The Cumulative Effect Coefficient Of Principal Component

Hyper-spectral infrared detectors (AIRS) brightness temperature data's contribution to numerical weather prediction is more significant at present. However, there are 2378 AIRS channels (the actual channels used in business are 324 only), putting all of the real-time data into assimilation system is impossible. Besides, taking the correlation between channels, the timeliness of assimilation and the effect of assimilation or inversion into account, channel selection of choosing the main channel of information is needed. Common method of foreign in channel selection is iterative method based on information entropy. But each time this method will choose one channel which contains the largest amount of information, and the previous selected channels will affect the later ones. Iterative method has two defects main:the effect of iterative methods in channel selection depends on the background and observation error covariance matrix, but on many occasions specific information on the background filed is lack; iterative method select one channel only from unselected channels at each iterative, which is time-consuming. In business iterative method's efficiency is not so well. In this paper, a method of channel selection based on the cumulative effect coefficient method is introduced. Firstly, it was the pre-processing of the channels. Then, for the channels'combination during the daytime, the principal component analysis of temperature and humidity's Jacobi was made respectively to get the cumulative effect coefficient of each channel on the principal component. Finally, for the channels'combination during the nighttime, sunlight's effect to channels was added in. Thus, the subset of selected channels both during the daytime and nighttime was obtained. Comparing to the iteration method, the cumulative effect coefficient method keeps the most important information and gets channels of most sensitivity. Besides, without too much prior knowledge, Jacobi matrix's structure can be got directly by this method. Lastly cumulative effect coefficient method can process a number of channels at the same time. It is shown that the method is feasible according to the profile retrieval experiments of temperature and humidity.