Predictive properties of an anisotropic ETAS Space-time model applied to Chilean seismicity

Seismic activity can be often described by a space-time ETAS (Epidemic Type Aftershock Sequences) model, which is composed of a background seismicity component (large scale) and a triggering one (small scale). Typically the large-scale component is a spatial inhomogeneous Poissonian process, whose intensity is usually estimated through non-parametric techniques: in the case of Chilean seismicity, the majority of events, with a greater magnitude, occur along the Nazca plate, due to the subduction process, so that the  anisotropic kernel estimates should better describe the background seismicity than the classical isotropic kernel estimates. Similar considerations could be made for triggered events.
In previous papers, we used the ETAS model, with the Forward Likelihood Predictive approach (FLP), with the triggered seismicity modeled with a parametric space-time function, using also some covariates together with the magnitude of the triggering events. From a statistical point of view, a forecast of triggered seismicity can be made in the days following a big event. In this work, we will explore the predictive properties of a new proposal of anisotropic ETAS model, with an extension of the semiparametric approach of etasFLP proposed by Chiodi, et al. (2021).
We used open-source software (R package etasFLP, Chiodi and Adelfio 
(2017, 2020)) to perform the semiparametric estimation of the ETAS model with covariates.