Compression of Marine Environmental Data Using Convolutional Attention Autoencoder

Ocean temperature data is fundamental to the study of ocean dynamics and climate change, and its efficient compression and storage are critical for large-scale data analysis and transmission. However, traditional compression methods based on Fourier transform struggle to balance compression ratio and fidelity when confronted with the complex characteristics of marine environments. This study proposes a convolutional attention autoencoder (CAAE) to compress and reconstruct three-dimensional temperature fields and evaluates its performance across different depths and compression ratios. The experimental results indicate that although reconstruction error slightly increases with higher compression ratios, the proposed model achieves near-perfect compression and reconstruction of marine environmental data, performing robustly across various depths and spatial locations. This work offers a viable solution for the efficient and accurate compression of three-dimensional ocean data and provides valuable insights for the management of large-scale marine datasets.