Concep tual M odels for Aquitards Neither hydrogeochemical (Figure 7) nor hydraulic head data (Figure 5) in the Maquoketa aquitard are consistent with the conventional conceptual model that flow is predominantly downward in aquitards overlying heavily pumped aquifers (Remenda et al. 1996; E aton and Bradbury 2003). In the traditional conceptual model, vertical flow under steady-state conditions is associated with a linear head decrease with depth, and under transient conditions, a monotonic head decrease with depth, both characteristic of uniformly downward gradients. Instead, we find that flow between monitoring intervals is bidirectional, and there is an abrupt decline in head across the base of the formation at both field sites (Figure 5), indicating nonequilibrium and a very low effective vertical hydraulic conductivity (E aton and Bradbury 2003). High-transmissivity channels (Tsang and Neretnieks 1998) within fracture planes greatly increase the effective hydraulic conductivity in a porous medium depending on fracture orientation, extent, and especially connectivity (Robinson 1984 ; L ong and Witherspoon 1985). In the P ierre Shale, through-going vertical fractures spaced at kilometer intervals may explain the discrepancy between effective vertical hydraulic conductivity values estimated from small-scale measurements on rock core and relatively higher values based on results of regional flow modeling (Bredehoeft et al. 1983; Neuzil 1986, 1994 ). A similar discrepancy in estimates from field data (this paper) and regional model values (Hart et al. 2006) exists for the Maquoketa aquitard. However, in contrast to the P ierre Shale, the Maquoketa aquitard in southeastern Wisconsin is penetrated by numerous, old multiaquifer wells (SE WRP C/WG NHS 2002; Hart et al. 2006). The Maquoketa Formation is usually cased off in these wells, but many wells are uncased above and below the shale, allowing flow through the borehole from the surficial aquifer to the deeper aquifer. In regional flow modeling under transient pumping conditions, borehole flux through such wells can be accommodated using an apparent higher vertical hydraulic conductivity of the aquitard. A major structural discontinuity whose hydraulic properties are poorly known (the Waukesha Fault) occurs directly between the field sites (Figure 1). Y et, a conceptual model of through-going vertical fractures is inconsistent with available field evidence at the tens of kilometers scale encompassed by our two sites. Flowpaths through such vertical fractures, if they existed, would intersect the numerous bedding-plane fractures that have been observed in the Maquoketa aquitard (E aton 2002; E aton and Bradbury 2003) and are known from similar hydrogeologic settings. In such a hypothetical high-diffusivity interconnected fracture network, hydraulic head must respond rapidly to more than 100 m of drawdown caused by pumping in the underlying Cambrian-O rdovician Aquifer system. The hydraulic response time to equilibrium (Alley et al. 2002; E aton and Bradbury 2003) in such a fracture network would be on the order of days. Furthermore, flow within a through-going fracture network would cause vertical head profile equilibration to a monotonic pattern and homogeneous hydrogeochemistry with depth in the formation, contrary to observations (Figure 7). 610 T.T. Eaton et al. GROUND WATER 45, n o. 5: 601–615 Measurements over a period of 3 to 4 years indicated that there is no drawdown trend in head change or systematic head fluctuation within the Maquoketa Formation at our sites (Figure 5). Several large municipal supply wells within a few miles of the field sites were concurrently being pumped on various schedules, and the underlying aquifer system has been pumped for over a century. Together, the hydraulic head data (Figure 5), the stratified hydrogeochemistry (Figure 7), and limitations on vertical fracture propagation due to heterogeneous mechanical stratigraphy provide important evidence against continuous large vertical fractures in the Maquoketa aquitard, at least at the scale of our two field sites. A new geological conceptual model for the aquitard (Figure 2) accounts for all the available hydrogeological observations in the Maquoketa aquitard at our study sites and is consistent with recent understanding of the mechanisms and patterns of fracture formation and porosity development in similar, relatively undeformed, heterogeneous sedimentary rocks. The major elements of the new conceptual model are as follows: 1. O rders of magnitude contrast between hydraulic diffusivity (K/S s) of horizontal bedding-plane fractured zones and relatively intact aquitard rock matrix. In the case of the Maquoketa aquitard, this accounts for the rapid responses in observation wells to the interwell pumping tests and horizontal fracture flow, resulting in the vertically stratified ground water chemistry (Figure 7). 2. L aterally extensive bedding-plane fracture zones extending over distances of at least 10 km and probably more. Development of bedding-plane fracture porosity in carbonate rocks is similar to evolution of preferential flowpaths in karst through progressive dissolution (Kaufmann and Braun 1999; G abrovsek and Dreybrodt 2001; Bloomfield et al. 2005). These studies have shown that fracture porosity is highly dependent on boundary conditions. For example, given a carbonate rock subcrop in an aquifer system, increased fracture porosity and transmissivity develop progressively over time with distance away from the subcrop. If numerous interbeds of different lithology impose a mechanical stratigraphic control on vertical fractures but allow the development of bedding-plane fractures, the result would be laterally extensive bedding-plane fracture zones extending over distances of up to tens of kilometers in carbonate rocks. 3. V ery low vertical hydraulic conductivity in thick unfractured shale intervals. In this case, effective hydraulic conductivity of the apparently unfractured shale near the base of the Maquoketa aquitard appears to be similar to values from laboratory rock-core testing (E aton and Bradbury 2003), a type of scale independence suggested for some aquitards by Neuzil (1994 ). 4 . Hydraulic influence of the aquitard subcrop, where hydraulic head in the shallow aquifer system is controlled by numerous surface water bodies. This last element of the conceptual model does not excessively limit its general applicability because unrecognized edges and proximity to areas where the aquitard has been eroded through to the aquifer are common uncertainties about aquitards (Cherry et al. 2006).
N umerical M odeling Results and S ensitiv ity Simulation results are shown using vertical profiles of head (Figure 9A) for comparison to hydraulic head data collected at the field sites (Figure 5). V ertical head distribution for the Maquoketa aquitard and below is illustrated from predevelopment steady-state conditions (1900) to present-day transient conditions (2000), with head profiles at two intermediate time periods: 1960 and 1980. Simulated head values in different layers (curves) are most different from measured head values (points) at the base of the formation, where head changes rapidly over a short vertical distance. The discrepancy is likely due to inadequate vertical resolution of the model grid. The simulated 2000 head profile with depth generally matches the observed abrupt loss in head across the bottom contact of the Maquoketa Formation. The match between simulated transient heads at present (year 2000) and the observed head distribution with depth (Figure 5) suggests that the flow field in the Maquoketa aquitard is not at steady state with respect to pumping in the underlying aquifer system, as previously inferred (E aton and Bradbury 2003). Since long-term changes in head values were not observed within but only below the Maquoketa Formation, model parameters represent an upper bound on aquitard vertical diffusivity, and hence indicate a long time constant to equilibrium. The model was also used to investigate the transient nature of the ground water flow system. The evolution of head is dependent on inherently unpredictable patterns of pumping for municipal supply. However, if the current head in the deep aquifer system is assumed to remain at the present value of 14 0 m above msl, the evolution of head in the Maquoketa aquitard can be projected. Results of predictive simulations (shown in Figure 9B for years 2200 and 2500, and steady state), using initial heads from the present-day (year 2000) simulation, show that the current hydraulic head profile is far from steady state. The steady-state head profile shows a distinct change in vertical gradient below the upper fractured portion of the Maquoketa aquitard, where the hydraulic head gradient with depth becomes constant and linear. In all simulations, head in the transmissive bedding-plane fracture layers tends to maintain the vertical head distribution close to the head values at the lateral model boundary (260 m above msl). The difference between present-day (year 2000) and future profiles suggests that a steady-state flow system within the Maquoketa aquitard might not occur for many hundreds of years. In order to test the effect of both the constant-head boundary condition at the aquifer subcrop (represented by P ewaukee L ake) and the bedding-plane fracture zones on the head profile in the aquitard, we arbitrarily reduced the value of the western constant-head boundary condition by 80 m and reran the steady-state and transient simulations. Bedding-plane fracture zones that are laterally extensive over scales of kilometers can provide highdiffusivity flowpaths, causing distant lateral boundary conditions to become more important than much closer lower boundary conditions for rock aquitards. The resulting effect on the head profile (Figure 9C) indicates that hydraulic head at the lateral boundary can play a more important role in controlling the Maquoketa vertical head distribution than the lower boundary condition in the immediate underlying heavily pumped aquifer. Although it is well known that boundary conditions affect steadystate hydraulic heads, the dominant influence of lateral boundary conditions here, under transient conditions, is Figure 9 . Field- measured v ertical head distribution (points) within the Maquoketa Formation (shaded) and modeling results (curv es) at the DOT field site (A ) after different stress periods (dates) to present; (B) from present (20 0 0 ) to years 220 0 and 250 0 and steady- state future head distribution; (C ) with constant- head boundary condition v alues reduced by 80 m to demonstrate role of head at western subcrop. Dashed lines indicate approximate elev ation of layers representing bedding- plane fracture zones. Open and solid circles as in Figure 5. The simulated decrease in head across the bottom of the Maquoketa Formation is discretized by the number of layers used— it would be more abrupt with ev en thinner layers. T.T. Eaton et al. GROUND WATER 45, n o. 5: 601–615 611