Estimating ocean subsurface thermohaline information from satellite data is vital for understanding ocean dynamics and El Niño. This paper presents a double-output Residual Neural Network (DO-ResNet) to estimate ocean subsurface temperature (OST) and salinity (OSS) in the tropical Western Pacific using multi-source remote sensing data. The model, trained and validated with Argo data, shows strong performance, with average RMSE and R² values of 0.34°C (0.05 psu) and 0.91 (0.95) for OST (OSS). DO-ResNet outperforms other models, effectively capturing spatial features and displaying strong seasonal adaptability. This research introduces a new AI method for estimating OST and OSS.

Mesoscale eddies are swirling motions with a radius ranging from several tens to a hundred of kilometers. The mesoscale eddies can cause strong vertical displacement of isopycnals, yet the vertical structure of such displacement has not been well understood. This study examines the vertical structures of mesoscale eddies in the Kuroshio-Oyashio Extension region based on the observational and reanalysis datasets. Mesoscale eddies with three distinct kinds of vertical structures are identified.