We propose the use of statistical measures to quantify robustness, uncertainty, and significance for Markovian models of large magnitude seismic systems, and we also propose a method for choosing the best of different models by using the normalized measures in a discriminant function. We tested the proposed methods on earthquakes occurring in an area around Japan, divided into four regions; modeling the system as having four states, where each state corresponds to the region where the latest large earthquake, larger than a given threshold moment magnitude, has occurred. Our results show that for the 7.0–7.3 threshold magnitude range the seismicity of this region does occur according to a Markovian process, with optimum results for threshold magnitude 7.1, whereas for magnitudes outside this range seismicity is less Markovian.

我们建议使用统计量度来量化大震度系统的马尔可夫模型的鲁棒性，不确定性和重要性，并且还提出一种通过在判别函数中使用归一化量度来选择最佳模型的方法。我们对日本周边地区（分为四个地区）发生的地震测试了建议的方法。将系统建模为具有四个状态，其中每个状态对应于发生了大于给定阈值矩大小的最新大地震的区域。我们的结果表明，对于7.0–7.3阈值震级范围，该区域的地震活动确实按照马尔可夫过程发生，对于7.1阈值震级的最佳结果，而对于超出此范围的震级，则马尔可夫效应较小。