Grinnell and Terra Nivea Ice Caps are located on the southern Baffin Island,
Nunavut, in the Canadian Arctic Archipelago. These relatively small ice caps
have received little attention compared to the much larger ice masses
further north. Their evolution can, however, give valuable information about
the impact of the recent Arctic warming at lower latitudes
(i.e.
With a glacierized area of
Located in the southeastern part of the CAA, Baffin Island is the largest
island of the archipelago (Andrews et al., 2002) and has a total ice-covered
area of
Study area.
GRIC and TNIC (Fig. 1) are located on Meta Incognita Peninsula, 200 km south
of Iqaluit, Nunavut. GRIC (62.56
Observations from various expeditions in the 1950s revealed that the western margin of GRIC was relatively stable, but that coastal outlet glaciers (eastern margin) were shrinking moderately when compared to photographs from 1897 (Mercer, 1954, 1956). Moraines studied near both ice caps in the early 1980s indicated that the most recent phase of recession dated from the last 100 years, and that both ice caps probably reached their largest areal extent during the Little Ice Age cold climate interval (Muller, 1980; Dowdeswell, 1982, 1984; Andrews, 2002). Dowdeswell (1982) estimated that the outlet glacier of GRIC that calves into Watts Bay extended much further out a few centuries earlier, but also reported that another outlet glacier to the south of the ice cap was advancing.
Launched on 17 December 2011 and 2 December 2012 respectively, the Pléiades 1A and 1B satellites have recently shown their high potential for glacier DEM extraction and thus, for mass balance estimations (Wagnon et al., 2013; Berthier et al., 2014; Marti et al., 2015). The two satellites follow the same near-polar sun-synchronous orbit and provide panchromatic and multispectral imagery at a very high ground spatial resolution, 0.7 m for panchromatic and 2.8 m for multispectral images, respectively (Astrium, 2012). Both satellites have independent stereoscopic capabilities. The fact that the panchromatic band images derived from Pléiades satellites are coded in 12 bits represents a clear advantage on a glacier surface (especially over the low contrast accumulation area), given the fact that a large radiometric range provides better contrast and reduces the risk of image saturation (Berthier et al., 2014).
Three stereoscopic pairs were acquired over our study area (Table 1): one
for GRIC (3 August 2014) and two for TNIC (14 August 2014
for the eastern part and 26 August 2014 for the western part, with
an overlapping area of 84 km
Elevation data sets used in this study with the acquisition date and the purpose of their use for each ice cap.
Historic Canadian Digital Elevation Data (CDED, Natural Resources Canada), provided at a scale of 1 : 50 k, were acquired for the two ice caps. These elevations were derived by stereo-compilation of aerial photos acquired during the summers of 1958 and 1959. Raw elevations are orthometric and referenced to the Canadian Gravitational Vertical Model of 1928 (CGVD1928). The average elevation differences and their standard deviation (SD) between CDED and ICESat laser altimetry were previously calculated off-glacier for 340 CDED maps tiles covering Baffin Island and were reported to be 1.1 and 5.1 m, respectively (Gardner et al., 2012). Here, CDED were used (1) as historical elevations for TNIC and (2) elevations of the surrounding ice-free terrain were used for absolute co-registration for both ice caps (see Sect. 4.3). Artefacts (unrealistic elevations) located in the accumulation area of TNIC were manually identified and deleted using a shaded relief image derived from the DEM. These artefacts were likely due to the poor contrast and low texture of the 1958/59 aerial photos used to generate the CDED.
Historic aerial photos covering GRIC were obtained through the Canadian
National Air Photo Library (Natural Resources Canada). We used 24 photos
acquired at the end of the ablation season, on 21 and 22 August 1952.
A Williamson Eagle IX Cone 524 camera type with a focal
length of 152.15 mm was used and the flight altitude was 16 000 ft (
Surface elevation profiles (GLA14, Release 634) collected by the Geoscience
Laser Altimetry System (GLAS) onboard ICESat were acquired (Zwally et al.,
2002). Each laser pulse-derived footprint corresponds to field-of-view with
a diameter of
Products derived from the ASTER satellite mission have been widely used for
glaciological studies (e.g. Kääb, 2008; Nuth and Kääb,
2011; Das et al., 2014). To estimate the recent mass balance for TNIC, we
used a DEM (product AST14DMO) generated from an ASTER stereo pair acquired
on 3 August 2007. The DEM was automatically derived from bands 3N
(nadir-viewing) and 3B (backward-viewing) that have an intersection angle of
27.6
In April 2004, a team from the Geological Survey of Canada measured three surface elevation profiles at 50 m horizontal intervals using a Trimble® high-precision Real-Time Kinematic GPS system on the southeast, west and northwest sides of GRIC, and at the front of one of its outlet glaciers (Zdanowicz, 2007). Data acquisition was made using a fixed base station on a geodetic benchmark monument, and GPS positions were subsequently processed with the Canadian Center for Remote Sensing's Precise Point Positioning (PPP) System to obtain an accuracy of a few centimeters. For this paper, those transects were used for recent elevation change calculations. Elevations derived from those GPS measurements are referenced to the GRS80 ellipsoid which can be assumed equal to the WGS84 ellipsoid (sub-mm differences).
Various data sets have been used to extract the areal extent of the two ice
caps at the end of the ablation season (August/September). For GRIC, three
data sets from different dates were used. The 1952 outline was derived
manually from the orthorectified historical aerial photos. For 1999, we used
the ice cap contour from the Randolph Glacier Inventory (RGI 3.2; Pfeffer et
al., 2014), which originates from the Canadian CanVec data set for this
region, itself derived from a September 1999 Landsat 7 image. We manually
digitized the 2014 margin from the orthorectified panchromatic Pléiades
image. For TNIC, outlines were derived for four different dates. We used the
raw vectors from the 1 : 250 k Canadian National Topographic Data Base as the
1958/59 boundary. Anomalies were found in the delineation of the 1999 margin
from the RGI 3.2 (i.e. off-glacier snow patches erroneously included). As an
alternative, we manually digitized the ice cap margin using a 30 m
resolution Landsat 5 image acquired on August 1998. The August 2007 limit
was manually traced from an ASTER orthoimage (15 m resolution) provided with
the on-demand AST14DMO product, while the 2014 margin was extracted from the
orthorectified panchromatic Pléiades images (East and West). To overcome
the cloudiness on the Pléiades orthoimages
(
To quantify changes in the regional climate of the southern Baffin Island region over the period covered in our study, air temperature records were retrieved from the Adjusted and Homogenized Canadian Historical Climate Data of the Iqaluit weather station for the period 1950–2014 (Vincent et al., 2002). This is the permanent weather station in the eastern Canadian Arctic with the most continuous records, extending back to 1946. In addition, time series of sea ice cover area for Hudson Strait and Davis Strait were obtained from the Canadian Ice Service archives over the 1968–2014 period.
The Pléiades DEMs were generated using the OrthoEngine module of Geomatica 2013. No GCP were available for the geometric correction so we relied on the RPCs provided with the images. Adding GCP does not improve the vertical precision of the Pléiades DEM, but can reduce the vertical bias (Berthier et al., 2014). The latter bias can be corrected over ice-free terrain when a good reference data set, such as ICESat, is available (Nuth and Kääb, 2011).
The following steps of DEM extraction were repeated for the three Pléiades
stereoscopic pairs. First, we collected 20 tie points (TPs) outside and six on
the ice cap. Collecting well-distributed TPs was found to improve the
relative orientation between the two images providing increased coverage
(Berthier et al., 2014). The following processing parameters were used for DEM extraction: the relief type was set to
Since the ice-free zones on our Pléiades DEM were not large enough to calculate an elevation accuracy with a sufficient number of ICESat points, we report here the vertical precisions obtained in recent glaciological studies. Wagnon et al. (2013) measured a precision of 1 m (SD) on a glacier surface in Himalaya using Pléiades DEM. Berthier et al. (2014) also obtained a precision ranging between 0.5 and 1 m (SD), highlighting the consistent precision over glacier surfaces. This precision was shown to be mostly correlated with slope. For the small Ossoue Glacier (French Pyrénées), the precision was slightly lower at 1.8 m (Marti et al., 2015). A similar vertical precision is expected here.
Photogrammetry is widely used in glaciological studies for reconstructing glacier surface prior to the modern satellite era (Fox et Nuttall, 1997; Barrand et al., 2009). In this study, a 1952 DEM of GRIC was created from historical air photos using OrthoEngine. This software uses a mathematical model compensating for both terrain variations and inherent camera distortions (PCI Geomatics, 2013). A typical photogrammetric procedure was then followed to compute the model, solving the least-square bundle adjustment.
Collecting effective GCPs for photogrammetry in mountainous or polar regions
remains one of the main difficulties, especially for archive photos (Barrand
et al., 2009). To overcome this difficulty, Pléiades-derived products
(DEM and orthoimage) were used to collect GCPs. For each aerial stereoscopic
model partially covering the surrounding ice-free terrain, 3 to 7 GCPs were
collected outside the ice cap on topographic or geomorphologic structures
visible on both the Pléiades orthoimage and the aerial photographs. In
order to strengthen the mathematical model, every GCP was collected as
stereo GCP (i.e. was identified in all possible aerial photographs). A total
of 39 stereo GCPs were collected resulting in 106 GCPs. Also, 6 to
10 widely-dispersed TPs were collected for each aerial stereoscopic model. For
the models situated in the middle of the photogrammetric block and covering
only the ice cap (no ice-free terrain), only TPs were collected in order to
connect them to the photogrammetric block. After the final bundle
adjustment, the resulting residual averages of all the GCPs were 2.85 m in
Validation of the resulting DEM (before co-registration) against 76 ICESat
points on ice-free terrain showed a mean offset of
DEM co-registration is of primary importance before performing any DEM-based
volume change calculations (Nuth and Kääb, 2011). This 3-D
co-registration method uses the relationship between aspect, slope and
elevation differences over ice-free terrain (Nuth and Kääb, 2011).
The Pléiades images only included a small corridor (
The two independently co-registered Pléiades DEMs of TNIC (14 and 26 August)
were compared in their overlapping zone of 84 km
For both Grinnell and Terra Nivea Ice Caps, recent elevation changes were measured between 6 ICESat tracks from different laser overpass periods (autumn 2003 to winter 2007) and the 2014 Pléiades DEMs. For GRIC only, elevation changes were also calculated between the April 2004 in situ GPS transects and the 2014 Pléiades DEM. We did not attempt to compute glacier-wide volume or mass changes from those recent elevation changes measurements since (1) they are sparse and only cover a small fraction of the two ice caps and (2) the GPS and some of the ICESat data were acquired at the end of winter, and limited data were available to apply a seasonal correction. Nevertheless, those recent elevation changes along selected tracks were used to complement the differential DEM analysis described below.
The geodetic method was applied in order to calculate glacier-wide elevation and mass balances from the DEMs. The following steps were performed for each calculation.
First, the co-registered DEMs were subtracted to obtain maps of elevation
changes (d
The area-averaged change in elevation over the entire ice cap
(glacier-wide), d
Finally, the area-averaged specific geodetic mass balance rate (m a
The main sources of uncertainty in our mass balance estimates are related to uncertainties in the elevation change measurements, the ice cap limits and the density used to convert volume to mass changes. For historical measurements, elevation change uncertainty was assumed equal to the standard deviation over stable terrain between the two coregistered DEMs (GRIC 1952–2014: 13.8 m; TNIC 1958–2007: 9.6 m; TNIC 1958–2014: 9 m), assuming that elevation errors were 100 % correlated. This is a conservative approach that takes into account both the highly correlated CDED elevation errors (Gardner et al., 2012) and the possible errors associated to the aerial photos-derived DEM (i.e. artefacts and low coverage at higher altitudes).
Spatial autocorrelation between the ASTER 2007 and Pléiades 2014 DEMs
was analyzed on ice-free terrain to better characterize the elevation change
uncertainty in the recent mass balance estimation on TNIC. A low
autocorrelation distance (
For ice caps outlines of 1998 and later, we estimated an error of 3 %.
This estimate includes possible image interpretation errors (
Areal changes measured for Grinnell and Terra Nivea ice caps since the 1950s
are shown in Fig. 2. GRIC experienced a mean rate of areal change of
Maps of historical and recent elevation change rates (d
The glacier-wide rates of elevation change (d
Elevation change rates sharply increased in recent years for both ice caps.
On TNIC, the recent (2007–2014) d
On GRIC, changes in d
Elevation change rates (m a
Additionally, elevation changes measured between ICESat repeat track
transects and the Pléiades DEMs over both GRIC and TNIC between 2003 and
2014 are shown in Fig. 7. This analysis reveals a similar range of
variability of annual elevation changes between both ice caps during the
2003–2007 interval and a coherent pattern of seasonal to inter-annual
fluctuations. The absolute difference in elevation change between 2003 and
2014 for the two ice caps (total thinning of
Mass balances for both ice caps are summarized in Table 2. Between 1952 and
2014, a mass balance of
Historical and recent glacier-wide mass balances for both ice caps.
Elevation change rates (m a
Recent elevation differences on GRIC and TNIC measured between the
Pléiades DEMs (2014) and ICESat altimetric points (2003 to 2007). Only
the complete ICESat tracks available for both ice caps were used. The trend
lines indicate the mean rate of elevation changes along these two ICESat
reference tracks and are
Representation of the same geomorphological feature on ice-free terrain surrounding GRIC using three different technologies, namely an aerial photography (August 1952), a Pléiades panchromatic band (3 August 2014) and a Landsat 8 panchromatic band (15 August 2014). Note the very fine resolution of the Pléiades panchromatic band (70 cm), in comparison to the Landsat 8 panchromatic band (15 m), allowing to retrieve bedrocks and ice-free features on archives aerial photos and thus to collect GCPs (e.g. at the bedrock localised by the yellow arrow).
In many regions of the world, vast archives of historical aerial photographs
represent a potential gold mine for glaciologists in order to document
multi-decadal volumetric change of glaciers and ice caps (e.g. Soruco et
al., 2009; Zemp et al., 2010). DEMs generated from these aerial photographs
allows to put the recent glacier variations (satellite era) in a longer-term
perspective. However, these data remain difficult to exploit due to the
logistical difficulties involved in the field collection of accurate and
well-distributed GCPs in the remote high latitude/altitude regions.
Field GCPs were also lacking for the two ice caps studied here. Instead, we
took advantage of the very high resolution of the Pléiades imagery (0.7 m)
and the vertical precision of the derived DEMs (
Our estimates of shrinkage for GRIC and TNIC can be compared with other studies from Baffin Island to verify the coherence of results and get a more complete picture of the pattern of glacier changes across this vast region.
Sharp et al. (2014) reported rates of areal change for TNIC of up to
Gardner et al. (2012) estimated that the average mass loss rate of all
glaciers and ice caps on southern Baffin Island (South of 68.6
The accelerating recession of glaciers and ice caps across the CAA in recent decades, and the concurrent increase in surface melt over the Greenland Ice Sheet, have been ascribed to a sustained atmospheric pressure and circulation pattern that favors the advection of warm air from the northwest Atlantic into the eastern Arctic and over western Greenland (Sharp et al., 2011; Fettweis et al., 2013). This situation has led to warmer, longer summer melt periods on glaciers of the eastern CAA, and this largely accounts for their increasingly negative mass balance (Weaver, 1975; Hooke et al., 1987; Koerner, 2005; Sneed et al., 2008; Gardner and Sharp, 2007; Gardner et al., 2012).
In the southern Baffin Island region, annual and seasonal mean air
temperatures have generally increased since 1948 (except in the spring), but
not monotonically (Vincent et al., 2015). At Iqaluit, seasonal trends from
1948 to
A factor that may have indirectly contributed to the accelerating rate of
glacier recession on southernmost Baffin Island is the decline in summer sea
ice cover in this region (Fig. 9b), one of the steepest observed across the
entire CAA (up to
This paper highlighted historical and recent trends in area, elevation and mass changes for the two southernmost ice caps of the Canadian Arctic Archipelago, Grinnell and Terra Nivea Ice Caps. Our analysis is based on multiple data sets and uses an original approach where ground control points for the photogrammetric processing of old aerial photographs are derived from sub-meter resolution Pléiades satellite stereo-images. This approach takes full advantage of the highly precise Pléiades products and represents an important advance for eventually unlocking the vast archives of historical aerial photographs.
Results show that the areal extent of TNIC is 34 % smaller in 2014 when compared to the end of the 1950's extent, while GRIC shrank by nearly 20 % between 1952 and 2014. Both ice caps also experienced an acceleration of their shrinkage rates since the beginning of the 21st century.
The historical glacier-wide mass balance for GRIC was estimated to be
The 2007–2014 mass balance of TNIC is among the most negative multi-annual glacier-wide mass balances measured to date, comparable to other negative values observed in the southern mid-latitudes (e.g. Willis et al., 2012; Berthier et al., 2009) or in southeast Alaska (Trüssel et al., 2013). Given the absence of calving glaciers for TNIC, its high rate of mass loss can only be explained by negative surface mass balance due to an ELA that, for most years, is above the maximum ice cap altitude. Nonetheless, this similarity in rate of mass loss underlines the strong sensitivity of maritime low-elevation ice bodies to the currently observed climate change at mid-latitudes and in polar regions (Hock et al., 2009). The recent acceleration of ice cap wastage on Meta Incognita Peninsula is linked to a strong near-surface regional warming and a lengthening of the melt season into the autumn that may be reinforced by sea ice cover reduction and later freeze-up in Hudson Strait and nearby marine areas.
This paper is dedicated to Gunnar Østrem (PhD Stockholm Univ., 1965) a tireless pioneer in the study of mountain and Arctic glaciers across Canada, who surveyed GRIC in the early 1990s. Charles Papasodoro acknowledges support from the Fond Québécois de Recherche en Nature et Technologies (FQRNT) fellowship program and the Centre d'Études Nordiques (CEN) for an internship at LEGOS (Toulouse, France). The 2003-04 field surveys on GRIC were conducted with the able assistance of J.C. Lavergne and C. Kinnard, and logistical support from the Geological Survey of Canada, the Polar Continental Shelf Project, and the Nunavut Research Institute. D. Scott, F. Savopol, C. Armenakis and P. Sauvé (Geomatics Canada) assisted with the GPS data reduction back in 2004. This research was supported by the Natural Sciences and Engineering Research Council of Canada, by the French Space Agency (CNES) through the ISIS and TOSCA programs (Pléiades data) and by the Geological Survey of Canada (field campaign). ASTER and Landsat data were obtained free of charge thanks respectively to the GLIMS program (NSIDC) and USGS. Edited by: J. O. Hagen