NASA has been providing global sea-surface temperatures (SST) from MODIS on daily to decadal periods, and these are extensively used for a wide range of atmospheric and oceanic studies. However, the retrieval quality and cloud detection are somewhat problematic. We will present a new physical deterministic algorithm based on truncated total least squares (TTLS) using multiple channels for SST retrieval from MODIS measurements in conjunction with a new cloud detection scheme using a radiative transfer model atop a functional spectral difference method. The TTLS method is a new addition to improve the information content of the retrieval compared to our earlier work using modified total least squares (MTLS). A systematic study is conducted to ascertain the optimal channel selection from the 16 channels in the thermal IR of MODIS. Our new algorithm can reduce average RMSE of ~50% while increasing the average data coverage by ~50% compared to the operationally available MODIS SST.
The MODIS advanced sensor contains 16 channels in the thermal infrared band, makes it an attractive instrument for atmospheric and oceanic sciences. Even for satellite-derived sea surface temperature (SST) retrievals, the dynamics of atmospheric conditions are intended to be characterized by the satellite measurement sufficiently to retrieve good quality SST. The Group for High Resolution SST (GHRSST), which is an international scientific body, provides MODIS SST to date using only two and/or three channels by employing regression method. The few coefficients used in regression based retrieval methods are unable to compensate for wide atmospheric variability and as a result, significant errors are embedded in the retrieved SST. We will demonstrate in this work that the MODIS SST can be retrieved with approximately double the accuracy compared GHRSST operational SST, by using more channels and our physical deterministic-based modified total least squares (MTLS) method. This study also includes the SST4/NLSST and optimal estimation based SST retrieval for comparison purposes. The information content and noise analysis of these retrievals, and the retrieval error due to the quality of cloud detection is discussed.