本文基于主成分分析（Principal Components Analysis, PCA）和集合经验模态分解（Ensemble Empirical Mode Decomposition, EEMD）降噪的基本思想，利用改进的自适应噪声总体集合经验模态分解（Improved Complete Ensemble Empirical Mode Decomposition With Adaptive Noise, ICEEMDAN）对原EEMD的条带降噪方法进行了修改，并利用该方法对风云三号C星、D星(FY-3C、3D)微波湿度计(MWHS-2)实际观测亮温中的条带噪声进行了分析。其中主成分分析对数据进行降维，得到各扫描线的主成分分量，模态分解方法分解对应分量，利用各模态能量密度的差异提取出其中的噪声并去除，随后组合剩余模态重构出观测亮温，实现噪声的抑制。通过对原始算法和各类改进后的模态分解方法降噪效果的对比，结果表明使用ICEEMDAN可有效避免EEMD中残余噪声等问题，减少重构误差。数值分析结果表明，改进后的方法使方差进一步降低0.020K2，信噪比提升0.031dB，进一步提升了算法的降噪能力。

Noise analysis and mitigation plays an important role in the processing of meteorological satellite data. This paper is based on the idea of mitigate noise by using Principal Components Analysis (PCA) and Ensemble Empirical Mode Decomposition (EEMD) algorithm, and using Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) as an improvement. The modified method is used in the observation data of the Microwave Humidity Sounder (MWHS-2) of the Fengyun-3C and 3D satellites (FY-3C, FY-3D) to analyze the striping noise in its observed brightness temperature. In this paper, the effectiveness of this method for MWHS-2 data is confirmed, and an analysis of the performance of the improved method for data processing and noise mitigation is given. As the striping noise has a very high correlation with scan line, using principal component analysis can not only effectively isolate the noise related principal components, but also reduce the dimension of the processed data. When the noise contained principal components are extracted, the empirical mode decomposition method can be used to adaptively separate each component into multiple modes with different frequencies. Using the differences of energy between noise and signal modes, noise can be easily separated from the signal by a method calculating and comparing the average period and energy density. Finally, the remaining modes are combined to reconstruct the principal components, and all the principal components reconstruct the observed brightness temperature data. When this method is applied to the data of MWHS-2, we used the hourly global reanalysis data of ERA5 with RTTOV model to generate the simulated brightness temperature data to compare with the observed brightness temperature before and after processing. The result shows that the algorithm successfully extracts the striping noise in the signal, and the histogram of noise exhibits a Gaussian distribution. By comparing the noise mitigation effect between the original EEMD algorithm and various improved mode decomposition methods, the results show that the use of ICEEMDAN can effectively avoid some problems in EEMD, such as the residue noise, and can reduce reconstruction errors. Numerical analysis results show that compared with EEMD method, this improved method further reduces the variance by 0.020K2 and the signal-to-noise ratio (SNR) increases by 0.031dB, which further improves the noise mitigation capability of the algorithm. Principal component analysis combined with ensemble empirical mode decomposition is an effective method to mitigate striping noise. Although the mode decomposition method is affected by some of its own properties and has certain limitations in accuracy, it has more convenient operation and higher adaptability. And the test result shows that this method can also achieve satisfactory results. The improvement of using ICEEMDAN and calculate the energy density with average period can also be helpful to noise analysis and mitigation, and can improve the reconstruction accuracy. This may have certain value for the further improvement of noise reduction algorithm, and may be helpful to improve the accuracy of meteorological data’s analysis and forecasting.