Application of the RI model to forecasting future large earthquakes in Japan
Kazu Z. Nanjo (ERI, Univ. of Tokyo)
International symposium “Toward constructing earthquake forecast systems for Japan”
27 May 2009 at ERI, Univ. Tokyo
RI & PI RI (Relative Intensity of Seismicity)
- Future large earthquakes regions with high seismic intensity
- More specifically, count past earthquakes for each node
PI (Pattern Informatics)- Future large earthquakes regions with high rate change
(activation and quiescence) of seism city
- More specifically, the change of number of events based on past earthquakes for each node
Studies for CA, China, and Japan show- Both are similar for their forecast accuracy
RI and PI need to be optimized
Forecast models using PI and RI
forecasting 2000-2009 M≥5 events
based on 1965-1999 M≥3 events
PIRI
•PI method: find seismic activation and quiescence•RI method: find seismic intensity
•PI method: find seismic activation and quiescence•RI method: find seismic intensity
Nanjo et al. (2006a,b)
Log10 P Log10 P
As of Aug. 2005
Molchan testA test to measure of matching between forecast map based on EQs. in ≤1999 and EQs. in ≥ 2000
A test to measure of matching between forecast map based on EQs. in ≤1999 and EQs. in ≥ 2000
•PI method: find seismic activation and quiescence•RI method: find seismic intensity
•PI method: find seismic activation and quiescence•RI method: find seismic intensity
Application of RI to Japan JMA catalog CSEP testing region (Bin size: 0.1 deg) Retrospective test: m≥5 events in the last 3 years
Optimization- Change t0 and minimum magnitude Mmin:
• To see the effect of catalog completeness on forecasting
- Nondeclustered and declustered catalogs: • To see aftershock effect on forecasting
tt 0 (v
ariable)
2005/04/01
2008/03/31
Forecast period58 m≥5 targets
Learning periodMmin: a variable
Likelihood test
m>=2.5m>=3.0m>=4.0
ND D
•Aftershock locations are important information of forecasting future events•Catalog completeness and maximizing data need to be considered for optimization
•Aftershock locations are important information of forecasting future events•Catalog completeness and maximizing data need to be considered for optimization
Summary
Results- Aftershock location
• Important information to forecast the location of future large earthquakes
- The need of optimization for RI forecast• Catalog completeness
• Maximize used data
• (Non)declustering
Current status for submission - Under test for the testing since 2008
- Ready for submission to the 1 day forecast class if there is any proposed one-day model!