I n s t r u m e n t s

a n d

M e t h o d s

Fig. 6: Statistical distributions of avalanche activity according to windspeed and snowdrift.

W0: probability that windspeed < 10 knots (snowdrift threshold), W1: probability that windspeed >

_[4]

Byapplying[4],wefindl AS[!]0.44sothatknowing
whether or not snowdrift occurred reducesthe probability
of errorin predicting avalanche activity by 44 percent on
average.
What about wind?
Using the same sample,we findlAW = 0.If we calcu-
latelAW(j-1),using the higherwind speed values meas-
ured the day before avalanchesoccur,we findlAW(j-1) =
0.04.This means that we expect an improvement of ava-
lanche activity prediction by using snowdrift data instead

of wind data. This also means that wind speed andsnow-
drift intensity are not strongly correlated. Field measure-
ments confirm this deduction (fig.8).

2.3.DISCUSSION

Itmustbeemphasizedthat theseconclusions onlycon-
cernsituations without snowfall, and that this experiment
showsthatsnowdriftmeasurementsaremore predictive
than non-continuous wind measurements when no other
information is available. Windmeasurements areprobably
more predictive when they are used in combination with
otherdata.Ofcourse,comparingtheperformancesof

132