Understanding ocean temperature structure and its spatiotemporal variability is essential for studying ocean circulation, climate, and marine ecosystems. While previous approaches using observations and numerical models have advanced our understanding, they face limitations such as sparse data coverage and computational bias. To address these issues, we developed an ensemble of data-driven neural network models trained with in situ vertical profiles and daily remote sensing inputs. Unlike previous studies that were limited to open-ocean regions, our model explicitly included coastal areas with complex bathymetry. The model was applied to the East/Japan Sea and reconstructed 31 years (1993–2023) of daily three-dimensional ocean temperature fields at 13 standard depths. The predictions were validated against observations, showing RMSE < 1.33 °C and bias < 0.10 °C. Comparisons with previous studies confirmed the model’s ability to capture short- to mid-term temperature variations. This data-driven approach demonstrates a robust alternative to traditional methods and offers an applicable and reliable tool for understanding long-term ocean variability in marginal seas.
Coastal bottom dissolved oxygen (DO) depletion poses a serious threat to marine ecosystems and aquaculture, and hypoxic events in the semi-enclosed Jinhae Bay, Korea, repeatedly cause large-scale damage to fish farms. Accurate DO prediction models are therefore crucial for ecosystem management and loss mitigation. This study analyzes how different tidal input representations affect the performance of data-driven DO prediction models in a tide-dominated coastal environment. Using time-series data of oceanographic and meteorological variables from nearby observation sites, we develop an long short-term memory (LSTM)-based neural network ensemble model with four experimental configurations. These include not only water level but also tidal envelope, tidal-intensity proxy, and temporal differences in water level and DO (Δtide, ΔDO) as additional inputs. Compared with the baseline configuration, the full tide-informed input case reduced the 72 h mean root mean square error (RMSE) from 1.16 to 1.12 and increased the Pearson correlation coefficient from 0.873 to 0.883. It also improved the representation of intraday variability and prediction stability. These results show that tide-derived variables help the model more effectively capture tidal-phase-locked DO fluctuations, while temporal-difference inputs further strengthen short-term variability and sensitivity to DO changes. These results indicate that properly representing tidal forcing is essential for learning the temporal structure and variability of coastal bottom DO.
25 February 2026 in Journal of the Korean Society for Marine Environment & Energy