Abnormally High Water Temperature Prediction using Machine Learning

Hyun Yang, MinKyu Kim and Hee-Jeong Han, KIOST Korea Institute of Ocean Science and Technology, Busan, South Korea
Abstract:
Over the past few years, abnormally high water temperature (AHWT) phenomena have damaged the maritime economy of Korea because it causes a mass stranding of farmed fish and an illness by the propagation of Vibrio pathogens. To reduce damages caused by AHWT occurrences, we need to respond as quickly as possible or forecast in advance. In this paper, therefore, we proposed a deep learning-based approach to forecast AHWT occurrences using the recurrent neural network (RNN). First, high-performance computing and storage system were employed in order to rapidly train the RNN model from the water temperature dataset over the past twenty years. Then the water temperatures after 7-days were estimated from the trained model. As a result, the proposed model achieved 0.98, 0.75, and 2.84 in terms of R-squared, root mean square error (RMSE), and mean absolute percentage error (MAPE), respectively. We are expecting that this approach will contribute to effectively mitigate the damages from AHWT occurrences and protect against the destruction of aquaculture industry environments.