I n s t r u m e n t s

a n d

M e t h o d s

AcousticDetectionSystemforOperationalAvalanche
V.Chritin¹,M.Rossi¹andR.Bolognesi²

1 _{EPFL, De-Lema, CH-1015 Lausanne}

2 _{SFISAR, Weissfluhjoch, CH-7260 Davos}

Forecasting

Key Words :Avalanche forecast,Avalanche warning,Sen-

sor, Acoustics

ABSTRACT

Applied acoustics has recentlycome to the fore as a useful
tool for improving avalanche forecasting by supplying the
means of automatically detecting, in continuous realtime,
avalanche activity overa mountain range. Such informa-
tion,whichisimpossibletoprovidefrommanual
observationsalone,duetoweather,lackofvisibilityat
night, masking relief's, etc., is of major importance to fore-
castingsystems - in particular for thoseusinganalogy-based
reasoning models.
Theacoustic systemARFANGconsistsoffourmicro-
phones combined in such a way as to constitute an acous-
ticgoniometer:theincident directionofsoundwaves-
azimuth, elevation - areobtained from the calculatedtime-
delays of sound waves between pairs of microphones.
Special microphonescalled ECHOwere builtandin-
stalled at Anzère ski resort (Switzerland). ECHOs are dedi-
cated toinfrasoundsand aresuitableforhighmountain

wintertopographical and meteorological conditions.The
system including automatic signal recognition procedures
demonstrated the possibility of using sounds to detect and
localizeavalanchesoverareasextendinguptoseveral
square km.
ARFANG is going to be interfaced to the avalanchefore-
casting system NXLOG - which uses avalanches observed
in the past to produce predictions - with the objective of
building an automatic avalanche forecasting tool.

1. OBJECTIVE
   
   Case-based forecasting systems have shown that they can
   perform excellentavalanchepredictions(Bolognesi, 1996).
   The principleofthese systemsis:'samecauses produce
   same effects'.According to this postulate,daily forecasts
   can beinferredfromavalanche activityrecorded during
   the most similardays stored in a data base (fig.1).
   It is easy to understandthat predictionscan only be right
   if the data are reliable. Input data are to models what fuel
   is toengines:acriticalperformancefactor.Thusitis
   meaningful to try to obtain the best input data possible.

Fig.1:Ourelectronic friend must be informed about avalanche activityas precisely as possible!

129