Applications of Deep Learning-Based Super-Resolution for Sea Surface Temperature Reconstruction

Deep learning-based super-resolution (SR) methods have been widely used in natural images; however, their applications satellite-derived sea surface temperature (SST) not yet fully discussed. Hence, it is necessary to analyze the validity of deep SR SST reconstruction. In this study, an model, including multiscale feature extraction and multireceptive field mapping, was first proposed. Then, proposed model four other existing models were applied reconstruction analyzed. First, compared with bicubic interpolation method, can improve accuracy. Compared models, achieve lowest mean squared error (MAE) East China Sea (ECS), northwest Pacific (NWP) west Atlantic (WA), second-lowest MAE southeast (SEP); root (RMSE) ECS WA, RMSE NWP SEP. Additionally, ODRE acquire highest or second-highest peak single-to-noise ratio structural similarity index ECS, NWP, Moreover, number missing pixels variety are two essential factors performance. The process enhance performance, especially for small regions stable regions. Finally, while a deeper network be helpful achieving approach simply adding more dilation convolutions may