Genetic waveform modeling for the crustal structure in Northeast Japan

• Genetic algorithm and generalized ray theory are used for waveform modeling.
• A 1-D crustal velocity model in NE Japan is obtained by waveform modeling.
• Waveforms and later-phase travel times of a local crustal earthquake are used jointly.
• Travel-time residuals obtained reflect the crustal heterogeneities.

We propose a genetic algorithm (GA) search procedure for waveform modeling of local crustal earthquakes for optimal one-dimensional (1-D) crustal velocity model. Both waveforms and travel-time data are used for the structure determination. The use of travel times in model evaluation improves the waveform modeling performance in the sense of computation speed and accuracy. We applied this method to broadband waveforms of a local crustal earthquake (M 4.2) in Northeast Japan. P-wave velocities of the crustal model are found to be 4.95 ± 0.30, 5.9 ± 0.02, and 6.51 ± 0.20 km/s for a surface layer, upper crust and lower crust, respectively. The surface layer thickness and the Conrad and Moho depths are found to be 3.01 ± 0.8, 17.77 ± 0.4 and 34.59 ± 1.0 km, respectively. For epicentral distances <200 km, our synthetic waveforms match the observed ones generally well. Early arrivals are mainly observed at stations near the Pacific coast in the forearc area having a thinner crust. In contrast, delayed arrivals appear at stations near the volcanic front and back-arc areas where low-velocity anomalies exist due to the effect of the Pacific slab dehydration and the hot upwelling flows in the mantle wedge. In general, our results agree well with the main tectonic setting of the study area, which confirms the reliability of the proposed approach. Despite a 1-D velocity model is too simple to represent the complex crustal structure, it is still required for the conventional routine analysis of seismology, such as earthquake location and source parameter studies. The current approach is considered as a step toward the genetic full waveform modeling for the 3-D velocity model estimation.