ABSTRACT

In 2009, the global Collaboratory for the Study of Earthquake Predictability (CSEP) launched three experiments to forecast the distribution of earthquakes in Italy in the subsequent 5 yrs. CSEP solicited forecasts for seismicity tomorrow, in the next three months, and for the entire 5 yrs. In those 5 yrs, the Istituto Nazionale di Geofisica e Vulcanologia (INGV) recorded 83 target earthquakes with local magnitude 3.95M<4.95, and 14 larger shocks. The results show that 1‐day forecasts are consistent with the number and magnitudes of the target earthquakes, and one version of the epidemic‐type aftershock sequence (ETAS) model is also consistent with the spatial distribution; ensemble forecasts, which we created for the 1‐day experiment, are consistent with the number, locations, and magnitudes of the target earthquakes, and they perform as well as the best model; none of the 3‐month time‐independent models produce consistent forecasts; the best 5‐yr models account for the fault distribution and the historical seismicity; and 5‐yr models based on instrumental seismicity and b‐value spatial variation show poor forecasting performance.

You do not currently have access to this article.