Quantitative Analysis and Interpretation of Transient Electromagnetic Data via Principal Component Analysis

Transient electromagnetic (TEM) data in applied geophysics are invariably contaminated by random and coherent noise associated with acquisition, geology, and the environment. Advanced unexploded ordnance (UXO) detection and discrimination using TEM data require the suppression of the random noise and reliable separation of UXO signal from other coherent signals. The random noise is usually easier to remove, with coherent signal being more difficult to identify as well as separate. We have developed a method based on principal component analysis (PCA) to achieve both objectives. As a data-adaptive linear transformation, PCA is a fast and reliable method for the removal of random uncorrelated noise as well as for the separation of coherent undesired signals from those due to UXO and UXO-like anomalies. In this paper, we outline the PCA method, including the choice of data organization, construction of covariance matrices, and choice of principal components in reconstruction. We then show both synthetic and field data as examples of the efficacy of the method.