3 Methods and Data
3.1 Remotely Sensed Imagery
High-resolution aerial imagery was captured during the summers of 2016,
2017 and 2018 using the DJI Phantom 4 drone. A total of between 600 and
1000 usable aerial images were collected in each survey. Ten black and
white markers were distributed throughout the site to use as ground
control points. These were georeferenced using real-time kinematic
differential global navigation satellite system to produce centimeter
scale locational accuracy. Drone images were processed using the SfM-MVS
method to generate high-resolution DEMs and georegistered orthomosaics.
SfM-MVS is a photogrammetric range imaging approach that allows for high
resolution 2.5D surface reconstruction through the analysis of
overlapping 2D images. Images can be captured using consumer grade
digital cameras (Carrivick et al., 2016; Westoby et al., 2012), and can
produce results with similar levels of accuracy to terrestrial laser
scanners, but with a fraction of the cost (James and Robson, 2012;
Westoby et al., 2018). The 2016 drone images were processed using Pix4D
software. The remaining data were processed using Agisoft Photoscan
1.2.4© (Agisoft, 2016), with all point clouds finely co-registered using
the CloudCompare (CC) software (CloudCompare v2.7.0, 2020).
3.1.1 Model Co-Registration
The 2016-point cloud was chosen as the base model on which the other
SfM-MVS models were registered, accuracy assessed and rates of HWR
determined. Horizontal accuracy was calculated by comparing the position
of discreet ground surface features across the three years. For 2017 and
2018, 10 distinct features were used for comparison, resulting in a root
mean square error (RMSE) for 2017 of 0.12 m for X, and 0.14 m for Y and
an RMSE for 2018 of 0.15 m in X and 0.11 m in Y.
Vertical accuracy was assessed using eight 25 m long elevation profiles
in the undisturbed terrain. For 2017 and 2018 the vertical RMSEs were
0.10 m and 0.20 m, respectively.
3.2 Massive Ice Elevation and Overburden Thickness
Determination of the depth to MI followed to process of Lim et al.,
(2020), but is briefly described here. Seismic noise is present
throughout the lithosphere and can be used to gather data on subsurface
properties. Monitoring of this noise has been used in various geological
settings to determine features such as the depth of subsurface strata or
the different seismic wave velocities of the buried rock and soil
(Cultrera et al., 2012; Scheib, 2014; Tallett-Williams et al., 2016).
Subsurface strata exhibit differences in the amplitude of their vertical
and horizontal motions. This occurs due to impedance contrasts between
layers and differences in the shear and compressional wave velocities of
the materials. This allows for the data to be processed using the H/V
(horizonal to vertical) ratio method (Nakamura, 1989) to identify the
natural resonance frequencies of the subsurface strata. For a simple
two-layer system (in this case the two layers are overburden and MI),
the resonance frequency can be used in the following formula to
calculate the depth of the layer boundary:
\begin{equation}
f_{r}=\ \ \frac{V_{s}}{4h}\nonumber \\
\end{equation}Where \(f_{r}\) is the resonance frequency, \(V_{s}\) the shear wave
velocity and \(h\) is the depth.
In August 2017, passive seismic data were collected in nine locations on
Peninsula Point, as three separate North to South transects with three
measurements each, all using the Tromino® device
(http://moho.world/en/tromino/). One transect was along a
low-ground segment that had recently been affected by RTS activity and
where the active layer extended to the surface of the MI (indicated with
“A” label in Figure 3). Two further transects were along undisturbed
high ground to both the east and west of the first transect that
contained permafrost layers between the ground surface and the MI
(indicated with “U” labels in Figure 3). A further five MI height
measurements were taken from headwall exposures (indicated with “O”
labels in Figure 3).
Processing of the seismic signal allowed for the \(f_{r}\) to be
calculated but to determine the depth to the ice it was necessary to
calculate the \(V_{s}\) of the overburden layers. This was achieved by
selecting two measurement points near the headwall edge, one for the low
ground without permafrost and one for the high ground that contained
permafrost. The observable depth to ice near these measurement points
provided \(h\) values, while the H/V processing method provided the\(f_{r}\) values. By using these control measurement points, the\(V_{s}\) for the two overburden layers could then be calculated. For
the high ground, this produced a \(V_{s}\) of 1059.38 m
s-1, which was then be applied to the five other high
ground points to determine their respective depths to the MI surface.
For the low ground section, doing the same produced a \(V_{s}\) of 213.8
m s-1, a significantly slower velocity, due to the
greater water content of the thawed soil. This value was then applied to
the two other low-ground segments to calculate their respective MI
depths. Finally, a continuous MI surface layer was then created using
the IDW interpolation within ArcGIS 10.3.