Wide-Range Ocean Current Speed Estimation From Buoy Measurement Data Using Machine Learning

Abstract

The conventional Doppler-based single point current meters (SPCM) which are used for the measurement of surface ocean current speed and direction in the moored data buoy systems face challenges as their output is susceptible to biofouling. SPCM requires noticeable power for the operation and it is expensive. A recently reported innovative approach which involves integrating load cell in the mooring line of a data buoy, is a viable option for ocean current measurement with advantages such as lower power requirements, less cost, and resistance to biofouling. However, the reported method is useful only during extreme weather conditions when the mooring line is sufficiently stretched. In this article, an effort is made to overcome this limitation, by incorporating machine-learning (ML) techniques with the additional measurement data, such as wind, wave, and buoy position along with the mooring load. The new approach was developed and tested for two data buoys deployed in the Arabian Sea and the Bay of Bengal over nearly a year duration. This study compares six different ML models, ultimately identifying random forest (RF) as the top-performing model with a correlation value of 0.92 between the observed and estimated current for both the buoys and the root mean square error (RMSE) of 0.072 and 0.042 m/s for BD08 and AD07 buoy in the Bay of Bengal and Arabian Sea, respectively. The study shows that the proposed method is capable of estimating a wide range of ocean currents reliably with very good accuracy.