GEOG 414/514

Lab Exercise 4: Principal Components Analysis

Principal Components Analysis (PCA) is a powerful data reduction technique often used with multispectral or hyperspectral imagery. PCA can be used to re-organize the distribution of variance in the multidimensional data space in order to maximize the information "content" of a three-band image composite that can be viewed on a typical color monitor. The PCA procedure produces a set of Eigen-images (also called Principal Component images) resulting from a transformation ("rotation") of the data space axes from their configuration in the original multispectral dataset.

The sample images below were derived from Landsat ETM+ image data acquired on September 23, 1999. The area shown is a portion of Philadelphia, Pennsylvania, separated by the Delaware River from neighboring Camden, New Jersey.

Natural Color composite (TM Bands 3, 2, and 1)

Color-IR composite (TM Bands 4, 3, and 2)

Eigen-Image composite composed of the first three Principal Components
(representing a more "information-rich" 3-band composite than what is possible with the original TM image bands)

What follows next are the six individual grayscale Principal Component images. Note that the information "content" of the first few PC images is noticably greater than the last few PC images, which contain mostly noise.

PC1                                                                 PC2       

PC3                                           PC4                    

PC5                                                                 PC6

The source for this Landsat imagery was the Global Land Cover Facility (GLCF), www.landcover.org

Processing by Matthew Ramspott, using ENVI 4.2.