Chunlin Ning

Accurate and timely detection of emergency event holds significant practical value for mitigating potential risks and losses. However, such emergency event often exhibits continuous, gradual evolutionary patterns in the temporal dimension, and this non-stepwise transition introduces inherent ambiguity to the detection task. To address the ambiguity issues in emergency event detection tasks within complex systems, this paper proposes a Multi-Scale Fuzzy Learning method with Weak Supervision (MS-FuLW). The proposed method first constructs a dynamic label mapping mechanism based on fuzzy set theory, extending traditional binary labels into fuzzy labels to characterize the gradual transition process of emergency events. Next, a Multi Scale Convolutional Network (MSC) is designed to extract both local abrupt features and global trend features from time series data in parallel, thereby enhancing the model ability to capture complex patterns, along with a specialized fuzzy loss function to improve compatibility with fuzzy labels. Finally, a weakly super vised learning mechanism is introduced to optimize the fuzzy label modeling process and reduce dependency on precise anno tations. The experimental results show that compared to baseline methods, MS-FuLW steadily improves detection real-time performance by 5%–25% in real environments, providing an effective solution for real-time risk warning in complex systems.

Accurate monitoring of internal solitary waves (ISWs) is crucial for understanding their influence on ocean mixing and energy transfer. Satellite observations offer a powerful tool for large‐scale ISW monitoring, but they mainly detect surface signatures and thus require nearby in situ measurements for validation and amplitude estimation. However, buoy data with sparse sampling intervals cannot resolve the narrow waveforms of ISWs, resulting in systematic underestimation of amplitude parameters and reducing their usefulness for remote sensing‐based inversion models. To overcome this issue, we introduce IWSR, a physics‐guided generative adversarial network designed to reconstruct 1‐min ISW signals from 10‐min buoy data. The model uses a thermocline‐weighted loss function to focus on ISW‐relevant features and is trained with high‐frequency buoy records from the Andaman Sea. Validation results show that IWSR effectively restores detailed soliton structures that are missed in coarse‐resolution data, achieving a peak signal‐to‐noise ratio of 38.83 dB, an RMSE of 0.22°C, and a thermocline‐specific masked RMSE of 1.59°C. When applied to buoy data from 2012 to 2014, IWSR increases estimated ISW amplitudes by an average of 5.60%, reaching up to 18.96%, and uncovers more high‐amplitude events. The statistics of the reconstructed data also reflect seasonal changes linked to thermocline depth fluctuations. By reducing the resolution‐duration trade‐off, IWSR improves the accuracy of buoy‐based ISW observations and offers better input for remote sensing‐based ISW amplitude inversion models, enabling a more precise analysis of subsurface ocean dynamics.

To address the limitations in identifying complex anomaly patterns and the heavy reliance on manual labeling in traditional oceanographic data quality control (QC) processes, this study proposes an intelligent QC method that integrates Gated Recurrent Units (GRU) with a Mean Teacher–based semi-supervised learning framework. Unlike conventional deep learning approaches that require large amounts of high-quality labeled data, our model adopts an innovative training strategy that combines a small set of labeled samples with a large volume of unlabeled data. Leveraging consistency regularization and a teacher–student network architecture, the model effectively enhances its ability to learn anomalous features from unlabeled observations. The input incorporates multiple sources of information, including temperature, salinity, vertical gradients, depth one-hot encodings, and seasonal encodings. A bidirectional GRU combined with an attention mechanism enables precise extraction of profile structure features and accurate identification of anomalous observations. Validation on real-world profile datasets from the Bailong (BL01) moored buoy and Argo floats demonstrates that the proposed model achieves outstanding performance in detecting temperature and salinity anomalies, with ROC-AUC scores of 0.966 and 0.940, and precision–recall AUCs of 0.952 and 0.916, respectively. Manual verification shows over 90% consistency, indicating high sensitivity and robust generalization capability under challenging scenarios such as weak anomalies and structural profile shifts. Compared to existing fully supervised models, the proposed semi-supervised QC framework exhibits superior practical value in terms of labeling efficiency, anomaly modeling capacity, and cross-platform adaptability.

Abnormal tidal levels pose a serious threat to maritime navigation, coastal infrastructure, and human life and property. Therefore, it is crucial to accurately predict tidal levels. However, due to the influence of topography and meteorology, tidal levels exhibit complex and non-stationary characteristics, making high-precision prediction a significant challenge. This study proposes a tidal prediction model, named SVMD-BiLSTM-Residual Decomposition (SBRD), which combines Successive Variational Mode Decomposition (SVMD) and Bidirectional Long Short-Term Memory (BiLSTM) networks. SBRD decomposes non-stationary tidal signals into simpler intrinsic mode functions (IMFs) using SVMD. Each IMF is then independently predicted using a BiLSTM network, and the final prediction is obtained through signal reconstruction. Experimental results show that SBRD accurately predicts tidal levels within a 24 h horizon and maintains robust performance during abnormal tidal events, such as acqua alta. Compared to other models, SBRD achieves the highest prediction accuracy and the lowest error, with a Coefficient of Determination (R²) exceeding 99%, a Mean Absolute Error (MAE) of 1.33 cm or less, and a Root Mean Square Error (RMSE) within 2.13 cm for tidal forecasts within a 24 h horizon. These results demonstrate that SBRD effectively enhances the accuracy of tidal level prediction, contributing to the advancement of marine economic technologies and the prevention and mitigation of marine disasters.

Significant wave height (SWH) is a key physical parameter influencing the safety of shipping, fisheries, and marine engineering projects, and is closely related to climate change and marine disasters. Existing models struggle to balance a high prediction accuracy with low parameter counts, and are challenging to deploy on platforms such as buoys. To address these issues, this study proposes an innovative method for SWH prediction by combining Singular Spectrum Analysis (SSA) with a residual correction mechanism in a Long Short-Term Memory (LSTM) network. This method utilizes SSA to decompose SWH time series, accurately extracting its main feature modes as inputs to the LSTM network and significantly enhancing the model’s ability to capture time-series data. Additionally, a residual correction module is introduced to fine-tune the prediction results, effectively improving the model’s 12 h forecasting accuracy. The experimental results show that for 1, 3, 6, and 12 h SWH predictions, by incorporating SSA and the residual correction module, the model reduces the Mean Squared Error (MSE), Root-Mean-Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) by 60–95%, and increases the coefficient of determination (R²) by 2–60%. The proposed model has only 10% of the parameters for LSTM based on Variational Mode Decomposition (VMD), striking an excellent balance between prediction accuracy and computational efficiency. This study provides a new methodology for deploying SWH prediction models on platforms such as buoys, and holds significant application value in marine disaster warning and environmental monitoring.

In January 2019, Tropical Storm Pabuk traversed the southern South China Sea, inducing significant upper‐ocean cooling with a maximum temperature drop of 3.8°C lasting over three weeks. As the storm passed directly over the Bailong marine meteorological buoy (5.843°N, 104.208°E), high‐frequency oceanic and atmospheric observations captured its immediate impact, providing valuable insights into air‐sea interactions during extreme weather events. Analysis of buoy data and FIO‐COM simulations reveals that Pabuk triggered a rapid sea surface temperature (SST) decline, initially driven by wind‐driven vertical mixing and heat flux loss, followed by prolonged cooling sustained by cold advection. The latter was primarily controlled by changes in the Vietnam Coastal Current (VCC), whose intensity and structure were significantly altered by the storm. Additionally, near‐inertial oscillations (NIOs) enhanced subsurface mixing contributing to the persistence of cooling, whereas winter monsoon winds further influenced post‐storm SST evolution. The cooling response exhibited strong spatial variability: in deep offshore regions, vertical mixing dominated, entraining colder subsurface waters into the mixed layer; along the Vietnam coastal shelf, cold advection played a leading role; in shallow waters, heat flux loss initiated cooling with residual cold advection and NIO‐driven subsurface mixing extending SST anomalies. These findings underscore the heightened sensitivity of shelf seas to tropical storms, where coastal currents, wind‐driven mixing, and bathymetric constraints critically influence SST evolution. A more accurate representation of these localized processes in oceanographic and climate models is essential for improving storm impact assessments and upper‐ocean thermal predictions.