Assessment of rapid earthquake source characterizations for local tsunami forecasting along the Cascadia subduction zone

For many environments, tsunami inundation can occur within 20 minutes, before the tsunami is directly observed at open-ocean gauges. The result is that the earliest tsunami forecasts lack complete data on the earthquake source and potential tsunami threat. This creates a gap in forecasting ability, where regional and far-field distanced communities have warnings with adequate information and time to act, while the near-field, which often sees the greatest threat, is left vulnerable. Currently, operational forecasts rely on seismic instrumentation, which often underestimate large magnitude earthquakes when using near-field and rapidly acquired data. Global Navigation Satellite System (GNSS) networks, which do not saturate in the near-field, can help. Displacement data, relayed at a high rate (1Hz) and in real-time have driven a number of recently developed source modeling codes. These codes estimate magnitude and fault extent within minutes of local M7+ events. The linkage of these outputs fully optimized, non-linear tsunami propagation codes can lead to estimates of expected amplitudes along the near-source coastline for forecasting purposes. One such module, Geodetic First Approximation of Size and Time (G-FAST) can produce an initial source estimate within 3 to 4 minutes and has been tested in simulated real-time with past events including the 2010 Maule and 2016 Kaikoura earthquakes. This estimate of the earthquake source can then be used to determine tsunami hazards rapidly at near-field coastlines. However, to assess its viability for local forecasting, a larger dataset of events is needed. Here we show results from testing of G-FAST for local tsunami forecasts using a large synthetic dataset (n > 1000) of earthquake ruptures focusing on the Cascadia Subduction Zone. We assess the accuracy of forecasted coastal tsunami amplitudes and arrival times for a range of M7+ thrust events, leveraging a network of GNSS stations located in the Pacific Northwest. We also determine the potential timing that forecasts would be available and how that can assist in localized warnings.