Drastic changes before the 2011 Tohoku earthquake, revealed by exploratory data analysis

Predicting earthquakes is of the utmost importance, especially to those countries of high risk, and although much effort has been made, it has yet to be realised. Nevertheless, there is a paucity of statistical approaches in seismic studies to the extent that an old theory is believed without verification. Seismic records of time and magnitude in Japan were analysed by exploratory data analysis (EDA). EDA is a parametric statistical approach based on the characteristics of data and is suitable for data-driven investigations. The distribution style of each dataset was determined, and the important parameters were found. This enabled us to identify and evaluate the anomalies in the data. Before the huge 2011 Tohoku earthquake, swarm earthquakes occurred before the main earthquake at improbable frequencies. The frequency and magnitude of all earthquakes increased. Both changes made larger earthquakes more likely to occur: even an M9 earthquake was expected every two years. From these simple measurements, the EDA succeeded in extracting useful information. Detecting and evaluating anomalies using this approach for every set of data would lead to a more accurate prediction of earthquakes.地震预测对于高地震风险国家而言至关重要。尽管学界已投入大量研究精力，但目前该目标仍未实现。当前地震学研究中统计方法的应用仍较为匮乏，以至于部分老旧理论未经验证便被直接采信。本研究针对日本地区的地震时间与震级记录，采用探索性数据分析（Exploratory Data Analysis, EDA）展开分析。探索性数据分析是一种基于数据特征的参数化统计方法，适配数据驱动型研究场景。研究确定了各数据集的分布形式，并提取出关键参数，借此得以识别并评估数据中的异常特征。在2011年东日本特大地震发生前，主震前的群震活动以远超常规的频率发生，所有地震的发生频率与震级均呈现上升趋势。这两项变化均增大了强震发生的可能性：在此异常情境下，M9级地震的预期发生周期甚至缩短至每两年一次。通过上述分析手段，EDA成功提取出了具有参考价值的异常信息。若将该方法推广应用于各类数据集的异常检测与评估，有望实现更为精准的地震预测。