“…In recent years, deep-learning-based phase pickers/event detectors (e.g., Kriegerowski et al, 2019;Mousavi et al, 2020;Ross et al, 2018;Soto & Schurr, 2021; have been gaining increasing attention due to their picking accuracy being comparable to human analysts (Chai et al, 2020) and high efficiency. Their application has surged in recent years, including for delineating seismicity in fault zones, subduction zones, oceanic transform faults, and volcanoes (e.g., Chen et al, 2022;Garza-Girón et al, 2023;Gong et al, 2023;Jiang et al, 2022;Liu et al, 2023;Liu et al, 2024;Tan et al, 2021;Wilding et al, 2023;Zhang et al, 2024). However, it can be difficult to predict deep-learning models' performance for outof-distribution data that are not well represented by training data (Teney et al, 2022;Wenzel et al, 2022).…”