Development of Nonlinear Regression Model to Estimate OLR Based on FY-3/IRAS

OLR (outgoing longwave radiation) is the radiative energy flux the Earth and atmosphere emit out into the outspace, which is one of three components of the Earth and atmosphere radiative budget system, reflecting the climate and weather characteristics. Since the invention of meteorological satellites, OLR products have been processed for more than 40 years. Numerous methods have been developed to estimate OLR from satellite observations, including the relationship between the window channel brightness temperature of AVHRR and the flux equivalent brightness temperature proposed by Arnald Gruber in 1977 and George Ohring in 1984, regression models relating OLR with narrow band fluxes of window channel and water vapour channel of geostationary meteorological satellites developed by Liu in 1988, the linear and none-linear models relating OLR with satellite multi-channel radiances developed by Enllingson in 1994 and Lee in 2010. At the same time, broadband instruments such as ERBE and CERES on board of NOAA, Nimbus, Terra, Aqua are designed to directly observe OLR from outspace. Due to the high quality, CERES OLR products become the best available data to validate other retrieved OLR products.The IRAS (infrared atmospheric sounder) on board of FY-3 polar meteorological satellites carry 26 channels, among which 20 channels are used to observe radiances at the top of the Earth atmosphere at the wavenumber between 669 cm^-1 and 2666 cm^-1.These narrow band radiances have high relations with the full wavenumber radiative flux (OLR) the Earth and atmosphere emit. Therefore, a formula is derived for calculating OLR with multi-channel radiances of IRAS through infrared radiative transfer simulation. Based on radiances at top of atmosphere simulated with LBLRTM (line by line radiative transfer model) software for 2521 atmospheric profiles and statistical regression, a nonlinear model which relates OLR with multi-channel radiances of FY-3/IRAS are developed. By applying the model into FY-3/IRAS L1 data, the global daily mean OLR and monthly mean OLR data in April 2016 are produced. Comparing the IRAS OLR data with the Aqua/CERES and Terra/CERES OLR products, the root mean square error is 7.5 W·m^-2, the correlation coefficient is 0.98, the mean bias is-0.2 W·m^-2 when comparing the IRAS daily mean OLR with that of CERES. The root mean square error is 2.22 W·m^-2, the correlation coefficient is 0.9982, and the mean bias is-0.2 W·m^-2 when comparing the IRAS monthly mean OLR with that of CERES. The accuracy indicates that both the calibration quality of FY-3/IRAS instruments and the OLR retrieval model all achieve at a high level. In addition, OLR retrieval models used by various satellites since 1970 are also reviewed in brief.