Satellite–Argo Reconstruction of the Mesoscale-Eddy Sound-Speed Field and Its Effect on Towed-Array Gain: A Differentiable Framework with Bounded Acoustic Feedback

Abstract

We reconstruct the four-dimensional sound-speed field of mesoscale eddies from satellite sea-level anomaly and sparse Argo profiles with a geostrophically constrained coordinate network, and connect it to array-level acoustics through a differentiable parabolic-equation model and a differentiable beamformer, so gradients flow from the array output back to the ocean estimate. On Kuroshio-Extension data (539 Argo profiles) held-out subsurface temperature reaches 0.50 degrees Celsius RMSE, 2.4 times better than an EOF regression on altimetry. Because the acoustics consume sound speed rather than temperature, we validate the sound-speed anomaly directly on 106 held-out in-situ profiles: 2.46 m/s, against 4.58 for the same baseline. With that measured anomaly as truth and the reconstruction as the replica, environment-aware processing reaches a median array gain of 13.58 dB against a perfect-replica bound of 15.05 and 10.58 uncompensated, recovering 67.3 percent of the eddy-induced loss per profile. Run in reverse, acoustic feedback rescales the anomaly to within 0.2 to 0.5 percent in intensity. The measured acoustic evidence is component-level: no dataset here is a measured horizontal eddy inversion.