According to the statistics of water transportation accidents, collision accidents are on the rise as the shipping industry has expanded by leaps and bounds, and the water transportation environment has become more complex, which can result in grave consequences, such as casualties, environmental destruction, and even massive financial losses. In view of this situation, high-precision and real-time ship trajectory prediction based on AIS data can serve as a crucial foundation for vessel traffic services and ship navigation to prevent collision accidents. Thus, this paper proposes a high-precision ship track prediction model based on a combination of a multi-head attention mechanism and bidirectional gate recurrent unit (MHA-BiGRU) to fully exploit the valuable information contained in massive AIS data and address the insufficiencies in existing trajectory prediction methods. The primary advantages of this model are that it allows for the retention of long-term ship track sequence information, filters and modifies ship track historical data for enhanced time series prediction, and models the potential association between historical and future ship trajectory status information with the current state via the bidirectional gate recurrent unit. Significantly, the introduction of a multi-head attention mechanism calculates the correlation between the characteristics of AIS data, actively learns cross-time synchronization between the hidden layers of ship track sequences, and assigns different weights to the result based on the input criterion, thereby enhancing the accuracy of forecasts. The comparative experimental results also verify that MHA-BiGRU outperforms the other ship track prediction models, demonstrating that it possesses the characteristics of ease of implementation, high precision, and high reliability.

Significant Wave Height (SWH) is a crucial parameter in ocean wave dynamics, impacting coastal safety, maritime transportation, and meteorological research. Building upon the TimesNet neural network, a recent advancement in the realm of time series prediction in deep learning, this study proposes an integrated approach combining Empirical Mode Decomposition (EMD) with TimesNet, introducing the EMD-TimesNet model for SWH forecasting. The TimesNet model’s multidimensional spatial mapping guarantees effective historical information extraction, while the EMD approach makes it easier to decompose subsequence characteristics inside the original SWH data. The predicted Root Mean Square Error (RMSE) and Correlation Coefficient (CC) values of the EMD-TimesNet model are 0.0494 m and 0.9936; 0.0982 m and 0.9747; and 0.1573 m and 0.9352 at 1 h, 3 h, and 6 h, respectively. The results indicate that the EMD-TimesNet model outperforms existing models, including the TimesNet, Autoformer, Transformer, and CNN-BiLSTM-Attention models, both in terms of overall evaluation metrics and prediction performance for diverse sea states. This integrated model represents a promising advancement in enhancing the accuracy of SWH predictions.

The sea surface temperature anomaly (SSTA) plays a key role in climate change and extreme weather processes. Usually, SSTA forecast methods consist of numerical and conventional statistical models, and the former can be seriously influenced by the uncertainty of physical parameterization schemes, the nonlinearity of ocean dynamic processes, and the nonrobustness of numerical discretization algorithms. Recently, deep learning has been explored to address forecast issues in the field of oceanography. However, existing deep learning models for ocean forecasting are mainly site specific, which were designed for forecasting on a single point or for an independent variable. Moreover, few special deep learning networks have been developed to deal with SSTA field forecasts under typhoon conditions. In this study, a multivariable convolutional neural network (MCNN) is proposed, which can be applied for synoptic-scale SSTA forecasting in the South China Sea. In addition to the SSTA itself, the surface wind speed and the surface current velocity are regarded as input variables for the prediction networks, effectively reflecting the influences of both local atmospheric dynamic forcing and nonlocal oceanic thermal advection. Experimental results demonstrate that MCNN exhibits better performance than a single-variable convolutional neural network (SCNN), especially for the SSTA forecast during the typhoon passage. While forecast results deteriorate rapidly in the SCNN during the passage of a typhoon, forecast errors in the MCNN can be effectively restrained to slowly increase over the forecast time due to the introduction of the surface wind speed in this network.

Sea surface temperature anomalies (SSTAs) and sea surface height anomalies (SSHAs) are indispensable parts of scientific research, such as mesoscale eddy, current, ocean-atmosphere interaction and so on. Nowadays, extended-range predictions of ocean dynamics, especially in SSTA and SSHA, can provide daily prediction services in the range of 30 days, which bridges the gap between synoptic-scale weather forecasts and monthly average scale climate predictions. However, the forecast efficiency of extended range remains problematic. With the development of ocean reanalysis and satellite remote sensing products, large amounts datasets provide an unprecedented opportunity to use big data for the extended range prediction of ocean dynamics. In this study, a hybrid model, combing convolutional neural network (CNN) model with transfer learning (TL), was established to predict SSTA and SSHA at monthly scales, which makes full use of these data resources that arise from delayed gridding reanalysis products and real-time satellite remote sensing observations. The proposed model, where both ocean and atmosphere reanalysis datasets serve as the pretraining dataset and the satellite remote sensing observations are employed for fine-tuning based on the transfer learning (TL) method, can effectively capture the evolving spatial characteristics of SSTAs and SSHAs with low prediction errors over the 30 days range. When the forecast lead time is 30 days, the root means square errors for the SSTAs and SSHAs model results are 0.32°C and 0.027 m in the South China Sea, respectively, indicating that this model has not only satisfactory prediction performance but also offers great potential for practical operational applications in improving the skill of extended-range predictions.

Accurately inverting global and regional subsurface temperature (ST) by multisource satellite observations is a challenging but hot topic. This study proposes a new method to invert daily ST from the sea surface information in China's marginal seas based on generative adversarial network (GAN) model. The proposed GAN‐based model can project the STs from sea surface information (SLA, SSTA, SST) with a high resolution of 1/12°. A traditional regression‐based model, Modular Ocean Data Assimilation System (MODAS), is set up same experiments for comparison. The results show that the averaged root mean square error results are less than 1.45°C in the upper 200 m and the highest averaged R² of 0.97 at the 70 m level, which is better than that of MODAS. Errors analysis and typical oceanographic phenomena analysis results show the superiority of the proposed GAN‐based model in this study. This study can provide high‐precision daily ST data from sea surface information, which can be expanded to further studies on the interior ocean variation characteristics.

The usage of four‐dimensional variational (4DVar) scheme is limited by the static background error covariance and the adjoint model. In a hybrid frame of the four‐dimensional ensemble‐variational data assimilation scheme (4DEnVar), being able to avoid the tangent linear and adjoint models in the 4DVar and nowadays developed into a cutting‐edge research topic of the next‐generation data assimilation methods, an analytical 4DEnVar (A‐4DEnVar) scheme is designed. First, an analytical expression for explicit evolution of the background error covariances is derived. The expression collects the innovation of observations over an assimilation window simultaneously and propagates information to the initial background field by temporal cross covariances. Second, to estimate the adjoint model, the temporal covariances are constructed with ensemble members being centralized with respect to the model states integrated from the initial condition. Third, an iterative linear search process is introduced to minimize the cost function to update the analysis field until convergence. Twin experiments based on the Lorenz chaos model with three variables are conducted for the validation of the A‐4DEnVar scheme. Comparisons to the conventional 4DVar show that the A‐4DEnVar is comparable in accuracy even with a long assimilation window and sparse observations. The assimilation results also show that the A‐4DEnVar scheme can be implemented with a very small ensemble size which means that under circumstances without the tangent linear and adjoint models it can be easily incorporated into data assimilation systems in use.