RESEARCH TOWARDS REAL SEISMIC ACTIVITY MONITORING SYSTEM BASED ON STATISTICAL MODELING

尽管有关于大地震前异常地震活动的各种研究报道，但目前还没有真正的地震监测系统可用。本研究由三部分组成：1)异常检测系统，2)地震先验概率估计，3)结果展示系统。首先，构建地震活动性监测系统，检测地震活动性异常。地震活动最显著的特征是空间和时间上的集中，这一点用Ogata于1998年提出的ETAS模型得到了最好的描述。根据该模型，所有的活动可以分为统计上独立的背景活动和统计上相依的余震。以往的分析大多避免客观地将两者分开，从而将正常活动判断为异常，或者在大地震前无法检测到异常。我们使用背景地震活动来检测异常活动，在随机分离独立和d…之后使用强度函数可以获得更多的从属均分。我们将检测系统应用于实际地震活动数据和ETAS模型产生的合成地震活动数据的结果进行了比较。我们使用两个模型之间的AIC(=SAIC)之差：一个假设某一时间段的异常活动，另一个假设整个时间段的正常活动。我们的分析表明，SAIC小于-8的地震平静期和SAIC大于10的强震活动是显着的异常。对于第二个主题，我们在过去各种建议的基础上，构建了八个地震潜力的时不变概率模型。并与历史地震资料进行了对比检验，认为基于活动断裂资料的空间平滑综合地震活动模型是最可靠的。第三，我们建议使用动画电影来展示地震监测的结果。作为一个例子，我们的报告显示了b值的空间分布随时间变化。较少

No real earthquake monitoring system is available, although various researches have been reported on anomalous seismic activity prior to a large earthquake. This study consists of three parts: 1) detection system of anomalies, 2) prior earthquake probability estimate, and 3) presentation system of results. The first of all, a monitoring system of seismic activity is constructed to detect anomalous seismicity. The most prominent feature of seismic activity is concentration in space and time, which is best described by the ETAS model proposed by Ogata in 1998. According to the model, all the activity can be divided into two: statistically independent background activity and statistically dependent aftershocks. Most of the previous analysis often avoided to separate the two objectively, thus judged ordinary activity as unusual, or could not detect anomaly prior to a large earthquake. We used background seismicity to detect unusual activity after stochastically separating independent and d … More ependent evens using the intensity function. We compare the results obtained by applying the detection system to the real seismicity data and also to synthetic seismicity data produced from the ETAS model. We use the difference in AIC (=SAIC) between two models: one assuming anomalous activity during a certain time-period and the other assuming normal activity throughout all the time period. Our analysis shows seismic quiescence showing SAIC smaller than -8 and high seismicity showing SAIC larger than 10 are significant anomalies. For the second subject, we constructed eight time-invariant probabilistic models of earthquake potential on the basis of various proposals made in the past. Then we test them against historical earthquake data and concluded that the spatially smoothed synthesized seismicity model based on active fault data is the most reliable. Thirdly we recommend to use animation movie to show the result of earthquake monitoring. As an example, the spatial distribution of b-value is shown to change in time in our report. Less