An attempt to calibrate a density-dependent groundwater flow model for a high energy subterranean estuary using particle swarm optimization and integrating salinity-, temperature- and 3H/He age observations

Subterranean estuaries below high-energy beaches are understudied, despite being potential powerful biogeochemical reactors at the land/sea transition zone affecting the quality of coastal waters. Highly transient hydro(geo)logical boundary conditions and density-effects lead to dynamic subsurface flow and transport patterns which are difficult to understand and hard to replicate by models. A comprehensive and unique 1-year dataset of hydraulic heads, salinity and temperature data in combination with apparent ³H/He ages was obtained at a beach research site on Spiekeroog Island in North Germany. The site includes 3 multilevel groundwater monitoring wells and a vertical electrode chain with 10 temperature sensors, all positioned on a transect aligned along the principal cross-shore flow direction and all reaching down to 24 m depth below ground surface. The data-set was used to set up and calibrate a site-specific groundwater flow and transport model, aiming to approximate the highly dynamic groundwater flow patterns on that transect. The simulation time needed to be 20 years because of the long model spin-up. Due to the complex and nonlinear nature of the system, model calibration was carried out via particle swarm optimization, which is superior to gradient-based optimization techniques with respect to finding a global minimum of the objective function. The calibration results were reasonable. The dynamics of hydraulic head data were well captured, however, simulated values were constantly higher than those observed. The observed salinities were best captured for the multilevel wells near the mean high water and low water line. At the highest multilevel well located at the upper beach right at the dune base, simulations matched observations less well. Similarly, groundwater temperatures and ages were best replicated at the location in the infiltration zone near the high-water line. Groundwater ages and their temporal dynamics at the dune base and mean low water line could only be replicated down to 12 m depth. Deviations between simulations and observations are likely due to 3D flow effects in longshore direction, which could not be captured with the 2D vertical cross-sectional model approach. However, long model run times hindered calibration of a full-blown 3D density-dependent, 20-year long-term groundwater flow and transport model. The next step is to estimating the importance of longshore hydraulic gradients. Finally, the model will be extended for hydrobiogeochemical reactions to assist in the analysis and understanding of the observed hydrochemical data at this site.