Sea surface temperatures (SST) have been operationally retrieved from AVHRR onboard NOAA and later MetOp satellites for over two decades. Customarily, SST accuracy and precision are evaluated by one global mean bias and one root mean square deviation (RMSD) of retrieved minus in-situ SST. For the purpose of quality control, SST in each pixel may further contain quality flags or confidence levels. The Global High-Resolution SST (GHRSST) community has become increasingly interested in supplying error estimates in each individual pixel since such information is critically important for blending different SST products. For SST developers, these estimates may be useful in identifying potential areas for improvement in the SST product. For SST users, they can provide comprehensive quality information of SST products and facilitate decisions on the domains of data suitable for their particular applications.

This study examines the new SST products available from the AVHRR Clear-Sky Processor for Oceans (ACSPO) recently developed at NOAA/NESDIS. Analyses were conducted with several weeks of global data from NOAA-17, -18, and MetOp-A data. The retrieved SST was referenced with respect to the Optimum Interpolation (OI v.2) SST analysis, and SST anomaly analyzed as a function of atmospheric and surface information available from the NCEP Global Forecast System (GFS) data, saved on ACSPO granules. Pronounced dependencies of SST anomaly on the number of ambient clear-sky pixels, column water vapor, and view angle were observed, and analytical fit functions, based on physical considerations, were tested to approximate the bias dependencies in the multidimensional retrieval space. To minimize the effect of outliers, robust least-square method was adopted in fitting. An independent evaluation using a training and test samples have shown an RMSD reduction from ~0.6 to 0.5K. In addition, cross-platform consistency was examined and stability monitored over time.

In future studies, such factors as the air surface temperature difference, wind speed, and aerosol should also be taken into consideration. RMSD dependencies will be analyzed and fitted in the same manner as the bias considered in this study. Different reference SST, including in-situ SST, will be used to minimize the effect of possible errors in the reference fields on SST error characteristics.

Keywords – Error characterization, Sea Surface Temperature, ACSPO