| Abstract |
This paper focuses on the quantitative interpretation of the existing seismic lines of the Canton of Geneva in the framework of the Innosuisse funded project GECOS (Geothermal Energy Chance of Success). The goal of GECOS is to reduce the costs and the exploration risk for geothermal exploration by integrating high-resolution data acquisitions such as gravity, S-waves reflection seismic and 3D DAS VSP (Distributed Acoustic Sensing Vertical Seismic Profiling) as well as vintage seismic data. The main geological challenges in geothermal exploration in the Geneva area are the characterization of the lithological heterogeneities and the fault zones affecting potential geothermal targets in the Quaternary sediments, Tertiary Molasse sequence and the Mesozoic Units. The subsurface risk and uncertainty quantification involve, for example, the characterization of fractures (fault), the facies interpretation and the evaluation of geomechanical parameters like brittleness, Poisson's ratio and Young's modulus from seismic data and well logs. Artificial intelligence and in particular machine learning (ML) are promising techniques which popularity is growing exponentially and have proven to be very useful for big data assimilation. In fact, standard techniques are limited to integrate few variables. However, ML algorithms perform very well with hundreds to thousands of variables. The aim of this work is to demonstrate how these quantitative techniques can be applied to fault detection, seismic facies interpretation, and to automatically identify lithofacies based on well-log measurements. Results show that these techniques are an additional tool that could help improve the knowledge and characterize a geological reservoir, thus reduce the subsurface uncertainty, if and only if they are applied together with a domain specialist such as experimented geologists and geophysicists. |