Toward the elimination of bias in satellite retrievals of sea surface temperature 2. Comparison with in situ measurements

We test a new scheme for retrieving skin sea surface temperatures (SSTs) from brightness temperatures observed by the satellite-borne along-track scanning radiometer (ATSR). The main improvement is in the accuracy of SST retrievals using dual-view observations at 11 and 12 mu m under conditions of high optical depth of stratospheric aerosol and high tropospheric water vapor loading. This is demonstrated by comparing 10 are min average ATSR SSTs with in situ measurements of the Tropical Atmosphere Ocean (TAO) array of buoys between September 1991 and May 1992 (when volcanic aerosol was present in the stratosphere after the eruption of Mount Pinatubo in June 1991). The 620 coincidences of ATSR and TAO SSTs are matched within 1 hour and similar to 20 km. The mean and standard deviation of the difference between satellite and in situ SSTs are -0.16 K and 0.37 K, respectively, compared to -1.39 K and 0.59 K, respectively, using the prelaunch scheme. Meteorological measurements recorded with most TAO SSTs allow calculation of skin SSTs, more suitable for comparison with the "radiometric" SSTs derived from ATSR. The bias and standard deviation compared to skin SST are +0.07 K and 0.27 K, respectively. About 5% of ATSR retrievals have errors which we attribute to residual cloud contamination. The robust standard deviation describing the uncontaminated data is 0.19 K. The aerosol-robustness of the new scheme is evident in the absence of any trend in SST error corresponding to the decay of the stratospheric aerosol loading (0.00 +/- 0.16 K yr(-1)). The new scheme is shown to be unbiased by water vapor loadings between 20 and 60 kg m(-2). The effects of matchup and buoy error and either skin-bulk variability or skin-bulk adjustment error contribute to the standard deviations given above, suggesting that the random retrieval errors intrinsic to use of the new coefficients are smaller still.