The Yellow Sea is known for its strong tidal and wind forcing that influence surface currents. However, traditional methods assume steady-state surface current, making it hard to capture effects of tide and typhoon. In this study, we developed a new method that considers inertia. By comparing our results with observations, we found that this approach provides improved accuracy compared to previous methods. This improvement can contribute to better understanding of dynamics in the Yellow Sea.

We created a new global dataset that reveals how ocean surface carbon dioxide has changed each month over the past four decades. By applying a deep learning model trained on both observational data and model simulations, we improved the representation of interannual variability and more accurately captured ocean responses to climate events like El Niño. This work supports global efforts to understand the ocean’s role in the carbon cycle and its response to climate change.

This study addresses the lack of comprehensive sea surface partial pressure of CO₂ (pCO₂) data in the North American Atlantic Coastal Ocean Margin (NAACOM) by developing the Reconstructed Coastal Acidification Database (ReCAD-NAACOM-pCO₂). The product reconstructed sea surface pCO₂ from 1993 to 2021 using machine-learning and environmental data, capturing seasonal cycles, regional variations, and long-term trends of pCO₂ for coastal carbon research.

Storm surges pose a significant flooding risk to coastal areas. This research, taking China's Daya Bay Petrochemical Industrial Zone as a case study, addresses the dynamic nature of flooding events and the limitations of traditional evacuation plans for individuals with restricted real-time information. By combining the hydrological model and artificial intelligence, the method proves highly effective in optimizing evacuation routes, providing invaluable guidance during actual storm surges.

Ocean surface waves forced by tropical cyclones can cause tremendous damage to offshore structures and coastal communities. As such, forecasting their evolution is of utmost importance. This study investigates the usage of a long short-term memory neural network to forecast hurricane-forced waves in the Caribbean Sea. Results strongly suggest that forecasts can be performed to a high degree of accuracy up to 12 h at minimal computational expense and even outperform a wave model.