Prediction of subsurface oceanographic parameter using machine learning technique based on long term historical in-situ measurements

Ocean Observation systems group of the National Institute of Ocean Technology maintains deep-sea moored buoy systems for real-time measurement and transmission of met-ocean and sub-surface oceanographic data from remote locations in Bay of Bengal and Arabian Sea. The OMNI (Ocean Moored Buoy system in the Northern Indian Ocean) is equipped with suite of sub-surface oceanographic sensors for measurement of sub-surface salinity, temperature, water current speed and direction. The data from the buoy systems are used for real time cyclone warning, long term climate studies etc. Innovative engineering aspects were considered for consistent operation of buoy system in the harsh marine environment and the average transmitted data return from the buoy system is more than 95%. The real time data from the sub-surface sensor provides critical data which helps in understanding the pattern and intensification of cyclonic events. The continuous data from the sub-surface sensors are inevitable for providing crucial information on the state of oceanographic conditions. However, the sub-surface sensors are subjected to continuous wave forces, marine biofouling, aging, drift in measurement and the data is prone to discontinuities which may reduce the performance of models predicting the accurate information of cyclones and other phenomenon. The present study explores the application of machine learning (ML) technique for prediction of missing sub-surface data using historical in-situ data. Salinity data is considered in the present study and the predicted results are validated with archived data set. The salinity and temperature data are measured at discreet depths and the required depth can be selected as target depth and the ML model can be tuned for predicting the value at target depth. In this paper, data from 5 buoy locations for the year 2017 are filtered for training the model. This study will help in the accurate prediction of salinity values and complement during data gaps for improving the performance of cyclone forecast models.