We investigate the effect of extended faulting processes and heterogeneous wave propagation on the early warning system capability to predict the peak ground velocity (PGV) from moderate to large earthquakes occurring in the southern Apennines (Italy). Simulated time histories at the early warning network have been used to retrieve early estimates of source parameters and to predict the PGV, following an evolutionary, probabilistic approach. The system performance is measured through the Effective Lead‐Time (ELT), i.e., the time interval between the arrival of the first S‐wave and the time at which the probability to observe the true PGV value within one standard deviation becomes stationary, and the Probability of Prediction Error (PPE), which provides a measure of PGV prediction error. The regional maps of ELT and PPE show a significant variability around the fault up to large distances, thus indicating that the system's capability to accurately predict the observed peak ground motion strongly depends on distance and azimuth from the fault.

The last strong earthquake that occurred in the southern Apennines, the Irpinia earthquake on 23 November 1980 (M 6.9), was characterized by a complex rupture mechanism that ruptured three different faults (Bernard and Zollo 1989). This earthquake was well studied, and the quantity of data available has allowed a very detailed definition of the geometry and mechanisms of faults activated during this seismic event (Westaway and Jackson 1987; Pantosti and Valensise 1990).

When accompanied by appropriate training and preparedness of a population, Earthquake Early Warning Systems (EEWS) are effective and viable tools for the real‐time reduction of societal exposure to seismic events in metropolitan areas. The Italian Accelerometric Network, RAN, which consists of about 500 stations installed over all the active seismic zones, as well as many cities and strategic infrastructures in Italy, has the potential to serve as a nationwide early warning system. In this work, we present a feasibility study for a nationwide EEWS in Italy obtained by the integration of the RAN and the software platform PRobabilistic and Evolutionary early warning SysTem (PRESTo). The performance of the RAN‐PRESTo EEWS is first assessed by testing it on real strong motion recordings of 40 of the largest earthquakes that have occurred during the last 10 years in Italy. Furthermore, we extend the analysis to regions that did not experience earthquakes by considering a nationwide grid of synthetic sources capable of generating Gutenberg‐Richter sequences corresponding to the one adopted by the seismic hazard map of the Italian territory. Our results indicate that the RAN‐PRESTo EEWS could theoretically provide for higher seismic hazard areas reliable alert messages within about 5 to 10 s and maximum lead times of about 25 s. In case of large events (M > 6.5), this amount of lead time would be sufficient for taking basic protective measures (e.g., duck and cover, move away from windows or equipment) in tens to hundreds of municipalities affected by large ground shaking.

Earthquake early warning systems (EEWSs) are nowadays contributing to seismic risk mitigation actions, both in terms of losses and societal resilience, by issuing an alert promptly after the earthquake origin and before the ground‐shaking impacts the target to be protected. In this work, we analyze the performance of network‐based and stand‐alone (on‐site) early warning systems during the 2016–2017 central Italy sequence, characterized by events with magnitude as large as 6.5. For the largest magnitude event, both systems predict well the ground shaking nearby the event source, with a rate of success in the 85%–90% range, within the potential earthquake damage zone. However, the lead time, that is, the time available for security actions, is significantly larger for the network‐based system. For the regional system, it increases to more than 10 s at 40 km from the event epicenter. The stand‐alone system performs better in the near‐source region, still showing a positive albeit small lead time ($<2s$). Far away from the source ($>60km$), the performances slightly degrade, mostly owing to the large uncertainty in the attenuation relationships. This study opens up the possibility for making an operational EEWS in Italy, based on the available acceleration networks, provided that the delay due to data telemetry has to be reduced.