[A-TT30] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT30-P04] Subsurface temperature estimation using artificial neural networks in the East/Japan sea

*Eun-Joo Lee¹, Hyunju Oh², Young-Taeg Kim², Boonsoon Kang², Jae-Hun Park¹ (1.Inha Univ., 2.KHOA)

Keywords:Prediction of 3-D temperature field, Artificial Intelligence

Generally, three-dimensional (3-D) sea temperature fields can be predicted through numerical models and data assimilation applications. However, there are limitations to this approach. The first is the substantial computing resources required for data assimilation, and the second is that most of the data used in this process depends on surface remote sensing data. Recently, to address these issues, artificial neural network (ANN) models have been developed for predicting 3-D sea temperature. Verified ANN models have an advantage in terms of reduced time requirements for temperature estimation compared to numerical models. Additionally, unlike numerical models, which require data smoothing for the stability of computations during data assimilation, ANN training prioritizes the reproduction of observed data. This study concentrates on coastal seas instead of open oceans, which means that ocean dynamics in the nearshore environment need to be considered in the ANN model, and temperature profiles at various depths with varying lengths should be used as the target data. To address these challenges, our ANN model's loss function is designed to account for the geometric temperature structure, and masking is applied to the bottom boundaries. The newly developed ANN model is successfully applied in the East/Japan sea. Compared to data-assimilated numerical models, It shows better performance in estimating 3-D subsurface temperature fields.