We propose a method of real‐time tsunami detection using ensemble empirical mode decomposition (EEMD). EEMD decomposes the time series into a set of intrinsic mode functions adaptively. The tsunami signals of ocean‐bottom pressure gauges (OBPGs) are automatically separated from the tidal signals, seismic signals, as well as background noise. Unlike the traditional tsunami detection methods, our algorithm does not need to make a prediction of tides. The application to the actual data of cabled OBPGs off the Tokohu coast shows that it successfully detects the tsunami from the 2016 Fukushima earthquake (M 7.4). The method was also applied to the extremely large tsunami from the 2011 Tohoku earthquake (M 9.0) and extremely small tsunami from the 1998 Sanriku earthquake (M 6.4). The algorithm detected the former huge tsunami that caused devastating damage, whereas it did not detect the latter microtsunami, which was not noticed on the coast. The algorithm was also tested for month‐long OBPG data and caused no false alarm. Therefore, the algorithm is very useful for a tsunami early warning system, as it does not require any earthquake information to detect the tsunamis. It detects the tsunami with a short‐time delay and characterizes the tsunami amplitudes accurately.

中文翻译：

---

基于集合经验模态分解的实时海啸检测方法

我们提出了一种使用集成经验模式分解（EEMD）的实时海啸检测方法。EEMD将时间序列自适应地分解为一组固有模式函数。海底压力表（OBPG）的海啸信号会自动与潮汐信号，地震信号以及背景噪声分开。与传统的海啸检测方法不同，我们的算法无需预测潮汐。在将电缆OBPG的实际数据应用到托胡湖沿岸的情况下，表明它已成功检测出2016年福岛地震（M 7.4）中的海啸。该方法还应用于2011年东北地震造成的特大海啸（M 9.0）和1998年Sanriku地震造成的极小海啸（M 6.4）。该算法检测到造成毁灭性破坏的前者巨大海啸，而未检测到海岸上未发现的后者微小海啸。还对该算法进行了为期一个月的OBPG数据测试，未引起任何误报。因此，该算法对于海啸预警系统非常有用，因为它不需要任何地震信息即可检测到海啸。它以短时延迟检测海啸，并准确表征海啸振幅。