I. Introduction
Matched-Field PROCESSING (MFP) combines ocean acoustics and signal processing techniques to solve the problem of passive source localization and/or environmental parameter inversion [1]–[6]. It incorporates the mathematical representations of ocean environmental models, sound propagation models, measurement models (hydrophone arrays), and noise models (discrete noise sources, distributed ambient noises, and sensor noise) into a sophisticated inverse processing scheme, which exhaustively computes the replica fields coming from hypothetical models and finds the best match with the pressure field received on a hydrophone array [3], [7]. MFP attempts to exploit the sound propagation complexities to maximum advantage for parameter inversion. However, there are also significant disadvantages with the technique. The most serious liability is the sensitivity to mismatch. Empirically, the mismatch can be divided into three categories [6], [8]: 1) environmental mismatch, where one uses an incorrect environmental description in the propagation model, e.g., sound-speed profile errors, geoacoustic model mismatch; 2) statistical mismatch, where one uses an inaccurate data covariance matrix, sampled from a finite number of snapshots, to represent the joint probability density function of the received array data and to adaptively control the sidelobes; and 3) system mismatch, where the hydrophone array is not calibrated or positioned accurately.