MACHINE LEARNING FOR FAST AND RELIABLE SOURCE LOCATION ESTIMATION IN EARTHQUAKE EARLY WARNING

Abstract

Our goal in creating this random forest (RF) model is to help earthquake early 
warning (EEW) systems make quick decisions when it comes to earthquake location. 
Utilizing P-wave arrival times at the first five seismic stations, this technique 
calculates the relative arrival times of each station in relation to a reference station, 
which in this case is the first recording station. In order to determine the approximate 
position of the epicenter, the RF model categorizes these differential P-wave arrival 
timings and station locations. The suggested technique is trained and tested using a 
Japanese earthquake database. With a Mean Absolute Error (MAE) of 2.88 km, the 
RF model generates very accurate earthquake location predictions. Notably, the 
suggested RF model may acquire sufficient knowledge using only 10% of the 
information and far fewer recording stations (i.e., three) while still attaining 
acceptable outcomes (MAE<5 km). An effective new tool for quick and trustworthy 
source-location prediction in EEW is provided by the method, which is accurate, 
generalizable, and responds quickly.