DeepWave TurbNet: a Hybrid CNN-LSTM Framework for Ocean Turbulence Prediction

Accurate ocean wave turbulence prediction is essential for maritime safety, offshore operations, and climate modeling. However, traditional models often overlook key meteorological factors, struggle with long-range dependencies, and fail to integrate spatial-temporal patterns effectively, leading to reduced predictive reliability. Extensive preprocessing further limits their adaptability across varying sea conditions. This study introduces DeepWave-TurbNet, a scalable and adaptive model for real-time ocean wave turbulence classification. Sensor readings from accelerometers, gyroscopes, and wave height sensors often suffer from misalignment due to varying sampling rates, while extreme wave conditions and sensor malfunctions introduce outliers. To address these issues, propose Z-Score KNN-Based Adaptive Filling (Z-KAF), which normalizes wave height, acceleration, and gyroscope data while handling missing values and removing outliers, ensuring robust feature representation. Following preprocessing, Pearson Correlation Coefficient (PCC) extracts meaningful relationships between sensor signals, eliminating redundancy. Feature selection is performed using DeepTurb-CNN-LSTM, where CNN captures local dependencies, and LSTM refines sequential learning, selecting the most relevant PCC-extracted features. A Softmax activation function classifies turbulence into Low, Moderate, and High levels, enabling precise decision-making. The proposed model achieves high performance, with RMSE of 0.0223 and MAE of 0.0153, significantly enhancing predictive reliability. DeepWave-TurbNet effectively mitigates traditional challenges, ensuring real-time turbulence assessment with improved accuracy, robustness, and efficiency.