Monitoring of landslide affecting factors using automatic microseismic location technology in Deda town, Tibet

Landslides are one of the most dangerous natural hazards. Despite several efforts to study this phenomenon, there is still little clarity regarding dynamic processes associated with landslides. Recently, seismic signals are used to analyze the dynamic properties of landslides, because seismograms provide time-series recordings of sliding during run out. One of the important steps is to achieve a microseismic location. By analyzing the source, the events and the magnitude of the microseismic generated by landslides can indicate the risk of the slip surface. In previous studies, many people have outstanding performance at one aspect of microseismic localization. But those methods often need too many specialist operations and are difficult to achieve real-time and automatic operation. In this paper, we proposed the automatic microseismic location technology by CNN and applied it to landslide monitoring at Deda town, Tibet.

This automatic microseismic location technology is mainly divided into four steps: signal classification, first picking, phase connection, and hypocentral location. Both signal classification and first picking are based on CNN, which can automatically extract waveform features and avoid the tedious parameter setting from traditional detection technology. CNN is also a key point of intelligent processing of microseismic signals. In our study, different network architectures are used to improve the accuracy of these two tasks. Signal classification focuses on the difference between microseismic signals and noise, while time pick-up is the identification of the first starting point of microseismic signals after obtaining effective microseismic signals. In microseismic phase connection, we modify the “coincidence_trigger” function provided by Obspy(a Python library for seismic) to adapt to CNN predictions. Events identified by CNN were saved as waveform fragments containing multiple stations after phase connection. Meanwhile, timestamps were also saved. In the last step, the Newton method was adopted for source location, which proved to be very reliable in accuracy and stability through experimental comparison. By loading the time of microseismic events and station positions, we can achieve location. Since the number of stations detected by each microseismic event was not the same, dynamic processing was also carried out here. Therefore, the whole process of microseismic positioning only needs to input waveform data obtained by geophone and corresponding station information, without the operation of experts.

We applied this scheme to the field data are collected from Deda town, which is located in Tibet. There are faults on both sides of the mountain slopes. A total of 76 microseismic events were detected in 27 days by using our automatic microseismic location technology. All the events were located near the faults, and some events happened near the slip surface. But the magnitude of almost all of the events is less than 0 so we think these events are related to the landslides' energy release.