PLEASE POST

---

MMM SEMINAR COSMIC

---

A Quality Control (QC) Procedure for CHAMP Radio Occultation (RO) Data Using Multivariate Statistical Methods

Zhen Zeng

Florida State University

The principal component analysis (PCA) and singular value decomposition (SVD) analysis methods are applied for QC of CHAMP level-2 RO Data provided by CDAAC. PCA is performed on a square symmetric matrix of vertical correlation matrix of RO profiles to extract important modes that explain most variations of data, and to evaluate the quality of RO data via the norm of the primary PC scores of each profile. Taken 4884 RO profiles in March 2004 as an example, PCA QC recognizes 261 RO profiles as suspected samples. Compared with 374 erroneous ROs identified by CDAAC QC, 203 ROs overlap. In addition, the first leading mode of global refractivity is associates with the latitudinal characteristics of the atmosphere, which explains 60% of the total variance. 16% of the total variance is explained by the second leading mode, which shows a dipole pattern with positive anomalies in the North pole, a negative anomalies in the South pole, and significant positive anomalies over predominant convective areas in the western Pacific, South America and Africa in the tropics. Due to different local characteristics of various modes of RO dataset revealed from PCA, the RO observations are categorized into several subsets. As to each subset, SVD analysis is applied for QC of RO data. The outlier diagnosis is performed in three steps: (1) outliers detection; (2) outliers isolation; and (3) outliers-free data reconstruction. Procedures of and results from such a QC procedure will be presented and compared with a biweighting method.

Thursday, 1 September 2005, 10:30 AM

Refreshments 10:15 AM

NCAR-Foothills Laboratory

3450 Mitchell Lane

Bldg 2, Room 1001