Introducing Groundwater Dynamics into the ECLand Land Surface Model: Implementation and Effects

Land surface models still lack a realistic representation of groundwater, often relying on a free drainage condition at the bottom of the unsaturated soil column as in the current version of ECLand. This unrealistic assumption places the groundwater infinite depth below the surface, thus limiting the model’s ability to simulate realistic soil–vegetation-groundwater interaction.

To address this limitation, we implemented a Dirichlet boundary condition at the bottom of the unsaturated soil to enable a fully implicit numerical scheme for coupling with groundwater. First, we prescribed the water table depth (WTD) using global scale estimates to allow for the computation of realistic water fluxes between the unsaturated zone and the underlying aquifer. In a second step,  a dynamic WTD (hereafter the DYN configuration) was  developed by defining the water stored in the  unconfined aquifer, which evolves prognostically according to drainage (groundwater recharge) and subsurface runoff (groundwater discharge).

The effects of these developments were preliminarily evaluated through offline land-only simulations forced by station data from the PLUMBER2 project, which includes observational networks such as FLUXNET2015, La Thuile, and OzFlux. We validated the DYN configuration against the model setup with free-drainage conditions (CTRL). Our results show a systematic improvement in both latent and sensible heat fluxes, as quantified by the reductions in the error metrics  across most stations, with runoff scoring the best performances. 

The results of the global simulations largely corroborate and expand upon those of the station-based evaluation experiments conducted using PLUMBER2. The DYN configuration provides a more accurate representation of WTD, both spatially and temporally. This is evident in global climatological maps and independent observational datasets. Additionally, latent and sensible heat fluxes are consistently better represented in DYN than in CTRL, showing closer agreement with DOLCE and GLEAM products. Improvements are also evident in runoff simulations, with DYN exhibiting greater consistency with GLOFAS observations. Model performance was further evaluated against multiple observational datasets, such as GRACE/GRACE-FO to verify temporal variability in total water storage and to assess long-term mean conditions.

This work demonstrates that incorporating  groundwater dynamics significantly improves the realism of land-surface processes, particularly in the representation of the flux exchange of water and energy with other components. These results provide a foundation for the enhancement of the representation of land-climate interactions and hydroclimatological behaviour in next generation of reanalysis and climate predictions.