A comparative assessment of different machine learning algorithms for estimating near realistic salinity in the North Indian Ocean

Salinity affects the density of water and hence plays a significant role in the transport of mass, heat, and salt across the globe. Though temperature observations are abundant over the North Indian Ocean (NIO), the salinity observations are generally sparse. Accurate estimation of sound speed profiles is vital for naval operations, particularly for acoustic communication and underwater target detection. Typically, in naval missions, operators rely on in-situ point measurements from Expendable Bathythermographs (XBTs), supplemented by climatological salinity data for sound speed profile derivation. In this study, we explore the effectiveness of three machine learning techniques: Light Gradient Boosting (LightGBM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN) for deriving salinity profiles from temperature data. All three models performed well, but XGBoost outperformed the others with a strong correlation coefficient of 0.975 and a very low root mean square error value on the testing dataset. To gain further insights into the model’s behavior, a sensitivity analysis was conducted to identify the most influential parameters for salinity profile estimation. The model’s performance was then evaluated in the coastal and open ocean areas. While generally performing well, a slight decrease in accuracy was observed in areas with limited training data, particularly near coastlines. Furthermore, we evaluated the model using the time series observations from independent ocean buoys located in the NIO, and the results clearly recommended the model’s ability to fill the gaps in buoy-based salinity measurements. In conclusion, this supervised learning model presents a robust alternative for deriving salinity profiles using in-situ temperature data within the NIO region. Its implementation in the NIO region with limited salinity data holds significant promise for enhancing data availability and furthering oceanographic research.