In order to study climate change, high-precision measurement for carbon dioxide (CO2) by remote sensing satellite needs to be achieved. Scattering of aerosol and thin cirrus with high transmittance is the main environmental factor which affects CO2 retrieval precision. We combine statistical method of principal component analysis (PCA)with optical path probability density distribution function (PPDF). Priori values for CO2 retrieval can be obtained by PCA method, and large deviation which causes the problem that calculation results deviate from the truth value can be avoided. We also use three-layer PPDF model to approach the change of absorption spectral lines caused by variation of photon path because of scattering of thin cirrus and aerosol. The results show that the combination of PCA method and PPDF method improves the retrieval precision significantly. From the retrieval results of GOSAT data in Taklamakan desert in 2013, we can find that the variance is 3.5 by PPDF method, while the combination of two methods can obtain the variance of 1.4, which is superior to the variance of 1.6 provided by NIES.然科学基金青年科学基金(４１６０１３９３)