The accurate estimation of earthquake magnitude is crucial for assessing seismic hazards and ensuring effective disaster mitigation strategies. EqShake is a deep-learning model for accurate earthquake magnitude estimation using single-station raw waveforms. As opposed to other available models, EqShake is designed to be independent of waveform length, applicable to events recorded at local scales (<3 deg). EqShake was trained on 140k samples from STEAD dataset (https://github.com/smousavi05/STEAD).
EqShake takes advantage of both P and S waves for magnitude estimation. As only the first 3 seconds of the body waves are used as the input, EqShake works for waveforms with P-S separation larger than 3 seconds. Furthermore, event-station distance is fed to the model to account for the attenuation.
The test dataset included here consists of 30k samples. EqShake demonstrates excellent performance with R2-score= 0.93, MSE= 0.037, Mean= 0.0, and std= 0.22.
