Reconstruction of significant wave height distribution from sparse buoy data by using deep learning.

Significant wave height plays a crucial role in influencing marine ecosystems, ocean shipping, and other maritime activities. The distribution of buoy observation data tends to be sparse. Gridded wave data obtained through numerical simulation typically offer broader applicability, albeit with higher computational demands. In this paper, a deep learning model based on Full Connected and Convolutional Neural Networks is proposed, utilizing sparse buoy observation data as input to reconstruct the distribution of significant wave height in the sea area. The model reconstruction results are validated using ERA5 data, demonstrating excellent performance. Additionally, we explore the influence of the model's spatial boundaries and the number of input buoys on reconstruction accuracy, as well as the adaptability of the model to different sea areas. This study provides a novel method and approach for the rapid and cost-effective retrieval of regional significant wave height. • A significant wave height reconstruction model based on deep learning method is proposed. • The model utilizes sparse buoy data to reconstruct the distribution of significant wave height in the sea area. • The model shows rapid and accurate computational performance, with excellent adaptability to the sea area. [ABSTRACT FROM AUTHOR]