Jökull - 01.01.2016, Síða 8

Eyjólfur Magnússon et al.
imated by assuming the thickness to increase from 0
by 1 m for every 50 m to the nearest point of the mar-
gin (Figure 2). Even though this is likely to result in
significant underestimation of the thickness at some
locations it fits fairly well with the RES profiles on
the main ice patch.
The assembled glacier thickness data was next
used to calculate a continuous thickness grid with
20 m×20 m cell size, using kriging interpolation
method in Surfer 12 (©Golden Software, LLC). All
referred DEMs are grids in the metric ISN93 conic
conformal Lambert projection (see http://cocodati.-
lmi.is/cocodati/cocodat-i.jsp for definition). The de-
cision to interpolate the thickness rather than the bed
elevation is based on a test where every second pro-
file was removed in such a way that a typical inter-
val between the profiles was 500–600 m instead of
250–300 m. Applying the same kriging interpolation
scheme as above using less dense profiles for: i) the
bed elevation and ii) the thickness, resulted respec-
tively in 11.1 m and 10.6 m RMS error when com-
pared with the omitted profiles.
The thickness map, with zero thickness outside
the digitized margin was subtracted from the 2011
Lidar surface DEM to obtain the first draft of the
bedrock DEM. The first draft of the bedrock (Fig-
ure 5a) was displayed as a contour map and each 10
m elevation contour revised and manually redrawn if
needed. The purpose of the revision was to minimize
features that are artefacts of the kriging interpolation
scheme rather than realistic landforms. The kriging
method is designed for data sets with randomly dis-
tributed data points. The set of RES profiles consti-
tute a data set with high data density along the pro-
files but is undersampled perpendicular to the sound-
ing lines. This results in artefacts such as contour
lines having a tendency to be perpendicular to sound-
ing lines, and less variability in surface slopes in ar-
eas between survey profiles than along them. Figure
5c-d shows two of the most important type of modi-
fications typically carried out during the manual edit-
ing. The former artefact is associated with profiles of
the subglacial mountain sides, which typically show
stepwise structure, indicating edges of eroded Tertiary
lavas (Harðarson et al., 2008) as commonly observed
in the mountains near Drangajökull (Figure 6). The
kriging interpolation tends to smooth out the slopes
in between the profiles. The manually edited con-
tours were however drawn such that the variability
in the slope is much less smoothed, which makes the
stepwise mountain sides much more apparent in the
final product and more similar to the topography of
the mountains nearby. The latter artefact is associated
with both local highs and lows of neighbouring pro-
files. The kriging tends to produce series of isolated
depressions or peaks rather than connecting the lows
and the highs of neighbouring profiles to produces
continuous troughs and ridges. In better agreement
with landforms generated with erosional and tectonic
processes, the continuous option was favoured dur-
ing the manual edition of the contours. Crossover
point locations showing significant discrepancy be-
tween bed elevations obtained from the crossing RES-
profiles were also revised specifically. Such discrep-
ancy was, as explained above, related with areas of
steep bed slope. Contours at such points were gen-
erally edited more in accordance with the profile that
was better aligned along the bed slope direction.
In general the edition of the contours corre-
sponded to less than 10 m elevation modification at
a given location, with few exceptions (e.g. the in-
terpreted continuous ridge in Figure 5d). The largest
change is however in the area above the glacier tongue
of Kaldalónsjökull, between the end of the RES pro-
files (not extended further due to crevasses) and the
ice free area of 1994 (Figure 2). The kriging in-
terpolation resulted in relatively thin ice for all this
area with the two valleys surveyed further upstream
on Kaldalónsjökull diminishing towards the glacier
tongue. Here we interpret more continuous valley
structures. The interpreted topography of these val-
leys in the data gap should however be considered the
most uncertain part in the presented bedrock DEM of
Drangajökull (Figure 7).
To complete creation of the bedrock DEM the
contour lines with 10 m elevation interval were ex-
tracted from the modified bedrock contour map, re-
sulting in x,y,z-coordinates with 25 m spatial inter-
val along each contour line. At a few locations, con-
tour lines with 10 m elevation interval poorly repre-
8 JÖKULL No. 66, 2016