Stange, Jens (1997) EOF-Analyse der Feuchteprofile. (Diploma thesis), Christian-Albrechts-Universität Kiel, Kiel, Germany, 91 pp.

Text Dipl. 1997 Stange, J.pdf - Published Version Restricted to Registered users only Download (9MB) | Contact

Abstract

In this study a semi-statistical process is applied to derive humidity profiles over the ocean from passive microwave measurements in a range of 1 to 30 Ghz. The procedure is based on the vertical empirical orthogonal functions (EOFs) of the specific humidity extracted from a large sample of radiosonde measurements over the global ocean surface during the period from January to December of the year 1989. The measurements are divided up into seven classes, which have different total precipitable water content. The principal component analysis (PCA) is carried out for each class separately. The first three eigenvectors explain up to 95% of the total variance of humidity. The explained variance of the first three eigenvectors decreases with increasing total precipitable water content. In classes with a total water vapor content of more than 30 kg/m2 the cumulative explained variance of the first three eigenvectors reaches values less than 80% only. The reason for this fact may be the lack of the specific saturation humidity observations. It is shown that the first three eigenvectors are statistically significant and stable. The first eigenvector describes variations of the total precipitable water W, while the second and the third EOFs are related to the ratio of the precipitable water of the planetary boundary layer to the total precipitable water Wg/W and the sea surface temperature Ts, respectively. These physical facts are used to discuss the computed EOFs. The verification of the method with independent data in each class indicates that there is a significant improvement in describing the humidity profile if the degree of approximation of the model is increased. Further on the specific humidity is normalized with the analysis is carried out for each class and the vertical empirical orthogonal functions are computed again. The purpose of this normalization is the consideration of the upper atmospheric layers. Accordingly, the first four eigenvectors explain up to 76% of the total variability. It is shown that they are statistically significant and stable. The explained variance of the significant eigenvectors yields mainly unsatisfactory results. Therefore the conclusion is that because of the nonnalization too much of the explained variance is lost. Besides, there is no significant improvement when the degree of approximation is increased. An analysis of the complete setoff measurements without any classifications yields an improvement of the results. The first three eigenvectors explain up to 97% of the total variability. But the climatic variability of the humidity field cannot be considered. The study shows that the EOFs are superior to the frequently used analytical structure functions (e.g. "power-law") especially in the surface layer. The application of EOFs for different classes takes into account the large climatic variability of the humidity field. Furthermore the computer storage and compution time requirements are much less than those of the other methods. This is a great advantage for future operational applications.

Actions (login required)

View Item