Time-lapse ensemble-based electrical resistivity tomography to monitor water flow from managed aquifer recharge operations

• The effectiveness of ERT for monitoring recharge processes via drywells is demonstrated. • The wetting front was imaged until reaching the main aquifer at depths greater than 70 m. • Salt flushing and a perched layer are captured by both ERT and point sensor data. • Resistivity model ensembles provide stable model differences across time. • Model ensembles are used to provide uncertainty estimates. Various managed aquifer recharge strategies, such as drywells, are being used in the California Central Valley (CCV) to replenish groundwater resources that have been depleted by over-pumping, especially during droughts. Drywell technology allows recharge water to bypass shallow impermeable layers and possible contaminated soils near the land surface. Understanding water flow in the vadose zone is crucial for assessing the performance of drywells regarding the amount of water that reaches the groundwater table and the fate of solutes. In this study, we demonstrate the applicability of time-lapse electrical resistivity tomography (TL-ERT) for imaging the water flow and subsequent aquifer recharge at a drywell site in the CCV with a thick (67–72 m) vadose zone. Additionally, TL-ERT results were compared to point-scale observations from a collocated monitoring well. To invert our TL-ERT data sets, geostatistical constraints were applied to favor layered models as expected due to the alluvial deposits in the study area. By considering different correlation lengths, an ensemble of resistivity model solutions was generated per time-step instead of a single model solution (as typically performed). Model differences between the mean model of the baseline data set and the models from the subsequent time steps allowed us to image the wetting front development until reaching the regional aquifer, a perched water table, and flush of salts that were otherwise not visible or blurred when using single model solutions from standard deterministic TL-ERT inversion approaches.