17:15 〜 19:15
[SSS13-P04] Operational Bayesian network to assist probabilistic earthquake prediction
キーワード： Bayesian Network, probabilistic earthquake prediction
We designed and trained a Bayesian Network (BN) and calculated the probability of future earthquakes in different strong earthquake hazard zones in mainland China based on the long-term, medium-term, and short-term historical earthquake catalogs. First we collected the Global 10-Year M7 Earthquake catalog,the mainland China 3-Year M6 catalog as the long-term catalogs and use them as the first layers' input of the network,then we devide the mainland China into 8 strong earthquake risk zones according to the historical seismicity,and collected medium- and short-term period earthquake catalogs for each region from 1 to 12 months.We use all these data to build the Directed Acyclic Graph (DAG) of the BN network,the target output is whether an earthquake of magnitude 5 or above will occur in each of these eight regions next month.About 600 effective training samples were established from 1970,and we also build the test dataset with 684 samples. We use the accuracy rate, false alarm rate and R value to valuate the predictive performance of the model. Overall,the accuracy rate of the 8 region is from 75% to 100%.However, the false alarm rate is still relatively high,from 25% to 91%.The average R value is 0.67. Our results show that neural network plus Bayesian decision making is an operational probabilistic earthquake prediction solution,and is suitable for short- to medium-term forecasts, with objective conclusions.Through retrospective testing, the strategy threshold is automatically optimized, and the forecast accuracy is ensured, which is better than human prediction.