HR: 09:45h
AN: S41D-08 [Abstracts]
TI: Regional Earthquake Likelihood Models: A realm on shaky grounds?
AU: * Kossobokov, V
EM: volodya@mitp.ru
AF: International Institute of Earthquake Prediction Theory and Mathematical Geophysics, Russian Academy of
Sciences, 79-2 Warshavskoye Shosse, Moscow, 117556
Russian Federation
AU: * Kossobokov, V
EM: volodya@mitp.ru
AF: Institute de Physique du Globe de Paris, 4 Place Jussieu, Paris, 75252
France
AB:
Seismology is juvenile and its appropriate statistical tools to-date may have a "medievil flavor" for those who hurry up to
apply a fuzzy language of a highly developed probability theory. To become "quantitatively probabilistic" earthquake
forecasts/predictions must be defined with a scientific accuracy. Following the most popular objectivists' viewpoint on
probability, we cannot claim "probabilities" adequate without a long series of "yes/no" forecast/prediction outcomes. Without
"antiquated binary language" of "yes/no" certainty we cannot judge an outcome ("success/failure"), and, therefore, quantify
objectively a forecast/prediction method performance.
Likelihood scoring is one of the delicate tools of Statistics, which could be worthless or even misleading when inappropriate
probability models are used. This is a basic loophole for a misuse of likelihood as well as other statistical methods on
practice. The flaw could be avoided by an accurate verification of generic probability models on the empirical data. It is
not an easy task in the frames of the Regional Earthquake Likelihood Models (RELM) methodology, which neither defines the
forecast precision nor allows a means to judge the ultimate success or failure in specific cases. Hopefully, the RELM group
realizes the problem and its members do their best to close the hole with an adequate, data supported choice.
Regretfully, this is not the case with the erroneous choice of Gerstenberger et al., who started the public web site with
forecasts of expected ground shaking for `tomorrow' (Nature 435, 19 May 2005). Gerstenberger et al. have inverted the
critical evidence of their study, i.e., the 15 years of recent seismic record accumulated just in one figure, which suggests
rejecting with confidence above 97% "the generic California clustering model" used in automatic calculations. As a result,
since the date of publication in Nature the United States Geological Survey website delivers to the public, emergency
planners and the media, a forecast product, which is based on wrong assumptions that violate the best-documented earthquake
statistics in California, which accuracy was not investigated, and which forecasts were not tested in a rigorous way.
DE: 3238 Prediction (3245, 4263)
DE: 3245 Probabilistic forecasting (3238)
DE: 4263 Ocean predictability and prediction (3238)
DE: 7223 Earthquake interaction, forecasting, and prediction (1217, 1242)
SC: Seismology [S]
MN: Fall Meeting 2005