The Comparison of Several Statistical Retrieval Methods Used in Retrieving Temperature and Moisture Vertical Profiles

This paper aimed at studying the statistical retrieval of atmospheric temperature and moisture vertical profiles from satellite observations. Observations of FY-3 meteorological satellite were used in training and testing the traditional statistical regression, the back propagation (BP) neural network, and the support vector regression (SVR). Results indicated that compared with the traditional statistical regression, the BP neural network and the SVR method both have a higher precision in terms of temperature and moisture profiles. And compared with the BP neural network, the SVR method is slightly better in retrieving of temperature profiles over the land with a cloudy sky and over the sea with a cloudy but not rainy weather. When it comes to the retrieval of moisture profiles, the SVR method has a higher precision over the land with a clear sky, and a lower precision over the ocean with a not rainy weather, and the precision of these two methods are similar over the land with a cloudy sky.