Framework

Currently under development are plans for urban expansion for the city of Hilversum. Main focus is an area east of the city. This area however over the past 100 years has been severely contaminated by several types of pollutants (i.e. sewage disposal, industrial waste disposal, etc). Knowledge about the spreading of groundwater contaminants on the other hand is scarce. Complicating factor is the relative fast groundwater flow by which as a consequence contaminants move fast and reached to great depths. In addition individual plumes are difficult to identify, since contaminant sites are located next or even on top of each other, and are known to have merged.

The spreading of groundwater contaminants (some 68 substances) are predicted by a combined triwaco groundwater and inverse solute transport model.

Groundwater flow

The primary force responsible for contaminant transport is advection, i.e. the movement of pollutants in groundwater. A groundwater model therefore should be able to accurately simulate groundwater flow. Factors governing groundwater flow in this particular case are on the one hand anisotropy and permeability and on the other hand groundwater abstraction (for drinking water) and infiltration from ponds (used in the past waste water discharge). Contamination of the site took place over a period of approximately 100 years. The groundwater abstraction rate and infiltration rate from the ponds has varied strongly over time. For this reason a transient groundwater model was uses for the simulation. The model is also capable to account for the complex hydrogeological setting of which a strongly varying anisotropy. The results of the simulated groundwater flow show that the contaminated site is in fact an infiltration area (with ponds having a hig infiltration rate up 40mm/d) and that all this water ultimately reaches the groundwater abstraction.

Inverse modelling step 1: Correlate groundwater flow with contaminated sites

The next step in the process is to correlate the simulated groundwater flow with the contaminated sites. Based on simulated groundwater flow and path line calculations (particle tracking) the influence areas of individual contaminated sites are identified wit TRIWACO-SYSAL. These calculations are carried out as part of the inverse solute transport modelling. The influence areas already provide information about the maximum spreading of contaminants from the source areas that may be expected (worst case).

Inverse modelling step 2: Determine solute transport parameters

With TRIWACO-TRACE path line calculations for every observation well (filter) is used to determine the corresponding source area. The resulting travel time and measured concentration through time is used in the inverse modelling to calculate the duration of each source, initial concentration, retardation factor and dispersion coefficient for each substance. To do so the inverse modelling translates the 3D flow model to a 1D transport model, the method was first developed at the University of Amsterdam and University of Arizona and is called Shuffled Complex Evolution Metropolis UA (SCEM-UA).

Forward modelling step 3: Calculate spreading of contaminants in 3D

Based on simulated groundwater flow and path line calculations (particle tracking) the influence areas of individual contaminated sites are identified using TRIWACO-SYSAL. Every influence area is based on a large amount of path lines for which an end point (X,Y,Z), starting point (source area), travel time and travelled distance is known. This data combined with the data from inverse modelling was used to calculate the concentration at the end point of each path line. The result is a 3 dimensional map showing the spreading of contaminants to this date.

Conclusion

The method presented here is much more efficient than usually applied methods like MT3D. For the simulated 68 substances it took some 80 hours, which includes the inverse modelling by which unknown parameters necessary for transport modelling are determined. Whereas with MT3D it would have taken approximately 840 hours to complete. Not only is the method faster it also gives better results since it uses measured values to optimise model parameters.