Predicting the combined effect of model mismatch and noise in matched-field inversion

Abstract

Matched-field inversion undertakes to match the sound-field of a distant source recorded by a hydrophone array with the predictions of a computer model, the best-match model being the best estimate of the overall propagation scenario (source position and environmental parameters). However, the match is never perfect because the model is an ideal (hence mismatched) approximation of the real environment, and because background noise affects the quality of the received signal. Here it is shown how model mismatch and noise together degrade the correlation between the modeled and measured field using the Bartlett processor. Model mismatch is included by analysis of the normal modes of sound propagation, assuming the adiabatic mode approximation for weakly range-dependent media. The result is a fairly simple predictor of the Bartlett correlation expected in practice given estimates of the SNR and model mismatch—the mismatch being cast either in terms of realistic uncertainties for model parameters, or more simply by a single figure of merit quantifying the mean effect of mismatch through its influence on the modes.