A quasi-real-time system for automatic local event monitoring in Germany

Catalina Ramos , Stefanie Donner , Klaus Stammler

We present TieBeNN, a wrapper that integrates open-source, state-of-the-art seismic monitoring tools, including advanced machine learning--based approaches, to enhance the German Federal Seismological Survey’s (EdB) automatic real-time earthquake monitoring system. TieBeNN extends the existing workflow by adding automatic, probabilistic focal depth estimation using NonLinLoc and introduces a Location Quality Score (LQS) to quantify location reliability with a single metric. In testing, TieBeNN’s automated locations approach the accuracy of human analyst solutions, demonstrating comparable performance in well-instrumented regions. By automating depth determination and providing immediate quality assessment, the system reduces analysts’ daily workload, allowing them to focus on events flagged as low-quality or complex. The LQS effectively distinguishes well-constrained event locations from those with large uncertainties or poor network geometry, enabling rapid identification of high-quality automatic results versus those requiring review. However, events below the Moho depth (i.e., deeper than approximately 30~km), which are rare in Germany, remain challenging: their uncertainties are larger, and LQS values tend to be lower, indicating limitations in the current calibration. Overall, these enhancements significantly advance real-time local seismic event monitoring in Germany by increasing both the speed and reliability of automatic event characterization.