Experimental Site

A large, north-south oriented pass (Col du Lac Blanc, 2700 m a.s.1.), where the wind is generally similarly channelled, was chosen for this research program. Close by (at Dôme des Petites Rousses), avalanche activity on two east-facing slopes has been followed and observed: on one slope, avalanches are artificially released according to the snow-cover conditions; on the other slope, there has been completely natural avalanche activity (Reference Guyomarc’h., Castelle. and ArmstrongGuyomarc’h and others, 1992).

Instrumentation

Several parameters (air temperature, wind direction and velocity, snow depth, snow cover, surface temperature and water equivalent of precipitation) are recorded hourly by a sheltered computer and transmitted to Grenoble on request. For 4 years, we have complemented these measurements with data from another automatic meteorological station at the top of the Dôme des Petites Rousses (2800 m).

In addition to these instruments, a new device has been developed and tested. This apparatus is based on a set of snow-depth sensors set up on two cables along a slope; it has been called a “snow profiler” (Reference Guyomarc’h., Mérindol and Sudul.Guyomarc’h and others, 1997). It follows the real-time snow-depth evolution and obtains, every 15 minutes, a snow profile along a slope at a site subject to snowdrifting.

Field Measurements

Our goal is to measure precisely the different wind-velocity thresholds allowing for snow transport and relate these to the morphological and physical features of the snow particles. As often as possible, the automatic measurements have been supplemented with manual observations in order to describe, in maximum detail, each period of snowdrifting. The following observations were carried out:

A snowdrifting-event forecasting model

Wind Speed and Direction Estimation

The first part of the application was developed to estimate wind velocity and direction al Col du Lac Blanc. For this, data from the European Centre for Medium-Range Weather Forecasts were used. The statistical relationships between the model parameters on a 3 × 3 grid (covering the north of the French Alps) and wind measurements ai the site have been searched for. A multi-linear-regression method was used to calculate equations for the daily estimation of wind direction and velocity.

A comparison between measured wind parameters (over the 1993–94 season but not included in the learning file) and the forecast from this method is shown in Figure 1. In most cases, wind velocity is well estimated (correlation coefficient R = 0.77). The highest difference in wind direction appears only with low velocity (less than the threshold of the start of snowdrifting); in the other eases, the forecast direction is quite correct.

Fig. 1.

Comparison between the measured wind velocity (continuous line) and the analysed one (dotted line) at Col du Lac Blanc. The graph shows the absolute value of the difference between forecast and observed wind direction (dotted line represents the mean).

Snow-Particle Mobility

To reach our objective, it was necessary to know precisely the daily state of the snowpack surface. For this. SAFRAN (Reference Durand, Brun., Mérindol, Guyomarc’h, Lesaffre and Martin.Durand and others, 1993) - Crocus (Reference Brun, David, Sudul and Brunot.Brun and others, 1992) was used to simulate the evolution of the snowpack at the experimental site. The first model (SAFRAN) provides the second one (Crocus) with an estimate of relevant meteorological parameters (such as air temperature, wind velocity, humidity, precipitation and its phase, etc.). Then, Crocus calculates a detailed stratigraphy of the snow cover for the site considered. One of the most important problems is having a good description of snow-grain characteristics, so a modification of the formalism used (as described by Reference ColbeckColbeck and others, 1990) has been necessary in order to describe snow evolution as a function of continuous parameters (Fig. 2). These variables are dendricity, sphericity and grain-size. Dendricity varies from 1 to 0, according to the part of the original crystal shape still remaining in a snow layer. Sphericity varies from 0 to 1 and describes the ratio of rounded vs angular shapes.

Fig. 2.

Snow-grain symbolism in Crocus (as a function of continuous Parameters).

Using an empirical approach, we combined grain morphology, and grain-size to define a function describing a “driftability index”. This approach allowed us to take into account indirectly the differences in cohesion between snow particles, notably the sintering process. The influence of temperature on the snow cover is also indirectly taken into account through the metamorphisn laws used by Crocus. For example, it considers that, with the same dendricity, faceted crystals can be eroded more easily than small rounded grains, because of their sphericity difference. This hypothesis was confirmed by our field measurements.

In PROTÉON’s analysis, a Crocus-simulated snowpack was analysed from the top. For each layer, the driftability index (di) was calculated For each type of snow particle, it was necessary to describe the snow using the same parameters as the Crocus model. The analysis was separated into two cases:

We decided that a crust layer > 3 mm or a wet layer prevented snowdrifting and consequently stops the analysis. The driftability index takes a value between 1 for fresh recent snow fallen without wind and -1 for spherical melted snow.

Snow-Driftability Index

The last stage of the study consisted of a combination of previous information (di) and wind velocity) in order to forecast the occurrence of snow 1 day in advance in 6 hourly time-steps. Using field measurements and observations over 4 years, we determined, for each type of snow, the wind-velocity threshold which allowed snowdrifting. We compared at 6 hourly time-steps, the driftability-index value with the forecasted wind velocity. Then, it was possible to define a function using this index and wind velocity which discriminated between no-transport cases and the others. We calculate the driftability index (si) as:

where w is wind velocity and di is the previous snow-drift-ability index.

If the value of this function is positive, it is considered that snowdrifting will occur at the site (Fig. 3) and the thickness of the พ ind-eroded layer can be calculated.

Fig. 3.

The two domains of the snowdrifting index values.

The evolution of the “driftability index” can be displayed on a graph for the previous 7 days. This information is available to an avalanche forecaster on his computer screen (Fig. 4).

Fig. 4.

Evolution of the profile analysed by PROTÉON at the site from 23–30 November 1996. They axis gives the snow depth that could be eroded by wind. The green and blue colours describe a positive snowdrifting index for each snow layer (snowdrifting should occur). Yellow and red colours describe a negative snowdrifting index for snow layers which cannot be eroded by the forecast wind velocity. These layers are drawn below the zero line.

Test on a Whole Season

For validating this forecasting tool, a simulation of the whole 1992–93 winter season was made. The snowdrifting index was calculated at 6 hourly time-steps, using the measured wind velocity. Then, it was compared with snowdrifting periods at Col du Lac Blanc. The lack of a reliable sensor specific to the delection of snowdrifting in real-time led us to use the difference between the hourly maximum and minimum values of the depth sensor. When snowdrifting occurs, the snow particles in motion cause interference, near the snowpack surface, to the ultrasonic waves used by the snow-depth sensor. So, when this difference (averaged over an interval of 6 hours) increases above 2 cm, it is assumed that a snowdrifting event occurred at the site (snowfall is easily distinguished using records from the heating precipitation gauge).

The results from this comparison (made every 6 hours 776 times) are shown in Figure 5.

Fig. 5.

Comparison between observed snow events and simulated ones (PROTÉON) for 776 periods of the 1992–93 winter season

During this period, about 80% of the snowdrifting events were detected. These periods represent only 10.3% of the time during the winter season; 14% of the time false positive predictions were made.

Focus on a Period of the 1996–97 Season

The beginning of the 1996–97 season was rich in snowdrifting events. During this period, a new snow accumulation of about 2 m was observed at Col du Lac Blanc. In order to achieve a qualitative validation, a graphical comparison was made as a first approach. For this, it was necessary to display two indices: the first one synthesized the information of the analysed profile calculated by PROTÉON and the second one summarised the temporal behaviour of the snowpack thickness between two time-steps (6 hours). By using the SAFRAN-Crocus simulation (four times a day) and the PROTÉON analysis, a compound index (indc) was calculated for each simulated snow profile (Fig. 6). The principle is the following:

where n is the number of driftable layers, ind_i is the snow-drifting index of the layer i, ep_i is thickness of the layer i.

Fig. 6.

Comparison between the compound index and the EC_h index.

To define the snowdrifting period, hourly data from Col du Lac Blanc were used. In order to optimise the available data, the following function (Ec_h), named the normalized difference, has been defined:

where diffmx is the maximum difference for the period, hfnmx is the hourly maximum of snow depth, htnmn is the hourly minimum of snow depth. By definition, the Ec_h index is in the range (0,1).

A first validation could be made by comparing the values of the two previous indices on a graph. Figure 6 shows that, taken as a whole, the compound index and the normalized difference are in phase.

If we focus our attention on period 1, it appears that, when the driftable snow is thin, these layers disappear quickly due to the efficiency of wind erosion. However, the Crocus model is not yet able to remove snow layers, so snowdrifting conditions are still forecast by PROTÉON.

In period 2, another kind of disagreement is shown between EC_h and the results of the PROTÉON analysis. This is explained by underestimation of the forecast wind velocity (the hourly mean of the measured wind velocity is 5 m s⁻¹ stronger than the forecast one).

In period 3, a better agreement is observed. During these 20 days, snow depth increased by more than 150 cm and the wind blew with an average velocity greater than 10 ms⁻¹. For these reasons, there was always an erodable snow layer.

The following contingency tables (Figs 7 and 8) show the comparison between observed snowdrifting events and the simulated ones for the same period. In calculating these tables, the information delivered by each index has been transformed to a simulated or an observed snowdrifting event when the index is positive. In the whole period, about 62% of the snowdrifting events have been well forecast. These events represent 28.3% of the time during this period. But, 15.6% of the time (false alarm), snowdrifting was forecast when none occurred. In period 3, about 87% of the snowdrifting events were detected. These events represent 39% of the time. In this period, only 6.1% of the time was the forecast false positive.

Fig. 7.

Test for the whole period (1 November - 31 December

Fig. 8.

Test for the reduced period (23 November - 12 December 1996).