In this study, processes occurring during hypothetical biological in situ remediation measures in heterogeneous aquifers are investigated. The key question is whether or not it is possible to predict the success of the remediation in advance based upon tracer test data.

Monte Carlo simulations of reactive transport are performed. The conductivities on the rectangular grid of the two-dimensional model domain are assumed to be a stochastic variable. The model domain is initially saturated with organic pollutant to a certain level. Then oxygen is introduced from the left-hand side. The reaction model comprises oxygen, organic carbon, and microorganisms. They are coupled via Monod kinetics and stoichiometry. The organic carbon can sorb linearly on the matrix. Transport of a non-reactive tracer is also calculated in order to distinguish purely physical transport effects from biochemical ones.

For the tracer, heterogeneities in the conductivity field result in macrodispersion. This leads to a flattened breakthrough curve (measured in the remediation wells) of a heterogeneous realization as compared with the homogeneous case. The heterogeneity effects are more complex for the reactive system: Degradation only takes place if both oxygen and carbon are simultaneously present in one node. Mineralization of carbon is fast in the more permeable zones of the aquifer and oxygen supply is limited in less permeable zones. Consequently, it takes much longer to clean up a contaminated heterogeneous aquifer than a homogeneous one. In a homogeneous aquifer, the oxygen breakthrough curve is steeper than the one of the tracer. This can be explained as the effect of a self-sharpening front: The simultaneous presence of oxygen and organic carbon leads to microbially catalyzed mineralization of the contaminant. Therefore, the oxygen concentration in one node is low until all of the contaminant is removed. After that, the oxygen concentration rises quickly at that point.

Two different homogeneous upscaled models are studied as equivalent homogeneous media. The first approach is the biofilm model: An immobile biophase is introduced which is coupled with the mobile pore water by a linear diffusive exchange term. Mineralization of carbon only takes place in this biofilm. Its volumetric fraction is set to be very small compared with the mobile porosity. Therefore, the exchange process does not influence transport of the non-reactive tracer. It does not "see" the biofilm. In the first step of the fit procedure, the breakthrough curve of the tracer can be reproduced by adjusting the longitudinal dispersivity in an advection/dispersion formulation. This does not hold true for the reactive components such as oxygen, since an increased dispersivity leads to fast spatial spreading of oxygen and, consequently, degradation of contaminant. In order to restore the realistic, limited contact between oxidant and substrate, the exchange between pore water and biofilm is limited. In a second step of the fit procedure, the exchange coefficient is varied until the homogeneous oxygen breakthrough curve matches the heterogeneous one. The homogeneous medium constructed in this manner alos shows he same behavior as the heterogeneous realization with respect to the mass reduction of the contaminant. As the exchange coefficient cannot be determined from the tracer breakthrough curve, it is impossible to predict the remediation time on the basis of a tracer test. By fitting the oxygen breakthrough curve, it is feasible to simulate the contaminant mass reduction. However, this can be done only during the remediation.

As it is impossible in the biofilm model to deduce the effective parameter for the reactive system (i.e. the exchange coefficient) from the tracer breakthrough curve, a dual-porosity model was investigated as a second approach to a homogeneous medium: In this model, a mobile and an immobile fraction of pore water and matrix is defined. It is assumed that microbial degradation takes place in both the mobile and the immobile phases, which are no longer interpreted in terms of "pore water" and "biofilm" but rather as simplifications of the more and less permeable regions of the heterogeneous aquifer. There are three effective parameters, macrodispersivity, exchange coefficient between mobile and immobile regions, and mobile porosity. In this model, both tracer and reactive components undergo the exchange between mobile and immobile pore water. All three parameters are obtained by fitting a homogeneous tracer breakthrough curve to a heterogeneous one. It turns out that these parameters, when transferred to the reactive system, lead to a significant underestimation of the actual remediation time. Although the tracer "sees" the immobile phase, these parameters cannot be used for the reactive system. Hence, with this model it is not possible either to predict the remediation time on the basis of tracer test data.

Even by fitting the curve of the contaminant mass reduction of the homogeneous medium to the one of the heterogeneous case using the same effective parameters as mentioned above, these curves cannot be brought to the same degree of agreement as it is possible by using the biofilm model. Furthermore, the oxygen breakthrough curves show a completely different behavior in the homogeneous and in the heterogeneous case, respectively. This observation is valid in the case of tracer breakthrough curves as basis of the fit as well as in the case of fitting the contaminant mass. This leads to the conclusion that the biofilm model is more appropriate to simulate the biochemical processes in a heterogeneous aquifer than the dual-porosity model. The sole advantage of the latter model is the fact that the heterogeneous tracer breakthrough curves can be reproduced almost perfectly by homogeneous media.

This thesis concludes with other methods to predicting the success of a planned bioremediation measure based upon geostatistics and other interpretations of the tracer breakthrough curves. The most promising approach seems to be the method of evaluating tracer arrival time quantiles. However, other measurements have to be taken into account to make reliable predictions on the behavior of the reactive system.