Joint identification of contaminant source and non-Guassian conductivities via the restart normal-score Ensemble Kalman filter

The normal-score ensemble Kalman filter (NS-EnKF) has been proven to work well for the identification of a contaminant source in an aquifer together with the estimation of the spatial variability of hydraulic conductivity, even for highly heterogeneous, non-Gaussian conductivity distributions, in synthetic aquifers. In this work, the NS-EnKF is tested with real data coming from a sandbox experiment with a binary spatial distribution of conductivities. The sandbox contains patches of high conductivity mixed with patches of low conductivity. The main difference with the synthetic cases in which the technique had been tested previously is that no piezometric head data are available, and therefore, an important amount of information regarding the spatial distribution of conductivities will be missing. After some preliminary tests, it was found that for the filter to meet the goal of jointly identifying the contaminant source parameters and the non-Gaussian conductivities, it was needed to use a large number of members in the ensemble or with inflation techniques to prevent filter inbreeding. The best results were obtained with a large ensemble size or with the Bauser’s inflation method. The results show the ability of the NS-EnKF method to jointly identify contaminant source and conductivities in a sandbox experiment.