Microearthquakes detection in the Korean Peninsula using machine learning algorithm

Since the instrumental earthquake detection in 1978, there were few large earthquakes in the Korean peninsula. However, historical literatures suggest the occurrence of major earthquakes in the Korean Peninsula. The distribution of microearthquakes suggest valuable information about the area where major earthquakes may occur. We detect microearthquakes in the Korean peninsula using machine learning algorithms. We use two model architecture that include the convolutional neural network with six layers of four convolution layers and two fully-connected layers (Ross et al., 2018) and PhaseNet (Zhu and Beroza, 2019). We use STEAD (Stanford Earthquake Dataset) and phase data in the Korean peninsula to train the models. We determine the best trained model. We detect the seismic events in the Korean Peninsula. We pick P and S phases. We identify microearthquakes from coherent phase arrivals across stations. The detected phases are aligned for the P and S traveltime curves. In the Seoul metropolitan area, about 30 microearthquakes are identified every month. In this study, we newly find dozens of microearthquakes in regions of low instrumental seismicity. The applied method enhances the earthquake-detection ability in urban region, allowing us to monitor microearthquakes.