Jökull - 01.01.2019, Page 110

Remote sensing of snow patches on Tröllaskagi Peninsula, N-Iceland
be ≥0.11 (Paul et al., 2016). For Landsat-5/-7 and
-8 the blue band (Band 1 and 2 respectively) must
be ≥0.15–0.2 (usually 0.16) and the SWIR 1 band
≤0.2. The blue bands and NIR (Band 5 for Landsat-
5/-7 and Band 6 for Landsat-8) distinguishes snow
from clouds. The threshold values indicate spectral
reflectance (Dozier, 1989; Paul et al., 2016).
PSPs believed to be induced by avalanches were
identified and excluded using a digital elevation
model (DEM is50v, 2009), provided by LMI. For
avalanches, the slope angle is one of the most im-
portant parameters and it is well-known that most
avalanches are released from slopes between 30◦ and
50◦ (e.g. Bühler et al., 2013; Schweizer et al., 2003;
McClung and Schaerer, 2006). These angles might be
slightly different depending on snow and weather con-
ditions and therefore we considered slopes between
25◦ and 55◦ as potential avalanche areas. Besides
slope angle, slope curvature also had to be considered
for identification of avalanche channels. We classi-
fied slope curvature in convex and concave areas and
excluded convex shaped areas from the classification:
i) Plan concave areas: plan curvature < -0.2, ii) plan
convex areas: plan curvature > 0.2 and iii) flat areas:
-0.2 < plan curvature < 0.2. The resolution of the
DEM was 25 m which is suitable for detecting most
avalanche channels, except very small ones (Mag-
gioni and Gruber, 2003). Furthermore, the perime-
ter of the channels was determined and areas with a
small perimeter (smaller than 500 m) were excluded,
because they represent very small channels that are
most likely not avalanche channels.
Glaciers were identified based on the Randolph
Glacier Inventory 6.0 (RGI) and, where needed, man-
ually corrected, because the debris covered part was
not always identified correctly. Glacier areas were ex-
cluded from further analyses. To investigate if snow
patches occur in the same areas over a longer time
period, snow patches from different years were in-
tersected. The snow patches were grouped into five
time periods: (i) including Landsat-5 images from
the 80’s and 90’s, (ii) including Landsat-7 images
from around the turn of the millennium, (iii) includ-
ing Landsat-8 and Sentinel-2 images from 2013 until
2016, (iv) including all available years until 2016 and
(v) including all available years until 2017. Unfortu-
nately, in May 2003 one of the sensors of Landsat-
7 broke, therefore subsequent images were not suit-
able for this study (Allen et al., 2010), and subse-
quently we missed a period of 11 years (2003–2013).
Further, the classified PSPs were compared with or-
thophotos, aerial images and field photos for valida-
tion purposes. In a final stage the likelihood of oc-
currence of seasonal overlapping snow patches was
determined. The PSP distribution of each year was
intersected and where snow patches occur more often
the probability that permafrost occurs in this area is
higher.
In the field, we have analysed and photographed
the snow patches. We measured the thickness of the
snow pack and of the underlying ice layer and checked
for the occurrence of ice in the soil surrounding the
snow patch.
RESULTS
In the following section results of the PSP classifica-
tion and the comparison with aerial images, orthopho-
tos and field work photos are presented. Further re-
sults are in appendix A.
Perennial Snow Patch Classification
In the Brimnesdalur valley, only a few snow patches
were identified during 1986–2017 period (Figure 2
and Table 3). The snow patches in Brimnesdalur were
about twice as large from 1999 until 2002 (Figure
2b) and 2013 until 2016 (Figure 2c), compared to the
other periods. Between 2016 (Figure 2d) and 2017
(Figure 2e) the snow fields in Brimnesdalur decreased
in extent by 0.0918 km2 or 25%, suggesting intensive
ablation during the summer/autumn 2017. A similar
temporal pattern is observed for Kerling in the same
periods (Figure 3), where the PSP maximum snow ex-
tent occurs between1999 and 2002 (Figure 3b) and
from 2013 until 2015, but is strongly reduced in 2016
and 2017 (Figure 3d, 3e).
In Almenningar snow patches do not intersect be-
tween any of the investigated periods, but snow occurs
in some years. In both Almenningar and Úlfsdalir
the largest snow-covered area is determined between
1986 and 1990 but is almost as large in the period
JÖKULL No. 69, 2019 109