Realtime Risk: An Automated Real-Time Global Earthquake Rate Forecasting Tool

Abstract:

Aftershocks following a large earthquake occur not only on the fault rupture surfaces, but also in regions well off the fault that is best explained by Coulomb stress transfer. Since active faults can be located in these off-fault areas, an equal or larger event becomes possible after a mainshock, as occurred in the 2016 M 6.0 and M 7.0 Kumamoto sequence, 28 hr apart, and the 2019 M 6.4 and M 7.1 Ridgecrest sequences, 34 hr apart. Even moderate off-fault aftershocks cause significant damage when they occur in populated areas. These paired quakes, just days apart, remind us that the risk is greatest immediately after a mainshock, and so an automated real-time evaluation is needed.

‘Realtime Risk (RtR)’ forecasts the seismicity rate in space and time in the framework of rate/state Coulomb stress transfer. RtR is intended to evaluate large worldwide events so that the method can be rapidly tested and improved. Unlike Epidemic Type Aftershock Sequence (ETAS) models, RtR forecasts not only sites of seismicity rate increase, but also rate drops, which are also widely observed after large, well instrumented earthquakes.

To rapidly compute stress transfer after a large event, RtR first assumes a source fault model using the nearest fault plane solution of Kagan and Jackson (KJ, 2014), then switches to any rapid CMT solution, and then switching to the ‘finite fault’ model issued by USGS when as soon as it becomes available. Rather than using idealized mapped fault surfaces or optimal planes, RtR resolves Coulomb stress onto nodal planes of background gCMT focal mechanisms or KJ gridded solutions, utilizing local catalogs with more numerous focal mechanisms where available. RtR uses a ‘learning period’ to seek the best rate/state parameters with observed aftershocks, and also employs spatially variable b-values since larger events are forecast from smaller ones. Our retrospective test for the 2019 Ridgecrest sequence yields a spatial regression coefficient of 0.71 and a slope of 0.81 (Toda and Stein, BSSA, 2020).

We have carried out annual forecasts in Japan, Taiwan, Mexico, New Zealand, and California. By expanding RtR to forecast strong ground motion, we believe the method could provide the public with guidance on post-quake decisions and vigilance.