An improved regression method for the retrieval of trace gas profiles from ultra-spectral infrared data

With tens of thousands of channels, infrared ultra-spectral data are expected to improve the accuracy of trace gas profiles, especially carbon monoxide (CO), which possesses a weak adsorption intensity and strong interference signals at the adsorption band. The greatly increased number of channels will generate a considerable amount of redundant information. It is necessary to develop a new method to exclude redundant information, thereby enhancing the retrieval accuracy of ultra-spectral data. In this paper, regarding the ill-posed nature induced by these interference and strong correlation of multi-factors, an improved statistical regression retrieval method is proposed. The ratios of the retrieved gas signal to the interfering signals and the noise equivalent temperature differences (NE Delta T) are first analyzed as the weighting factors for the selected channels. By applying the weighting factors in the retrieval process, the proposed method amplifies the contribution of the channels whose information content are more efficient. The proposed method is assessed by the application on the retrieval CO profiles from the simulated ultra-spectral data. The result shows that the root mean square errors (RMSE) of the proposed method for CO profiles is smaller than traditional statistical regression method, and the accuracy is improved by 2.85%.