Sequential Estimation of Aquifer Parameters Using Water Table Elevations

Groundwater level time series contain valuable information about aquifer storage properties and the external source–sink terms to which aquifer systems are subjected. However, estimating these source–sink terms from water table fluctuations remains challenging due to the difficulty of quantifying specific yield at the field scale and uncertainty associated with it. Strong correlations among recharge, pumping, and specific yield further complicate inference; without explicit uncertainty quantification, water table fluctuation-based models are prone to parameter non-uniqueness and substantial uncertainty propagation. Moreover, such models are commonly calibrated by estimating all parameters simultaneously, and the relative performance of simultaneous versus sequential parameter estimation for reliable parameter recovery has not been systematically investigated in such models. In this study, we develop an inverse modelling framework to quantify parameter uncertainty and correlation in water table-based groundwater models and to assess whether sequential, iterative parameter estimation yields more reliable recovery of aquifer parameters than conventional simultaneous estimation. The methodology is first tested using synthetic water table fluctuations generated with known model parameters, enabling direct comparison between simultaneous and sequential estimation strategies. Simultaneous estimation exhibits strong equifinality and large errors, particularly in recharge and inferred pumping (up to 93%), whereas the sequential approach markedly improves parameter identifiability and recovers recharge, pumping, and specific yield with errors below 12%. The framework is subsequently applied to groundwater-level time series from 73 field borewells, demonstrating its ability to recover physically plausible aquifer parameters and groundwater forcings. The results further show that the framework can diagnose model structural errors, particularly those arising from neglecting groundwater abstraction when it is present. Leveraging 46.5 million simulations executed using parallel processing, the approach efficiently addresses key uncertainties in water table fluctuation–based models and provides a practical means of inferring aquifer parameters and forcings directly from groundwater observations.