Relocation of a reference seismic events catalog: influence of 1D and 3D velocity models and location methods

Earthquake locations derived from seismic phase arrival times are highly dependent on the velocity model used to predict theoretical travel times. In operational monitoring, simplified 1D velocity models are commonly used to ensure rapid processing, but such approximations may introduce systematic biases in hypocentral estimations, particularly in regions characterized by strong lateral heterogeneities. This issue is especially relevant in mainland France, where complex crustal structure challenges standard localization strategies.

In this study, we focus on quantifying the impact of different velocity models on earthquake locations, with particular emphasis on the contribution of a national 3D velocity model derived from passive seismic imagery. To minimize the influence of network geometry and isolate the effect of velocity structure, we rely on a high-quality reference catalog extracted from more than 88 000 seismic events reported in the CEA catalog since 1963 (Mazet-Roux, 2025). Event selection is based on Ground Truth (GT) criteria, which assess epicentral location accuracy solely from network geometry (Bondár et al., 2001; Bondár et McLaughlin, 2009). The 22 000 events satisfying the GT5 95% criteria, associated with an epicentral accuracy better than 5 km with a 95% confidence level, define our reference event set.

We relocate this reference catalog using two advanced location algorithms capable of incorporating complex velocity structures. NonLinLoc performs a fully probabilistic search of the solution space and estimates full hypocentral uncertainty distributions (Lomax, 2008; Lomax et al., 2009), while iLoc applies a hybrid, iterative approach optimized for high-precision earthquake location (Bondár and McLaughlin, 2009; Bondár and Storchak, 2011). These methods are first compared using a common 1D velocity model (. The impact of a regional 3D velocity model, obtained through the homogenization of multiple regional models constrained by passive seismic tomography (Arroucau, 2020; Arroucau et al., 2021), is then assessed using the probabilistic NonLinLoc approach.

Regardless of the location algorithm or velocity model considered, more than 95% of the reference events exhibit epicentral differences smaller than 5 km, reflecting the robustness of the GT-based event selection. The differences mainly concern depth estimates and the associated uncertainties.

Indeed, using a common 1D velocity model, significant differences arise in depth determination and associated uncertainties. NonLinLoc systematically converges toward free-depth solutions with quantified uncertainties, whereas iLoc fixes the depth for more than 55% of the events when depth is poorly constrained.

The introduction of the 3D velocity model leads to systematic changes in hypocentral depths, with inversions yielding statistically deeper events on average and uncertainty ellipses becoming better constrained compared to 1D velocity model, despite a slight statistical increase in depth uncertainty for reference events. Differences in depth uncertainties seem to reveal regional variability and possible dependence on event depth. Comparisons with well-documented seismic sequences and previous studies are discussed to better interpret the observed differences between velocity models.