Machine learning earthquake prediction method based on seismic activity images and its application in North China

This paper presents a study on machine learning methods for predicting earthquakes in North China based on seismic activity images. According to the average time interval between moderately strong earthquakes （magnitude 5.0 or above）, strong earthquakes, and major earthquakes in North China, a large number of M_L≥3.0 seismic activity images at different time scales before the occurrence of such earthquakes since 1970 are used as input datasets. The machine learning method for predicting earthquakes based on seismic activity images is proposed and tested using case studies. Using the method proposed in this paper, we select “unexpanded image dataset” and “expanded image dataset” to predict moderate-strong earthquakes in North China, it is found that increasing the sample size of the input image dataset is beneficial to improve the accuracy of the earthquake prediction models. The accuracy rate of earthquake activity image prediction in North China using the expanded image dataset can reach 77%. For M≥5.0 earthquakes happened in aseismic zones or low-seismic regions in North China, seismic activity images with larger time intervals than one year （such as three years or more than seven years） with M_L≥3.0 seismic activity images can be used to verify the target earthquakes in aseismic zones or low-seismic regions. This research can provide important guidance for earthquake prediction in North China.