Speaker
Description
The feasibility of muography studies depends on accurate estimates of the atmospheric muon flux at the site of interest, as well as on its interaction with the target structure and the detection system. However, many existing frameworks rely on simplified parameterizations restricted to specific angular and energy ranges, which do not include geographic and geomagnetic effects. This limitation hinders realistic exposure-time estimates for detecting density anomalies in geological targets.
To address this problem, we present \textit{ARTI-muon}, a modular framework for end-to-end muography simulations. In our approach, the surface-level muon flux is obtained from ARTI-based simulations, which incorporate geographic and geomagnetic effects in the secondary-muon flux. In addition, we implement a conditional normalizing flow model to reduce the computational cost associated with repeated flux simulations. Trained on a dataset of ARTI simulations, this generative surrogate reproduces the angular and energy spectra of muons, including at sites not used during training, with a reduced Poisson deviance (D_\nu<0.2) relative to the reference. It also reduces computation time from hours (5--9\,h per run, equivalent to one hour of ARTI-simulated flux) to approximately 2 minutes per scenario.
Unlike conventional parametric approaches, our framework integrates, within a single pipeline, a site-dependent stochastic flux, transport through the object of study (Python stage based CSDA, and a backward/adjoint propagation), and the detector response (detailed MEIGA/Geant4 simulation). This approach achieves substantially lower computational cost through the use of generative AI models. Finally, we illustrate the applicability of the framework with case studies at Cerro Mach\'{\i}n (Colombia) and Mt.\ Etna (Italy). The methodology generalizes to any location defined by its latitude/longitude, target geometry, and detector configuration.