Bayesian source localization with uncertain Green’s functionin an uncertain shallow water oceana

International audienceMatched-field acoustic source localization is a challenging task when environmental properties of theoceanic waveguide are not precisely known. Errors in the assumed environment (mismatch) can causesevere degradations in localization performance. This paper develops a Bayesian approach to improverobustness to environmental mismatch by considering the waveguide Green’s function to be anuncertain random vector whose probability density accounts for environmental uncertainty. Theposterior probability density is integrated over the Green’s function probability density to obtain ajoint marginal probability distribution for source range and depth, accounting for environmentaluncertainty and quantifying localization uncertainty. Because brute-force integration in high dimensionscan be costly, an efficient method is developed in which the multi-dimensional Green’s functionintegration is approximated by one-dimensional integration over a suitably defined correlationmeasure. An approach to approximate the Green’s function covariance matrix, which represents theenvironmental mismatch, is developed based on modal analysis. Examples are presented to illustratethe method and Monte-Carlo simulations are carried out to evaluate its performance relative to othermethods. The proposed method gives efficient, reliable source localization and uncertainties withimproved robustness toward environmental mismatc