高光谱大气红外探测器(Atmospheric Infrared Sounder，AIRS)主要覆盖CO2和H2O吸收带光谱区.区别于CO2通道，H2O通道亮温偏差非高斯性较强.为了充分有效地利用AIRS通道光谱信息，本文采用两种新算法开展应用研究，一是基于变分同化后验估计－观测误差重估计重新估算光谱通道误差，以更好地“符合”光谱亮温对变分同化目标泛函的权值分配；二是将M—估计法(L2—估计、Huber—估计、Fair—估计和Cauchy—估计)权重函数耦合到经典变分同化目标泛函中，得到广义变分同化目标泛函，使其具有非高斯性，其核心是在每次极小化迭代过程中重新估计观测项对目标泛函贡献率.在新算法研究基础上开展高光谱AIRS模拟亮温试验，结果表明观测误差重估计和Huber—估计广义变分同化AIRS资料效果优于经典变分同化.并基于信号自由度(Degrees of freedom for signal，DFS)开展观测资料对分析场影响诊断，得到该两种方法在同化过程中能够提高H2O通道亮温使用的信息量.通过对文中算法(观测误差重估计和Huber—估计)得到的分析场与探空资料温度场对比分析，得到Huber－估计广义尺度设定为1.345 K时效果最好，整体误差最小，2.5K次之，且观测误差重估计也优于经典变分同化结果.200~750 hPa效果较为显著，基于Huber－估计广义同化在对流层顶表面和周围(80~200 hPa)温度反演小于2 K.研究结果可为我国风云四号A星和风云三号D星高光谱资料变分同化提供新的方法思路和技术支撑.

Hyper-spectral Atmospheric Infrared Sounder (AIRS) mainly covers the CO2 and H2O absorption bands. Different from CO2 channels, the brightness temperature bias of water vapor channel follows non-Gaussian statistics. In order to use AIRS channel spectral information effectively, new algorithm research is needed, two methods are presented in this paper: (1)Different from the observation error of the given spectral channel remains unchanged during the classical variational assimilation minimization iteration, the paper based on the posterior estimate of variational assimilation, namely, observation error re-estimation, re-estimating the channel observation error, which is then regarded as the weight of observation to the objective function of classical variational assimilation; Observation error re-estimation can be used to identify the reasonable observation errors which can fit variational assimilation model better. By using the weight function of M-estimators (L2-estimator, Huber-estimator, Fair-estimator and Cauchy-estimator) to couple the classical variational assimilation, and then obtain the generalized variational assimilation, make it Non-Gaussian. Re-estimated the contribution rate of observation terms to the objective function during each minimization iteration. The simulated brightness temperatures of AIRS are used to conduct ideal experiments. It is shown that two methods of observation error re-estimation and Huber-estimator can provide better results than the classical method. We diagnose the impact of observations on the analysis with degrees of freedom for signal (DFS). The result of diagnosis shows that two methods can increase the available information of brightness temperatures of water vapour channels during the assimilation process. Furthermore, the analysis field obtained by using the algorithm (observation error re-estimation and Huber-estimator) in this paper is compared with the temperature field of sounding data, and it is obtained that the Huber-estimator, which generalized scale is set as 1.345 K with the best effect, which is set as 2.5 K latter, and the observation error re-estimation is better than classical variational assimilation. The effect of 200~750 hPa was relatively significant. The retrieval temperature at the surface and around the tropopause (80~200 hPa)is less than 2 K based on Huber-estimator variational assimilation. The results of this paper can lay the theoretical foundation and provide the algorithm reference for the variational assimilation of hyper-spectral data of Feng-Yun 4A and Feng-Yun 3D satellite.