Abstract:Spatial-temporal current forecasting is indispensable for ocean engineering and marine science exploration, for instance aiding in the conservation and protection of marine ecosystems, planning shipping-routes and determining the length and fuel consumption of sea-going voyages, obtaining deeper insights into the distribution of heat flux within the ocean, which is vital for better understanding climate changes, and so on. Most present related-studies primarily focused on single location or grid-cell-based forecasting, such methodologies are site-specific and neglect the importance of spatial-temporal fidelity. Furtherly, the Recurrent Neural Networks-based methods previously employed exhibit low efficiency in terms of model convergence concerning practical engineering purposes, and numerical weather models are time-consuming and computational expensive. A newly improved Unet-based model using residual-learning with attention strategy is proposed for 2D sea surface current (SSC) velocity predictions with a more efficient perspective. Several machine-learning methodologies were adopted for a better performance comparison. The final predictions demonstrated its superiorities that the proposed neural-learning method outperforms the other established approaches with spatial-resolved mean RMSE less than 0.009 m/s and 0.006 m/s. As a promising surrogate for SSC predictions, the proposed methodology has strong potential in operation marine monitoring and engineering constructions.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the problem of spatiotemporal prediction of Sea Surface Current (SSC). Specifically, the paper proposes a pure Convolutional Neural Network (CNN) method based on residual learning and attention mechanisms to improve the accuracy and efficiency of 2D sea surface current velocity prediction in nearshore areas. #### Main Issues and Background 1. **Limitations of Single Point or Grid Cell Prediction**: - Most existing studies focus on predictions at a single location or grid cell, neglecting the spatial correlation between adjacent grids, leading to information loss. 2. **Inefficiency of RNN/LSTM-based Methods**: - Traditional RNN/LSTM-based methods are inefficient in model convergence and have high computational costs, making them unsuitable for practical engineering applications. 3. **Complex and Time-consuming Numerical Weather Models**: - Although numerical weather models are accurate, they have high computational complexity and are time-consuming. #### Proposed New Method 1. **Pure CNN Method Combined with Residual Learning and Multi-Head Attention Mechanism (MHA-ResUnet)**: - By introducing residual learning and multi-head attention mechanisms, the new method can better capture spatial correlations and temporal dependencies, thereby improving prediction accuracy. 2. **Sliding Window Strategy in Data Preprocessing**: - A sliding window strategy is introduced in the data preparation stage to retain short-term temporal dependencies in the input tensors. 3. **Simplified Model Structure**: - The original Unet's BatchNormalization layers, MaxPooling2D layers, and UpSampling2D layers are removed, and Conv2DTranspose layers are used instead, improving model efficiency. #### Experimental Validation 1. **Experimental Area and Dataset**: - The dataset from the southern North Sea is used for experimental validation. This dataset is derived from reanalysis data and generated through data assimilation. 2. **Experimental Results**: - The experimental results show that the new method performs excellently in predicting the northward and eastward current velocity components, with average root mean square errors (RMSE) of less than 0.009 m/s and 0.006 m/s, respectively. #### Application Prospects 1. **Marine Engineering and Monitoring**: - Accurate sea surface current predictions help optimize route planning, save fuel consumption, and ensure navigation safety. 2. **Marine Renewable Energy Development**: - The prediction results can be used to deploy and adjust wave and tidal energy facilities. Through the above improvements, this method demonstrates high accuracy and efficiency in ocean dynamics prediction, providing strong support for marine engineering and scientific research.

Spatial-temporal Offshore Current Field Forecasting Using Residual-learning Based Purely CNN Methodology with Attention Mechanism

Short-term Prediction of Offshore Wind Power Based on CNN and LSTM

SDPNet: A Novel DeepLearning Method for Ocean Surface Current Prediction

Spatiotemporal Prediction of Monthly Coastal Upwelling Scenario in SST Fields Using Deep Learning Based Models

Flow Field Analysis and Development of a Prediction Model Based on Deep Learning

A Deep U-Net-ConvLSTM Framework with Hydrodynamic Model for Basin-Scale Hydrodynamic Prediction

A Physics-Guided Attention-Based Neural Network for Sea Surface Temperature Prediction

A deep-learning model for rapid spatiotemporal prediction of coastal water levels

Application of deep learning in predicting suspended sediment concentration: A case study in Jiaozhou Bay, China

Improving wave height prediction accuracy with deep learning

A short-term flood prediction based on spatial deep learning network: A case study for Xi County, China

Estimating Ocean Surface Currents With Machine Learning

Predicting temporal and spatial 4-D ocean temperature using satellite data based on a novel deep learning model

Advancing Spatio-temporal Storm Surge Prediction with Hierarchical Deep Neural Networks

Tidal Current Prediction Based on a Hybrid Machine Learning Method

Comparative Investigations of Tidal Current Velocity Prediction Considering Effect of Multi-Layer Current Velocity

Surface current prediction based on a physics-informed deep learning model

Ocean reanalysis data‐driven deep learning forecast for sea surface multivariate in the South China Sea

CAS-Canglong: A skillful 3D Transformer model for sub-seasonal to seasonal global sea surface temperature prediction

STCANet: Spatiotemporal Coupled Attention Network for Ocean Surface Current Prediction

A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data