Matrix partitioning and EOF/principal component analysis of Antarctic Sea ice brightness temperatures
A field of measured anomalies of some physical variable relative to their time averages, is partitioned in either the space domain or the time domain. Eigenvectors and corresponding principal components of the smaller dimensioned covariance matrices associated with the partitioned data sets are calc...
| Main Authors: | , , |
|---|---|
| Format: | Other/Unknown Material |
| Language: | unknown |
| Published: |
1984
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/2060/19840019251 |
| Summary: | A field of measured anomalies of some physical variable relative to their time averages, is partitioned in either the space domain or the time domain. Eigenvectors and corresponding principal components of the smaller dimensioned covariance matrices associated with the partitioned data sets are calculated independently, then joined to approximate the eigenstructure of the larger covariance matrix associated with the unpartitioned data set. The accuracy of the approximation (fraction of the total variance in the field) and the magnitudes of the largest eigenvalues from the partitioned covariance matrices together determine the number of local EOF's and principal components to be joined by any particular level. The space-time distribution of Nimbus-5 ESMR sea ice measurement is analyzed. |
|---|