Excessive-rate part affiliation with journey time neural fields

Earthquake science and seismology depend on part affiliation: grouping seismic arrivals recorded throughout a number of stations into the earthquakes that generated them. Modern deep-learning detectors now routinely determine many extra small occasions than earlier than, offering details about fault dynamics on more and more tremendous spatiotemporal scales. But the ensuing high-rate arrival sequences are more and more troublesome to affiliate, particularly when native wave speeds are heterogeneous or poorly recognized. Here we introduce HARPA, a phase-association framework that represents noticed and predicted arrival sequences as chance distributions and compares them utilizing an optimal-transport metric. HARPA collectively estimates earthquake places, origin occasions and a low-dimensional illustration of the wave-speed area utilizing travel-time neural fields, neural networks that map coordinates to journey occasions. On commonplace low-rate datasets with easy wave velocity, HARPA performs comparably to state-of-the-art associators. At excessive charges and with laterally heterogeneous or unknown wave velocity, HARPA outperforms present strategies. Our outcomes present that adaptive travel-time modeling turns into vital when seismicity is dense and counsel a route towards joint affiliation and passive-source tomography from unassociated arrivals.