Super resolution for downscaling on oceanographic fields
The study of Ocean physics is a major interest for the understanding of Earth climate. In with project, we address the problem of retrieving a geophysical field at a higher resolution from various low-resolution satellite data. High-resolution data allow studying sub-grid processes.
Recently, many algorithms of super-resolution have been proposed in the literature. The downscaling of oceanographic fields may be viewed as a super-resolution problem.
We propose to investigate this literature and transpose it to our environmental task. Data are provided by NATL60 simulator (a realistic simulation of North Atlantic Ocean). Several state variables are available: sea surface temperature (SST), sea surface circulation (SSC), sea surface height (SSH)… In the first step, we will try to retrieve SST high resolution from a lower resolution of SST. In a second step, we want to fuse SST, SSC and SSH are low resolutions to improve the SST high-resolution field.
- propose several suited architectures that will be trained and tested on NATL60 data