Mahalanobis clustering, with applications to AVO classification and seismic reservoir parameter estimation

Brian H. Russell, Laurence R. Lines

A new clustering algorithm, Mahalanobis clustering, is proposed as an improvement on traditional K-means clustering. We present applications of this method to both AVO classification and seismic reservoir parameter estimation using multiple attributes. In the latter application, we use the radial basis function neural network (RBFN) with centres, and apply Mahalanobis clustering to find the cluster centres that are used in the training of the network. We also show that this method allows us to improve the estimate of the covariance matrix parameters used in the general form of the RBFN approach.