Super-resolving ocean dynamics from space with computer vision algorithms: training datasets

We provide here the datasets used for the development of the dilated Adaptive Residual Network for the super-resolution of ocean Absolute Dynamic Topography described in Buongiorno Nardelli et al. (2022). The model is designed to combine satellite altimetry and thermal observations and provides super-resolved dynamic topography. The training/test datasets have been built starting from the data originally prepared for an Observing System Simulation Experiment carried out in the framework of the European Space Agency CIRCOL project [Ciani et al., 2021]. They consist of one year of synthetic daily Absolute Dynamic Topography (ADT), surface geostrophic currents and sea surface temperature data  obtained from Copernicus Marine Service Mediterranean Forecasting System (MFS) (Product ID: MEDSEA-ANALYSIS- FORECAST-PHY-006-013) [Clementi et al. 2021]. Synthetic Altimeter-derived ADT maps were obtained by first sampling the model output along the actual tracks of a synthetic constellation composed of 4 Radar Altimeters: Jason-3, Sentinel-3A, SARAL/Altika, and Cryosat-2 missions  (this step is achieved by running the SWOT simulator software [Gaultier et al., 2016]) and successively applying the DUACS (Data Unification and Altimeter Combination System) mapping method. The original input images cover the entire Mediterranean domain at 1/24° spatial resolution, leading to an individual image size of 380x1000 pixels. Here, we have randomly chosen 40 dates (~11% of the total) to be kept aside as fully independent test data, and successively re-sampled the original images extracting much smaller tiles (76x100), which are used as input to the network training. The tiles are extracted by going through a double loop on latitude and longitude, imposing a spatial overlap of 50%. Full details on data pre-processing (e.g.normalization strategies) are given in the paper:

Buongiorno Nardelli, B.; Cavaliere, D.; Charles, E.; Ciani, D. Super-Resolving Ocean Dynamics from Space with Computer Vision Algorithms. Remote Sens., 2022, 14, 1159. https://doi.org/10.3390/rs14051159