Carbonate aquifers are often assumed to behave like porous media. A justification given is that karst features may be present, but it is suggested that they are of only local scale and may be ignored at the scale of the whole aquifer. There are copious data on the carbonate aquifer in the Mammoth Cave area, which make it an ideal place to test this equivalent to porous medium (EPM) hypothesis.

A numerical simulation of flow in the aquifer was performed using the popular MODFLOW code. The first realization assumes that the aquifer behaves as a homogeneous EPM. Model calibration gives a hydraulic conductivity (K) of 1.1 x 10^-3 m/s, with a mean absolute error of 12 m for heads. The model behaves poorly in several key aspects: discharge at the major spring is underestimated by 94% and tracer velocities are underestimated by two to three orders of magnitude. The second realization assumes that there are tributary conduit networks discharging at springs. The conduits are modeled as lines of high K cells. The calibrated non-conduit K of 4 x 10^-5 m/s is midway between slug test and pumping test values, and conduit K ranges over four orders of magnitude. There is a good fit to water level data in both wells and cave streams and discharge to springs is accurately simulated. However, it is not possible to accurately model groundwater velocities.

The problem with using the EPM hypothesis for this aquifer is that the assumption that karst features are only of local scale is incorrect. Instead, an interconnected network of high-permeability channels extends over the scale of the aquifer, which consequently does not behave as an EPM.