Deep learning approach for downscaling of significant wave height data from wave models

The generation of high-resolution wave data remains a challenging task due to limitations in measurement techniques and computational resources. In the present study, we attempt to address the downscaling of ocean wave data from wave models using deep learning model, specifically convolutional neural networks. We conducted a comprehensive performance evaluation by comparing traditional methods, including bilinear interpolation, bicubic interpolation, and Kriging interpolation, against an improved Wave-Super-Resolution Convolutional Neural Network (W_SRCNN) in terms of high-resolution processing of significant wave height. Evaluation metrics including peak signal-to-noise ratio, root-mean-square error, structural similarity, cosine similarity, and no-reference image quality assessment were employed to assess the effectiveness of the methods. The results suggest that W_SRCNN outperforms other approaches in terms of overall performance, local detail preservation, and computational efficiency. Our findings demonstrate the successful application of deep learning techniques to achieve high-resolution processing of ocean data. The obtained results underscore the significance of tailored deep learning architectures to the analysis of ocean wave data, and thereby encouraging future investigations into their potential for high-fidelity reconstruction of sea surface geophysical fields from multisource data, including ocean models, satellite observations, and buoy measurements.

Automated summary

This study attempts to address the challenge of generating high-resolution ocean wave data using deep learning models. Comparing with traditional methods, an improved Wave-Super-Resolution Convolutional Neural Network (W_SRCNN) shows better performance in terms of overall quality, detail preservation, and computational efficiency.