Acoustic-oceanographic data fusion for prediction of oceanic acoustic elds

Tese de doutoramento, Engenharia Electrónica e Telecomunicações (Processamento de Sinal), Faculdade de Ciências e Tecnologia, Universidade do Algarve, 2014

Maritime rapid environmental assessment exercises became rather important. An underlying objective is to predict the evolution of the acoustic eld due to an underwater target. Main contributions to this end have relied on accurate models of acoustic propagation, which receive baseline properties and ocean forecasts as input. Intuitively, the most accurate oceanographic forecast should imply the most accurate acoustic forecast. This can fail, due to at least two reasons: 1) the full set of (space-time-variant) environmental properties are rarely known with enough accuracy; 2) even the most sophisticated propagation model cannot handle the full environmental detail in solving propagation equations, forcing the experimenter to reduce complex environmental features to a simpli ed representation. Acoustic modeling errors appear then as inevitable. Little possibility of error minimization exists, if the propagation model is simply run in a `forward' manner. The results presented in this work show that the acoustic error can be minimized, if the propagation model is fed with an environmental parameter vector containing two distinct sets: one, xed and formed by the environmental parameters with uncontrolled errors; the other, variable and with errors determined in a controlled way, adapted to the errors in the rst subset. Here, the second set is treated as a distinct quantity, labeled as \equivalent model". It can be determined by acoustic inversion. The equivalent model is employed for two objectives: to estimate the acoustic eld at a given present time (nowcast), and a given future time (forecast). Synthetic acoustic elds, and oceanographic measurements and predictions (with the Navy Coastal Ocean Model) obtained for the Maritime Rapid Environmental Assessment 2003 sea trial, drive the simulations. For the problem of nowcast, the equivalent model is determined at sparse transect points, and interpolated to points with no acoustic measurements. For the problem of forecast, the equivalent model is furthermore `extrapolated' to future time. The `extrapolation' consists of a mapping between sound speed pro le and equivalent model. When providing an estimate of the future sound speed at the mapping input, the estimate of the future equivalent model is obtained. The proposed method led to a decrease of 3{5 dB in transmission loss estimation error, as compared to standard procedures.

FCT - Fundação para a Ciência e Tecnologia: SFRH/BD/9032/2002