Conference Agenda

Approaches of monitoring earthquakes have evolved from conventional ground-based networks to include space. In the areas of seismology, geology and geophysics, scientists believe that the events leading up to earthquakes goes through a complex process and the process is somehow chaotic. Understanding earthquakes requires a breakthrough from traditional approaches to utilizing advanced technology. In fact, the seismology discipline has expanded the scope of earthquake study from conventional ground-based observations to space. In particular, since the Swarm satellite mission lunched in 2013, they have paved a way to provide a wide range of measurements in space by Vector Field Magnetometer, Absolute Scalar Magnetometer, Electrical Field Instrument, etc. instrumental sensors.

The measurements delivered by the three satellite are very valuable for a range of applications, including earthquake prediction study. However, for more than 5 years, relatively little advancement has been achieved on establishing a systematic approach for detecting anomalies from the satellite measurements for predicting earthquakes before they occur. This report presents a continuous effort, describing essential functional components of a system for automatic anomaly detection. Through a case study we demonstrate the functionality of the system in detecting anomalies, and the process of data processing and analysis along with experience in developing a viable tool for precisely discovering seismic anomalies from the observed data by the Swarm satellites.

Earthquake is one of the most severe natural disasters. More than half of lives lost is caused by earthquake among natural disasters in the world. The earthquake also causes huge economic losses, e.g., about 800 billion RMB were lost during the Wenchuan EQ (Mw=7.8, 2008). Therefore, the governments and scientist of many countries, especially of those with frequent earthquakes, pay great attention to the study of earthquake prediction. It was in the 1960s that the study on the earthquake prediction started and national projects for EQ prediction designed in several countries e.g., China, Japan, USSR and USA (Chen, et al, 2000, Uyeda, 2015).

As well known, the earthquake prediction is a difficult scientific problem in the world. It is often debated and doubtful about that earthquake could be predicted and about whether there is any observable anomalous precursor prior to the earthquake. During the last decades, a lot of example of observable short-term precursors was published letting more scientists considering that there indeed exists precursor before an earthquake and it can be observed only if the reasonably effective method is used. The electromagnetic (EM) method is believed by scientists to be one of the methods that can be used to first reach success in short-term prediction. In this report, we will introduce the first EM observation network built recently using the alternate EM field and its exploitation for monitoring earthquakes. As an example, we will present how to use natural EM source to capture the precursors before the 5.1 Ms Yangbi earthquake in Yunnan province, China, along with a comparative study with the result detected from the Swarm electromagnetic data in the corresponding period of time.

The study is supported by NSFC (41674081,41374077）and NDICC (15212Z0000001). The members of EM group of Institute of Geology, China Earthquake Administration and colleagues in the Yunnan Earthquake Administration joined the construction of the network and data observation.

Since the Swarm satellite constellation launched November 2013, three satellites have delivered immense amounts of geomagnetic field measurements. Currently it is very extremely difficult for researchers to track observed data based on the instrumental sensors, such as Vector Field Magnetometer (VFM), Absolute Scalar Magnetometer (ASM), and conduct anomaly detection, it is imperative to develop more effective data analytics for detecting and discovering abnormalities in the electromagnetic time series data for earthquake studies.

In this report, we propose a Martingale framework which could be adequate for assessing abnormal changes within electromagnetic data streams. The martingale method becomes essential as traditional statistical approach are inappropriate for the high dimensional electromagnetic dataset (Vapnik, 1998). The first step using the framework is to categorically obtain a practical data model in the machine learning standard scenarios through the use of strangeness measures. The strangeness measures establish a way of testing the exchangeability assumption of the dataset using a hypothesis test which drives the martingale process. The Martingale model will also involve the use of machine learning smoothing techniques to reduce noise and other interference efficiently making the framework more sensitive in detecting change point/anomaly. And finally, the model will be evaluated over Swarm electromagnetic data based on the two selected earthquakes, compared against a benchmark method and studied on its effectiveness in detecting abnormal changes before the earthquakes.