first classifying the data as one of three categories: clear, probably clear, or questionable. These categories were determined by analyzing three months of SST data matched to buoys. From that analysis we determined which parameters were most important in discriminating between classes. We chose as the first parameter the difference between the produced SST and a combination of climatology and of an interpolated 100-km SST field. The second parameter is the inter-comparison test: the difference between the SST from the operational equation and that from a second equation. The third parameter uses a pseudoprobability of sun-glint based on a combination of the solar zenith, solar azimuth, and satellite zenith angles. Several other parameters were surprisingly found not to significantly affect reliability estimates. The thresholds for the first, second, and third parameters, as well as the initial RMS values associated with each class, were derived from the initial three months of SST data matched to buoys. Reliability estimates have been monitored daily since they were added to the operational processing. During that period RMS error values have hovered around 0.45degreesC and 0.70degreesC for categories I and 2, respectively. Category 3 is more erratic because of the small number of samples in that class. We observed that about 90 percent of the daytime samples were classified as clear, while less than 10 percent were classified as probably clear. About 98 percent and 2 percent of nighttime samples were similarly classified as clear and probably clear.