Stochastic multi-fidelity joint hydrogeophysical inversion of consistent geological models

In Quaternary deposits, the characterization of subsurface heterogeneity and its associated uncertainty is critical when dealing with groundwater resource management. The combination different data types through joint inversion has proven to be an effective way reduce final model uncertainty. Moreover, it allows in agreement a wider spectrum available on site. However, integrating them stochastically very time-consuming expensive, due important number physical simulations needed. use multi-fidelity models, by combining low-fidelity inexpensive less accurate models high-fidelity expensive one time needed for converge. This multiscale logic can applied generation models. Most sedimentological considered as geological units (large scale), populated facies (medium finally completed parameters (small scale). this paper, both approaches are combined. A simple fast time-domain EM 1D geophysical direct problem used first constrain simplified stochastic geologically consistent model, where each stratigraphic unit homogeneous terms parameters. ensemble smoother multiple assimilation (ES-MDA) algorithm generating plausible realizations. Fast identification large-scale structures main point step. Once generated, transient flow incorporated. heterogeneous their parameter distribution. ES-MDA also task directly inferring property values (hydraulic conductivity resistivity) from generated model. To preserve consistency, hydrogeological inverted jointly. workflow ensures that therefore subject artifacts localized poor-quality data. It able robustly estimate Finally, simplification geology during part inversion, greatly reduces required converge complex while preserving consistency.