Robust Imaging of Deep Seismogenic Fault Geometry through Earthquake Relocation: The 2019 Durrës Sequence

Large earthquakes in slowly deforming collision zones often occur in regions where dense near-fault seismic networks are absent at the time of rupture. As a result, the most critical events for understanding deep seismogenic processes are frequently those for which observational constraints are intrinsically limited. The Mw 6.4 Durrës earthquake of 26 November 2019, at the eastern front of the Adria–Eurasia collision in Albania, is a prime example: it occurred beneath a complex foreland basin system, with a sparse and asymmetric station geometry that challenges conventional earthquake location methods.

In this study, we address the central question: how robustly can the geometry and depth of a deep seismogenic source be constrained when observational conditions cannot be improved retroactively? We relocate the full 2019–2020 Durrës sequence (foreshocks, mainshock, and aftershocks) using the hypoDD algorithm applied to catalog differential travel times. While no waveform cross-correlation data are available, the network of differential-time links is internally well-connected, allowing relative event positions to be resolved far more precisely than absolute hypocenters.

To obtain physically meaningful uncertainty estimates beyond formal inversion errors, we adopt a stepwise pre-relocation approach, including depth quality control and jackknife station weighting. A bootstrap resampling of the differential-time equations (200 realizations) is then applied to derive full spatial probability clouds for each event. This approach reveals a fundamental asymmetry in what the data can and cannot resolve: epicentral positions and along-strike geometry are highly stable, forming a compact NW–SE-oriented cluster, whereas individual event depths are less tightly constrained. Importantly, however, the bootstrap distributions are unimodal and consistently centered at ~18–23 km, demonstrating that the sequence is rooted in a deep seismogenic layer despite kilometer-scale depth uncertainty for single events. These results show that, even under unfavorable network conditions, a combination of differential-time relocation and uncertainty-aware resampling can robustly identify the depth range, orientation, and spatial coherence of an active fault system. In the case of Durrës earthquake, this supports a deep, NE-dipping blind fault associated with the collision-front architecture of Adria beneath the Periadriatic Depression.

Beyond the specific case study, our analysis provides a framework for translating limited coseismic datasets into actionable tectonic insight and for guiding the design of future seismic and geodetic monitoring strategies in regions where damaging earthquakes have long recurrence intervals but high societal impact.