Definition of a radiosounding database for sea surface brightness temperature simulations: Application to sea surface temperature retrieval algorithm determination

Abstract

The Ocean and Sea Ice Satellite Application Facility (O&SI SAF) project includes the development of Sea Surface Temperature (SST) retrieval algorithms for infrared radiometers. The algorithm definition stage presented in this paper includes three steps: building of representative simulated brightness temperature databases, definition and analysis of formalisms on simulated brightness temperatures, and tests of these formalisms on a Matchup Database (MDB). A cloud-free, latitude equally distributed simulation database has been built to determine SST algorithms, which were further checked on a NOAA-14 latitude equally distributed validation database build from the Pathfinder MDBs. The tests showed that the presence of clouds in the training database increases the bias, while an underrepresentation of low-latitude profiles in the training database increases the standard deviation. The use of a cloud-free, latitude equally distributed training database reduced both the standard deviation in the algorithms and the bias consistently over the whole range of latitudes. The effect of the instrumental noise has been assessed and algorithms using noisy simulated data have been tested: they appear much more useful at daytime than at nighttime. Indications suggest that such noise algorithms should be more adaptive in atypical situations and that they should perform better than noise-free algorithms in the presence of cloud contamination or aerosols. The systematic procedure allowed us to test a large number of different algorithms. We determined that the best possible theoretical accuracy that may be obtained with NOAA-14 derived SSTs against buoy measurements on a global scale is 0.4 K. The successive steps that were performed in this study (cloud-free training database, algorithm determination, and noise scheme) allowed us to obtain retrievals with a 0.57 K standard deviation at both daytime and nighttime. These performances appears to be comparable or better than those obtained with algorithms directly derived by regression against in situ measurements. This proves that simulation-based SST retrievals are feasible and accurate when sufficient care is taken in the constitution of the training database. The effect of maritime aerosols has also been assessed and was found to be generally of little consequences for SST retrievals: a positive bias found to be lower than 0.08 K and no effect on the standard deviation. This bias is overestimated when maritime aerosols are used in the training of the algorithms using radiative transfer models. Current SST products (North Atlantic Regional and high latitudes) based on NOAA-14 and NOAA-16 data and routine validations against drifting buoys are available at 〈 http://www.meteorologie.eu.org/safo〉.