Approximately 30% of the largest groundwater systems (i.e., aquifers) are being depleted due to sharp increases in groundwater pumping. However, when physics-based groundwater models are used to predict the evolution of groundwater levels, pumping behavior is often treated as an external forcing to the models. Such treatment simplifies the modeling effort by neglecting the non-stationarity of irrigation behavior, which can lead to erroneous forecasts of groundwater levels when actual pumping data is not available. To address this issue, we develop an agent-based model (ABM) of groundwater pumping using a data-driven approach; the ABM accounts for the impact of various factors on pumping behavior and is then integrated into the modeling of groundwater dynamics. As such, the coupled ABM and groundwater models can capture the two-way feedbacks between water users and groundwater systems. The primary results show that the coupled models can better describe the non-stationarity of pumping behavior and thus reduce forecast uncertainty in the prediction of groundwater levels in a long run, which allows the models to better inform groundwater management and governance.
