[SSS07-P06] Kanto 3 month testing class results of CSEP Japan
Keywords:Earthquake Predictability, CSEP
From the standard test results of CSEP, RI10K did not pass the S-test in the round of 2017/5 / 1-8 / 1. In addition, there were some rounds that did not pass N-Test in all models when the number of observed earthquakes was relatively small. Calculating the average of information gain values per earthquake for each model in all rounds was 1.419, 1.168, 1.162 for HISTETAS5PA1205, HISTETAS5PA, and RI10K respectively. Regarding the forecast of the number of earthquakes, HISTETA5PA1205 predicts the number of observation earthquakes well.
From the results of multiple rounds, forecast number of earthquakes improvement directly leads to improvement of an earthquake forecast model. Moreover, a model characteristic can be showed well by evaluating the spatial forecast performance and the performance of forecast number of earthquakes.