Modelling groundwater systems often involves high uncertainty in the conceptual assumptions made and the forcing inputs or parameterization used. This greatly affects the spatial and quantitative accuracy of model products, particularly for remote areas. In shallow water table systems, the groundwater dynamics are influenced by soil evapotranspiration (ET) and recharge to the water table (WT). The assimilation of remotely sensed evapotranspiration values has the potential to reduce the spatial uncertainty related to the groundwater dynamics.
We present a framework for assimilating ET data into a coupled unsaturated-saturated model through the Ensemble Kalman Filter. The model coupling, realized through the net-recharge variable, allows for the direct update of the state variables of heads and soil moisture from the assimilation of the diagnostic variable ET. The method is applied to a losing stream system in the south-east of South-Australia. The improvements are evaluated for different conditions of WT - vegetation interaction in areas of recharge (deep WT) and areas with high root extraction (shallow WT).
The filter showed improvements of the model states over all the domain when the ensemble spread was adequately generated. As hypothesised, best results for the states variables were found for the extraction area. However, the greatest error reduction was seen for the recharge areas, suggesting that the benefits of the assimilation are not reduced only to shallow water table areas. Consistent improvements were also obtained in the calculation of modelled fluxes of actual ET and net-recharge. Because net-recharge is the linking variable between the two models, the error reduction of it translates into improved modelling of the entire groundwater system.
This indicates that the assimilation of ET is suitable for reducing uncertainty in large scale groundwater simulation.