Study area and rock glacier inventory
The French Alps cover an area approximately 50–75 km wide and 250 km long,
located between 44 and 46∘ N and 5.7 and 7.7∘ W (Fig. 1). Apart from the
noticeably high Mont Blanc massif (peaking at 4810 m a.s.l.), mountain ranges
commonly peak between 3000 and 4000 m a.s.l. The lithology is heterogeneous
across the region. The northern French Alps can be roughly divided into the
west side, dominated by granite and gneiss (ranges of Mont Blanc, Belledonne,
Écrins and Grandes Rousses), and east side, where ophiolites and schists are
more common (ranges of Vanoise, Thabor and Mont Cenis). In the southern
French Alps ophiolites, limestone and mica schists are the most common
lithology (ranges of the Ubaye), while the crystalline range of Mercantour
can be found at the southernmost end of the region. Dominant geology is
described in the at 1/1000000 scale, and the vectorial
version of this map is used in this study to observe destabilization
occurrence in relation to lithology.
Identification of the study area in the European Alps and overview
of the periglacial environment. Permafrost distribution is represented by the
PFI map . Black dots identify active rock glacier
locations .
![]()
In this region permafrost was estimated to cover up to 770 km2
. The 0 ∘C annual isotherm at the end of the
20th century ranged from 2500 m a.s.l. in the south to 2300 m a.s.l.
in the north . The periglacial landforms of the region
were inventoried by the “Office national des forêts” (ONF: the National
Forest Office) and revealed the high presence of
active rock glacier in the region (i.e. 493 mapped rock glaciers). This
inventory was compiled between the years 2009 and 2016 by inspecting aerial
imagery and revised by . This inventory was used in the
present study to identify active rock glacier locations and to investigate
the occurrence of destabilization.
Description of surface disturbance features that could be observed
in the field or from orthoimagery to identify signs of rock glacier
destabilization.
Feature
Description
Cracks
These are shallow linear incisions in the surface of an active rock glacier where a strain is applied (called “scars” in ). Cracks can be several tens of metres long and occur either individually or in a great number, being spaced out by only a few metres. In this case we define the feature as a “crack cluster” (translated from ). Their proximity and shallowness led to the assumption that they affect only the active layer of the landform. Nevertheless, this feature was found to be largely predominant on the Lou , Signal de l'Iseran and Tsaté-Moiryl rock glaciers and therefore considered of interest in the context of the study.
Crevasses
These deep transverse incisions on the rock glacier surface can range in length from several metres to the entire landform width . Their depth is substantially larger than the active layer thickness, suggesting the presence of a shear plane sectioning the frozen body. Crevasses may be isolated or grouped. Spectacular crevasses can be found on Pierre Brune rock glacier (Fig. 1), where they are up to 7 m deep and 10 m wide, cutting across the entire landform (about 150 m). Similar dimensions are reported in the Furggwanghorn rock glacier .
Scarps
Scarps are described by and as steep slopes (30 to 40∘) several metres high, transversally cutting the entire rock glacier. Scarps are associated with deep shear planes that disconnect the rock glacier into two bodies that creep at different rates. Their activation is associated with a sudden acceleration of the downstream portion of the landform. One of the biggest scarps observable in the region is the one on Roc Noir rock glacier . This S-shaped scarp, 20–30 m high and 40–45∘ steep, transversally cuts the whole landform (120 m) and the downstream lobe creeps about twice as fast as the upper part.
According to , mean annual air temperature increased by up to
1.4 ∘C in the French Alps during the 20th century, and this rate has been
increasing in recent decades. This climate warming is suspected to have
caused some noticeable effects on the permafrost characteristics in the
region. The only deep permafrost borehole in the region, located in the
Écrins massif in temperate permafrost (-1.3 ∘C) with low ice content, showed a
temperature increase rate of 0.04 ∘C decade-1 between 2010 and 2014
, similar to other sites in Switzerland where data
series are longer . Increasing air temperature was also
addressed to be responsible for the acceleration since the late 1990s of the
active Laurichard rock glacier located in the Combeynot massif of the French
Alps . Several cases of rock glacier destabilization, such as the collapsed Bérard rock glacier
and the Pierre Brune rock glacier , were
also observed in the region.
mapped destabilized rock glaciers in the Maurienne
valley, Vanoise National Park and Ubaye valley, highlighting the high
incidence of destabilized rock glaciers in these areas.
Mapping rock glacier destabilization
The first step to identify destabilized rock glaciers was mapping surface
disturbances on rock glaciers. Previous studies that described destabilized
rock glaciers showed that these landforms present a wide variety of
geomorphological features e.g.. Here, we followed a
methodology similar to that of , which consisted of defining a
catalogue of typical surface disturbances that can be found on destabilized
rock glaciers. Surface disturbances on rock glaciers were classified in three
distinct categories, depending on their morphology: cracks, crevasses and
scarps. Surface disturbances are described in detail in Table 1 and
illustrated in Fig. 2.
In this study, surface disturbances were mapped for the inventoried rock
glaciers based on interpretation of a set of multi-temporal high-resolution
aerial imagery for the French Alps. This orthoimagery collection was obtained
from the Institut géographique national (IGN, National Institute of
Geography), which is freely available from the official website
(https://www.geoportail.gouv.fr/, last access: 10 December 2018) or can be accessed as a Web Map Service
. The IGN orthoimagery collection consists of
orthomosaics covering all of France for three different collection periods.
The first orthomosaic is composed of images taken from 2000 to 2004, the
second from 2008 to 2009 and the third from 2012 to 2013. All images are of
high resolution: 50 cm × 50 cm for the most recent mosaic and slightly lower
values (1 m × 1 m at its lowest) for the older mosaics, depending on the
location. This resolution was sufficient to identify the smallest features to
be mapped, i.e. the surface cracks (Fig. 2a). Nevertheless, several
limitations during the mapping process, such as image distortion
or illumination, were encountered and will be discussed in Sect. 4.4.1.
Examples of surface disturbances observable in the available
orthoimages of 2013 in comparison to field observations on (a) Roc
Noir and (b) Pierre Brune
destabilized rock glaciers. The black arrows indicate the rock glacier
displacement direction. A scarp (1) and cracks (2, 3) have been observed on
the Roc Noir rock glacier. Large crevasses (4) can be seen on the Pierre
Brune rock glacier. The dotted black lines indicate how the surface
disturbances were mapped on these orthoimages.
![]()
Using a single orthoimage to map surface disturbances can lead to
misinterpretations in the case of poor illumination of the terrain and snow
patches covering the ground . Indeed, as the surface
morphology of a rock glacier is naturally shaped according to spatially
varying creep patterns, it is easy to mistake actual surface disturbances
related to compression features, such as furrows, depending on image quality.
Therefore, surface disturbances, i.e. those morphological features not
related to the creeping of the ice-rich permafrost, were mapped using all
three available orthoimages in order to check that actual strain occurred
where surface disturbances were located and to overcome limitations related
to poor quality of an individual image.
Rating the degree of destabilization
After the rock glacier surface disturbances were mapped, a
rating of the degree of destabilization was assigned to each rock glacier.
This rating was given not only to provide some insight into the observed levels
of destabilization in the French Alps, but also to provide a confidence
rating to describe a rock glacier as stable or unstable for the spatial
distribution modelling of rock glacier destabilization.
Assigning a rating to quantify the degree of destabilization of a rock
glacier required the definition of the characteristics of the “typical”
destabilized rock glacier that can be observed in multiple orthoimages. To do
so, we investigated the features of destabilized rock glaciers reported in
the literature that could be observed by orthoimagery interpretation. At
first, it was observed that the presence of surface disturbances was a
necessary but not sufficient condition to the occurrence of destabilization,
as rock glaciers may present surface disturbances but be stable for decades.
For example, in the Pierre Brune, Roc Noir and Hinteres Langtalkar rock
glaciers, although crevasses could be observed in aerial imagery since the
1940s to the 1960s, destabilization occurred only in the late 1990s
. Second, the destabilization
process can be linked to an increase in surface disturbance occurrence (see
Fig. 3). Also, surface disturbances on destabilized landforms were observed
to create a discontinuity in the creep pattern. For example, the Plator,
Grosse Grabe and Gänder rock glaciers have gone through a sharp transition
from displacement speeds on the order of 0.1–0.9 m yr-1 to displacements
speeds of the order of several metres per year
. Finally, a high displacement rate may not be
a necessary feature, as some destabilized rock glaciers, e.g. Lou and
Furggwanghorn, moved at a “normal” rate of around 2 m yr-1
.
These observations suggest that destabilization may be spotted in orthoimages
if the landform has surface disturbances increasing over time time by
frequency and/or magnitude, as well as if disturbances also create a strong
discontinuity in the deformation pattern of the landform. Nevertheless, rock
glaciers were observed to show a wide variety and combination of these
features, making it unrealistic to construct a binary classification of
stable versus destabilized landforms. In order to acknowledge this, we
proposed a rock glacier destabilization rating based on four rates that
varied from 0 (stable rock glaciers) to 3 (rock glaciers potentially
destabilized), which is explained in more detail in Table 2. For each active
rock glacier, a rating of the degree of destabilization was assigned by
observing the combination of surface disturbances and a qualitative
assessment of recent deformation patterns. This rating was applied using a
standardized workflow (Fig. 4). A comparison of the available IGN
multi-year orthoimagery was used to observe the temporal evolution of the
surface disturbances and surface deformation patterns.
Rating classes used to describe rock glacier destabilization.
Rating
Label
Description
3
Potential destabilization, potentially destabilized rock glaciers
Surface disturbances are well recognizable and evolve in time, increasing in number and/or size. The deformation pattern of the rock glacier is discontinuous and some sectors move significantly faster than others. The source of the discontinuity may be located at the rock glacier's root and the whole landform may be affected by destabilization. Deformation pattern discontinuities are sharp and coincide with the presence of surface disturbances. Sectors moving appreciably faster may also present a series of surface disturbances. If the dominant surface disturbances are deep (i.e. crevasses and scarps), then it is attributed the rating 3a. If the dominant surface disturbances are shallow (i.e. crack and crack clusters) then it attributed the rating 3b
2
Suspected destabilization
In these landforms the surface disturbances are well recognizable and evolve in time, by increasing in number and/or size. The velocity field is continuous, i.e. there are no abrupt spatial differences in the velocity field. If there are sectors moving faster than others, their transition is smooth
1
Unlikely destabilization
In these landforms surface disturbances do not appear to evolve in time. The rock glacier presents a continuous deformation pattern, with no sectors moving substantially faster than others.
0
Non-observable destabilization
Active rock glaciers not presenting surface disturbances are considered as stable.
The evolution of the destabilization of the Pierre Brune rock
glacier. The destabilization evidence, in this case a crack observable since
1952, evolved to a crevasse, observable in 1970. Afterwards, the landform was
stable for 20 years as destabilization evidence did not further evolve.
Between 1990 and 2003 the rock glacier experienced severe destabilization
with the formation of new crevasses and a scarp at the location of the 1952
crack.
![]()
Potentially destabilized rock glaciers were then classified into two
different categories according to the type of surface disturbances observed.
Most of the destabilization cases observed by previous studies described rock
glaciers characterized by surface disturbances that may reach several metres
of depth, i.e. crevasses and scarps, and therefore suggested splitting the
permafrost body. These surface disturbances were mostly observed in coarsely
grained (i.e. blocky; sensu ) rock glaciers. Nevertheless, in
the French Alps many active rock glaciers are finely grained, and some
destabilization cases, e.g. the Lou and Iseran
rock glaciers, were observed to be characterized by the
presence of cracks only. These surface disturbances are shallower than
crevasses and scarps and are therefore suggested to affect only the upper
layer of the rock glacier. As these observations were relatively recent, at
present there is still not enough knowledge concerning the significance of
these shallow cracks in the context of rock glacier destabilization. We
therefore decided to separate rock glaciers showing shallow surface
disturbances from rock glaciers showing deep surface disturbances. This
distinction was made to make the reader aware of this gap in knowledge.
General pipeline used to rate rock glacier destabilization by
observing surface disturbances and the qualitative displacement field. Higher
destabilization ratings indicate
potentially unstable rock glaciers, while lower ratings indicate stable rock
glaciers.
![]()
Modelling rock glacier stability
Modelling the rock glacier stability aims to identify the terrain attributes
that may precondition rock glacier destabilization. The modelling followed a
statistical approach similar to previous studies on landslides
and arctic permafrost slope failures that
used the GAM with logistic link function (R
package “mgcv”). The GAM was selected because of its flexibility in modelling
non-linear interactions between dependent and predictor variables. The
logistic link function allows us to model the occurrence of a categorical
response variable as a function of continuous variables (predictor
variables). All numeric predictors were represented using spline-based
smoothing, for which we chose a maximum basis dimension of 4 in order to
limit their flexibility and reduce overfitting. The actual degree of
smoothness of the splines was determined using a generalized cross-validation
procedure .
In this study, rock glacier stability was hypothesized to be preconditioned
by a series of local terrain attributes. In particular, rock glacier
destabilization grouped by either presence or absence was used as the
response variable, while terrain attributes describing local topography and
climate were used as predictor variables. Multiple-variable models were
computed using different combinations of predictor variables. Different
models were compared using the Akaike information criterion (AIC), which is a
measure of goodness of fit that penalizes more complex models. The best
multiple variable model was selected by iterating a backward-and-forward
stepwise variable selection, aimed at identifying which combination of
predictors was better at describing the response variable by means of a
lower AIC. Finally, the best model performance was estimated using the area under the receiver operating characteristic (AUROC) . The
AUROC estimates the ability of the model to discriminate stable and unstable
areas.
The predictive power of the model was estimated with spatial cross-validation
(R package sperrorest). The method selected was the k-means clustering,
which consisted of dividing the mapped data in k spatially contiguous
clusters . All but one cluster were used to train the model,
while the remaining cluster was used to test the predictive power of the
model. This process was repeated until each cluster was used at least once in
both training and test sets. Here, we divided the database into k=5
clusters of equal size per run and used 100 repetitions. Performance
indicators were evaluated for the respective test sets, and the overall model
performance was evaluated using the average and standard deviation over all
partitioning clusters.
The variable importance was assessed using permutation-based variable
importance embedded in the spatial cross-validation . This
method consisted of permutating the values of each predictor variable one at
a time and calculating the reduction in model performance caused by the
permutations. A total of 1000 permutations were performed for each spatial
cross-validation repetition. Predictor variables causing higher deviations
while permutated were considered the most important ones in the model.
Response variable
Surface disturbances of potentially destabilized rock glaciers were used as
evidence of creeping permafrost destabilization. This was performed under the
hypothesis that surface disturbances were the geomorphological expression of
rock glacier destabilization. Although many surface disturbances could be
observed on rock glaciers that were classified as unlikely destabilized or as
suspected of destabilization, potentially destabilized rock glaciers could be
observed to increase surface disturbances over time by number and size,
creating a discontinuity in the deformation pattern, which provided stronger
evidence of destabilization. Therefore, only surface disturbances located in
potentially destabilized rock glaciers were considered to be solid evidence of
rock glacier destabilization.
As surface disturbances were digitized as linear features, they were buffered
and merged into an “unstable areas” polygon database. A buffer distance of 30 m was chosen. The model was found to be insensitive to changes in buffer size
up to 90 m. All remaining areas within the polygons of stable and likely
stable rock glaciers were used as “stable areas”. Polygons of both unstable
and stable areas were sampled using a 25 m × 25 m point grid in order to
assign the response variable to the modelling database. The point values were
then used as binary response variables with values of 0 for stable areas of
(likely) stable rock glaciers, while 1 was assigned for unstable areas of
potentially destabilized rock glaciers in the modelling stage.
Since the rock glacier inventory counted a relatively small number of
potentially destabilized cases (46 individuals), selecting only one point per
rock glacier would have caused large uncertainty in the model outcome.
Therefore, a simple exploratory analysis was performed to identify a suitable
number of points per rock glacier to be used for modelling. Multiple points
from one to 10 were randomly selected within each rock glacier perimeter and
used to compute a model. This was repeated 10 times per point sample
size to measure the variability in the model performance in relation to the
point sample size. Since the model performances were found to stabilize for
more than five points selected per rock glacier, the number of points
randomly extracted per rock glacier used for modelling was five. Overall, the
model was computed using 225 points with evidence of instability and 1785 points with evidence of stability.
Predictor variables
Terrain attributes used in modelling needed to be selected to act as proxies
for processes that precondition destabilization. Although destabilization is
found to occur in different conditions, some topographical features seem to
be recurrent. Destabilization has been observed to occur on steep slopes, as
high slope angles tend to increase the internal shear stress
. Surface disturbances are often located in convex-shaped
bedrock surfaces, which causes an extensive flow pattern and a thinning of
the permafrost body . Solar exposure may also be
significant in the destabilization occurrence since all known cases of
destabilized rock glaciers in the French Alps are north facing. Solar
exposure can also be a proxy of the snow cover duration, as north-facing
slopes are more prone to conserve longer snow patches through the summer,
making meltwater available through the summer. Elevation and mean annual air
temperature can also be proxies of snow cover duration that have the
possibility of affecting permafrost characteristics. Considering this, slope
angle, profile curvature, potential incoming solar radiation (PISR) and
elevation were tested as predictor variables.
Map of potential thawing permafrost (PTP) distribution in the Mont
Cenis range, indicating the extent of the permafrost zone not in equilibrium
with the present climate (red coloured areas). Temperature warming to compute
the map is evaluated using HISTALP data between the end of
the Little Ice Age (light blue shaded period in the temperature anomaly plot)
and the current climate (red shaded period).
![]()
Terrain attributes were derived from the BD ALTI DEM, 25 m × 25 m
spatial resolution . Slope angle and downslope curvature
were evaluated using the Morphometry Toolbox in SAGA GIS
(version 2.2.2). Negative values of curvature indicate concave topography,
while positive values indicate convex topography. Also, PISR was calculated
using the Terrain analysis toolbox in SAGA as the sum of the computed direct
and diffusive components of the radiation . Clear-sky
conditions, a transmittance of 70 % and absence of a snow cover were
assumed in the calculation of the annual total PISR. Finally, it was decided
to evaluate the relation between rock glacier destabilization and the spatial
distribution of degrading permafrost in order to give insight into the
significance of the warming climate with respect to the destabilization
phenomena. The spatial distribution of degrading permafrost was evaluated
following the method already presented by other studies
, which consisted of artificially
shifting a permafrost map proportionally to the estimated climate warming
occurring between the period of validity of the map and the current climate.
Here, we used a permafrost favourability index (PFI) map
to act as a permafrost distribution map for the region. The PFI map was
calibrated using active rock glaciers as evidence of permafrost occurrence,
and it represents the permafrost conditions during the cold episodes of the
Holocene, e.g. Little Ice Age (LIA). The climate warming between the years
1850–1920 and 1995–2005 was determined using the HISTALP database
over the region. A permafrost distribution map was then
recomputed taking into account these temperature variations and represented
the theoretical permafrost distribution in equilibrium with the current
climate. By comparing this theoretical permafrost distribution and the PFI, a
map of the potential thawing permafrost zone (PTP, i.e. the so-called
“melting area” in ) was obtained. In order to use the
PTP as a predictor variable, it was represented by an index ranging between
0, i.e. no thaw expected, and 1, i.e. potential thaw.
It should be emphasized that PTP is only a proxy of permafrost degradation,
which occurs at all the elevations, while the PTP zone consists of a belt of
250 to 300 m in elevation that affects about 50 % of the lower margins
of the permafrost zone (Fig. 5). PTP is used under the hypothesis that
degradation is more intense at the lower margins of the permafrost zone where
permafrost conditions may be more temperate, richer in water and more
sensitive to climate variations.
Susceptibility modelling
The model of rock glacier stability was also used to predict the occurrence
of degrading permafrost over the French Alps by producing a susceptibility
map e.g.. This was carried out using the R package RSAGA and
the raster images of the predictor variable maps, which allowed
extrapolation of the relationships between rock glacier stability and terrain
attributes at the landscape scale. We would like to highlight that since the
model is constructed using data on destabilized rock glaciers, the
susceptibility map applies mainly for processes relative to destabilization
of ice-rich debris slopes. Therefore, in areas where creeping permafrost does
not exist, the extrapolated susceptibility may have high uncertainty. The
model predicted a DEFROST index, which was classified into five susceptibility
zones using the 50, 75, 90 and 95 percentiles .
These zones described very low (<50), low (50–75), medium (75–90), high
(90–95) and very high (>95) susceptibility to permafrost destabilization.
Number of rock glaciers per dominant lithology in relation to
destabilization rate.
Destabilization
Ophiolites
Schist
Sandstone
Mica-schist
Gneiss
Granite
Limestone
Totals
rate
0
47
88
21
11
31
3
32
233
1
39
37
11
3
13
2
22
127
2
33
28
5
0
1
1
18
86
3a
5
2
1
0
0
0
5
13
3b
18
7
1
3
0
0
4
33