Snow avalanche, climate, statistic modelling, occurrence probability, French Alps

French Alps

Maurienne Massif

Valloire valley

East-west orientation (for the valley)

2400-3200 m asl (triggering zones)

1978-2003

Daily basis results:
For high-frequency avalanche (384 avalanches from 5 tracks), the independent variables yielding the best fit were the precipitation on the day before an event and daily mean air temperature during the event. When the mean air temperature is close to 1°C during the day of the event, the probability of avalanche occurrence is above 0.3 when precipitation exceeded 25 mm the day before of the event, and over 0.5 when precipitation reached 35 mm. When precipitation exceeded 70 mm, the role of temperature was negligible.

For low-frequency avalanche (196 avalanches from 7 avalanche tracks), the independent variable that yielded the best fit was the precipitation the day before the event. The probability of avalanche occurrence seemed not to be influenced by the meteorological conditions during the day of the event. The probability of avalanche occurrence was above 0.4 when precipitation exceeded 40 mm the day before the event. Moreover, the precipitation effect was not constant. The triggering probability increased strongly when precipitation ranged between 30 and 60 mm the day before the event.

Finally, among high-frequency avalanche tracks, the role of meteorological parameters according to the type of avalanches surveyed was tested. 378 avalanches from 2 types (melt and dry) were selected from 5 avalanche tracks. Specific models were obtained for each type of avalanche which were both highly significant statistically. The probability of occurrence of snowmelt avalanche increased strongly when precipitation exceeded 40 mm. The effect of temperature became clear when precipitation was above 50 mm, although for dry snow avalanches, this parameter was negligible. Finally, total precipitation on the day of a given event and the day before were most significant, especially when precipitation totalled 60 mm.

Annual basis results:
For both low- and high-frequency avalanche tracks, frequencies correlated closely with the number of times in the winter that precipitation was recorded during at least 3 successive days. The relationship was even stronger when calculated separately for each of the two groups of avalanche tracks.

For high-frequency avalanche tracks, the model fitted best with the number of times there were more than 3 successive days with rain in a year and the number of times that annual air temperature was above mean ± 1 standard deviation. The annual probability of high avalanche activity increased strongly with the occurrence of 3 successive days of high precipitation per winter, especially when 6 such triplets were counted per winter.

For low-frequency avalanche tracks, the independent variable that yielded the best fit was a maximum occurrence of 3 successive days with high precipitation per winter. The annual probability of high avalanche activity was around 0.5 when the occurrence of 3 successive days with high precipitations exceeded 8 per winter. Moreover, the probability of avalanche activity associated with this variable increased linearly.

Link with the NAO:
No correlation with any specific circulation patterns or with the NAO Index was found and no cyclical trend in the temporal distribution of avalanche occurrences was evidenced in this part of the Alps.

Data used in this study are from a survey of avalanches longer than 200m called EPA (Enquête Permanente des Avalanches).

The statistical analysis was conducted over the 1978-2003 period (high-quality records and homogeneity of survey methods) from December of year n-1 to April of year n. From 1236 observations in the Valloire valley, 12 avalanche tracks (576 events) were selected and classified into 2 groups: high-frequency (more than 40) and low-frequency (between 20 and 40). Because of a lack of information on physical or ecological characteristics, site parameters were not considered.

Daily precipitation and temperature data from the Valloire meteorological station (1460 m asl) were used. 32 meteorological parameters were tested on a daily basis: precipitation; mean, minimum and maximum temperatures; and thermal amplitude on the day of a given avalanche event, as well as 1, 2, and 3 days before a given event cumulated or not. 12 parameters were tested on an annual basis: winter mean precipitation, intense precipitations, number of times it rained during 2 or 3 successive days in a winter, freeze-thaw alternation, temperature anomalies… The weather classification at 500, 700, and 850 hp was also used to evaluate the relationship between avalanche occurrence and any special synoptic situation. The NAO Hurell Index (pressure ratio between Reykjavik and Lisbon) was used to test the teleconnection hypothesis.

The probability model used has evaluated uncertainties associated with discrepancies between the climate at meteorological stations and at avalanche sites. The effects of climatic variables on avalanche initiation were simulated using a logistic regression model. All variables considered on a daily and annual basis were systematically tested. All statistically significant variables were retained for further analysis. Finally, bootstrap analyses were performed to test the global sensitivity of models. This method employs sampling with replacement from unique original data set. The values of the different variables of the new samples are identical to the original but the frequencies may be different. If the mean of parameters in the simulated samples was close to the parameters in the original data set, the statistical stability of the model was deduced.

On a daily basis, the purpose was to identify meteorological factors responsible for avalanche initiation. On an annual basis, correlation matrices were calculated in order to detect any possible relationships between climatic parameters and avalanche frequency.

(4) - Remarques générales