Development of fast LLR algorithms to retrieve atmospheric profiles from IRAS/FY3B infrared measurements

This paper addressed the development of fast locally linear regression (LLR) algorithms to retrieve atmospheric temperature and humidity profiles from the measurements of Infrared Atmospheric Sounder (IRAS) on FengYun 3B (FY-3B), in which the matching samples were collected from IRAS/FY-3B measurements and AIRX2RET V5 product. Taking 2011 as example, first, the observation samples were obtained with collocation, the absolute observation time difference less than 15 minutes, and the absolute view zenith angle difference less than 2 degrees. Then, based on the matching samples, the fast LLR algorithms were developed and evaluated in contrast to the LLR algorithm, D matrix algorithm and neural network algorithm. Finally, the fast LLR algorithms were applied to retrieving atmospheric temperature and humidity profiles from IRAS/FY-3B measurements in 2011 and the first quarter of 2012, and then the results were, respectively, validated with the ECMWF reanalysis data, RAOB sounding data and AIRX2RET V5 product. The results indicate that, 1) the fast LLR algorithms is not only fast but also accurate, the retrieving errors for atmospheric temperature and humidity profiles are, respectively, reduced ∼0.8 K and ∼0.5 g/kg in contrast to the D matrix algorithm, and are comparable to the neural network algorithm. 2) the root mean square (rms) errors of the derived atmospheric temperature and humidity profiles in 2011 are, respectively, less than 2.5 K and 2.3 g/kg in contrast the ECMWF reanalysis data, and they are, respectively, 3.5 K and 2.0 g/kg in contrast to the RAOB sounding data. The rms errors of the derived atmospheric temperature and humidity profiles in 2011 are, respectively, 2.5 K and 1.6 g/kg in contrast to AIRX2RET V5 product, which are consistent with the algorithm.