Estimation of Seasonal Representation of the Sea Water Temperature Profile Using Machine Learning and Its Effect on the Prediction of Underwater Acoustic Detection Performance

Abstract

Seawater temperature and salinity profiles are important physical properties that represent oceanic environments and affect underwater acoustic detection prediction performance. Average ocean data can be used to predict the SONAR detection area in areas where obtaining real-time ocean data or instantly predicting the SONAR detection area is difficult. However, it can yield distorted results. In this study, representative temperature profiles reflecting properties of the vertical structure at various temperatures in the study area were obtained using K-means clustering, an unsupervised machine learning technique. K-means clustering was applied to the temperature profiles obtained from the three stations of the Ulleung Basin in the East Sea. In addition, the physical characteristics of the representative profiles obtained were compared, and the representativeness of the acoustic detection area obtained from the representative profiles was evaluated. In summer, when the mixed layer was thin, each cluster was classified according to the vertical temperature gradient of the thermocline. In winter, the clusters were classified according to the mixed layer and thermocline depths, rather than the vertical temperature gradient of the thermocline. For each obtained cluster, the acoustic detection area was calculated using all the profiles and displayed as a histogram. The acoustic detection area calculated from the representative profile of the cluster was generally close to the average of the acoustic detection area. Thus, K-means clustering can effectively classify temperature profiles physically and acoustically and can potentially be applied in other regions for the classification and analysis of seawater temperature and salinity profiles.