Nansen gjesteforelesning 17.02.2026

Dr. Matteo Broccoli, CMCC, Italia

«Data-Driven Eddy Detection from Wide-Swath Altimetry»

Satellite altimetry has been fundamental to the understanding of ocean circulation for more than three decades. Recently, the wide-swath altimetry product delivered by the SWOT mission has extended the capability of nadir altimeters to two-dimensional mapping and sampling of the ocean surface at an unprecedented spatial resolution, with the opportunity to better constrain the mesoscale variability. Indeed, SWOT enables the observation of a larger population of eddies including smaller and sharper structures that conventional altimetry cannot resolve. Given the shorter timeseries of SWOT data compared to conventional altimetry, how much can we benefit from it for data-driven eddy detection?
In this work we developed and compared two eddy detection algorithms based on convolutional neural networks that work on global satellite altimetry maps at 1/10deg. The first exploits conventional nadir-altimetry derived maps, and it is trained on SSALTO/DUACS Sea Surface Height L4 product and related geostrophic velocities, together with ESA CCI SST product. The second instead uses the version of L4 MIOST altimetry product which also incorporates wide-swath observations from SWOT, and the related velocity fields. As targets, eddy segmentation maps are obtained with a dynamical detection algorithm from the two altimetry products.
In the first case, about three years of data are enough to reach a global F1 score of 0.87 out of 1, with even higher skills in areas of strong mesoscale activity like the Agulhas and Gulf Stream. In the second case, the shorter timeseries of SWOT L4  data(less than a year) limits the achievable performance, making transfer learning required. We discuss the beneficial impact of including SST in the altimetry-only detection, and leverage the use of a mixed dataset with conventional altimetry in input and SWOT detection maps as output to extend the benefit of high-resolution altimetry to the whole past timeseries. These results underscore the need of a longer wide-swath altimetry product and hence the extension of the SWOT mission lifetime.

Når og hvor?

Tirsdag, 17.02.26 kl. 14:15 – 15:00.

Copernicus forelesningsrom, 1. etasje, Nansensenteret, Jahnebakken 3, Bergen

Om Dr. Matteo Broccoli

Dr. Matteo Broccoli is an early career scientist at CMCC. His research focuses on the application of machine learning techniques for ocean sciences, and in particular to data assimilation, with emphasis on the exploitation of remote sensing data.

He conducted his doctoral research at the Max Planck Institute for Gravitational Physics in Potsdam and was awarded the PhD degree in Theoretical Physics from Humboldt University of Berlin. He studied Physics and Theoretical Physics at the University of Bologna.