DDSync : Graph-Based Denoising of Differential Travel-Time Observations with Applications to Pick Reconstruction and Path-Difference Tomography

Differential travel times from waveform cross correlation are among the most precise routine measurements in earthquake seismology. In practice, however, double-difference pair files often contain cycle skips, outliers, and internally inconsistent links that can bias downstream relocation and other inversions. This is increasingly consequential for modern, very large catalogs, where automated quality control is essential and sequential relocation workflows can be brittle to a small fraction of bad measurements. We introduce DDSync, a lightweight graph-based conditioning method that treats differential times at each station–phase as a weighted event graph and solves a synchronization problem to recover a self-consistent set of relative arrival-time potentials (one scalar per event per station–phase). DDSync combines (1) a baseline weighted least-squares synchronization that implicitly averages over many redundant paths in dense graphs, (2) residual-based pruning using a median absolute deviation threshold to remove internally inconsistent links, and (3) robust reweighting with a Huber loss. Using a published synthetic benchmark with known ground truth, we show that synchronization removes severe outliers and reduces the scatter of retained differential times to a few milliseconds. We further show that the recovered potentials act as highly precise relative arrival proxies that support long-baseline differential reconstruction, improve downstream relocation, and might form a basis for path-difference tomography formulation.