Echogram Overview
Amplitude across the full ~310-day deployment and full depth range, with amplitude and phase zoomed in at four representative depths, from the specular near-surface layers down into the echo-free zone near the bed.
Interactive Application
Explore the full analysis dashboard below. Navigate between tabs to view the echograms, phase coherence, the vertical- and horizontal-velocity estimates, the interpretation and forward models, and the full methods write-up. The app may take a moment to load on first visit.
Hosted on Hugging Face Spaces. If the embed doesn't load, open it directly.
Overview
ApRES is an FMCW radar system used to measure internal ice deformation with millimeter precision. Deployed on the Mercer Ice Stream in Antarctica, it yields a depth profile of internal layers that can be tracked over time.
This project focuses on the echo-free zone, the region near the bed where clear layers vanish. Rather than relying on visible reflectors, we use amplitude-independent phase estimators that recover both the vertical and horizontal velocity even where no layer can be tracked, revealing ice dynamics that are otherwise hidden.
Methods & Key Findings
Rather than relying on visible layers, the velocity analysis rests on two amplitude-independent estimators: one for the vertical velocity, one for the horizontal-velocity distribution. Both defer every nonlinear phase operation until after coherent integration over time, so noise is suppressed before it can corrupt the phase.
CW-MLPR · Vertical Velocity
Coherence-Weighted Multi-Lag Phase Regression estimates the vertical velocity in every depth bin. It combines phase differences across many time lags with coherence weighting and a robust median, sidestepping phase unwrapping entirely. The result is a continuous velocity profile that extends smoothly into the echo-free zone.
MDI · Horizontal Velocity
Multi-band Decorrelation Inversion recovers the distribution of horizontal velocities. As scatterers advect laterally through the antenna beam, the return decorrelates at a rate set by horizontal speed; measuring this across several frequency bands and inverting the resulting integral equation yields the velocity distribution.
Validation · Forward Simulator & Neural Cross-Check
A 2-D forward simulator builds synthetic echograms from first principles, summing the coherent (Huygens) returns of moving reflectors with the same phase convention as the real data. It reproduces specular layers, dipping reflectors and volume scatter, and serves as a shared test bench for both estimators. An independent simulation-based inference model, a neural density estimator trained on the simulator, provides a second, learned estimate of the horizontal velocity, and a Viterbi optimal-path layer tracker gives a further amplitude-based cross-check wherever layers remain visible.
Key Finding · Coherent Signal in the Echo-Free Zone
The "echo-free zone" is not entirely echo-free. Coherent phase signals persist well below the deepest visible layers, producing velocity estimates that smoothly continue the profile. This provides new constraints on ice dynamics closer to the bed.
Key Finding · The EFZ Return Is Volume Scattering
What produces that coherent return? The weight of evidence points to volume (Rayleigh) scattering from distributed englacial inhomogeneities, rather than intact specular layers or broken ones:
- The scattering "support" recovered by sparse (multiple-measurement-vector) reconstruction stays in place over time.
- The return decorrelates as scatterers sweep through the antenna beam, exactly the signature MDI exploits.
- A forward simulator reproduces the echo-free-zone texture from distributed scatterers alone, with no loss of phase information.
Key Finding · A Subglacial Lake Leaves a Kinematic Fingerprint
Over a known subglacial lake the apparent vertical velocity runs slightly fast. A small (~0.2°) slope of the internal layers, combined with the fast horizontal flow of the ice stream, accounts for the excess. This lets a joint kinematic inversion tie the vertical and horizontal velocities together into a single consistent picture of the flow near the bed.