Introduction
The ice masses on top of Kilimanjaro (Tanzania, East Africa),
the “white roof of Africa”, are the most recognized among the sparse
glaciers in Africa. Three major ice bodies are found on the summit area of
Kilimanjaro today (cf. entire mountain), Furwängler Glacier and the
Northern and Eastern ice fields, which are remnants of a former ice cap which
encircled the Kilimanjaro plateau at the end of the 19th century. Present-day
climatological conditions are not favourable for maintaining these glaciers
and result in an overall negative mass balance of Kilimanjaro's glaciers
.
The recent decline of Kilimanjaro's glaciers is well documented, with changes
in glacier geometry derived from terrestrial and aerial photogrammetry as
well as satellite imagery
.
Ground-based observations document ice loss by terrestrial laser scanning,
comprehensive automatic weather stations (AWS) and a
network of mass balance stakes
; these data serve as input for
modelling mass and energy balance
. In contrast to the extensive
data sets from surface and aerial measurements, little is known so far about
the underlying bed conditions and topography as well as ice thickness
. Consequently, mapping ice thickness complements
monitoring glacier decline and glaciological modelling of the past and future
response of Kilimanjaro's glaciers to climate variability. This especially
concerns the Northern Ice Field (NIF, Fig. ) because of two
competing interpretations that exist regarding the maximum basal ice age and
the mechanism of glacier formation. Ice cores have been drilled at several
locations on NIF's central flat area . The two ice cores
that we refer to in the following, called NIF2 and NIF3, have been
interpreted as continuous paleoclimate records, extending as far back as
11.7 kaBP . Based on
observational as well as modelling considerations, arrived
at an alternative hypothesis, suggesting a cyclic build-up and decay of the
tabular glaciers, with the ice likely coming and going repeatedly throughout
the Holocene. New insights to resolve this ongoing controversy may come from
utilizing NIF's vertical walls to sample directly the glaciers' stratigraphy
for radiometric ice dating and ultra-high-resolution sampling techniques.
Previous attempts at radiocarbon dating of basal ice and also dust layers
from vertical wall sampling have not yet definitively constrained NIF's
glacier age . Notably, only an undisturbed
stratigraphy would allow a seamless link between results from wall samples
and ice cores drilled in the interior of NIF. Complicating the situation is
that the visible stratigraphy at some sections of the ice margin reveals
inclined, converging layers. In addition, basal melting features attributed
to isolated fumarole activities have been observed under plateau ice
, making stratigraphic disturbance by basal melting a
possibility. It is thus not a priori evident to what degree stratigraphic
integrity is preserved at NIF.
Spatial coverage of GPR common-offset profiles. Shown
in (a) are all 100 and 200 MHz profiles in black and grey lines,
respectively. The zoomed-in window (b) (corresponding to the black
rectangle in a) shows the 200 MHz profiles only. The black arrow
indicates the approximate position of the two profiles compared in
Fig. . Orange dashed lines highlight the tabular cliffs of
NIF. Coordinates are in UTM 37M.
Example of processed radargrams recorded with
100 MHz (a) and 200 MHz (b) over the same
profile (position shown in Fig. ). Increased near-surface
reflectivity at 100 MHz coincides with incoherent noise near surface
at 200 MHz (especially between 80 and 100 m) and is interpreted as
meltwater. Note that the horizontal lines in 200 MHz around 550 and
570 ns are artefacts.
In this context, ground-penetrating radar (GPR) offers a powerful tool to
investigate the geometry and internal structure of glaciers and ice sheets,
making GPR nowadays a standard tool in glaciology
e.g.. For non-polar glaciers, GPR is
typically applied to study ice thickness, accumulation distribution and ice
flow . GPR has also
been used successfully for mapping internal reflections in connection to ice
cores on mountain glaciers .
On tropical glaciers, GPR has already been utilized successfully to determine
ice thickness e.g.. However,
to our knowledge this is the first time a ground-penetrating radar survey was
conducted at Kilimanjaro's NIF. The NIF split in two separate
ice bodies in 2012 . We solely focus on the southern
portion remaining on the summit (Drygalski and Great Penck) comprising the
former ice core drilling sites. Hence we use the abbreviation “NIF” in the
following to refer to this southern portion only. Typical for the tabular
glaciers on Kilimanjaro's summit (cf. slope glaciers) the NIF topography is
characterized by a central flat plateau area and near-vertical ice margins
.
Our main objectives are to (i) map bed topography and ice thickness and
(ii) study the internal stratigraphy of NIF through internal reflection
horizons (IRH). In so doing, we devote special attention to evaluating the
stratigraphic integrity of NIF between the ice core drilling area in NIF's
interior and a sampling site on the vertical wall. Although not further
discussed here, samples for radiometric age determination were obtained at
this site on the vertical wall in a previous field campaign led by two of the
authors (M. Schwikowski and D. R. Hardy); results will be published elsewhere. Finally, we
estimate the total ice volume presently remaining at NIF by spatially
extrapolating the GPR-derived ice thickness.
Data and method
The basic principle of a pulsed GPR system is to send an electromagnetic
signal into the ground and to record the signal reflections as a function of
their two-way travel time (TWT). Partial reflections of the electromagnetic
wave recorded as IRH occur at vertical discontinuities in the dielectric
material. From polar studies, IRH are known to coincide with variations in
density, acidity , liquid water content
and changes in crystal orientation fabric . Only
IRH connected to density and acidity variations are typically regarded as
isochrones e.g.. In view of its visible dust bands,
potential presence of liquid water near surface and base and the absence of a
firn column, it is not self-evident what physical causes of IRH can be
expected to dominate at NIF.
Survey set-up and data acquisition
GPR profiles were obtained over the course of 3 days during our expedition to
the Kilimanjaro ice fields in September 2015. We used multiple GPR systems
with centre frequencies of 100, 200 and 800 MHz. Details of the
technical settings and data acquisition are summarized in
Table . GPR profiles were obtained as common-offset (CO)
profiles; i.e. transmitter and receiver are kept at a fixed distance while
being moved over the glacier surface. Positioning was provided by
conventional GPS receivers at approximately 5 m horizontal accuracy.
The spatial extent of the GPR survey was constrained by the tabular structure
of NIF and by its rough surface terrain. The flexible 100 MHz antenna could
be used over rough terrain found at large parts of NIF, especially close to
NIF's northern margins. The flat central area around the AWS and ice core
drilling sites allowed us to also use sled-mounted systems. The 200 MHz sled
antenna was used for mapping the spatial variation of the bed reflection as
well as IRH within the flat central area. We also used an 800 MHz system for
mapping shallow IRH (detected by 800 MHz within roughly the uppermost
10 m). Relative to the 200 MHz profiles, however, the 800 MHz
profiles did not provide additional information and are not further discussed
here. An overview of the spatial coverage provided by all CO profiles is
presented in Fig. . The glacier outline shown in
Fig. was digitized based on an ortho-rectified GeoEye-1
satellite image acquired on 23 October 2012 , consistent
with the methodology described in . Since the corresponding
satellite image was recorded in October 2012, this procedure includes a minor
overestimation of the ice margins (estimated in Sect.
below), which have been continuously retreating.
Overview of GPR systems used and data acquisition settings. In case
of the common-midpoint profile (CMP), number of samples refers to the number of shots to obtain the
profile.
Frequency
Manufacturer
Platform
Shielded
Profile
Trigger
Shot distance
Time window
Samples
(MHz)
(yes/no)
(ns)
100
Malå Geoscience
Rough terrain
No
CO
Time
0.5 s
996
2024
200
Ingegneria Dei Sistemi
Sled
Yes
CO
Wheel
0.05 m
650/800
8192
200
Malå Geoscience
Separable
No
CMP
Manual
0.1, 0.3,..., 3.5 m
100
18
800
Malå Geoscience
Sled
Yes
CO
Time
0.5 s
138/277
2024
Using an additional 200 MHz system with separable receiver and transmitter,
a common-midpoint profile (CMP) was performed at a central location within
the drilling area (Fig. ). Due to technical difficulties in
the field, only a maximum antenna separation of 7 m could be achieved
and only a single CMP was recorded. With one-sided distances, 0.1,0.3,…,3.5 m, a number of N=18 shots were obtained starting from
centre, symmetric and synchronous.
Post-processing of GPR data
We used the standard routines to process GPR data (using Reflexw, Sandmeier
Geophysical Research) including static correction, bandpass filtering and
adding a gain (to compensate for geometrical divergence losses). As opposed
to the wheel-triggered 200 MHz CO measurements, the 100 MHz measurements
were acquired by time triggering. Thus, they were interpolated to equidistant
shots (0.25 m) based on co-registered GPS data. We employed Kirchhoff
migration using a summation width of five traces. Because of the
insignificant amount of firn at NIF, we used the pure-ice value of v=0.168 mns-1 for the electromagnetic wave speed
e.g.. The same constant wave speed of v=0.168 mns-1 was used for travel time–depth conversion. Different
settings were tested and this processing scheme provided the best results
regarding visibility of internal and bed reflection. An illustration of the
processed GPR data for a direct comparison of 100 and 200 MHz CO profiles is
shown in Fig. . In the 200 MHz CO profiles, IRH were traced
visually and additionally supported with a semi-automated phase-following
routine. The bed reflection horizon was tracked visually. For the CMP, a
semi-automated feature tracking routine was used to pick the signal of two
internal reflections (Fig. ). A hyperbolic fit to these
picked reflections yielded an estimated near-surface permittivity of 3.22±0.17 and 3.21±0.35 for the uppermost (3.15±0.11) and (4.62±0.30) m depth, respectively. This corresponds to a wave speed of (0.167±0.004) and (0.167±0.009) mns-1, respectively.
Evaluation of common midpoint profile. Picking was executed by a
semi-automated algorithm tracing a centre feature of the respective pulse
throughout the radargram. Green lines show picks for direct air and ground
wave (no symbol and dots, respectively) and internal reflections (crosses and
diamonds). The lower two reflectors (crosses, diamonds) were used for
permittivity and wave speed estimation. For this purpose a hyperbola fit
(dashed line) was calculated based on the picked values using the time zero
offset as derived from the direct air wave.
Uncertainty considerations
Major contributions to uncertainty in depth of an IRH come from (i) the
vertical resolution provided to determine the TWT of the IRH and
(ii) uncertainty in the electromagnetic wave speed, in our case especially
related to the presence of near-surface meltwater.
Direct comparison of processed GPR profiles with visible
stratigraphy at the vertical wall. The location of the two GPR profiles is
shown in (e). Profile 1 extends to about 1 m before the ice
cliff. Top row: comparison with the full GPR profile (a). A local
trace was extracted by averaging over 2 m at the end of
profile 1 (b) and compared to a deskewed photo of the vertical
wall (c). Note the reflection hyperbola of an open crevasse around
78 m in the profile. The colour-coded lines indicate manually picked
internal reflectors, named IRH 1–5 and discussed in the
text.
Ice surface elevation change at NIF derived from ablation stakes
with at least two consecutive measurements (increasing from n=1 to n=19
stakes, in 2000 and 2015, respectively). The AWS and spatial coverage of
stakes at NIF are shown next to the legend in the upper left (black and red
triangles, respectively). In the top plot, grey box plots represent the
distribution or change in ice height (median, quartiles) at vertical or
near-vertical stakes (< 30∘ tip; height measured along stake).
Thick horizontal blue markers show the mean height change or height change
values for only one measurement (i.e. 2001–2004). When the sample size is
big enough, outliers are shown as black circles above/below the box plot
whisker caps (90th and 10th percentiles). Stakes leaning 30–45∘,
buried by accumulation, and those lying down due to ablation are shown as
“x”, orange and green inverted triangles, respectively; inverted triangles
are thus minimum estimates of surface lowering. The lower plot shows
cumulative ice height changes based on medians (thick blue line) and
quartiles (thin blue) of the box plot data set. Also shown (red line) is
cumulative height change at the AWS; any snow overlying the ice is included
in these heights (e.g. February 2001), accounting for some of the apparent
discrepancies.
Contribution (i) depends on the extension of the GPR pulse and is typically
assumed as half the wave period, or 5 and 2.5 ns for 100 and
200 MHz, respectively . For picking an IRH
(200 MHz) an additional uncertainty component stems from potentially
losing track of the individual coherent phase used for tracing the
reflection. Based on the typical separation in phases of the IRH the related
error was estimated as 4 ns, and twice as much in case of the bed
reflection. The combined uncertainty of the picked travel times is then 5 and
8 ns for IRH and bed reflection at 200 MHz, respectively, and
9 ns for bed reflection at 100 MHz (calculated by error
propagation, root sum of squares). Regarding contribution (ii), wave speed
values derived from the shallow CMP are within 1 % of the pure-ice value
0.168 mns-1. The negligible amount of firn reported in the NIF
ice cores suggests that it is a valid assumption to
neglect firn and snow layers. However, a spatially variable amount of
percolating meltwater (visible in the CO profiles and further discussed in
Sect. below) implies locally increased material
permittivity and hence lower wave speed values. Additional quantitative
information on water content within the ice column would be needed for a
precise calculation of wave speed variability due to meltwater. At the
position of the CMP, the CO profiles do not show exceptionally strong
meltwater presence (corresponding to point “intersection” in
Fig. ). Accordingly, the difference of 1 % between the
CMP estimate and the wave speed of pure ice is regarded as an adequate
uncertainty for sections without meltwater and only as a lower uncertainty
limit where meltwater is present. Based on these considerations, typical
uncertainties are around 1 m for detecting an IRH and around
1–2 m for the bed reflection (and ice thickness). In addition, in
case of a strong surface or bed
inclination the accuracy of the GPR-ice thickness can be limited to less than
16 % if only a 2-D migration is performed . A full 3-D
migration based on a dense survey set-up was beyond the scope of this work,
given that a mostly planar bed is expected at NIF.
Shot distances in data acquisition were chosen less than one-quarter
wavelength apart in order to avoid spatial aliasing (Table ).
This also holds for the 100 MHz measurements that were recorded at a
constant time interval of 0.5 s while pulling the antenna at a
walking speed of about 0.5 ms-1. Horizontal resolution of the
properly migrated radargrams can thus be estimated as half of the wavelength,
independent of reflector depth . This corresponds
to approximately 80 and 42 cm of horizontal resolution for the 100
and 200 MHz profiles, respectively.
Validation of travel time–depth conversion at vertical ice cliff
We used the vertical wall at the southern margin to directly compare the ice
thickness derived from GPR with a photogrammetric estimate. The 200 MHz
CO profile running towards the vertical wall (profile 1 in
Fig. ) ends within about 1 m from the cliff and shows
an ice thickness at the cliff of (37.0±1.5) m
(Fig. ). The height of the ice cliff was independently
estimated by hanging a 16 m long rope with a weight at the end from the
edge. Using the known length of the rope as a reference in a deskewed picture
of the cliff, obtained by photogrammetric processing of 17 multi-view oblique
photographs (using Agisoft Photoscan), yields an independent estimate of the
total height of the cliff of 38 m. To derive a lower estimate of
uncertainty, we assumed 0.3 m uncertainty in the length of the rope at
16 m (resulting from knots tied into the rope) and neglected
stretching of the rope. This translates to (38.0±0.7) m. Further
uncertainty is introduced by the image stitching and deskewing routines. The
software estimates the internal and external camera orientation from the
image data alone. Hence, the quality of the results strongly depends on the
camera positions, image overlap and the object shape . In
comparable applications, related errors in the millimetre and low centimetre
range were found e.g. In our case they
cannot be quantified and were assumed to be negligible.
Interpolation of ice thickness
To derive the ice thickness distribution over the NIF from our 100 and
200 MHz profiles, we essentially followed the approach previously developed
by , first interpolating the bed topography and then
computing the difference between surface and bed elevation. This method (here
referred to as “grid”) allowed us to use not only the GPR-ice thickness
measurements for the spatial interpolation, but also additional topographic
information from the existing digital elevation model (DEM) KILISoSDEM2012
and the position of the glacier margins . The DEM provides
high-resolution (0.5m×0.5m) data of the 2012
surface at Kilimanjaro summit area with 2.12 m LE90 (90 % percentile
linear error) accuracy. No densely distributed information is available
regarding changes in surface altitude over the entire NIF surface between the
acquisition of the DEM (2012) and our radar survey (2015). At the NIF central
flat area, however, ablation stake measurements show almost no systematic
change in mean surface elevation between 2012 and 2015
(Fig. ). We did not use the vertical coordinate of our
conventional GPS measurements for estimating ice thickness, since no
differential GPS was used and thus the uncertainty in altitude is likely
larger than the expected altitude change between 2012 and 2015.
The contour lines of bed elevation below NIF were drawn in 20 m equidistance
constrained by point values calculated by subtracting the GPR measured ice
thickness from the surface elevation (DEM). This was thus done at around 100
different positions using only a minor subset of all GPR data points.
Next, the subsurface bed topography was interpolated within the glacier
outline using the Topo2raster tool in the ArcGis 10.2 software
(parameters: no enforcing, 5 m grid size). Based on
the interpolated bed topography, distributed ice thickness was calculated as
the difference between surface DEM and interpolated bed.
We derived an estimate of the total ice volume by multiplying the mean ice
thickness by the total surface area. For estimating the uncertainty in mean
ice thickness, we used the GPR data points that were previously not used for
interpolation. Calculated at the positions of the respective GPR data points,
the mean of the difference in ice thickness between GPR and interpolation is
(0.63±2.24) m (small insert in Fig. ). To
estimate the surface area lost between October 2012 (satellite image) and
September 2015 (our expedition), the rate of area change reported for NIF by
for the time period 2011.46–2003.08 (Table 2 in
) was used. With an annual surface area loss of
-0.447 ×10-2 km2yr-1 this leads to a correction
of the surface area from 0.525680 km2 (October 2012, from satellite
image) to 0.512643 km2 (September 2015). The latter value was used
to calculate the 2015 ice volume (Table ).
For comparison with our combined GPR–DEM approach, we also considered
interpolation based solely on the DEM and GPR, respectively. For the latter,
we applied ordinary kriging directly to the GPR-ice thickness profiles. To
allow the interpolation of the sparse data, a large grid size had to be
chosen (100 m) and the ice thickness at the outline vertices was set
to zero. Although clearly suffering from these restrictions, we included the
use of kriging for comparison as a method based on the GPR data sets only. In
an approach similar to the grid method, interpolation based on the DEM
only was done by removing all data points over the NIF (including a 10 m
buffer) and interpolating the void using Topo2raster (in
this case without additional constraints from GPR, however). Notably, the
DEM-based approach includes no data from 2015, thus resembling conditions in
October 2012.
Results and discussion
Figure shows processed radargrams from two parallel
sections of 100 and 200 MHz CO profiles. Both profiles display a clear
reflection of the underlying bed (e.g. at TWT of 550–650 ns in
Fig. ) and some near-surface signal disturbance due to
meltwater (e.g. between 80 and 100 m along the profile in
Fig. ). Coherent internal reflectors are well represented in
all 200 MHz profiles (included as Supplement) but appear only to a limited
extent in 100 MHz profiles (due to the coarser vertical resolution at lower
frequency). The following discussion of results focuses on three main
features of the radar profiles, namely (i) bed reflection and ice thickness
estimation, (ii) internal layer architecture within the NIF central flat
area and (iii) meltwater disturbance.
Mapping ice thickness at NIF
The validation of the GPR-derived ice thickness at the ice cliff with a
photogrammetric estimate confirms a reliable estimation of ice thickness when
using the constant pure-ice wave speed for the travel time–depth conversion.
For intersecting or overlapping 100 and 200 MHz profiles, the TWT of the bed
reflection and hence also values for ice thickness are consistent within
their uncertainty, typically within less than 1 m
(Figs. and ).
Ice thickness derived from spatial interpolation using a combination
of GPR and DEM data (“grid”), GPR only (“kriging”) and DEM only
(“DEM”). Uncertainties for grid and kriging are estimated from comparison
with unused GPR data points (see text). For the DEM method uncertainties are
reported as 1 standard error.
Method
Grid
Kriging
DEM
Range (m)
2.0–54.0
0.0–53.5
0.0–55.5
Mean thickness (m)
23.3±0.6
21.2±1.0
27.2±2.5
Ice volume (106 m3)
12.0±0.3
10.9±0.5
14±1
Area (m2)
512 643
512 643
525 680
Date
Sep 2015
Sep 2015
Oct 2012
Figure shows the colour-coded ice thickness along all
acquired CO GPR profiles. Ice thickness ranges from around (6.1±0.5) m at the western margin to a maximum ice thickness of (53.5±1.0) m on the eastern part of the central flat area. Ice thickness
within the central drilling area is typically around 46 m. At the ice
core drilling sites NIF2 and NIF3, ice thicknesses of 50.8 and 49 m,
respectively, were reported by for the year 2000,
without uncertainty. In 2015, our GPR-derived ice thickness at NIF2 and NIF3
is (44.7±1.7) and (42.4±1.5) m, respectively. This
corresponds to a loss in ice thickness of (6.1±1.7) and (6.6±1.5) m at NIF2 and NIF3, respectively, between 2000 and 2015. Since
neither NIF2 nor NIF3 feature large surface or bed inclination (migration issues) or pronounced presence of
meltwater (Fig. ), the uncertainty in GPR-ice thickness
seems to be well represented by our previous considerations. For the time
period 2000–2015, ablation stakes at the NIF plateau show an average change
in surface elevation of around -4.0 m, with an uncertainty range
between -3.4 and -5.3 m (Fig. , bottom plot).
For February 2000 to 15 September 2015, the cumulative surface height change
measured by two ultrasonic sensors at the AWS, close to NIF2, is
-4.24 m.
Ice thickness derived from the bed reflection in 100 and 200 MHz
GPR profiles (a). The NIF outline is highlighted (red). Shown
in (b) is the NIF on an ortho-rectified GeoEye-1 satellite image
acquired on 23 October 2012 . Note the crater rim likely
extending below the NIF. Coordinates are in UTM 37M.
Interpolated ice thickness based on GPR profiles and the digital
elevation model (a). The maximum ice thickness is around
53.5 m in approximately the centre of the glacier, in association
with the highest elevation. The bottom row (b) shows the result of
the alternative interpolation method by kriging. Due to sparse coverage by
GPR, kriging had to be applied at coarse resolution. Black squares indicate
zero ice thickness. Shown on the right side is the value distribution of the
difference calculated between GPR measured and interpolated ice thickness at
GPR data points unused for interpolation (see text). Coordinates are in
UTM 37M.
Multiple views of the stratigraphy at NIF's vertical walls. The top
row (a) shows the ice front near the vertical wall sampling site
with a person standing next to the AWS. Note the exceptional inclined dark
layer merging with another horizontal layer close to the crater surface. The
ice front reveals distinct layers that are predominantly horizontal, seen
both on the north and south side (b and c, respectively).
Examples of the distinct layers are highlighted as green lines.
Tracing IRH in a closed course along all 200 MHz profile. Shown in
the colour-coded reflector depth of IRH 3 (a) and IRH 4 (b).
Except for IRH 4 at the eastern end, reflectors can be connected at all
intersections of two profiles. Coordinates are in UTM
37M.
Although ice loss values obtained from the GPR-ice core comparison and
ablation stakes agree within their estimated uncertainties, it seems worth
mentioning that GPR-ice core derived ice loss is systematically larger than
the ablation stake measurements. In this context we also note that, in
principle, a contribution to the difference between ice thickness and surface
elevation change could result from basal melting. Basal melting caused by
fumarole activity has been observed at NIF and by two of
the authors (D. R. Hardy and N. J. Cullen) at multiple locations. In contrast, our GPR
data generally shows no clear evidence of basal cavities that could result
from pronounced subglacial fumarole activity. To match the observed
difference between ice thickness and surface elevation change, basal melting
would need to occur below the central flat drilling area on NIF, at a slow
rate and without large spatial gradients. In the absence of GPR evidence for
basal fumarole activity and lacking quantitative information on basal
melting, it seems more likely to attribute the observed systematic difference
in the two ice loss estimates to the uncertainties involved in GPR and
ablation stake measurements, combined with spatial variability of ablation
rate and, to a minor extent, a potential discrepancy in the ice core length.
The interpolated ice thickness distribution is shown in
Fig. . Ice at NIF reaches a maximum thickness of
54.0 m at approximately the highest-elevation area of the glacier,
along an east–west trending ridge roughly at the centre of the remaining ice
field. A second area exceeding 40 m in thickness was identified
towards the eastern end. A large area of ice thicknesses less than
10 m is found towards the western margin. The low ice thickness is
likely a result of the surface gradually sloping off towards the west outside
the caldera. A distinct rise in the local GPR bed reflection appears where
the location of the crater rim below the ice is suggested by satellite images
(Fig. b). Accordingly, the assumption of a generally flat
bed topography does not hold everywhere below NIF. This finding supports the
idea that local bed relief features may have affected past ice build up and
decay through limiting exposure to solar radiation and wind
.
Table summarizes our estimates of mean ice thickness and ice
volume based on combining GPR and DEM (grid) and the interpolation based on
GPR (kriging) and DEM only, respectively. The assumption of zero ice
thickness at the margins used for kriging certainly does not apply to the
western margin. This results in an underestimation of volume as compared to
the grid approach. In addition, considering the coarse resolution used in the
kriging approach, we interpret the ice thickness derived from this method
with caution only. The estimates of total ice volume obtained from the grid
approach and DEM-only are (12.0±0.3) and (14±1)×106 m3, respectively. The main contribution to the difference in
ice volume comes from different mean ice thickness values as opposed to
surface area (using the 2012 surface area the mean ice thickness obtained
from the grid method gives a volume of (12.3±0.3)×106 m3). The decrease in mean ice thickness suggested by the
comparison of the two interpolation methods is not supported by surface
height change measurements 2012–2015. Since both interpolation methods use
the same surface topography supplied by the DEM as input, the difference in
mean ice thickness has to come from differences in determining subglacial bed
topography. Consequently, the difference in ice volume estimates is not used
to infer a rate of ice loss. Integrating both the DEM and GPR as constraints,
the grid method provides the most reliable ice volume estimate. We
acknowledge that (i) one volume estimate does not allow us to infer retreat
rates and (ii) for predicting the expected lifetime of NIF under ongoing ice
loss conditions, a simple linear extrapolation based on current rates of
lateral and surface retreat likely produces unrealistic values. Nonetheless,
this first-ever quantification of NIF's ice volume based on direct GPR
measurements of ice thickness provides an important context to the discussion
of ongoing glacier retreat .
Internal layer architecture within the NIF plateau
All 200 MHz profiles contain a number of coherent internal reflection
horizons except for the lowermost depths. Below typically about 30 m,
reflections still appear in intervals but cannot be traced continuously.
Sections lacking echoes from deep layers often coincide with a large amount of
near-surface scattering, presumably due to the presence of near-surface
meltwater. Absorption from meltwater causes less energy to be returned from
deeper layers and hampers the detection of deep IRH. This explanation implies
that coherent internal layers may still exist at greater depths but cannot be
detected continuously anymore by GPR. It is worth noting that the vertical
cliffs show instances of tilted and converging layers in close proximity to
bed (Fig. ) which can also hamper the detection of deep
reflectors. We believe that this stratigraphic convergence is an ablation
feature rather than due to rheology (e.g. localized shearing at the glacier
margin), as localized shearing appears evident only near the snout of the
steepest slope glaciers, and features such as that shown in
Fig. occur elsewhere on Kilimanjaro glaciers, particularly
on the south side. The GPR profiles towards the western end are the only case
in which adjacent IRH (representing boundaries to a layer of ice) are found
merging together (see Fig. S1 in the Supplement, profile D). While we find no
evidence of converging IRH in the central flat area of NIF, it is not
possible to generalize this result also for the lowermost metres of basal ice
where distinct IRH are absent.
Two-way travel times and (absolute and relative) depths of internal
reflections (IRH) traced between the ice core drilling sites NIF2 and NIF3
and the vertical wall sampling site (“wall”). Horizontal distances are
measured along the profiles from their intersection. Profile numbers refer to
the legend in Fig. .
Profile
1
1
1
2
2
Position
wall
NIF3
intersection
intersection
NIF2
Distance (m)
35
60
0
0
45
IRH 1 (ns/m/%)
74/6.2/17
84/7.1/17
86/7.2/17
87/7.3/17
91/7.6/17
IRH 2 (ns/m/%)
108/9.1/25
118/9.9/23
119/10.0/23
121/10.2/24
131/11.0/25
IRH 3 (ns/m/%)
187/15.7/42
198/16.6/39
203/17.1/40
202/17.0/40
221/18.6/42
IRH 4 (ns/m/%)
266/22.3/60
270/22.7/54
280/23.5/55
283/23.8/56
325/27.3/61
IRH 5 (ns/m/%)
324/27.2/74
317/26.6/63
331/27.8/65
333/28.0/66
396/33.3/74
bed (ns/m/%)
440/37.0/100
505/42.4/100
507/42.6/100
503/42.3/100
532/44.7/100
The comparison of the GPR signal with the photo of ice cliff shows that
distinct reflectors occur at depths where dark dust bands are visible
(Fig. ). The glacio-chemical analysis performed by
on the NIF ice cores shows that dust layers coincide
with a strong increase in the concentrations of nearly all ion species,
including ammonium and chloride. It is plausible that the according change in
the electrical conductivity of the ice layer produces a strong reflector seen
in the GPR data . Accordingly, this strongly suggests dust
layers being a main physical cause of IRH at NIF. and
report visible dust layers in the NIF2 and NIF3 ice
cores (one layer at 32.5 m in NIF3, four layers at 26 and 29 m and
two each around 32 m in NIF2, all in reference to 2000). With the
upper ice surface thinning estimated (from difference between the borehole
depth, 2000, and the GPR-derived ice thickness, 2015) as (6.1±1.7) and
(6.6±1.5) m for NIF2 and NIF3, respectively, these layers are
expected at around (25.9±1.5) m (NIF3) and (19.9±1.7),
(22.9±1.7) m and (25.9±1.7) m (NIF2) depth in 2015.
We visually identified five prominent reflectors, consecutively labelled
IRH 1–5 with increasing depth (Table ), in profile 1 shown
in Fig. .
In order to trace IRH 1–5 along multiple profiles, they are linked at the
intersections of the profiles by checking for consistent TWT, or depths
(Fig. and Table ). Connecting all 200 MHz
profiles in this manner, IRH were followed proceeding along a closed course,
successfully demonstrating that it is possible to return to the same
depth–travel time after a complete round course. While it is possible to trace
IRH 5 between the vertical wall and the drilling sites NIF2 and NIF3, IRH 4
is the deepest reflector that is traceable almost uninterruptedly throughout
all profiles, with a short exception towards the eastern end and below the
crater rim towards the west (Fig. ). This spatial extension
of IRH within NIF suggests that, at least within the area mapped by 200 MHz
profiles, IRH stem from continuous reflecting surfaces that can be associated
with a corresponding dust layer. We thus conclude that the internal
stratigraphy within the NIF central flat area is generally composed of
uninterrupted, spatially coherent layers (as opposed to deformed,
macroscopically disturbed layers). A potential exception to this finding is
ice just above the bed where GPR can neither support the existence of
disturbances nor their absence.
The continuous layering mapped by GPR demonstrates that, in general, the
internal layering is intact between the ice margin and the interior of the
NIF plateau area. More specifically, the link between IRH and major dust
layers implies that IRH represent isochrones and, thus, can be used to
extrapolate and compare age–depth information. This GPR-based tracking of
isochrones has been employed successfully not only on polar ice sheets but
also at small-scale mountain drilling sites . At
NIF, the tracing of IRH provides a quantitative link between isochrone depths
at existing sampling sites, thereby revealing important constraints for
future efforts at integrating age–depth information obtained from the NIF ice
cores and the vertical wall. Table summarizes the respective
isochrone depths obtained from tracing IRH 1–5 between NIF2, NIF3 and the
vertical wall sampling site (cf. Fig. ).
With respect to the two ice core drilling sites, related isochrone layers are
found at lower relative depth at NIF3 than at NIF2 (Table ).
Comparing the main features of the stable water isotope records of the NIF2
and NIF3 ice cores, developed a matching of the two ice
core depth scales that is qualitatively going in the same direction (i.e.
Fig. d and Table of this study in
comparison with Fig. 2 in ). On a quantitative level,
however, tracing IRH between NIF3 and NIF2 yields tie points that are
systematically at lesser depth in NIF2 as compared to the ice core stable
isotope matching. For instance, the thick layer at (25.9±1.7) m
(for 2015) in NIF3 (reported as 30 mm thick by )
appears to correspond with IRH 5 found at NIF3 around (26.6±0.6) m. interpreted this layer in NIF3 to be
aligned with the base of NIF2 (17.5 m deeper). Our findings raise
questions about this interpretation (Table ). In this respect
it is worth noting that our findings do not change significantly if the
average change in surface elevation of around -4.0 m is used in the
above correction for the 2000–2015 surface thinning.
Effects of near-surface meltwater
As illustrated by the 100 and 200 MHz profiles in Fig. ,
incoherent near-surface noise in 200 MHz radargrams coincides with
increased near-surface reflectivity in the 100 MHz data. This characteristic
is observed throughout all profiles at great spatial variability and is
interpreted as backscatter due to meltwater. This effect can extend to
substantial depths (at times more than 10 m), probably where
meltwater percolates through cracks or small crevasses. The abundant presence
of englacial meltwater was confirmed by shallow mechanical drillings at the
NIF central area during the 2015 expedition and has been observed
intermittently since February 2000 by one of the authors (D. R. Hardy). Even during
the early morning hours, shallow boreholes (around 0.6 m depth) filled with
meltwater in 15–20 min. Hence our GPR profiles demonstrate a highly
heterogeneous presence of meltwater near the surface, apparently a
widespread feature at NIF related to spatial and temporal variability in
surface characteristics and processes . This finding is of
relevance for any new ice core drilling efforts at NIF in the future,
suggesting that chemical and isotopic records of the upper 10 m or
more could be potentially corrupted by meltwater. The widespread presence of
near-surface meltwater also needs to be considered in future energy and mass
balance modelling efforts . Further quantifying the
generation and evolution of the near-surface meltwater distribution points to
important future research questions at NIF.
Conclusions and outlook
The application of ground-penetrating
radar at Kilimanjaro's Northern Ice Field provided a direct estimate of the
remaining ice volume. For the central former drilling area, the radar
profiles reveal macroscopic coherent, uninterrupted ice layering for at least
the upper 30 m, and demonstrate abundant meltwater in the top
10 m. The latter finding suggests that the upper part of future
chemical and isotopic ice core records could potentially be corrupted by
meltwater. The association of internal reflections seen by GPR with dust
layers becomes evident from using NIF's vertical walls to compare the local
GPR signal to the visual stratigraphy. The internal reflections were traced
consistently within our 200 MHz profiles, indicating that the stratigraphic
integrity is preserved. Tracing internal reflections provided a link of
isochrone depths among the former ice core drilling sites and the vertical
wall sampling site. This link implies valuable constraints for future efforts
at integrating age–depth information obtained from the NIF ice cores and the
vertical wall. Accordingly, our results contribute to future attempts at
resolving the ongoing debate on NIF's age structure and glacier history.
For the first time on NIF, our GPR measurements provided widespread ice
thickness soundings. In combination with the existing DEM this allowed us to
estimate the total ice volume remaining at NIF's southern portion as
(12.0±0.3)×106 m3. These data contribute to the
understanding of ongoing glacier loss and will support existing glacier
monitoring databases. Regarding future drilling efforts at NIF, the presented
data can aid the selection of potential coring sites through the newly gained
information on ice thickness and bed topography as well as the heterogeneity
in the presence of liquid water near the surface. Although connected to
substantial logistical effort, repeat measurements of ice thickness would
offer a precise method to support future studies on the ice loss at NIF,
especially in terms of spatial variability. Moreover, the application of GPR
could be extended with great benefit also to monitor ice thickness at the
other major ice bodies remaining on Kilimanjaro.