Abstract
Sea surface temperature (SST) retrievals from satellite imager measurements are often performed using only two or three channels, and employ a regression methodology. As there are 16 thermal infrared (IR) channels available for MODIS, we demonstrate a new SST retrieval methodology using more channels and a physically deterministic method, the modified total least squares (MTLS), to improve the quality of SST. Since cloud detection is always a part of any parameter estimation from IR satellite measurements, we hereby extend our recently-published novel cloud detection technique, which is based on both functional spectral differences and radiative transfer modeling for GOES-13. We demonstrate that the cloud detection coefficients derived for GOES-13 are working well for MODIS, while further improvements are made possible by the extra channels replacing some of the previous tests. The results are compared with available operational MODIS SST through the Group for High Resolution SST website–the data themselves are originally processed by the NASA Goddard Ocean Biology Processing Group. It is observed the data coverage can be more than doubled compared to the currently-available operational product, and at the same time the quality can be improved significantly. Two other SST retrieval methods, offline-calculated coefficients using the same form of the operational regression equation, and radiative transfer based optimal estimation, are included for comparison purposes.
Highlights
Most operational sea surface temperature (SST) retrievals from satellite measurements are still performed using regression-based methods and, over the years, there has been only limited progress.Such approaches were justifiable in the interest of time and computational resources when they were formulated (e.g., [1,2])
Cloud mask data, we use the subset of pixels, which has the quality level of 3 and above from the operational database as representative of the total cloud-free pixels obtained from the MODIS cloud (MC) algorithm for the comparison of our cloud scheme
This is based on a single-channel retrieval of total column water vapor (TCWV) from the 11 μm measurement, using a linear assumption, and rtv3.9 as we described earlier: Figure 2a,b shows that 145 good measurements can be saved and additional 1844 cloudy pixels can be discarded by using the new double difference test instead of the previous one
Summary
Most operational sea surface temperature (SST) retrievals from satellite measurements are still performed using regression-based methods and, over the years, there has been only limited progress Such approaches were justifiable in the interest of time and computational resources when they were formulated (e.g., [1,2]). Most such operational forward model-based methods ( SST) are stochastic approaches, which are based on Bayesian probability theory, or the related one-dimensional variational principle Such stochastic techniques differ from each other both in the procedure for solving a set of spectrally-independent radiative transfer (RT) equations (e.g., matrix inversion, numerical iteration) and in the choice of ancillary data. We will show in this paper that good retrievals can be achieved using a deterministic method when the truth is far from the initial guess
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
