[A-TT35] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT35-08] Artificial Neural Networks for Predicting Sea Surface Currents Around the Korean Peninsula

Jeong-Yeob Choi^1,2, *Jae-Hun Park¹, Ho-Jeong Ju¹, Young Taeg Kim³ (1.Department of Ocean Sciences, Inha University, Incheon, Republic of Korea, 2.Graduate School of Oceanography, University of Rhode Island, USA, 3.Ocean Research Division, Korea Hydrographic and Oceanographic Agency, Busan, Republic of Korea)

Keywords:Artificial Neural Network, Sea surface current prediction, 3-D U-Net structure, High resolution

Accurate prediction of sea surface currents is crucial for various marine activities, including disaster monitoring, fisheries, and search and rescue operations. Advances in numerical models, particularly through data assimilation and high-resolution simulations, have enhanced ocean forecasts. However, these models require significant computational resources, making near-real-time forecasting challenging. To address this, efficient computational approaches that integrate numerical model outputs are needed. Artificial neural networks (ANNs) offer a promising solution due to their ability to generate predictions with relatively low computational costs by leveraging pretrained networks. In this study, we present a prediction framework for sea surface currents around the Korean Peninsula using a three-dimensional (3D) convolutional neural network (CNN). The model is based on a 3D U-Net structure, modified to incorporate oceanic and atmospheric variables for current prediction. The forecasting process is optimized to minimize the next-day prediction error at a spatial resolution of 1/24°, and its recursive structure enables multi-day forecasts. To evaluate performance, we test various input configurations to identify the optimal ANN model for surface current prediction. The model achievements demonstrate its strong potential for practical application in near-real-time ocean forecasting.