Ensemble Retrieval of Atmospheric Temperature Profiles from AIRS

ABSTRACT Satellite-based observations provide great opportunities for improving weather forecasting. Physical retrieval of atmo spheric profiles from satellite observations is sensitive to the uncertainty of the first guess and other factors. In order to improve the accuracy of the physical retrieval, an ensemble methodology was developed with an emphasis on perturbing the first guess. In the methodology, a normal probability density function （PDF） is used to select the optimal profile from the ensemble retrievals. The ensemble retrieval algorithm contains four steps： （1） regression retrieval for original first guess; （2） perturbation of the original first guess to generate new first guesses （ensemble first guesses）; （3） using the ensemble first guesses and nonlinear iterative physical retrieval to generate ensemble physical results; and （4） the final optimal profile is selected from the ensemble physical results by using PDE Temperature eigenvectors （EVs） were used to generate the pertur- bation and generate the ensemble first guess. Compared with the regular temperature profile retrievals from the Atmospheric InfraRed Sounder （AIRS）, the ensemble retrievals RMSE of temperature profiles selected by the PDF was reduced between 150 and 320 hPa and below 400 hPa, with a maximum improvement of 0.3 K at 400 hPa. The bias was also reduced in many layers, with a maximum improvement of 0.69 K at 460 hPa. The combined optimal （CombOpt） profile and a mean optimal （MeanOpt） profile of all ensemble physical results were improved below 150 hPa. The MeanOpt profile was better than the CombOpt profile, and was regarded as the final optimal （FinOpt） profile. This study lays the foundation for improving temperature retrievals from hyper-spectral infrared radiance measurements.

Ensemble retrieval of atmospheric temperature profiles from AIRS

Satellite-based observations provide great opportunities for improving weather forecasting. Physical retrieval of atmospheric profiles from satellite observations is sensitive to the uncertainty of the first guess and other factors. In order to improve the accuracy of the physical retrieval, an ensemble methodology was developed with an emphasis on perturbing the first guess. In the methodology, a normal probability density function (PDF) is used to select the optimal profile from the ensemble retrievals. The ensemble retrieval algorithm contains four steps: (1) regression retrieval for original first guess; (2) perturbation of the original first guess to generate new first guesses (ensemble first guesses); (3) using the ensemble first guesses and nonlinear iterative physical retrieval to generate ensemble physical results; and (4) the final optimal profile is selected from the ensemble physical results by using PDF. Temperature eigenvectors (EVs) were used to generate the perturbation and generate the ensemble first guess. Compared with the regular temperature profile retrievals from the Atmospheric InfraRed Sounder (AIRS), the ensemble retrievals RMSE of temperature profiles selected by the PDF was reduced between 150 and 320 hPa and below 400 hPa, with a maximum improvement of 0.3 K at 400 hPa. The bias was also reduced in many layers, with a maximum improvement of 0.69 K at 460 hPa. The combined optimal (CombOpt) profile and a mean optimal (MeanOpt) profile of all ensemble physical results were improved below 150 hPa. The MeanOpt profile was better than the CombOpt profile, and was regarded as the final optimal (FinOpt) profile. This study lays the foundation for improving temperature retrievals from hyper-spectral infrared radiance measurements.