Virtual sensing of ocean currents and hydrodynamic drift for underwater gliders via adjoint-based dynamics inversion

Underwater gliders, as key autonomous subsea observation platforms, are widely used in ocean environmental monitoring owing to their long endurance, ease of deployment, and wide-area coverage. The coupled effects of time-varying ocean currents and progressive biofouling, compounded by restricted underwater communication, constrain glider observations. The same coupling also leaves signatures in the onboard records, which can be used to reconstruct environmental fields and assess glider health. This paper proposes a physics-informed, adjoint-based dynamics inversion framework. A six-dimensional lumped disturbance vector is introduced to reformulate disturbance identification as a full-window optimal-tracking problem, and an adjoint-based windowed variational estimator is constructed to recover physically consistent environmental perturbations, thereby inferring quantities beyond those directly measured by the native sensor suite. Validation on operational glider data demonstrates that the framework reconstructs sparse, impulsive disturbances correlated with non-coordinated maneuvers, whose spectral and intermittency signatures are consistent with Kolmogorov inertial-subrange scaling and She–Lévêque anomalous scaling ( ζ 6 / ζ 3 = 1.77 versus the predicted 1.78). These disturbances also survive surrogate-data, noise-only, and model-error falsification tests of the full inversion pipeline. By coupling the identified disturbances with a viscous-moment inversion model, depth-resolved current retrieval is demonstrated without acoustic Doppler hardware. The depth-averaged component is cross-checked against the glide-angle method and two independent ocean reanalyses under an explicit uncertainty budget. Multi-profile analysis further reveals a statistically significant, monotonic drift in the glider’s effective hydrodynamics over the deployment. The framework enables multi-dimensional environmental sensing and glider health assessment from low-cost onboard sensor data.