The CM module is built upon the CLAVRx and mainly based on general knowledge of cloud emission and reflection. The output of the CM is a Cloud Flag (CF), which classifies each ocean pixel into 2 major categories with 2 sub-categories: clear (confidently and probably) and cloudy (probably and confidently). The CM is a fully autonomous module of ACSPO, outside and upstream of the SST module. The major objective of the CM is to reliably identify the two cloudy subcategories, which will not be considered any further in the SST algorithm.
The SST module of ACSPO version 2 contains two SST retrieval algorithms: regression and physical inversion, which are applied to all valid ocean pixels irrespective of CF value. Regression algorithms are based on well-established split- and triple-window techniques. The inversion retrieval, based on using the Community Radiative Transfer Model (CRTM) in conjunction with Global Forecast System (GFS) atmospheric profiles and Reynolds SST, is relatively new. It is expected to become primary SST algorithm, after completing a series of adjustments and sensitivity studies, whereas the regression SST will continue to be used as a fall-back. While the regression algorithm retrieves SST only, the inversion algorithm produces two additional quantities: (1) RMS residual of fitting TOA BT with CRTM; and (2) an estimate of the atmospheric optical depth (OD).
The QC module further examines the output of the SST module. Specific tests depend upon what SST algorithm was run.
If only regression SST is available, then only one QC test is possible
o The adaptive SST test checks if the retrieved SST belongs to a clear-sky cluster. Typically, cloud causes retrieved SST to be biased cold with respect to the first-guess SST. However, the boundary between cloud bias and a real SST anomaly from a first-guess is fuzzy. The threshold is thus set for each pixel dynamically within a surrounding sliding window. Initially, the line between the “clear” and “cloudy” clusters is drawn far enough from the first-guess SST, and then it is iteratively adjusted by analyzing statistics of retrieved SST in both clusters.
If physical inversion algorithm is available, then the following two QC tests are used
o The BT test checks measured BTs for consistency with simulated clear-sky BT using RMS BT residual as a predictor.
o The OD test uses the retrieved OD as a predictor. Analyses have shown a very high negative correlation between retrieved OD and SST deviation from first-guess, in the vicinity of cloudy pixels, identified with the CM tests. The threshold in this test is set to minimize the effect of ambient cloud on retrieved SST.
Note that the adaptive SST test can be also employed for inversion SSTs, although its information content is greatly reduced here by the BT and OD tests. Also, the availability of CRTM and GFS allows one to substitute the regression SST into CRTM and check if the simulated BTs are consistent with measured BTs. The BT test proved a valuable test for the quality of the regression SST, in the earlier version of ACSPO, although again its efficacy is greatly diminished compared to the physical inversions, where the CRTM clear-sky BT fit the measured BTs much more accurately.
Based on the QC results, some pixels, originally classified clear by CM, can be flagged as unusable for SST. Also, some pixels initially classified as “cloudy” (confidently or probably), can be “restored” into a “clear” category. This information is used to adjust the CM, which objective is to minimize the number of such “false alarms”. In addition the QC provides a continuous measure of SST quality, such as RMS BT residuals and, optionally, OD estimates.
In this study, the above methodology is tested with NOAA and MetOp-A AVHRR GAC (4 km) and FRAC (1 km), and MSG/SEVIRI data. Performance of the ACSPO CM and QC modules is evaluated based on the analyses of instantaneous SST images and global statistics of the SST anomalies (RSST minus reference SST), including the fraction of clear-sky pixels over oceans, mean bias and RMSD. Upon extensive testing with these heritage sensors, the methodology presented here is also planned to be applied for clear-sky ocean products generation from NPOESS/VIIRS and GOES-R/ABI, under the NOAA's NPOESS Data Exploitation Program and GOES-R Algorithm Working Group.