A Learning-Based Approach for Sensor Calibration in SWOT Sea Surface Height Observations

Abstract

Sea Surface Height (SSH) maps are crucial for observing and analyzing mesoscale ocean surface dy- namics. Traditionally, such maps have been generated using one-dimensional satellite observations. The upcoming Surface Water and Ocean Topography (SWOT) mission introduces a groundbreaking sensor capable of acquiring two-dimensional SSH profiles, enabling observations at previously unresolved spatio-temporal scales. However, a critical challenge lies in separating the true SSH signal from instrument or geophysical errors in the SWOT data. This paper presents a novel learning-based approach for sensor calibration, incorporating a scale decomposition step informed by the structural characteristics of different signals. In a supervised framework, our method achieves state-of-the-art residual error reduction while providing corrections across the full spectrum of observations and imposing weaker constraints on the modeled error signal. These results highlight the potential of advanced learning techniques for improving the accuracy and reliability of SSH measurements in the SWOT mission.