André Nuber

Water content estimates of a first-year sea-ice pressure ridge keel from surface-nuclear magnetic resonance tomography

Abstract The porosity of a sea-ice pressure ridge keel is an important but poorly known variable, relevant for determining the mass budget and evolution of the Arctic sea-ice cover. Determination of keel porosity from drillholes is time-intensive and only yields limited information because of their limited lateral extent. Since the porosity within a keel equals its liquid water content, surface-nuclear magnetic resonance (surface-NMR) methods can be used to estimate porosity within such features. Surface-NMR tomography measurements were made in April 2011 using seven surface coil positions across a first-year pressure ridge on landfast sea ice near Barrow, Alaska, USA. The inversion results indicate water contents of 30 ± 7% and 40 ±10% in the ridge’s shallow and deep parts, respectively. These values are much higher than those obtained from drillholes, which are ∼10% and ∼27%, respectively. In contrast to drilling, surface-NMR tomography yields average porosity values for the entire subsurface volume. However, the inversion process is sensitive to the electrical conductivity distribution; uncertain conductivity estimates limit the reliability of the inverted water contents. Nevertheless, the results suggest that ridge porosities obtained from invasive measurements such as drilling may lead to substantially overestimated sea-ice volume.

Towards Robust Acoustic Full Waveform Inversion

Seismic full waveform inversion (FWI) is a powerful method for obtaining high resolution subsurface images, but the objective function of FWI algorithms typically contains many local minima, which may lead to erroneous solutions. This can be avoided by choosing an initial model that is close to the true model, and by incorporating only low frequencies at an early stage of an inversion run. Here, we demonstrate that the local minima problem can be mitigated by removing, at each iteration, all data that have more than half a cycle mismatch in the traveltimes compared with those predicted for the current model. Using a simple two parameter example, we show that filtering waveform data by this traveltime criterion can successfully circumvent local minimum trapping. Inversions of synthetic 2D acoustic data show that our novel traveltime-directed FWI approach is more robust and far less dependent on having an accurate starting model.

Seismic Full Waveform Inversion for Characterizing Near-surface Structures - Potential Problems and Solutions

Full waveform inversion (FWI) of seismic data has great potential to image the shallow subsurface, but the specific nature of the associated data sets requires that several problems be first addressed. One problem is the predominance of high-amplitude surface waves, which enforces changes primarily in the shallowest part of the subsurface model when matching observed and predicted data. A more uniform model update over the full depth range of interest can be obtained using Jacobian matrix scaling techniques. A further issue with FWI concerns the occurrence of local minima in the model space. We demonstrate the use of joint inversion of travel times and waveforms to alleviate this problem. Finally, we address the problem of variable source and receiver coupling and propose an estimation procedure that allows such variations to be accommodated and subsequently determined during the inversion.

Combined Inversion of Seismic Waveforms and Traveltimes for Acoustic Waves

Seismic full waveform inversion (FWI) is a powerful method for obtaining high resolution subsurface images, but the objective function of waveform data typically contains many local minima that may lead to inappropriate solutions. This can be avoided by choosing an initial model that is close to the true model, and by incorporating only low frequencies at an early stage of an inversion run. Here, we demonstrate that the local minima problem can also be mitigated by simultaneously inverting waveforms and first arrival travel times. Using a simple two parameter example we show, how the addition of travel time data can remove a local minimum present in the waveform misfit function. Inversions of synthetic 2D acoustic data show that our novel combined travel time and waveform inversion approach leads to superior results compared with a pure FWI inversion.