Ocean current prediction in the gulf of Thailand using spatio-temporal deep learning on high-frequency radar

Abstract

Ocean surface current prediction is a crucial task for a variety of marine activities, such as disaster monitoring, search and rescue operations, power forecasting, and etc. There are three traditional forecasting approaches: (i) numerical based approach, (ii) time series based approach and (iii) machine learning based approach. However, their prediction accuracy was limited as they did not cooperate with spatial and temporal effects together, including oceanic knowledge is also not considered.

This paper introduces the ocean surface prediction model that accounts for spatial and temporal characteristics by a combination between CNN and GRU and also the incorporation of oceanic inputs which are month number, lunar effect, and hour number. The experiment compared the proposed model with an existing method, e.g., ARIMA, Perceptron, Temporal kNN and etc. by using RMSE as a metrics on both U and V components of dataset that was collected by high frequency (HF) radar stations located along coastal Gulf of Thailand by GISTDA from 2014 to 2016.