The Pacific Northwest (PNW) has substantial earthquake risk, partly due to crustal faults that lie under population centers such as Seattle. Models of earthquake rates are key to probabilistic seismic hazard assessment and aftershock forecasts, which support hazard mitigation and disaster response. The Epidemic-Type Aftershock Sequence model (ETAS) is a spatiotemporal point process model which parameterizes the rates of earthquakes and aftershocks within a seismic region, using a catalog of its past earthquakes. Estimates of ETAS parameters are directly used in planning the response to aftershock sequences. Typically, maximum likelihood estimation (MLE) is used to fit ETAS to an earthquake catalog; however, the ETAS likelihood suffers from flatness near its optima, parameter correlation and numerical instability. We present a spatiotemporal version of a Bayesian procedure to estimate ETAS parameters, such that uncertainty estimates can be ! resolved. The procedure is conditional on knowing which earthquakes triggered which aftershocks; this latent structure and the ETAS parameters are estimated stepwise, similar to the expectation-maximization algorithm. This procedure can also account for measurement error in observed earthquake magnitudes and locations. A PNW catalog is merged from three existing catalogs, using new automated procedures for duplicate detection and identification of earthquake swarms, which violate the time-stationarity assumed by the ETAS model. We present Bayesian ETAS results for simulated catalogs and the PNW catalog.

Seismicity rate estimates and the earthquake forecasts they yield vary spatially and are usually represented as heat maps. While visualization literature suggests that displaying forecast uncertainty improves understanding in users of forecast maps, research on uncertainty visualization (UV) is missing from earthquake science. We present a pre-registered online experiment to test the effectiveness of three UV techniques for displaying aftershock forecasts. Human participants complete map-reading and prediction tasks using an aftershock forecast displayed with its uncertainty. Task performance under the UVs is compared to a condition where no uncertainty is visualized (the current practice).