DoubleF: A Fast and Flexible Phase Association and Earthquake Location Method Using Adaptive Sobol Sampling

Abstract Earthquake catalogs play a vital role in seismological research. Recent advances in deep learning-based phase picking have greatly enhanced the efficiency and completeness of catalog construction. A crucial step in this workflow is phase association, which links detected phases to their corresponding seismic events. We present DoubleF, a fast and flexible phase association and earthquake location method based on adaptive Sobol sampling. We evaluated DoubleF using three datasets: the 2023 MS 6.2 Jishishan earthquake aftershock sequence, the Baihetan reservoir-induced seismicity from 1 to 31 October 2021, and synthetic data. These experiments were designed to assess the accuracy, robustness to noise, and sensitivity to parameter settings. In real-data experiments, DoubleF generally associated the most events and phases and achieved the highest recall among several widely used methods tested, improving recall by approximately 1% to 4% over the next best method. In synthetic experiments, DoubleF achieved nearly 100% event and phase recall, even when 80% of the phases were noise. Through Sobol-based sampling, matrix parallelization, and graphics processing unit acceleration, the algorithm substantially reduces computation time relative to the tested methods by up to several dozen times. Moreover, the insensitivity of the method to parameter settings ensures its broad applicability, from dense networks for monitoring induced seismicity to sparse regional networks, and makes it particularly well suited for real-time monitoring.