I n s t r u m e n t s

a n d

M e t h o d s

ReliableEstimationofAvalancheActivityUsingSeismicMethods B.Leprettre,J.P.Navarre,A.Taillefer,Y.Danielou,J.M.Panel,F.Touvier Météo-France, Centre d'Etude de la Neige 1441 Rue de la Piscine,F-38406 Saint-Martin-d'Hères Tel : +33 76 63 79 26 Fax : +33 76.51.53.46 e-mail : Benoi t.Leprettre@meteo.fr

Key Words: Avalanche, Seismology, Signal Processing, Data

Fusion

ABSTRACT

Inordertoimprovenatural avalanches forecast,Météo-
Francehasdeveloped anoperational system for automatic,
quasi real-timeestimation ofthe avalanche activity.It is
based on thedetectionof the seismic signal associatedwith
avalanches. Avalanches represent only about 10% of the
recorded seismic signals,all the others being unwelcome
signalsthatmustbeeliminated.Thistaskrequiresa
complete analysis of the signal. We have developed a sys-
tem(namedSARA)whichusestime-frequencyanalysis
and fuzzy logic to recognize avalanche signals.The aver-
age success rate ofoursystem isabout90%.The results
providedbytheSARAsystemarecompared withother
parameters relatedto avalancheactivity. SARAproves able
to follow the evolution of avalanche activity in almost real
time.Concrete applications, such as surveillance of ava-
lanche prone zones, are considered.

INTRODUCTION

The estimationof natural avalancheactivityis mainly based
onthevisualobservationofavalanchedeposits.These
require good visibilityconditions (i.e.daylight and clear
weather) as well as significant observation staff. A time lag
ofseveraldaysislikelybetweentheoccurenceofan
avalanche and itsreport.Furthermore,airbornepowder
avalanchesoften producealmost invisible depositsthat will
notbeobservedwhentheweatherclears.Thus,the
estimation of natural avalanche activity is biased in both
terms of time and intensity. A reliable measurement of the
avalanche activity would nevertheless help nivologiststo
forecast the short-term avalanchehazard. It could also lead
to concrete applications, such assurveillance of avalanche
proneareas, andlegitimatedecisionsconcerningtheclosing
orreopeningofroadsthreatenedbyavalanches.The
presentedsystem uses seismic detection in order to get rid
of the constraints of visual observations.

1. PRINCIPLE AND DIFFICULTIES OF THE SEISMIC
   
   DETECTION OF AVALANCHES (S.D.A.)
   
   When an avalancheoccurs,the descendingsnow packets
   generateseismic waves into theground.These wavestravel
   in the ground and can thus be detected and recorded by a
   seismic station. The effective range of SDA is typically 5-6
   km, although big avalanches have been detected up to 11
   km. The feasibility of SDAhas been proved byprevious
   studies (Saint Lawrence and Williams,1976) (Navarre et
   al., 1991).These studieshave alsohighlighted themain
   difficulty of SDA:in addition to avalanche signals,many
   unwelcomesignalsarerecorded:earthquakes,truckor
   helicoptersounds,thunder, mining blasts,animal orhu-
   man footsteps,... As a matter of fact, avalanchesignals rep-

resentonlyabout10% oftherecorded signals.To geta
reliable estimation of the avalanche activity, it is thus nec-
essary to separate avalanchesignals from the others.

2. TRAINING PHASE
   
   The obtained avalancheseismic signals arehighly depend-
   ent on several factors:quantity and type of moving snow,
   nature and profile of the slip surface, distance and nature
   of the ground between the avalancheand the seismic sen-
   sor. Consequently,avalanche signals are not typical:they
   vary a lot from one event to another. The automatic recog-
   nitionofavalanche signals can thereforeprovedifficult.
   We thus had to learn how to discriminate avalanche sig-
   nals.
   Thefirstphaseofourexperimenthasthereforebeen
   devoted to obtaininga setof unambiguously identifiedseis-
   mic signals.The signals are three-component seismic sig-
   nals recordedin Saint-Christophe-en-Oisans, Isère(French
   Alps,Oisansmassif)inWinters'91,'92and'93.A
   methodology fora posterioriidentification of the recorded
   signals has been developed. For example, earthquakes are
   identified by comparing the date andtime of our recorded
   signals to that of earthquakes listed in a bulletin published
   by the Laboratoire deGeophysique.For helicopterandtruck
   sounds,informationisgainedfromlocalcouncils,
   companiesorlocalresidents.Avalanches areidentified
   accordingtotestimoniesoflocalresidents,skiersor
   climbersaswellasnationalparkwardens.Aftera two-
   yearpractice, we finally got about 200 identified events,
   including about 15 avalanches.
   At thesametime,alltheidentifiedsignalshavebeen

analyzed in orderto find recognition criteria. We rapidly
realized that this problem was rathercomplex and that a
detailed analysis of the signals, involving several descrip-
tion domains, was necessary to get a reliableidentification
of the signals.

129