Leveraging Graph Neural Networks for Explainable and Adaptive Seismic Risk Modelling (Work in Progress))

Seismic risk modeling predicts the likelihood of earthquake occurrences in different regions over time. This presents a distinctive challenge as both the spatial structure of underlying geological features and the temporal evolution of seismic events must be accounted for accurately. In addition, for planners and decision-makers, it is essential that such models are not only accurate but also explainable. However, most existing models lack explainability, which limits their usefulness. This research proposes a novel graph neural network based seismic risk modeling methodology that is both explainable as well as adaptable, as it allows decision-makers the flexibility to choose the risk mapping region size as well as gain insights into the factors associated with a certain risk prediction.