|
identified signals. Wethus obtain,as an output,the truth
value associated to every possible signal origin. The truth
valueofthe"avalanche" classindicateshowmuchthe
signalisrelated toan avalanche signalaccording toour
knowledge.Ifthisvalueisgreaterthan thoseassociated
with the otherclasses, we can conclude that the signal is
an avalanche.
An example of signal identification is proposed in fig-
ure2. Theinputsignalis amid-rangeearthquake.The signal
featuresare listed together with their truth value. The right
signal origin is found by oursystem.However,due to the
complexityofthesignal,the"avalanche" conclusionis
proposed with a lower (but non-zero) truth value, indicat-
ingthat thesignal partiallyressemblesanavalanche ac-
cording to the recognition criteria we have set up.
3.2.PERFORMANCE OF THE SARA SYSTEM
The described signalrecognition system, named SARA for
"Systemed'AnalysepourlaReconnaissancedesAva-
lanches" (analysis system foravalanche recognition),has
been tested on a population of 280 unambiguously identi-
fied signals of various origin,including 13 avalanche sig-
nals.This population is of course different from the set of
signals used during the training phase of the experiment.
Theresultsofautomaticavalanche discrimination,con-
sidering only two classes (avalanche and non-avalanche),
are presentedin Table 1. Only one avalanche signal out of
13 has not been recognized, andonly 9 non-avalanchesig-
nals out of 267 have been erroneously recognized as ava-
lanches.Therefore,theglobal successrateoftheSARA
systemisbetween90%and95%.Moreprecisely,
teleseismicearthquakes,miningblastsandhelicopter
sounds are almostperfectly rejected by SARA.However,
about 10% of short- and mid-range earthquake signals as
well as thunder sounds are not rejected.
-
DEVELOPMENT OF A PRE-OPERATIONAL SDA
PROTOTYPE
Given the satisfactory performance of the SARAsystem,
we have developed a pre-operational version in orderto
evaluatetheinterestofSDAforconcreteapplications.
DuringWinters1995and1996,theSARAsystemwas
implementedonaPC computerinthe Saint-Christophe
recordingsite. The recorded signals (the number of which
gives the daily global seismic activity ) were analyzedevery
24 hours on the PC and the decision concerning each sig-
nal was transmitted to the laboratory.The numberof re-
corded signals which have been recognizedas avalanches
by
theSARAsystemgivestheso-calledSARAseismic
avalancheactivity.Usingthis pre-operationalconfigura-
tion,we have been able to follow up the evolution versus
time of the seismic avalanche activity.
4.1.COMPARISON BETWEEN THE SARA SEISMIC
AVALANCHE ACTIVITY AND OTHER DATA
We have compared, day by day, the information provided
by the SARA s ystem with other data related to avalanche
activity in order to check its consistency. The comparison
parameters are:
Numberofavalanches observedeach dayinthe Les
Deux-Alpes ski resort (close to the Saint-Christophe site).
|
Degree of natural avalanche risk forthe Oisans massif on
theEuropeanAvalanche HazardScale(EAHS),ranging
from 1 (very low) to 5 (very high). It is estimated daily by
the weather station at Saint Martin d'Hères,Isère.When
the risk is larger than 2, significant naturalavalanches are
likely.
Degreeofnaturalavalancheriskestimatedbythe
SAFRAN-CROCUS-MEPRAexpertsystem.Thisisbased
on a deterministic simulationof the snow mantle evolu-
tion (Giraud et al., 1994).
Figure 3showstheevolutionoftheseparametersfor
March 1996.The comparison between graphs (a) and (b)
highlights the needto sort the recordedsignals, in order to
retain only these associated with avalanches.The global
seismic activity can behighalthoughthe avalancheactivity
islow(seeMarch,12).Thecorrelationbetweenthe
estimated SARA seismic avalancheactivity andthe degrees
ofriskisfair.Thatis,theperiodswithhighavalanche
seismicactivitygenerallycorrespondtoperiodswhen
either the forecast (c) or the simulated (d) hazarddegreeis
3orhigher.However,thedayswithnon-zeroseismic
|
|
1
2
3
4
5
