the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Estimating ocean currents from the joint reconstruction of absolute dynamic topography and sea surface temperature through deep learning algorithms
Abstract. Our study focuses on the improvement of Absolute Dynamic Topography (ADT) and Sea Surface Temperature (SST) mapping from satellite observations. Retrieving consistent high resolution ADT and SST information from space is challenging, due to instrument limitations, sampling constraints and degradations introduced by the interpolation algorithms used to obtain gap free (L4) analyses. To address these issues, we developed and tested different deep learning methodologies, specifically Convolutional Neural Network (CNN) models that were originally proposed for single-image super-resolution. Building upon recent findings, we conduct an Observing System Simulation Experiments (OSSE) relying on Copernicus numerical model outputs and we present a strategy for further refinements. Previous OSSEs combined low resolution L4 satellite equivalent ADTs with high resolution "perfectly known" SSTs to derive high resolution sea surface dynamical features. Here, we introduce realistic SST L4 processing errors and modify the network to concurrently predict high resolution SST and ADT from synthetic, satellite equivalent L4 products. This modification allows us to evaluate the potential enhancement in the ADT and SST mapping while integrating dynamical constraints through tailored, physics informed loss functions. The neural networks are thus trained using OSSE data and subsequently applied to the Copernicus Marine Service satellite derived ADTs and SSTs, with the primary goal of reconstructing super resolved ADTs and geostrophic currents. A 12 years long time series of super resolved geostrophic currents (2008–2019) is thus presented and validated against in situ measured currents from drogued drifting buoys. This study suggests that CNNs are beneficial for improving standard Altimetry mapping: they generally sharpen the ADT gradients with consequent correction of the surface currents direction and intensities with respect to the altimeter derived products. Our investigation is focused on the Mediterranean Sea, a quite challenging region due to its small Rossby deformation radius (around 10 km).
- Preprint
(7504 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-1164', Anonymous Referee #1, 15 May 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-1164/egusphere-2024-1164-RC1-supplement.pdf
- AC1: 'Reply on RC1', Daniele Ciani, 14 Oct 2024
-
RC2: 'Comment on egusphere-2024-1164', Anonymous Referee #2, 06 Aug 2024
The authors are following their previous work on super-resolving and inpainting sea-surface height, this time focusing on a network learned through a new OSSE experiment. Their new study focuses on the Mediterranean Sea and is used to evaluate sea currents.
There are significant strong points in their approach. The article is well structured and scientifically sound, and can further a very active research field tied to the OceanChallenges data challenge.
I especially appreciate the care taken to evaluate the physical fields obtained.
In general, the authors are keenly aware of the bibliography in the field. However, given the many similitudes (there are differences too) I would like to see their positioning in regards to Archambault, Théo, et al. "Learning sea surface height interpolation from multi‐variate simulated satellite observations." Journal of Advances in Modeling Earth Systems 16.6 (2024): e2023MS004047.I have some reservations in regards to the validation procedure since it has some contradictory information in the paragraph that starts at line 183. There seems to be attention paid to avoid data leakage but at the same time, early in the paragraph, the 40 days seem to be selected randomly. A clarification of which is the actual approach in this paper, and how it guarantees a significant enough time lag between data used in train, validation, and test is important.
The choice of architecture is interesting, and I expect that a lot of other approaches were tested. It would be interesting to include them in the annexes, as negative results are often ill-represented in literature.
There is also a technical question of how the authors reconstruct the whole Mediterranean basin, and whether there are discontinuities in the full reconstruction.
More details would be useful as to the process of selecting the hyperparameters of the new loss function, and some ablations on its usefulness would not be remiss.
While the evaluation of the currents is extremely important, the method relies on geostrophic approximation, which holds in the Mediterranean, but should be discussed in the case of applying this approach to other basins closer to the equator where this approximation does not hold.
In general, should the authors address these small points I would be very pleased to see their work published.Citation: https://doi.org/10.5194/egusphere-2024-1164-RC2 - AC2: 'Reply on RC2', Daniele Ciani, 14 Oct 2024
Data sets
Mediterranean Sea Super Resolved Geostrophic Currents Daniele Ciani, Bruno Buongiorno Nardelli, and Elodie Charles https://zenodo.org/records/10727432
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
383 | 166 | 39 | 588 | 21 | 24 |
- HTML: 383
- PDF: 166
- XML: 39
- Total: 588
- BibTeX: 21
- EndNote: 24
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1