同化卫星资料能够得到模式较好的初始场,目前资料变分同化是基于误差服从高斯分布这一理论,因此在同化高光谱大气红外探测器(AIRS)资料前,必须进行资料质量控制。从通道选择、异常值剔除、偏差订正、云检测和数据稀疏化5个方面对AIRS资料质量控制研究现状进行分析与讨论。归纳总结了基于信息熵分步迭代法、主成分累计影响系数法和主成分—逐步回归法3种通道选择方法。经分析比较认为信息熵分步迭代法使用得较为广泛,但所选通道之间存在“弱相关”;主成分—逐步回归法能够获得信息量较大的通道组合,由于算法的原因,执行过程较耗时。探讨了莱茵达法则和稳健性较强的双权重法对异常离群值剔除,得到双权重法效果较好。介绍了离线和在线偏差订正方法,包括静态、自适应、回归法、变分、基于辐射传输模式、基于卡尔曼滤波偏差订正法和偏差订正的动态更新技术。对比发现静态法时效性较好;变分法能够解决数据漂移等问题;基于模式和卡尔曼法虽效果较好,但较耗时不适合业务化使用;综合而言,偏差订正动态更新技术的效果和时效性都较好。分析了晴空视场点、晴空通道、云辐射订正和不同仪器云产品的匹配4种云检测方法。从数值业务时效性角度出发,晴空视场点和晴空通道云检测法较为可行,但经过晴空视场点云检测后同化的资料量比晴空通道法少,会造成在气象敏感区如高层通道资料的丢弃,在一定程度上会影响分析场的质量。进一步分析了跳点跳线、box法和主成分分析法在AIRS资料稀疏化中的初步应用研究。从同化时效性和可操作性出发,得出box法可行;主成分分析法算法复杂度较高,但具有一定的应用前景。在综述质量控制部分基础上,进一步给出了该领域未来的相关研究方向。

Satellite data assimilation can provide accurate initial field for Numerical Weather Prediction (NWP) models. So far, data variational assimilation is based on the theory where error obeys Gaussian distribution, so as to apply the least square method. During classical variational assimilation, if the data contain outliers, the results of optimal parameter estimation is meaningless. Therefore, quality control is quite necessary for Atmospheric Infrared Sounder (AIRS) data before data assimilation. This paper made a comment of the advances in the quality control using AIRS data, which analyzed and discussed the research status from five aspects: channel selection, outliers elimination, bias correction, cloud detection and data sparseness. Three methods for channel selection were summarized, which are stepwise iterative method based on information entropy, the cumulative effect coefficient of principal component and principal components—Stepwise regression, respectively. Comparatively, stepwise iterative method based on information entropy is more widely used, but the selected channels are weak related; Channel combination with large amount of information can be obtained through the method of principal components—stepwise regression, but the implementation process is time-consuming due to the algorithm. Both the lane of law and the double weight method were used in outliers elimination, and the result shows that the latter one is better. Two kinds of bias correction method including off-line and on-line, were introduced, which contain static, adaptive, regression method, variational, method based on the radiative transfer model, bias correction with Kalman filter and dynamic update of bias correction technique. It is found that the timeliness of static method is better; while variational method could solve the problems of data drift and so on. The result is better when using bias correction based on the model and Kalman methods, but it is more time-consuming and not suitable for business application. Generally, the effect and timeliness of dynamic update one is the best among them. In this paper four kinds of cloud detection method are discussed here, including the sky field-of-view, sky channel, cloud radiation correction and different instrument cloud products matching. The first two methods are more feasible from the perspective of timeliness for numerical prediction, but the data quantity using could detection method of sky field-of-view is less than sky channel, leading to discarding of lots of channel data in climate sensitive area such as upper channel, and thus affecting the quality of analysis field. Further on, the methods of hops jumper, box and principal component applied to AIRS data sparseness were analyzed. From assimilation timeliness and operability, box method is feasible; although there is high complexity with algorithm of principal component analysis, which has a certain application prospect. After reviewing the quality control section, some further research directions in these fields were given respectively.