In this study, we propose a methodology to estimate the spatial
distribution of destabilizing rock glaciers, with a focus on the French Alps.
We mapped geomorphological features that can be typically found in cases of
rock glacier destabilization (e.g. crevasses and scarps) using orthoimages
taken from 2000 to 2013. A destabilization rating was assigned by taking into
account the evolution of these mapped destabilization geomorphological
features and by observing the surface deformation patterns of the rock
glacier, also using the available orthoimages. This destabilization rating
then served as input to model the occurrence of rock glacier destabilization
in relation to terrain attributes and to spatially predict the
susceptibility to destabilization at a regional scale. Significant evidence
of destabilization could be observed in 46 rock glaciers, i.e. 10 % of the
total active rock glaciers in the region. Based on our susceptibility model
of destabilization occurrence, it was found that this phenomenon is more
likely to occur in elevations around the 0
Warmer mean annual air temperatures
In this context, rock glaciers experiencing destabilization have recently
become of interest. While active rock glaciers commonly present moderate
interannual velocity variations that correlate with the ground temperature
The destabilization process can be triggered by either mechanical forces or
changes in climate. An overload on the glacier surface caused by a landslide
or glacio-isostatic uplift can cause a compressive wave that propagates
through the landform, increasing its displacement rates and leading to
destabilization
The purpose of this study was to obtain regional-scale insights into the
issue of destabilizing rock glaciers in the French Alps. Destabilization has
been observed by several studies in the region
The following step, i.e. modelling the destabilization occurrence, was
performed by using a statistical approach that has been used for mapping
landslide susceptibility (
The French Alps cover an area approximately 50–75 km wide and 250 km long,
located between 44 and 46
Identification of the study area in the European Alps and overview
of the periglacial environment. Permafrost distribution is represented by the
PFI map
In this region permafrost was estimated to cover up to 770 km
Description of surface disturbance features that could be observed in the field or from orthoimagery to identify signs of rock glacier destabilization.
According to
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
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
(
Examples of surface disturbances observable in the available
orthoimages of 2013 in comparison to field observations on
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
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
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.
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
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 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
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 predictive power of the model was estimated with spatial cross-validation
(R package sperrorest). The method selected was the
The variable importance was assessed using permutation-based variable
importance embedded in the spatial cross-validation
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
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.
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
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
Terrain attributes were derived from the BD ALTI DEM, 25 m
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.
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
Number of rock glaciers per dominant lithology in relation to destabilization rate.
More than 1300 surface disturbances were digitized, involving 259 active rock glaciers (Fig. 6). Overall, more than the 50 % of the active rock glaciers may be affected by some degree of destabilization as 46 rock glaciers (9.7 %) showed potential destabilization, 86 (17.0 %) were suspected of destabilization and 127 (25.7 %) were unlikely destabilized. Only 13 potentially destabilized rock glaciers presented deep surface disturbances. Location and destabilization rate of each active rock glacier in the region is provided as a shapefile in the Supplement.
Potentially destabilized rock glaciers were mainly located in the Vanoise National Park and in the Queyras and Ubaye mountain ranges. In these areas, densely jointed lithologies (i.e. ophiolites and schists) dominate. Rock glaciers in crystalline lithologies (i.e. gneiss and granite) were found to have low destabilization ratings. That is, only two rock glaciers were rated as possibly destabilized over a population of 55 (Table 3).
The predominant surface disturbance observed was cracks, which were present in 187 of the active rock glaciers (Table 4). Crack clusters also had a high number of observed cases (152), while the deep surface disturbances occurred in about 15 % of all the examined rock glaciers. In general, the occurrences of surface disturbances were dependent on the destabilization rating. Scarps and crevasses were found in about 10 % of unlikely destabilized landforms. The observation of each surface disturbance was highest for potentially destabilized rock glaciers with deep surface disturbances, indicating that in these landforms multiple surface disturbances coexist.
Map of active rock glaciers in France by rock glacier
destabilization rating, with focus on the
Transformation function plots of the GAM model showing the
relationship between each predictor variable and destabilization occurrence.
The data distribution with respect to predictor variables is indicated with
dots on top (destabilization evidence) and on the bottom (stability evidence)
of the plots. The
Following a stepwise backward and forward selection, the chosen model
included PISR, slope angle, elevation and curvature as predictors. The mean
cross-validated AUROC was 0.76 on the test set, indicating a good performance
Number of rock glaciers per destabilization rating showing a specific surface disturbance.
Examples of the susceptibility map in
The model transformation functions revealed the relations between terrain
attributes and rock glacier stability (Fig. 7). Higher predisposition to
destabilization was more likely to occur in an altitudinal range between 2700
and 2900 m a.s.l. and slope angles ranging between 25 and 30
The susceptibility map highlights creeping permafrost areas susceptible to destabilization based on regional-scale model predictions (examples shown in Fig. 8, and the full map is available in the Supplement). The susceptibility map reproduced the previously known cases of destabilization well. The destabilized areas of Iseran, Roc Noir and Pierre Brune were predicted to have a high susceptibility to destabilization, which matches field observations. In some cases, the susceptibility map predicted high destabilization susceptibility in areas belonging to stable rock glaciers.
Rock glacier surfaces were investigated with respect to each susceptibility
class (Table 5). About 75 % of the creeping permafrost was found at low or
very low susceptibility to destabilization. Creeping permafrost at high and
very high susceptibility to destabilization accounted for 10 % of the total
creeping permafrost surface, i.e. 2.9 km
The present study provided the first comprehensive assessment of rock glacier
destabilization for the French Alps and indicates the potentially high
prevalence of this phenomenon. Destabilized rock glaciers were more likely
located in the Vanoise, Queyras and Ubaye ranges. In these areas the densely
jointed lithology was suspected to generate mainly pebbly rock glaciers
The majority of rock glaciers showing potential destabilization were
characterized by shallow cracks (33 cases versus 13). Although this is
suggested to be partially due to the high incidence of rock glaciers located
in densely jointed lithology, there are a number of questions that still need
to be answered in this context. At present, we are unsure about the
significance of these surface disturbances in the context of destabilization.
Cracks may be either “mild” evidence of destabilization as they affect only
the upper layer of the landform, or a typical surface disturbance occurring
on destabilized pebbly rock glaciers. In the first case, using cracks as
destabilization evidence could lead to an over-interpretation of the
destabilization severity of the landform. Conversely, it was observed
that destabilization may occur when only these type of surface
disturbances occurred
Active rock glacier area per class of destabilization susceptibility.
Overall, rock glacier destabilization rating can be a relevant tool for the local authorities to focus monitoring efforts related to periglacial risk assessment, as we identified all rock glaciers presenting signs of destabilization in the region. The destabilization rating, if combined with an assessment of displacement rates and landform connectivity, could indicate the severity of the potential hazard and be used to help identify actions that should be undertaken to deal with the problem. In general rock glaciers with a low destabilization rating are currently evolving slowly or are stable, and consequently monitoring based on remote sensing may be sufficient. Suspected or potentially destabilized rock glaciers require more caution and in situ monitoring is recommended.
A potential source of uncertainty in this study was the subjectivity that can occur while mapping surface disturbances and rating the degree of destabilization. These activities were based on expert knowledge; however, it is possible that mapping and rating results vary depending on the operator. For example, the operators in charge of the digitization process were requested to interpret surface features that in many cases have small dimensions with respect to the resolution of the orthoimages, making the identification challenging. Orthoimages can have varying illumination from one year to another, causing surface disturbances to change their appearance. Orthoimages may also be distorted, creating unrealistic deformation patterns of the rock glacier surface. Also, although surface disturbances were inventoried into the catalogue in an attempt to standardize the classification, destabilized rock glacier morphology is complex, and its identification requires intense training. In many cases the boundaries between the different typologies proposed were not sharp. Personal knowledge of the process evolved through the inventory compilation, requiring various iterations to review the work.
Another issue was that the operator's metrics of judgment were subjected to
the “prevalence-induced concept change”
Although observing aerial orthoimagery or high-resolution DEMs could not
replace the relevance of a proper in situ survey, it provides us with data
and resulting insights that would normally not be possible with in situ
surveys alone, a characteristic that fitted with the aim of the study.
Additionally, the use of orthoimagery has been proven to be a useful approach
for mapping rock glacier surface disturbances by
Despite the various limitations of the data, the results were encouraging.
The spatially cross-validated model had a good performance, suggesting that
the method is valuable in the context of modelling rock glacier stability. The
relationships with predictor variables were found to be consistent with
topographic settings observed in known cases of destabilization. High slope
angles are suggested to increase internal shear, making the landform more
susceptible to destabilization
PISR had the most importance in the model, suggesting that rock glacier
destabilization was primarily more likely to occur on north-facing slopes. We
cannot offer a convincing explanation of this phenomenon since, at the
present state of the art, there is no systematic study comparing rock glacier
characteristics in relation to their solar exposure. Nevertheless, we suggest
that a possible explanation resides in the variability in meltwater input of
the rock glaciers with respect to solar exposure.
Modelling rock glacier destabilization using PTP instead of elevation revealed that an increasing potential in permafrost thaw was linked to an increase in susceptibility to destabilization, indicating that destabilization was more likely to occur where the permafrost zone was expected to be thawing. This seems to be consistent with the relationship between destabilization and elevation, as potentially destabilized rock glaciers are more often located around 2800 m a.s.l., which roughly coincides with the lower margins of the regional permafrost zone.
Overall, permafrost destabilization was adequately described, as indicated by
the cross-validated performance, in most of the observed cases of
destabilization. Although cases of potential destabilization were
inventoried, rock glaciers that have a low rating of destabilization and are
located in areas with high susceptibility should be identified as having a
high potential of future destabilization. Results indicated that these rock
glaciers had a large area of high predisposition to destabilization,
suggesting that there is a high potential for future destabilization in the
region. The map may therefore be used to spot rock glaciers that present a
predisposition to develop destabilization. In particular, the Laurichard rock
glacier is a site currently under monitoring and was found to present a low
to medium susceptibility to destabilization in this study
The present study aimed to give insights into the extent of destabilizing
rock glaciers in the French Alps. Mapping and modelling rock glacier
destabilization in this region was conducted using an orthoimagery
collection, a 25 m
Despite the limitations of this methodology, the study contributes to the knowledge related to permafrost degradation in the French Alps. Rock glacier destabilization potentially involves 46 active landforms, uniquely located in non-crystalline lithologies, which are typically densely jointed as ophiolites and schist. Shallow surface disturbances (i.e. cracks) had the highest incidence in potentially destabilized rock glaciers. At present, there are several questions concerning the destabilization of pebbly rock glaciers presenting these shallow surface disturbances, as only a few studies tackled the subject. Therefore, considering the high incidence of these landforms in the region, it is suggested to dedicate more attention to these issues in the future.
The destabilization of creeping permafrost was found to be a widespread
phenomenon that involves more than 10 % of the total surface of active rock
glaciers, i.e. 3 km
Shape files of destabilization ratings at rock glacier locations and destabilization susceptibility maps are available in the Supplement (.tiff format). Data are referenced in EPSG: 2154.
The supplement related to this article is available online at:
MM, CS, XB and PS conceived the project and collected and interpreted data. MM, AB and JG designed the modelling approach. MM wrote the paper. All authors contributed to discussion and editing. AB and JG provided feedback on the writing.
The authors declare that they have no conflict of interest.
The present study was funded by the region Auvergne-Rhône-Alpes through the ARC-3 grant and by the European Regional Development Fund (POIA PA0004100) grant. The Lanslebourg–Val Cenis municipality also contributed to the present study by funding internships within the PERMARISK project. We would finally like to acknowledge the two anonymous reviewers for their highly constructive feedback provided during the reviewing process. Edited by: Moritz Langer Reviewed by: two anonymous referees