[M-GI25] Near Surface Investigation and Modeling for Groundwater Resources Assessment and Conservation

[MGI25-P02] Unconsolidated deposits heterogeneity modeling by combining stochastic geological models and joint inversion

*Ludovic Schorpp¹、Alexis Neven¹、Philippe Renard^1,2、Julien Straubhaar¹ (1.Center for hydrogeology and geothermics, Neuchâtel, Switzerland、2.Department of Geosciences, University of Oslo, Oslo, Norway)

キーワード：Stochastic modeling, Joint inversion, Quaternary modeling, ES-MDA

The heterogeneity characterization of Quaternary deposits and its associated uncertainty is of critical importance in various situations such as groundwater resource management or underground geotechnical projects. Such formations often result from complex intertwined processes (e.g. glacial or fluvial) at different spatial and temporal scales, which lead to a high degree of heterogeneity. Their modeling requires to consider complex hierarchical relations between geological units. For this purpose, a recent open-source software, ArchPy, was developed to generate easily complex hierarchical stochastic geological models. Using this tool, boreholes and conceptual knowledge of an area can be integrated, and a large number of plausible and realistic geological models can be easily generated. However, these models are generally not sufficient and can be improved by data assimilation and inverse procedures. Indeed, multiple data types are generally often available on a site, such as geophysical soundings or groundwater level measurements. The final models must be consistent with these data while maintaining a realistic geology. In this study we propose to use the ensemble smoother with multiple data assimilation (ES-MDA) algorithm and apply it to complex geological models generated with ArchPy. Stochastic joint inversions are very time and resource consuming since numerous physical simulations are required. To overcome this, we rely on the use of hierarchical hyperparameters, where the hierarchy of the geological models is used as a way to diminish the number of parameters and the complexity of the forwards, without cost in terms of model complexity.

To prove the feasibility of our methodology, we applied it to the Upper Aare Valley, Bern, Switzerland, where large datasets have already been acquired for decades (boreholes, groundwater and geophysical measurements). The presented workflow produces a realistic ensemble of models compatible with conceptual geological knowledge, boreholes, geophysical and hydrological measurements. It is able to robustly estimate the associated uncertainty, while being greatly automated and data driven. This methodology could be easily applied to a wide variety of hydrogeological problems, and significantly improve decision making in Quaternary environments.