Abstract :
[en] Alluvial plains constitute essential geological bodies for environmental studies such as contaminated sites remediation, low-enthalpy geothermal energy or groundwater resources. The heterogeneity of these deposits governs flow processes and needs to be quantified. A proper description of such complex deposits requires an integrated approach combining geological, geophysical and hydrogeological data. Solving such spatial inverse problems in the Earth Sciences remains a considerable challenge given the large number of parameters to invert for, the non-linearity of forward models and, as a result, the ill-posedness of the problem. Geostatistics is therefore needed to specify prior models, more particularly, information to control the spatial features of the inverse solutions.
Two-point geostatistical approaches have been developed to describe the heterogeneity of one geological formation but fail to reproduce the heterogeneity of fluvial deposits with multiple facies. Multiple-point statistics (MPS) introduced the training image (TI) concept to replace the variogram within an extended sequential simulation framework. The use of geophysics to constrain such simulations has been studied in the petroleum industry with wave-based methods (seismic reflection), but little research has been done to assess the use of near-surface potential methods to condition MPS in environmental studies.
In this work, we propose to integrate geological (borehole logs), geophysical (electrical resistivity tomography (ERT) profiles) and hydrogeological (hydraulic heads) data within MPS models on the alluvial plain of the Meuse River, Belgium.
Potential-based geophysical methods being integrative, they suffer from a relatively poor resolution. We first study how we can improve the informative content of geophysical inversion by including prior information in the ERT inverse problem. Three methods are tested and compared in several field cases, namely the reference model inversion, the structural inversion and the regularized geostatistical inversion. If every method has advantages and drawbacks, the best suited method for the considered problem is the regularized geostatistical method. Electromagnetic borehole logs enable to derive the vertical correlation length of electrical resistivity in the deposits and to subsequently use it to constrain the inversion. In addition to the knowledge of the bedrock position, it enables to retrieve an electrical resistivity distribution of the deposits close to direct observations. This ensures that geophysical models will be informative to constrain MPS simulations.
Given the lack of geological and sedimentological data to build accurate TIs, a data base of TIs is built using several different parameters and scenarios. They are all based on a three facies description: clay/loam, sand and gravel corresponding to low, intermediate and high hydraulic conductivity. Then, we develop a methodology to verify the consistency of independently-built TIs with geophysical data. Our methodology starts by creating subsurface models with each TI. From these models we create synthetic geophysical data and from this synthetic data, synthetic inverted models. These models are now compared with a single inverted model obtained from the field survey, allowing for our definition of what is ``consistent''. To that extent, we calculate the Euclidean distance between any two inverted models as well as field data and visualize the results in a 2D or 3D space using multidimensional scaling (MDS). With this technique, it is possible to verify if field cases fall in the distribution represented by synthetic cases, and thus are consistent with them. In a second step, we present a cluster analysis on the MDS-map to highlight which parameters are the most sensitive for the construction of TI. Based on this analysis, a probability of each geological scenario is computed through kernel smoothing of the densities in reduced projected metric space.
The integration of hydrogeological data is made through a stochastic inversion method: the probability perturbation method (PPM), using MPS constrained with geophysical data to generate models. The PPM algorithm automatically seeks solutions fitting both hydrogeological data and training-image based geostatistical constraints. Only geometrical features of the model are affected by the perturbation, i.e. we do not attempt to directly find the optimal value of hydrogeological parameters (chosen a priori), but the optimal spatial distribution of facies whose prior distribution is quantified in a training image.
Tracing experiments may be used to further constrain hydrogeological models. ERT has proven its ability to monitor salt tracer tests, but few studies have investigated its performances in thermal tracing experiments. In this study, we demonstrate the ability of surface and crosshole ERT to image quantitatively temperature changes during heat injection experiments. Such resistivity data provides important information to improve hydrogeological models. Our study proves that ERT, especially crosshole ERT, is a reliable tool to follow thermal tracing experiments. It also confirms that ERT should be included to in situ techniques to characterize heat transfer in the subsurface and to monitor geothermal resources exploitation.
