Super-Resolution of Sea Surface Temperature Satellite Images

Availability of high-resolution maps of geophysical fields, devoid of data loss due to clouds, is an urgent requirement for operational forecasting. We develop a Bayesian algorithm for super-resolution (or downscaling) of lower resolution geophysical fields observed by satellites. The key novelty in the present algorithm is the development and use of a Generative Adversarial Network (GAN) to learn the prior probability distribution of the high-resolution geophysical fields from historical data and/or model forecasts. The trained GAN is used to sample from the high-resolution prior and a particle filter along with the low-resolution data (as observation) is used to obtain the posterior high-resolution geophysical field. The resultant algorithm has been named the Particle Filter Generative Adversarial Network super-resolution (PF-GAN-SR) algorithm. The new algorithm is applied to downscale sea surface temperature fields in the northwest Atlantic Ocean. Results show consistent performance across different downscaling ratios. Notably, the high-resolution fields obtained from the new algorithm has better similarity score with the true high-resolution field compared to those from bi-cubic interpolation (commonly used in the geophysical community) and the SR-GAN algorithm (used in the computer vision community).