Estimation of sea surface temperature using the AVHRR mid-wave IR band

We describe a method for estimating sea surface temperature (SST) using MWIR band data from the AVHRR polar orbiter. Currently, SST is routinely calculated with a split-window, nonlinear multichannel algorithm incorporating data from AVHRR Channels 4 and 5 (10.3-11.3 and 11.5-12.5 /spl mu/m, respectively). The accuracy of these results is dependent to a certain degree upon regional variations and is inherently limited by the spatial resolution of the measurements. Nevertheless, these SST maps are generally considered reliable, and are widely used for studying ocean currents and their effect on weather patterns. We are interested, however, in testing the feasibility of using MWIR data in the absence of LWIR measurements for estimating SST both at night, when reflected solar radiance is not an issue, as well as during the day, when it is. A MWIR SST algorithm of the type we discuss would be using data, for example, from a satellite without LWIR capabilities in order to calculate a parameter that is ancillary to the satellite mission (but which is nevertheless of high interest). The SST algorithms we describe are based upon the comparison of MODTRAN ocean radiance values, at a variety of surface temperatures and calculated over the aforementioned AVHRR bands, to the values of the collected pixels in these bands. These MODTRAN calculations are scene-specific, as viewing angle and atmospheric conditions are important input parameters. MODTRAN is therefore launched from within the main SST program architecture for a range of different temperatures. The results of such calculations could conceivably be implemented, however, as a look-up table for a grid of LZAs, standard atmospheres and temperatures. Before the temperature of the pixels can be assessed, the scene must be screened for clouds, which tend to lower the temperature estimation for contaminated pixels. We accomplish this screening using our CloudDI algorithm, a modified least squares template-matching approach. Finally, we test the validity of our results against the AVHRR SST algorithms as well as against available ground truth. Since the MODTRAN calculations require sensor geometry and atmospheric conditions as input parameters, it is possible, in theory, to correct for the effect of high levels of water vapor on the SST results in certain situations.