The reconstruction of sea surface currents from satellite altimeter data is a key challenge in spatial oceanography, especially with the upcoming wide-swath SWOT (Surface Water and Ocean and Topography) altimeter mission. Operational systems, however, generally fail to retrieve mesoscale dynamics for horizontal scales below 100 km and timescales below 10 d. Here, we address this challenge through the 4DVarnet framework, an end-to-end neural scheme backed on a variational data assimilation formulation. We introduce a parameterization of the 4DVarNet scheme dedicated to the space–time interpolation of satellite altimeter data. Within an observing system simulation experiment (NATL60), we demonstrate the relevance of the proposed approach, both for nadir and nadir plus SWOT altimeter configurations for two contrasting case study regions in terms of upper ocean dynamics. We report a relative improvement with respect to the operational optimal interpolation between 30 % and 60 % in terms of the reconstruction error. Interestingly, for the nadir plus SWOT altimeter configuration, we reach resolved space–timescales below 70 km and 7 d. The code is open source to enable reproducibility and future collaborative developments. Beyond its applicability to large-scale domains, we also address the uncertainty quantification issues and generalization properties of the proposed learning setting. We discuss further future research avenues and extensions to other ocean data assimilation and space oceanography challenges.

This paper addresses physics-informed deep learning schemes for satellite ocean remote sensing data. Such observation datasets are characterized by the irregular space-time sampling of the ocean surface due to sensors’ characteristics and satellite orbits. With a focus on satellite altimetry, we show that end-to-end learning schemes based on variational formulations provide new means to explore and exploit such observation datasets. Through Observing System Simulation Experiments (OSSE) using numerical ocean simulations and real nadir and wide-swath altimeter sampling patterns, we demonstrate their relevance w.r.t. state-of-the-art and operational methods for space-time interpolation and short-term forecasting issues. We also stress and discuss how they could contribute to the design and calibration of ocean observing systems.

Over the last few years, a very active field of research has aimed at exploring new data-driven and learning-based methodologies to propose computationally efficient strategies able to benefit from the large amount of observational remote sensing and numerical simulations for the reconstruction, interpolation and prediction of high-resolution derived products of geophysical fields. In this paper, we investigate how they might help to solve for the oversmoothing of the state-of-the-art optimal interpolation (OI) techniques in the reconstruction of sea surface height (SSH) spatio-temporal fields. We focus on two small $10°\times 10°$ GULFSTREAM and $8°\times 10°$ OSMOSIS regions, part of the North Atlantic basin: the GULFSTREAM area is mainly driven by energetic mesoscale dynamics, while OSMOSIS is less energetic but with more noticeable small spatial patterns. Based on observation system simulation experiments (OSSE), we used a NATL60 high resolution deterministic ocean simulation of the North Atlantic to generate two types of pseudo-altimetric observational dataset: along-track nadir data for the current capabilities of the observation system and wide-swath SWOT data in the context of the upcoming SWOT (Surface Water Ocean Topography) mission. We briefly introduce the analog data assimilation (AnDA), an up-to-date version of the DINEOF algorithm, and a new neural networks-based end-to-end learning framework for the representation of spatio-temporal irregularly-sampled data. The main objective of this paper consists of providing a thorough intercomparison exercise with appropriate benchmarking metrics to assess whether these approaches help to improve the SSH altimetric interpolation problem and to identify which one performs best in this context. We demonstrate how the newly introduced NN method is a significant improvement with a plug-and-play implementation and its ability to catch up the small scales ranging up to 40 km, inaccessible by the conventional methods so far. A clear gain is also demonstrated when assimilating jointly wide-swath SWOT and (aggregated) along-track nadir observations.