Snow slope stability evaluation requires considering weak
layer as well as slab properties – and in particular their interaction. We
developed a stability index from snow micro-penetrometer (SMP) measurements and
compared it to 129 concurrent point observations with the compression test
(CT). The index considers the SMP-derived micro-structural strength and the
additional load, which depends on the hardness of the surface layers. The new
quantitative measure of stability discriminated well between point
observations rated as either “poor” or “fair” (CT

Snow stability data are among the key ingredients when establishing avalanche forecasts. Snow stability can either be assessed from observations of instability such as recent avalanching (Jamieson et al., 2009), by stability tests performed in the field (e.g., Schweizer and Jamieson, 2010) or from stability indices derived from modeled snow stratigraphy (e.g., Durand et al., 1999; Schweizer et al., 2006). Whereas numerical modeling allows obtaining data with a high temporal and spatial resolution – though often of unknown accuracy – field tests are laborious and reveal partly subjective point information of low temporal and spatial resolution. Nevertheless, stability tests are presently the method of choice for estimating snow instability – despite the fact that the inherently variable nature of the mountain snowpack hinders extrapolation of point observations (Schweizer et al., 2008). One way to overcome the limitation of point observations is to perform many measurements in a given area within a couple of hours. This approach is only possible with a quick probing method, for example, with the snow micro-penetrometer (Schneebeli and Johnson, 1998) or potentially with remote sensing techniques.

The snow micro-penetrometer (SMP) is a probe with a high-resolution force
sensor at its tip, driven into the snowpack at constant speed. It provides a
penetration resistance (force)-depth signal that includes micro-structural
information (Johnson and Schneebeli, 1999; Marshall and Johnson, 2009).
Mechanical properties can be derived from the three basic micro-structural
parameters: element length (

In one of the first attempts to relate the SMP resistance to stability,
Kronholm (2004) found increasing stability scores with increasing weak layer
penetration resistance for four out of five investigated weak layers. Based
on the characteristics found in manually observed snow profiles (Schweizer
and Jamieson, 2007), Pielmeier and Schweizer (2007) tried to discriminate
between unstable and stable observations based on SMP-derived
characteristics of the weak layer and the adjacent layers. Pielmeier and
Marshall (2009) refined this approach and showed that the micro-structural
strength of the weak layer (manually identified in the SMP profile) was the
single best classifier to discriminate between unstable and stable
Rutschblock test results. Classification accuracy improved to about 85 %
when SMP-derived mean slab density (Pielmeier, 2003) was included in a
2-node classification tree. They pointed out the importance of signal
quality control and showed the improvement in classification accuracy that
can be obtained when several SMP measurements within an area of a few
m

Direct correlations of point measurements such as SMP penetration hardness to stability have proven challenging (Lutz et al., 2009). Previous studies of many point measurements at the slope scale using the SMP revealed, among other things, that typical weak layers are often continuously present but have clearly varying properties (e.g., Kronholm et al., 2004). However, relating spatial variations as derived from point measurements to slope stability has so far not been successful. For example, Bellaire and Schweizer (2011) stated that firm conclusions on the dependence of slope stability on spatial variations were not possible due to the limited range of snow conditions in the data set and the fact that the definition of slope stability is partly intangible. From a theoretical point of view, as supported by numerical modeling (e.g., Fyffe and Zaiser, 2004; Gaume et al., 2014; Kronholm and Birkeland, 2005), it seems clear that stability variations at the slope scale can either promote or hinder slope failure. Slope instability should increase with increasing coefficient of variation and increasing correlation length.

Whereas the above-mentioned studies indicate that considerable progress has been made towards objectively deriving snow stability information from the SMP resistance profile, a single measure of stability, combining slab and weak layer properties, is so far lacking. Also, relating point measurements to slope stability has not been successful. We will present a first attempt to directly derive an index of snow stability from the SMP signal and compare it to results of numerous small-scale stability tests (CT). The index will then be applied to the gridded SMP measurements collected on 11 slopes by Bellaire and Schweizer (2011) to explore the spatial structure of stability on these slopes and possibly relate it to slope stability. The index will at best be an estimate of the probability of initiating a failure in a weak layer, but it will not provide any information on the propensity of crack propagation.

We primarily used the data collected and described by Bellaire and Schweizer (2011).
They concurrently observed snow stability using the CT (Jamieson, 1999) and measured penetration resistance using the SMP on
15 slopes above Davos, Switzerland during the winters 2006–2007 to
2007–2008. During the winter 2008–2009 Bellaire (2010) sampled another eight
slopes. On each of the slopes, one manually observed snow profile, nine pairs
of CT and 45 SMP measurements were performed. In addition, other
observations relevant for assessing slope stability such as signs of
instability, snow surface conditions and ski penetration were recorded. For
the analysis described below, we used three different data sets (a, b, c) out
of the data from 23 slopes:

From the 23 concurrent observations of snow stratigraphy (snow profile)
and penetration resistance (SMP) after quality control, 19 cases were retained
for calculating the SMP-derived penetration depth (

From the concurrent observations of point stability (CT) and penetration
resistance (SMP), we analyzed, after quality control, in total 129 SMP
profiles with corresponding CT score (

From the gridded SMP data on 23 slopes, we analyzed 11 slopes in regard to their stability distribution. Most of the remaining data could not be used due to quality issues.

Our assumptions are tied to the CT as we aim at a stability
criterion for failure initiation, which can be validated with previously
collected field data. The stability index follows a simple strength to
additional stress criterion in the weak layer, still accounting for slab
layering. In the CT experiment, a snow column is loaded by
dropping the hand, the forearm or the arm (Jamieson, 1999). For simplicity
we consider a fixed weight corresponding to the weight of a forearm. Due to
the impact, the surface layers are compacted, or rather crushed (van Herwijnen
and Birkeland, 2014). The stress in the column is related to the braking (or
decelerating) distance. The softer the surface layers, the larger the
compaction (and also the braking distance) and hence the smaller the
stress – and vice versa. We assumed idealized elastic-plastic behavior of
the snow so that the initial potential energy (

Schematic of elastic-plastic deformation describing the inelastic
collision while loading the snow column in a compression test.

We assumed the braking distance

Cumulated penetration resistance (

For calculating the stability index, we assumed that the additional stress (derived from Eq. 2) would not decrease strongly with depth as the snow column is uniformly loaded at the top. Furthermore, we neglected the weight of the overlying slab (which is e.g., considered in the skier stability index introduced by Föhn, 1987) because we suppose that the dynamic load (rather than the static load) is essential for initiating a failure due to the well-known deformation rate dependence of snow strength (e.g., Reiweger and Schweizer, 2010). Finally, we did not consider the effect of slope angle on either stress or strength as its effect is largely unknown in the case of a CT.

The simple stability index was defined as

To relate continuous ratio data such as the stability index and
micro-structural strength, we used the Pearson correlation coefficient

Point stability classification based on CT score and CT fracture type.

As suggested by Bellaire and Schweizer (2011), CT scores were classified into
three point stability classes: “poor”, “fair” and “good” (Table 1). “Poor”
stability refers to CT scores

Similarly, all slopes were classified into one of three classes of slope stability “POOR”, “FAIR” and “GOOD”. The classification considered the presence or absence of signs of instability and the slope median CT score; in contrast to Bellaire and Schweizer (2011), we did not consider the profile classification because it is essentially based on a point observation (Table 2).

Slope stability classification based on the slope median point stability and signs of instability (recent avalanching, whumpfs of shooting cracks).

The stability distributions were characterized by the median, the
interquartile range (IQR) and the quartile coefficient of variation (QCV).
When comparing the distributions of the stability index from the three point
stability classes, the non-parametric Kruskall–Wallis

To explore the spatial structure, the experimental semivariogram for a linear trend model of the Cartesian coordinates was calculated. By fitting a spherical model to the experimental semivariogram we determined the range, which is a measure of the correlation length. Details are given in Bellaire and Schweizer (2011). For contour plots, data were interpolated by ordinary kriging.

Stability index vs. CT score (eight cases with CT score 35 (no
fracture) not shown,

The newly developed stability index was calculated for data set (b) of the
129 cases with SMP profile and CT score (Fig. 3). The Spearman rank
correlation coefficient between the CT score and stability index

Stability index grouped into the three classes of point stability
(based on CT score and fracture type) (

Grouping the stability index values according to the three classes of point
stability indicated that particularly the tests rated as “poor” or
“fair” can be discriminated well from those rated as “good” (

For data set (c), including the measurements on the 11 slopes, we will first present the result for those locations where concurrent SMP measurements and CTs were performed so that the performance of the new stability index can be directly compared. We then proceed to analyze the slope-scale stability index distributions based on all SMP measurements performed on the 11 slopes.

Distributions of stability index for the 11 slopes sampled during
winters 2006–2007 and 2007–2008. Only stability derived from those SMP
measurements with concurrent CT test are shown. Median CT score is indicated
in the middle of each histogram. Slopes are ordered with regard to slope
stability (“POOR”, “FAIR”, “GOOD”) and within rows based on median CT score.

Figure 5 shows the stability index values for the locations of SMP
measurements where a CT was performed concurrently. The
stability index distributions found on the 11 slopes were fairly different.
The four slopes that were rated “POOR” had low stability indices and a low
mean CT score. The two slopes rated as “FAIR” had similarly low stability
indices and mean CT scores but were not rated as “POOR” since no signs of
instability were observed. Three of the slopes rated as “GOOD” had
relatively low stability index values and intermediate CT scores, whereas
the other two slopes had high stability index values as well as high CT
scores. Overall, per slope, the median stability index was still positively,
but not significantly, correlated with the median CT score
(

Exemplary distributions of strength, stress and stability index for
four slopes (grids). Grids 0708_3 and 0607_6
are rather unstable whereas grids 0708_9 and
0708_6 are rather stable. Numbers indicate the median value.

The different stability index distributions were the result of various different stress-strength (slab-weak layer) configurations (Fig. 6). For example, grid 0708_9 had rather low strength (53 kPa), but the stability index was relatively high (355) due to the low additional stress (151 Pa). On the other hand, grid 0607_6 had a rather low median value of stability (167) though the weak layer strength was intermediate (95 kPa), but the additional stress was high (561 Pa).

Summary statistics for the 11 grids. For signs of instability,
“1” indicates the presence and “0” the absence of whumpfs, shooting cracks or
recent avalanches. For the stability index, the slope median, the
interquartile range (IQR) and the quartile coefficient of variation (QCV)
are given per grid. The range is a measure of autocorrelation and was
determined by fitting a spherical model to the experimental semivariogram.
Proportion weak describes the portion of point stability measurements with

Considering all SMP measurements in the 11 grids (data set c) (Table 3), the
grid median stability index tended to increase with increasing median CT
score, but the correlation was not significant (

In all grids, the distribution of stability indices showed clear tendencies
towards either primarily stable or primarily unstable values. The two grids
0708_6 and 0708_9 had 0 and 4 %
“unstable” stability values, while in the other cases more than 75 % of the
stability index values were below the stable-unstable threshold (

The variation within a grid, expressed as the quartile coefficient of
variation, was typically largest for the stability index (mean
QCV

The slope median stability index was positively related to the slope
median strength of the weak layer (

In most grids, the variogram indicated that the range was less than 5 m
(Table 3). The values of the range for stress (load), strength and stability
index varied on a given day. They were not significantly correlated,
though the range of the stability index tended to increase with
the range of strength (

Contour plots of strength, stress (load) and stability index for two exemplary grids in winter 2007–2008. Grid 0708_3 (left) represents a rather unstable slope, whereas grid 0708_6 (right) represents a rather stable one.

Figure 7 illustrates for two grids the variable spatial structure of strength, stress and the resulting stability index. For grid 0708_3, when stability index values were low, the stress values did not show any particular trend or clustering; only the values towards the lower left and right corners tended to be slightly higher. For stress, a slope scale trend with some higher values towards the left can be observed. The stability index was somewhat higher in the lower left corner than in the higher left one. This observation can be explained by the trend for higher stress in the upper left corner and higher strength in the lower left corner. For the second grid (0708_6) in Fig. 7, some slope scale trends were observed for strength (higher values towards the right), stress (higher values towards the left) and accordingly for stability index (higher values towards the right).

Based on concurrent SMP measurements and CTs we have derived a new stability index that may be used as a criterion to assess the propensity of failure initiation. The new index has then been applied to a previously published data set of spatially distributed SMP measurements on 11 slopes. The proposed stability index will at best be an estimate of the probability of initiating a failure in a weak layer, but it will not provide any information on the propensity of crack propagation.

As many SMP measurements with concurrent CT results were available, we used the CT as stability reference. Obviously this test is far from perfect (e.g., Winkler and Schweizer, 2009), but at least it is known that the CT score increases with decreasing probability of skier triggering (Jamieson, 1999). Some of the problems include the geometry (scale of length to width), unknown boundary effects and the stepwise loading with only three loading steps. Many factors that probably play a role (how important is mostly unknown) were not considered in our simple model for determining the additional stress acting at the depth of the weak layer. We considered the loading at the top of the column, the size of load according to the second loading step (tapping from the elbow) and the considerable compression of the surface layers. On the other hand, we did not consider the stratigraphy (apart from the surface layers), boundary effects, possible stress waves in the column or load dissipation with depth. As the column is loaded over the whole area we assumed no load dissipation, which is obviously important in the case of a point or a line load. Some recent measurements by Thumlert and Jamieson (2014) may question this assumption.

Nevertheless, the derived stability index was clearly related to the
stability reference we had at hand. The micro-structural compressive
strength has been shown to be related to stability (Pielmeier and Marshall,
2009), and our stability index is slightly better related to the CT score,
further suggesting that our index is indicative of point snow stability.
However, as has been previously shown by Jamieson (1999), the CT does not
differentiate well within the intermediate range (CT scores 11 to 20) and hence
our stability index is afflicted with the same problem. Accordingly,
the correlation between the slope median CT score and the slope median
stability index was rather poor, mainly since the sample size was small
(

Certainly, a better stability reference should allow developing a more sophisticated index along the lines of the skier stability index (SK38) (Jamieson and Johnston, 1998). The comparison to the various slope classifications has clearly shown that the index lacks any information about crack propagation propensity. The slopes rated as “FAIR” had mostly low values of the stability index but no signs of instability were observed. In the future, the initiation index should be combined with a measure of crack propagation propensity, such as the critical crack length that can be derived from the SMP signal (Reuter et al., 2013).

Non-spatial variations of strength, stress and stability index expressed as the QCV were similar to those found in previous slope-scale studies (Schweizer et al., 2008). Whereas Bellaire and Schweizer (2011) separately related weak and slab layer properties to slope stability, we jointly considered both properties by introducing a simple measure of stability. However, we were still not able to resolve the influence of spatial patterns on slope stability. One of the reasons we did not find any relation between spatial characteristics of point stability measurements and the slope stability estimate may be the lack of slopes exhibiting strong variations in stability index, i.e., about equal shares of high and low stability index values. In such situations one would expect that the spatial patterns of point stability characterized by its correlation length would control slope stability. A further reason may be that the proposed stability index only considers failure initiation and does not include crack propagation.

We developed a new measurement-based snow instability index combining weak layer and slab properties to predict snow instability in view of dry-snow slab avalanche forecasting. We analyzed a large data set of gridded field measurements acquired on snow slopes by Bellaire and Schweizer (2011) to first derive the index and then relate point stability to slope stability.

The new index combines weak layer strength (SMP-derived micro-structural strength) with a rough measure of the additional stress at the depth of the weak layer, depending on the properties of the surface layers (i.e., slab layers) (SMP-derived penetration depth). The index was positively correlated with the results of CTs performed concurrently with the SMP measurements. It discriminated well between point stabilities rated as either “poor” or “fair” and those rated as “good” with a 10-fold cross-validated classification accuracy of about 80 %. A rich variety of stress, strength and stability scenarios was found, indicating that the index, despite its simplicity, seems to be able to mimic at least some of the complex interactions between slab and weak layer properties. The well-known challenging problem of correlating variations in point stability to slope stability could not be solved, despite the fact that now at least a measure of stability exists. However, the target variable – slope stability – is not even well defined either.

In a next step we will seek a data set with reference stability better suited than the CT, possibly the Rutschblock, and combine a more sophisticated stability index, rather an initiation index, with a propagation propensity index, possibly the critical length from the propagation saw test (Reuter et al., 2014).

We thank Sascha Bellaire for providing the data and Christoph Mitterer for help with some of the analysis. We are grateful for the comments by two anonymous reviewers. Edited by: P. Tarolli Reviewed by: two anonymous referees