An integrated multi-physics Machine Learning approach for exploration risk mitigation

In this paper, we combine pre-stack depth migration of seismic data, cooperative modelling of Controlled Source Electromagnetic (CSEM) and gravity data, and constrained inversion of CSEM data, with Machine Learning (ML) classification approaches. Our objective is to obtain probability maps of hydrocarbon distribution aimed at mitigating the exploration risk and supporting the process of appraisal of hydrocarbon fields. We introduce a novel workflow divided into two linked branches: one consists of an iterative loop of modelling and inversion steps driving towards a multi-parametric Earth model; the other path of the workflow goes through the application of advanced statistical tools and takes the benefits of automatic learning and classification algorithms. These allow us combining the entire set of heterogeneous data/models into probabilistic maps of oil distribution at target depth. We applied our methodology to a complex data set in the Norway offshore, obtaining encouraging results.