Application Of Principal Component Algorithm To The Assimilation Of High-resolution Infrared Sounder Observations

In the recent years, the observations from high-resolution infrared sounders have become a major part of the assimilation systems in many numerical weather prediction (NWP) centers with the development of these sounders. Unlike the common sounders, high-resolution sounders can measure several thousands to a few thousands channels and can provide atmospheric information with high accuracy and vertical resolution. It is very hard to analysis the useful information from so many channels and, a large number of channels are improper for data assimilation for many reasons like instrument noise and etc., nearly only 200 channels are employed in the data assimilation system.The spectral of high-resolution infrared sounders contain many channels that have very similar spectral signatures. Channels that similar in content are highly correlated to each other. The complete high-resolution infrared spectral with thousands of channels can be compressed by some methods for data assimilation. Also the compressed information can involve the whole spectral information in direct data assimilation system efficiently and improve the calculation speed.Using principal component algorithm described in this paper can involve the channel information from the infrared hyper-spectral data efficiently which is compressed into several hundreds of principal components(PC). The principal components are generally calculated through the covariance of these spectral and they are in descending order according to the eigenvalues. These principal components are associated with real atmospheric variations rather random noise and the bulk of the atmospheric information in the spectrum is present in the first hundreds of PC. So only several hundreds of PC can represent the useful information involved in thousands of channels of the infrared hyper spectral data. This reflects the potential of principal components to reduce noise.The direct use of PC for data assimilation in an NWP assimilation systems requires an efficient radiative transfer (RT) model that can simulate PC directly given first-guess fields of temperature, water vapour, ozone, surface properties and other variables, or even reconstruct any spectrum by using the PC to a assumed accuracy in a multiple linear regression algorithm. A principal component algorithm based radiative transfer (PCRT) model is designed in this article. The PCRT model is formed by adding some models corresponding to principal component method to the fast radiative transfer model which has been developed after many training and testing, then it can simulate principal component scores and consequently reconstruct radiance. Compared to the fast radiative transfer model simulating the whole spectral the PCRT model can simulate the PC or reconstruct radiance for any atmospheric profile inputted to the system, and it can simulate the radiance more effectively and achieve higher computing efficiency for it have the characters derived from the principal component algorithm which can reduce the special dimension of the issue.This article describes the implementation of the PCRT model in the four-dimensional variational data assimilation (4D-Var) system by modifying some interfaces in the 4D-Var system and setting the method to call the PCRT model for users, thus the 4D-Var system is formed with the function to assimilate directly the principal component of the high-resolution infrared spectral. Also we need to set the namelist parameters for high-resolution infrared spectral data and prepare the corresponding input files to assimilate directly high-resolution infrared radiance. Based on the result of many experiences we can draw a conclusion that the impact of the assimilation of PC from high-resolution infrared spectral is statistically neutral to the forecast score but it is significant positive compared the experiences without high-resolution data. By setting the number of the PC we can gain 5 to 21 times of computational efficiency of PCRT model over the fast radiative transfer model for TOVS/ATOVS (RTTOV) developed by the European Centre for Medium-Range Weather Forecasts (ECMWF) on the basis of not reducing the assimilation impact.