Joint calibration and mapping of satellite altimetry data using trainable variaitional models

Abstract

<p>Satellite radar altimeters are a key source of observation of ocean surface dynamics. However, current sensor technology and mapping techniques do not yet allow to systematically resolve scales smaller than 100km. With their new sensors, upcoming wide-swath altimeter missions such as SWOT should help resolve finer scales. Current mapping techniques rely on the quality of the input data, which is why the raw data go through multiple preprocessing stages before being used. Those calibration stages are improved and refined over many years and represent a challenge when a new type of sensor start acquiring data.</p><p>We show how a data-driven variational data assimilation framework could be used to jointly learn a calibration operator and an interpolator from non-calibrated data . The proposed framework significantly outperforms the operational state-of-the-art mapping pipeline and truly benefits from wide-swath data to resolve finer scales on the global map as well as in the SWOT sensor geometry.</p><p> </p>

Deep Learning Improves Global Satellite Observations of Ocean Eddy Dynamics

“…Seeing a wide range of dynamical regimes during training forces the network to generalise, which should make it robust against local regime changes induced by the inter-annual variability of large-scale currents. While bespoke regional networks [38,40] in the Gulf Stream offer marginally improved SSH mapping compared to our global network [53] (Extended Data Table 3), fine-tuning on a smaller set of observations from the Gulf Stream Extension (Methods) brings our network's performance close to state-of-the-art regional networks (Extended Data Table 3).…”

“…Our neural network maps SSH across all regions despite their distinct regional dynamics, unlike prior studies which trained bespoke region-specific networks [35,36,37,38,39,40]. Seeing a wide range of dynamical regimes during training forces the network to generalise, which should make it robust against local regime changes induced by the inter-annual variability of large-scale currents.…”

“…This biases the diagnosed surface geostrophic currents, relative vorticity, and strain rate, potentially underestimating the strength of non-linear eddy interactions. There has thus been growing interest in developing improved mapping algorithms to create higher-resolution SSH maps [30,31,32,33,34,35,36,37,38,39,40].…”

“…Deep learning [41] is spurring rapid progress in climate modelling [42], numerical weather prediction [43,44,45], and increasingly in satellite oceanography [46,34,35,36,37,38,39,40,47,48]. Recent studies show deep neural networks can be trained to estimate the SSH field from along-track observations using supervised learning on either synthetic data from numerical simulations [34,35,36,37,40] or real-world satellite observations [38,39]. The advantage of this data-driven approach is that it allows the optimal mapping to emerge objectively from the data itself, unlike traditional linear methods [28,29].…”

“…For example, SST can be a powerful predictor of SSH [37,38,39], highlighted by the close relationship between SSH and SST in idealised ocean turbulence theories [49]. Regional proof-of-concept studies showed deep learning yields state-of-the-art SSH maps, especially when SST is used [35,36,37,38,39,40]. However, SSH mapping over the whole globe poses a significant challenge for deep learning algorithms as they need to generalise over a wide range of dynamical regimes, including strongly localised boundary currents, chaotic eddy fields, and fast-propagating large-scale waves.…”