ETD

• Deep Learning
• Earthquake Location
• Inverse Theory
• Seismology

Accurate earthquake location of seismicity is a longstanding challenge in seismology, particularly when analyzing clusters of seismicity recorded by sparse (i.e. one or two stations only) and/or asymmetric seismic networks. In such conditions, traditional location methods often struggle to correctly locate seismic events producing large uncertainties that may lead to worng interpretation of the results. Relative location techniques have improved precision by exploiting the similarity between close events. Among them, the HADES algorithm (eartHquake locAtion via Distance gEometry Solvers) marked a major step forward, reformulating the problem through distance geometry to reconstruct earthquake clusters even with very sparse seismic networks. However, its effectiveness depends on inter-event distances derived from P- and S-wave arrival-time estimates, which are sensitive to noise and inaccuracies in the velocity model.

Meanwhile, the rise of artificial intelligence (AI) is reshaping how science faces complex problems. By learning from data, AI handles noisy signals, reduces model bias, and finds pattern that traditional methods often miss. Seismology, with its data richness and natural uncertainties, is an ideal field to benefit from this new approach. This thesis introduces I-HADES, an AI-powered extension of HADES that integrates machine learning into the earthquake relocation process of seismicity clusters using distance geometry approaches. A fully connected neural network was trained on synthetic datasets, where true event separations are known, to directly predict inter-event distances from observed data. Two configurations were tested using two and four stations to evaluate the impact of network density. The results are clear: AI enhances distance prediction, mitigates the effects of noise and model errors, and makes the I-HADES framework more robust and reliable. By merging distance-geometry methods with the power of AI, this work demonstrates how artificial intelligence can unlock a new level of precision in earthquake location, opening the way for applications from sparse monitoring environments to complex seismic networks.