Feasibility of a National Earthquake Early Warning System for New Zealand: The Propagation of Local Undamped Motion Algorithm Evaluated on a Physics‐Based Synthetic Earthquake Catalogue

Earthquake early warning (EEW) can reduce injuries and damage by issuing alerts before incoming shaking. We evaluate the Propagation of Local Undamped Motion (PLUM) algorithm as a candidate for a national EEW system for Aotearoa New Zealand (NZ). We develop and demonstrate a novel methodology for evaluating EEW performance under a wide range of future earthquake scenarios using a synthetic earthquake catalogue. We used a 220,000‐year‐long physics‐based synthetic earthquake catalogue of NZ and generated synthetic seismograms using deterministic ground‐motion simulations. The algorithm was tested in the Canterbury and Wellington regions for alert correctness and timeliness, the effects of different warning radii and robustness to sensor/telemetry failures. The PLUM algorithm works well in highly populated areas of Canterbury, but not as effectively in some southern parts of Wellington City, due to the lack of stations in Cook Strait to provide warnings for earthquakes centred south of Wellington. Performance is also weaker in rural areas of both regions because the station density is too low. In Canterbury, PLUM provided correct timely alerts (warning time > 0 s) for more than 70% of moderate shaking events (probability of correct timely alert; PrCTA ≥ 70%) over 100‐year intervals in 88% of the high‐population areas. All grid cells with PrCTA < 50% were in low‐population areas. In Wellington, 89% of grid cells with PrCTA < 50% were in low‐population areas, and two‐thirds of high‐population grid cells achieved PrCTA ≥ 70%. Results suggest an appropriate choice of PLUM warning radius for the current NZ network would be 40–60 km, wider than the original 30 km radius designed for Japan. The algorithm performs well in Canterbury, but Wellington may require further enhancement to the sensor network. The study also demonstrates the robustness of the algorithm under simulated sensor or communication breakdowns, particularly in regions with high seismic instrument density.