comparisons of pattern extraction capability between the self-organizing maps and principal component analysis

Abstract

Although the principal component analysis has been widely used to analyze climate data, limitation of the method has been also recognized. This study compares the pattern extraction capability between the Self-Organizing Maps (SOM) and the Principal Component Analysis (PCA). The comparison was conducted for simulation data and actual climate data. For the simulation data, while the SOM succeeded in identifying all patterns, PCA failed to identify the simulated patterns. The comparison for actual climate data implied the geometric positions of reference vectors in the SOM can be explained by the leading modes of PCA analysis.