Research papersPreconditioning an ensemble Kalman filter for groundwater flow using environmental-tracer observations

- We estimate hydraulic conductivity and temporally variable groundwater recharge.
- A highly variable initial ensemble is considered within an EnKF framework.
- EnKF can well estimate temporally variable recharge if ensemble size is large.
- If the initial ensemble is preconditioned, smaller ensemble sizes can be used.
- Preconditioning using groundwater-age tracer observations.

Groundwater resources management requires operational, regional-scale groundwater models accounting for dominant spatial variability of aquifer properties and spatiotemporal variability of groundwater recharge. We test the Ensemble Kalman filter (EnKF) to estimate transient hydraulic heads and groundwater recharge, as well as the hydraulic conductivity and specific-yield distributions of a virtual phreatic aquifer. To speed up computation time, we use a coarsened spatial grid in the filter simulations, and reconstruct head measurements at observation points by a local model in the vicinity of the piezometer as part of the observation operator. We show that the EnKF can adequately estimate both the mean and spatial patterns of hydraulic conductivity when assimilating daily values of hydraulic heads from a highly variable initial sample. The filter can also estimate temporally variable recharge to a satisfactory level, as long as the ensemble size is large enough. Constraining the parameters on concentrations of groundwater-age tracers (here: tritium) and transient hydraulic-head observations cannot reasonably be done by the EnKF because the concentrations depend on the recharge history over longer times while the head observations have much shorter temporal support. We thus use a different method, the Kalman Ensemble Generator (KEG), to precondition the initial ensemble of the EnKF on the groundwater-age tracer data and time-averaged hydraulic-head values. The preconditioned initial ensemble exhibits a smaller spread as well as improved means and spatial patterns. The preconditioning improves the EnKF particularly for smaller ensemble sizes, allowing operational data assimilation with reduced computational effort. In a validation scenario of delineating groundwater protection zones, the preconditioned filter performs clearly better than the filter using the original initial ensemble.

255