Determination of reliable hypocenters of earthquakes is crucial to earthquake seismology and to evaluate hazards associated with earthquakes. There are many associated computer codes for this purpose; however, most of the location algorithms are designed to determine hypocentral parameters based on previously determined velocity models. In contrast, we employed a location method that is independent of the initial velocity model, using a genetic algorithm (GA) to determine an optimal 1D velocity model and the locations of earthquakes. Using this GA, we relocated earthquakes that occurred in the New Madrid Seismic Zone (NMSZ) in the central United States between October 1989 and August 1992. The goal of this work was to delineate the possible fault planes by reliable relocation of those earthquakes and to determine a 1D velocity structure for the NMSZ. A total of 502 earthquakes recorded by 37 Portable Array for Numerical Data Acquisition (PANDA) stations were used in the relocation study. In the relocation process, the root mean square travel‐time residuals were reduced by ∼35%, corresponding to an average of 2.3 km deeper in depth, 0.7 km shift in latitude, and 0.8 km shift in longitude compared with those in the initial catalog locations. The hypocenters of the earthquakes can be subdivided into four groups based on their spatial distributions. The group that corresponds to the Cottonwood Grove fault (CGF) in the southwestern NMSZ represents a very steep plane, whereas the other three groups fall into Reelfoot fault (RF). We inverted P‐ and S‐wave travel times from the new hypocentral parameters to determine 1D velocity models. The resulting eight‐layered velocity models consist of a 2 km thick surface layer followed by seven 2 km thick layers, with VP ranges from 5.36 to 6.74 km/s and VS ranges from 2.83 to 3.90 km/s for both CGF and RF regions.
Online Material: Interactive visualizations of hypocentral distributions.