Artificial Neural Network Based Estimation of Moment Magnitude with Relevance to Earthquake Early Warning

An Artificial Neural Network (ANN) based algorithm for rapid estimation of seismic moment magnitude (M-w) for Earthquake Early Warning (EEW) is proposed in the paper. The rapidness here refers to regional warning as fast as 3 s after the arrival of P wave in the seismogram. The amplitude spectrum computed in the frequency range between 0.0098 and 10 Hz from the vertical component of seismograms recorded by a single 3-component station have been considered as inputs to the ANN and Mw as desired output. The trained ANN has been found to estimate Mw of about 73% of the new events with absolute error less than 0.1 and of 100% of the events with absolute error less than 0.35. The method has been tested using vertical component broad band seismic data recorded at PALK (Pallekele, Sri Lanka) station provided by Incorporated Research Institutes for Seismology (IRIS) for earthquakes originating from Sumatra region of magnitude 6 and above. The fair agreement between the estimated Mw and that reported by United States Geological Survey (USGS) ensures that a rapid reliable EEW system could be developed based on proposed method.