A neural-attenuation model before Mexican extreme events

The most recent shaking experiences have demonstrated that the predictions of the seismic models are not always in agreement with the registered responses, especially in the face of extreme earthquakes. Records collected from 1960 to 2011 at a rock-like site are used to develop a neural network that permits to estimate peak ground accelerations via the magnitude, the focal depth, the site-source distance and a seismogenic zone. The neural model is applied to the 8th and 19th September 2017 events that hit Mexican territory and the obtained results show that the network is flexible enough to work appropriately to various conditions of intensity and sites-sources with remarkably predictive capacity. The neural-attenuation curves are compared with those obtained from Ground Motion Prediction Equations and their performance is assessed for events, in addition to the devastating Mexican events, from Japan, Taiwan, Chile and USA.