Assimilation de données hydrogéophysiques par filtre de Kalman d'ensemble pour l'estimation de la teneur en eau en restauration minière

For the mining industry, acid mine drainage (AMD) is a major environmental issue occurring when tailings containing sulfide minerals are exposed to water and oxygen. Building a cover with capillary barrier effects (CCBE) is generally an efficient way to prevent AMD because the moisture-retaining layer keeps a low gas diffusion coefficient which leads to low oxygen flux from the atmosphere to the reactive wastes. The performance of such covers relies on precise control of the saturation degree in the moisture-retaining layer, which should remain above 85%. Current methods for evaluating the water content uses point measurements (capacitance or TDR sensors) and their representativity is limited since the waste storage facilities often cover several hectares. Time-lapse electrical resistivity tomography (ERT) shows promise as it can monitor changes in the electrical conductivity (EC) of soil on a large scale. EC is directly related to saturation, but it is also affected by other factors such as temperature, porosity, and electrical conductivity of pore water. Also, some well-known limitations are associated with the inversion process (smoothed and non-unique solution), often leading to the use of ERT data in a qualitative way. The interpretation of ERT data could be improved by processing simultaneously other quan- titative information (e.g. point measurements of water content, hydrogeological properties of the materials) in order to estimate water content in the CCBE over large areas with good accuracy. To this end, the project put forward the Ensemble Kalman Filter (EnkF) which is a probabilistic data assimilation approach used for nonlinear problems that have been used successfully for various hydrogeological and geophysical applications, although not widely applied in mining engineering. It uses an ensemble of realizations to estimate the state variables (e.g. saturation degree, hydrogeological properties) by comparing a theoretical pre- diction (e.g. 1D unsaturated flow model) and some measurements (e.g. moisture probes data or resistance data). Prediction and hydrogeophysical data are combined in order to provide the best estimate of the water content distribution, minimizing its uncertainty. The accuracy of the water content estimation by the EnKF algorithm has been tested with synthetic experiments. We have discussed the effect of different parameters: the ensemble size, the assimilated data resolution, and the ERT data type. In general, the water con- tent distribution was evaluated with good accuracy (mean error < 0.01 m3/m3) and low uncertainty (standard deviation 0.02 m3/m3). In cases where the assimilated data had a low sensitivity to the water content state, an estimation was still available, but with significant uncertainty (standard deviation 0.1 m3/m3). Moreover, the algorithm is fairly robust against an inaccurate theoretical model, the average error remains below 0.02 m3/m3. This perfor- mance is achieved even when the number of assimilated data is significantly reduced, for example by keeping only one water content probe in the tailings. Lastly, the algorithm was modified to allow simultaneous estimation of the water content as well as the hydrogeological properties of the materials (porosity, permeability and Van Genuchten parameters). To test the approach under realistic conditions, an experimental column was built, repro- ducing the structure of a large-scale CCBE (geometry and materials) under investigation at Canadian Malartic Mine, a gold mine located in Quebec, Canada. Water content and resistance were measured by probes and electrodes during a 3-month infiltration test. As for the synthetic cases, the assimilation of experimental hydrogeophysical data was able to estimate the water content distribution over time as well as the hydrogeological properties of tailings and waste rocks. Moreover, the water content estimated by ERT data assimilation has highlighted a bias in the petrophysical relationship since the estimated water content was lower than the water content measured by the probes. Some solutions have been proposed to improve the relationship between water content and electrical conductivity such as the estimation of petrophysical parameters by EnKF. Finally, the study of the synthetic case and the simplified experimental case allowed to demon- strate the potential of hydrogeophysical data assimilation by EnKF for the water content es- timation in tailings. Several recommendations have been made to extend the methodology to larger-scale hydrogeophysical data under field conditions. For example, taking into account variations in temperature and electrical conductivity as well as parallelization to accelerate the calculation speed.