Inversion of Full-depth Temperature Profiles Based on Surface Temperature and AUV Measured Data

The sound speed profile has an important impact on sound propagation and underwater acoustic detection, and the water temperature has the most significant impact on the sound speed, so it is critical to obtain high-precision full-depth temperature profiles. With the rapid development of marine mobile platforms, it is possible to obtain depth-fixed temperature data using surface velocimeter or autonomous underwater vehicles (AUVs). This paper uses the measured thermistor chain data to carry out numerical simulation, and discusses the feasibility of reconstructed full-depth temperature profiles using measured temperature of few discrete depths. The back propagation (BP) neural network is used to generate the nonlinear mapping relationship between the temperature in a few discrete depths and the first two empirical orthogonal function (EOF) coefficients. The experimental results show that the temperature at two specially selected depths can reflect the full-depth temperature profiles to a certain extent. However, the information about the water temperature at different depths is diverse and the thermocline contains the most information. As the depth-fixed data measured by the AUV increases, the inversion accuracy of the full-depth temperature profiles increases accordingly. Results shows that, even in ocean regions that have solitary internal waves, when the depth of the depth-fixed data is selected the same as the depth of the surface layer and the two extreme points of the second EOF, the root mean square error (RMSE) of almost all reconstructed temperature profiles in the test set is less than 0.2°C, and the mean RMSE is about 0.12°C.