Sensitivity Study for Seismic Waves Guided in an Ice Pack: Influence of the Frequency Content and Snow Layer Thickness Covering the Ice

Studying Arctic sea ice is essential as it plays an important role in climate regulation, influencing weather
patterns, as well as impacting the local ecosystems and living conditions of people. Among different
methods for collecting data and studying sea ice, seismology has proved to be an efficient way to extract
the ice properties, from which the mechanical behavior of sea ice can be explored. Seismic data recorded on
sea ice, using 3-component geophones, are used as a starting point to derive useful information regarding
the ice. To devise an efficient inversion method for deriving ice properties, an effective tool would be
necessary to generate synthetic data in a way that encompasses the physics of sea ice. While there are
approximate solutions to wave propagation problem in a floating ice layer based on plate theory, which is
based on the assumptions of homogeneity of the ice layer and valid at low values of frequency×wavelength,
numerical counterparts such as wavenumber integration method and finite element method have been
also used to to create synthetic waveforms. The numerical methods have shown the limitations of these
approximate solutions in modeling wave propagation; nonetheless, the effects of these limitations on the
estimations of the location of icequakes and thickness of ice need to be investigated.

In this study, these limitations are explored. To do this, two possible scenarios that can happen in
practice are taken into account: (1) when there is high-frequency content in the source generating the
seismic data, and (2) when the physical model includes a snow layer overlying the ice layer. First, we
will show the limitations of the approximate solutions for these two cases by comparing the waveforms,
derived from these approximate solutions, with those of a numerical method at a given distance from
the source. The numerical used here is spectral element method. Then, the effects of these limitations
on the estimations of icequake location and ice thickness are explored in an inversion process, in which
synthetic data are created using the approximate solutions. Results indicate that when there are high-
frequency content in the data and a snow layer on top of the ice, the use of the approximate solutions
to generate synthetic data introduces bias in the estimation of ice thickness and source-receiver distance
in the inversion process. This bias is in the form of underestimations, smaller ice thicknesses and smaller
source-receiver distances. Furthermore, to tackle the biases associated with the inversion method based on
the approximate solutions, a novel strategy is adopted, where a database of simulations using the proposed
numerical method is built for various models of ice and snow. Here the inversion comprises of searching
in the database to find the best ice thickness and source-receiver distance for each icequake. In addition,
the database-based inversion reduces the computational cost. Thanks to this inversion strategy, and
using real data recorded on sea ice, the ice thicknesses along different source-receiver paths are estimated
efficiently, from which a 3D map of ice thickness is constructed.