Estimation of Sound Speed Profiles Using Artificial Neural Networks

The vast and complex oceans that are optically opaque are acoustically transparent, enabling characterization of physical and biological bodies and processes of sea using sound as a premier tool. Lack of direct observations of vertical profiles of velocimeters and/or temperature and salinity, from which sound speed can be calculated, limits specifications and investigation of temporal and spatial variabilities of the three-dimensional structure of the sound speed in the oceans. In this study, the authors demonstrate estimation of sound speed profiles (SSPs) from surface observations using an artificial neural network (ANN) method. Surface observations from a mooring in the central Arabian Sea are used as a proxy to the satellite observations. The ANN-estimated SSPs had a root-mean-square error of 1.16 m/s and a coefficient of determination of 0.98. About 76% (93%) of the estimates lie within
$pm$
1 m/s (
$pm$
2 m/s) of the SSPs obtained from in situ temperature and salinity profiles.