An Automated Workflow for Reliable Focal Mechanism Determination of Low-Magnitude Earthquakes in the Southeastern Alps

In regions dominated by low-magnitude seismicity, obtaining reliable focal mechanisms (FMs) at the catalog scale remains challenging, particularly when waveform-based inver- sions are limited by simplified velocity models and network geometry. Here, we present an automated workflow for polarity-based FM determination that combines deep-learning first-motion classification using the convolutional first-motion model with probabilistic inversion through SKHASH, accounting for uncertainties in source location, velocity structure, and polarity assignment. Once event locations and P-wave arrivals are available, the workflow requires no manual polarity review and provides quantitative quality metrics for solution robustness. The workflow is validated against a reference focal-mechanism catalog of 159 earthquakes with manually reviewed polarities, used here as a validation dataset. Automated polarity classifications agree with manual picks in 92% of cases, and 73% of the resulting FMs differ by less than 30° in Kagan angle from the reference solutions, demonstrating reliable performance for small-magnitude events. We apply the workflow to approximately 1500 earthquakes recorded in northeastern Italy between 2024 and 2025 (1:0 ≤ Md ≤ 4:6), yielding 122 quality-controlled FMs, con- sistent with the established seismotectonic framework of the southeastern Alps, and pro- ducing the first automated FM catalog for this region. Nearly 75% of the retained solutions correspond to earthquakes with Md ≤ 2:5, significantly densifying the available FM dataset at small magnitudes. An additional application to the 2025 Raveo earthquake sequence confirms the consistency of the automated solutions with independently derived moment tensor and manual focal mechanism results.