Bayes classification of subsurface electromagnetic responses


  • Dennis F. Moore,

  • Edmund A. Quincy


Statistical classification of geoelectromagnetic responses containing additive gaussian noise is considered. The classification is performed with minimum average probability of error criterion of goodness. The Bayes receiver structure for classifying three different but equally likely geoelectromagnetic responses is presented. Since each geophysical model is assumed to have one unknown parameter, the three hypotheses are composite and the received signals are treated as random. The composite hypotheses are developed in terms of frequency-domain model responses. The three conductive models selected for this research are a half space, a sheet, and a two-layer earth. Geoelectromagnetic responses are found to yield strongly correlated signals with correlation coefficients typically of 0.9999. The receiver performance in terms of average probability of error is compared with (1) a system having known and equally correlated signals (ρ = 0.9999), and (2) a system having known orthogonal signals. These comparisons illustrate the degradation of performance resulting from the random and correlated nature of geoelectromagnetic responses.