A new alarm-based earthquake forecast model: Application to M7+ events in Japan

Speaker: Talbi Abdelhak
We present an alarm-based earthquake forecast model that uses the ratio of the first to second order moments of earthquake inter-event times, namely mean ratio (MR), as a precursory alarm function to forecast large earthquakes. The proposed model is based on the idea that MR is associated with long term changes in background seismicity which may be anomalous before large events. The performance of the proposed model in forecasting M≧7 target earthquakes is tested using inter-event times from M≧6 events occurred in Japan during the period 1890-2012. Three retrospective tests are performed to evaluate long, intermediate and short term performance. To discuss the applicability of the proposed model, we use optimal spatial maps obtained by minimizing the forecasting error (miss rate + alarm rate). Using an alarm rate of about 20%, the forecast maps catch in hot spots 15 of the 18 shallow earthquakes occurred in the testing region during the last two decades. Comparing to the relative intensity (RI) method, the miss rate of shallow events is reduced by about 60 %. At short term, the proposed model succeeds in predicting the 2011 Mw 9.0 Tohoku earthquake for which very high MR values are calculated few time before the mainshock. Finally, we examine MR signals from 0.5°x0.5° grid cells, and find high MR values with eventually typical MR increase before many large events occurred between 1890 and 2011.