Derivation of Regression Coefficients for Sea Surface Temperature Retrieval over East Asia

Among the regression-based algorithms for deriving SST from satellite measurements, regionally optimized algorithms normally perform better than the corresponding global algorithm. In this paper, three algorithms are considered for SST retrieval over the East Asia region (15°–55°N, 105°–170°E), including the multi-channel algorithm (MCSST), the quadratic algorithm (QSST), and the Pathfinder algorithm (PFSST). All algorithms are derived and validated using collocated buoy and Geostationary Meteorological Satellite (GMS-5) observations from 1997 to 2001. An important part of the derivation and validation of the algorithms is the quality control procedure for the buoy SST data and an improved cloud screening method for the satellite brightness temperature measurements. The regionally optimized MCSST algorithm shows an overall improvement over the global algorithm, removing the bias of about 0.13°C and reducing the root-mean-square difference (rmsd) from 1.36°C to 1.26°C. The QSST is only slightly better than the MCSST. For both algorithms, a seasonal dependence of the remaining error statistics is still evident. The Pathfinder approach for deriving a season-specific set of coefficients, one for August to October and one for the rest of the year, provides the smallest rmsd overall that is also stable over time.