SCMLPick: A SeisComP Module Implementing Machine Learning Phase Picking for Real-Time Seismic Monitoring

Abstract Rapid and accurate detection of seismic phases is essential for effective seismic monitoring. Although various detection methods exist, recent advances in machine learning (ML) algorithms have demonstrated superior performance in automatically identifying seismic phases. However, the real-time deployment of such models remains challenging due to substantial computational demands and limited feasibility of integration with operational workflows. Since 2023, the Texas Seismological Network (TexNet) has utilized an ML-based picker, increasing the detection rate of low-magnitude earthquakes (primarily M < 1.0) by a factor of 50 compared to the conventional short-term average to long-term average method. Nevertheless, the existing implementation operates in postprocessing mode, introducing latencies exceeding 10 min relative to phase onset. To overcome this limitation, we developed SCMLPick, a novel SeisComP module that enables the real-time application of ML-based phase pickers within operational seismic networks. SCMLPick bridges the gap between state-of-the-art ML algorithms and the stringent demands of real-time seismic processing. Its development focused on integrating ML-based pickers into the SeisComP framework, optimizing runtime performance, and tuning region-specific parameters. SCMLPick continuously processes real-time waveform data, achieving average phase-picking latencies of less than 60 s, a significant reduction from the 600–3000 s observed in the previous postprocessing implementation. A comparative evaluation against TexNet’s existing postprocessing implementation demonstrated strong agreement in pick and event detections, with over 98% matching of events for M ≥ 1.0 and high consistency in arrival time estimation. Region-specific tuning further enhanced the detection accuracy and operational reliability of SCMLPick.