Evaluating earthquake location precision in simple and complex velocity models using Markov chain Monte Carlo sampling

Abstract Accurate and precise earthquake locations are essential for seismic catalogues and unbiased geophysical research, but their precision is often limited by assumptions about the velocity model used. The construction of a 3D model can improve precision compared to a previously available 1D model, although a quantitative assessment of this impact has been lacking. To address this, we conducted a comparison between earthquake locations in the Alto Tiberina (AT) study area, Northern Apennines of Italy, obtained using three elastic models: (1) a homogeneous half-space model, (2) a 1D elastic model developed for the broader Umbria-Marche region and (3) a 3D elastic model developed ad-hoc for the AT study area, where a dense seismic network operates. Using a Markov chain Monte Carlo algorithm, we evaluated the precision of the hypocentral parameters for each model. Multiple events were analysed at different positions within the study area to test the models performance across varying seismic network configurations. Our results quantify the improvement in earthquake locations, particularly hypocentral depth, achieved with the 3D velocity model, relative to those obtained using the homogeneous and 1D models. For events located at the centre of the seismic network, depth uncertainty decreased to one-third, while events at the network's periphery showed reductions of up to 90% from the homogeneous to the 3D model. Epicentral uncertainties also decreased: by 50% for events located at the border and by up to 90% for events outside the network shifting from the homogeneous to the 3D model. This quantitative analysis underscores the advantages of using a 3D model for earthquake location, particularly in improving hypocentral and epicentral precision across different network positions.