Our EIFFEL partner ICCS participated in Big Data from Space 2023 (BiDS), an in-person conference that was held from 6 to 9 November, at the Austria Center (Vienna). Big Data from Space 2023 (BiDS) brought together key actors from industry, academia, EU entities and government to reveal user needs, exchange ideas and showcase latest technical solutions and applications touching all aspects of space and big data technologies. BiDS provided a premier opportunity for practitioners, researchers, and educators to plunge into the EIFFEL world exploring our partner’s (ICCS) interesting paper presented as poster entitled: “Enhancing the spatial resolution of Sentinel 3 SYNERGY through the super-resolution via repeated refinement method”.
The rationale of this study underlines the fact that despite the abundance of open Earth Observation (EO) data from the Copernicus program and the GEOSS platform, their uptake in the context of Climate Change (CC) related applications is often limited due to inherent spatial and temporal resolution constraints. Super-resolution techniques aim to reconstruct high spatial resolution images from degraded low spatial resolution images, but current approaches are being tested mainly for high spatial resolution imagery and small scaling factors. To fill this gap, this study investigates the ability of the diffusion-based image super-resolution via repeated refinement (SR3) method in enhancing the spatial resolution of the near-daily Sentinel-3 SYNERGY images from 300 to 75 meters by taking advantage of the higher spatial resolution of the Sentinel-2 MSI images during training.
In particular, our EIFFEL partner ICCS investigated the performance of the novel diffusion-based super-resolution SR3 model in enhancing the spatial resolution of Sentinel-3 SYNERGY images. The model was first trained with a training dataset including randomly picked pairs of Sentinel-2 images and then further trained with image pairs similar to the test images. The two trained models used as an input only Sentinel-3 images during inference and produced qualitative consistent results. However, the retrained model achieved better scores in reference-based metrics (i.e., ERGAS < 3) and produced images with a higher level of detail. This can be explained by the fact that diffusion models create images that have the same distribution as the images they are trained on.
More info about the event: https://www.bigdatafromspace2023.org/
Read the Proceedings including ICCS’s paper for EIFFEL project here: https://op.europa.eu/en/publication-detail/-/publication/10ba86b1-7c63-11ee-99ba-01aa75ed71a1/language-en