HR: 09:00h
AN: S41D-05 [Abstracts]
TI: Practical problems in testing earthquake forecasts
AU: Kagan, Y Y
EM: ykagan@ucla.edu
AF: Department of Earth & Space Sciences, UCLA, 595 Young Drive E., Los Angeles, CA 90095-1567
United States
AU: * Jackson, D D
EM: djackson@ucla.edu
AF: Department of Earth & Space Sciences, UCLA, 595 Young Drive E., Los Angeles, CA 90095-1567
United States
AB:
In principle earthquake forecasts can be stated as probabilities, probability densities, or rate densities, and their
predictive value can be assessed using likelihood scores, receiver operating characteristics, or other statistical methods.
However, earthquake behavior and the limitations of earthquake recording pose some practical problems.
1. Some users need long-term forecasts, decades or more, for planning construction and for making long-term decisions.
But earthquakes are strongly clustered, and one big earthquake can change the prospects of other ones in minutes. In other
words, the probabilities are highly conditional and the conditions change rapidly. Some possible approaches include using
simulation of earthquake sequences to make more realistic forecasts, and testing only the largest earthquake in pre-assigned
zones.
2. Prospective testing of fully specified forecasts is the gold standard, but many forecasts involve large earthquakes,
and who wants to wait? Thus, retrospective testing will be inevitable. That implies posterior data selection, and complicates
or even prevents valid statistical testing.
3. Earthquake data vary in quality, and even now the uncertainties in earthquake location, magnitude, and completeness
are highly time dependent. Following a large earthquake, many moderate ones may still be missed because their waves overlap.
Statistical tests must allow for data errors, so rather sophisticated error models are needed. So far, we
don't have them.
4. Hypotheses that predict only a few events are unlikely to lead to definitive tests. To reject either of two typical,
distinct hypotheses (e.g., fault-based vs. seismicity-based) may require that the expected number of events by one or both
be about 10.
DE: 7200 SEISMOLOGY
DE: 7215 Earthquake source observations (1240)
DE: 7223 Earthquake interaction, forecasting, and prediction (1217, 1242)
DE: 7230 Seismicity and tectonics (1207, 1217, 1240, 1242)
SC: Seismology [S]
MN: Fall Meeting 2005