Super-resolution of sea surface temperature with convolutional neural network- and generative adversarial network-based methods

Abstract In this paper, we perform the super-resolution of sea surface temperature data with enhanced generative adversarial network (ESRGAN), which is a deep neural network-based single-image (SISR) method that uses (GAN). We generate high-quality ESRGAN and convolutional (SRCNN) residual-in-residual dense block (RRDBNet) methods, are based on networks (CNNs). The images generated these methods compared high-resolution optimum interpolation (OISST) using root mean square error (RMSE), learned perceptual image patch similarity (LPIPS), index (PI) evaluation methods. RRDBNet has better RMSE than SRCNN ESRGAN. However, CNN-based SISR do not provide faithful representation ocean currents OISST. LPIPS PI can represent complex distribution currents.