Abstract
A method of retrieving the basic vertical structure of water vapor profiles from satellite-observed radiances is presented. The statistical tools of empirical orthogonal function analysis and clustering were used to define classes of vertical structure of water vapor. As a result, any water vapor sounding can be assigned to one of four vertical structure classes. Each class was shown to be identified with certain types of weather features. Multiple regression was used to retrieve approximate total precipitable water by use of brightness temperatures simulated for the Defense Meteorological Satellite Program SSH-2 infrared sounder, resulting in explained variances of about 80%. In addition, discriminant analysis was then applied to retrieve the vertical structure class of each water vapor profile, giving percentages of correct discrimination near 60%. Selection from among the SSH-2 spectral channels was used to optimize both the total water regression and the structure class discrimination. Also, it was shown that separation of soundings by total water content generally improves discrimination skill by a few percent. The results suggest that this retrieval approach should be particularly useful for application to subjective weather forecasting.
