Time-lapse ERT monitoring and measures of model reliability for an injection/withdrawal experiment in a near-surface aquifer

To quantify performance of 3D time-lapse electrical resistivity tomography (ERT), a sequential injection/withdrawal experiment was designed for geophysical monitoring of the pump- and-capture remediation of a conductive solute in an unconfined alluvial aquifer. Borehole ERT data were collected over a period of several days and inverted for 3D electrical conductivity using both independent and time-lapse regularization techniques. The ERT-estimated electrical conductivity can be used to predict solute concentration and solute mass in the aquifer over time via post-inversion parameter differencing. At any experimental stage, the total solute mass in the aquifer can be predicted with a maximum accuracy of 60-85% depending on regularization protocol and survey geometry. However, sequential ERT models can also be used to estimate the withdrawn solute mass for every experimental stage (the change in mass between experimental stages). Withdrawn mass estimates are more reliable than total mass estimates, do not require background data, and do not exhibit systematic under-prediction or dependence on the regularization protocol. Withdrawn mass predictions are accurate provided that changes in mass are not too large, where how large is too large is data- and model-dependent. Measures of model relibilty including prediction accuracy, the volume of investigation, and the point-spread function can provide valuable insight into the performance of ERT and the applicabilty of time-lapse differencing.