Answers are grounded in this paper and cite their sources.
Deep-sea bathymetry inversion from geosynchronous synthetic aperture radar (GEO SAR) offers persistent wide-area monitoring capabilities but faces two fundamental challenges: the defocusing of topography-induced sea-surface waves due to their nonlinear space-variant motions during long integration times, and the ill-posed nature of retrieving seabed topography from surface signatures due to the nonlinear, depth-dependent filtering effect of the stratified water column. To address these issues, this article builds upon our previous GEO spatial-correlation (SC) SAR spatial-correlation imaging framework and existing sea-surface parameter retrieval modules, and introduces two new components for this task: a GEO dual-tree complex wavelet transform (DTCWT)-SC SAR imaging algorithm and a StratiFormer inversion network. On the imaging side, we develop the GEO DTCWT-SC SAR algorithm: oriented DTCWT subbands represent nonlinear, space-variant kinematics with a set of direction-separable and 2-D cross-directional residual modes that parameterize a motion-compensated spatial-correlation kernel. Applied point-by-point to the echoes, this kernel corrects the phase history, restores the slow-time spatial correlation of the sea-surface pattern, and enables closed-loop, correlation-driven motion refinement. On the inversion side, we introduce StratiFormer, a hierarchical spatiotemporal network that performs multiparameter time-series inversion from sea-surface height and 2-D current sequences to deep-sea bathymetry. Unlike static image-to-depth models, StratiFormer ingests the multiparameter time series into a three-stage hierarchical encoder with learnable interstage feature fusion. The hybrid-phase decoupling and current-retrieval steps are used as an integrated retrieval and observation–construction interface rather than as new contributions of this article. Controlled echo-to-depth simulations, reanalysis-informed surface-field experiments, and terrain-disjoint diagnostics support the algorithmic feasibility of the proposed workflow under controlled and reanalysis-informed settings, while operational validation with measured GEO-SAR observations and collocated bathymetric truth remains necessary.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: