Impacts of Pleistocene glaciation on large-scale groundwater flow and salinity in the Michigan Basin

Transcription

1 Geofluids (2011) 11, doi: /j x Impacts of Pleistocene glaciation on large-scale groundwater flow and salinity in the Michigan Basin J. C. MCINTOSH 1, *, G. GARVEN 2 ANDJ.S.HANOR 3 1 Department of Hydrology and Water Resources, University of Arizona, Tucson, AZ, USA; 2 Department of Geology, Tufts University, Medford, MA, USA; 3 Department of Geology and Geophysics, Louisiana State University, Baton Rouge, LA, USA ABSTRACT Pleistocene melting of kilometer-thick continental ice sheets significantly impacted regional-scale groundwater flow in the low-lying stable interiors of the North American and Eurasian cratons. Glacial meltwaters penetrated hundreds of meters into the underlying sedimentary basins and fractured crystalline bedrock, disrupting relatively stagnant saline fluids and creating a strong disequilibrium pattern in fluid salinity. To constrain the impact of continental glaciation on variable density fluid flow, heat and solute transport in the Michigan Basin, we constructed a transient two-dimensional finite-element model of the northern half of the basin and imposed modern versus Pleistocene recharge conditions. The sag-type basin contains up to approximately 5 km of Paleozoic strata (carbonates, siliciclastics, and bedded evaporites) overlain by a thick veneer (up to 300 m) of glacial deposits. Formation water salinity increases exponentially from <0.5 g l )1 total dissolved solids (TDS) near the surface to >350 g l )1 TDS at over 800 m depth. Model simulations show that modern groundwater flow is primarily restricted to shallow glacial drift aquifers with discharge to the Great Lakes. During the Pleistocene, however, high hydraulic heads from melting of the Laurentide Ice Sheet reversed regional flow patterns and focused recharge into Paleozoic carbonate and siliciclastic aquifers. Dilute waters (<20 g l )1 TDS) migrated approximately 110 km laterally into the Devonian carbonate aquifers, significantly depressing the freshwater-saline water mixing zones. These results are consistent with 14 C ages and oxygen isotope values of confined groundwaters in Devonian carbonates along the basin margin, which reflect past recharge beneath the Laurentide Ice Sheet (14 50 ka). Constraining the paleohydrology of glaciated sedimentary basins, such as the Michigan Basin, is important for determining the source and residence times of groundwater resources, in addition to resolving geologic forces responsible for basinal-scale fluid and solute migration. Key words: Michigan Basin, hydrology, modeling, glaciation, salinity Received 28 April 2010; accepted 30 June 2010 Corresponding author: Jennifer C. McIntosh, Department of Hydrology and Water Resources, University of Arizona, Tucson, AZ 85721, USA. mcintosh@hwr.arizona.edu. Tel: Fax: Geofluids (2011) 11, INTRODUCTION Tectonics and sediment compaction have driven fluid and solute transport in intracratonic basins over millennial timescales (e.g. Cathles & Smith 1983; Bethke 1985; Ge & Garven 1992; Garven et al. 1993). In contrast, there is a growing body of evidence for similar fluid migration events on much shorter timescales (thousands of years) driven by continental glaciation (e.g. Boulton et al. 1995; Piotrowski 1997a,b; Person et al. 2003, 2007; Lemieux et al. 2008a c). Melting of kilometer-thick ice sheets during the Late Pleistocene profoundly altered regional-scale groundwater flow in the low-lying stable interior of the North American and Eurasian cratons. Geochemical and isotopic studies of crustal fluids show that large volumes of glacial meltwaters penetrated to depths up to 1 km in underlying sedimentary basins and fractured crystalline rocks, significantly diluting and displacing remnant basinal brines, and storing freshwater resources on the continents (e.g. Clark et al. 2000; Grasby et al. 2000; Person et al. 2003; McIntosh & Walter 2005). Grasby & Chen (2005) and McIntosh & Walter (2006) showed that carbonate aquifer systems along the margins of Paleozoic basins were particularly important for transmission of meltwaters into the subsurface as they acted like subglacial drains (Fig. 1). In the Michigan and Illinois Ó 2010 Blackwell Publishing Ltd

2 Glacial impacts on basinal-scale fluid and solute transport 19 Fig. 1. Map of Paleozoic basins of North America that were glaciated during the Pleistocene [modified from Grasby & Chen (2005)]. The dotted black line highlights the maximum extent of the Wisconsin glaciation. The Paleozoic carbonate aquifers that Grasby & Chen (2005) suggest acted like subglacial drains for meltwaters are shown in blue. The approximate maximum total dissolved solids measured for brines in each basin are reported in mg l )1. Brine data from Wilson & Long (1993a), Grasby et al. (2000), Stueber & Walter (1994), Sanders (1991), and Connolly et al. (1990). basins (Fig. 2A), located in the Great Lakes region, these meltwaters migrated from Devonian carbonate aquifers into adjacent fractured Upper Devonian organic-rich shales and enhanced generation of economic reservoirs of microbial methane by significantly diluting brines and likely transporting in microbial communities (Martini et al. 1998; McIntosh et al. 2002; Schlegel et al. 2008). Differences in the fluid salinity, structure, and hydrostratigraphy of sedimentary basins and crystalline bedrock may have controlled the extent of meltwater invasion and redox processes. This study integrates finite-element modeling of variable density groundwater flow with previous geochemical results of Michigan Basin brines to better constrain the impacts of Pleistocene glaciation on fluid, heat and solute transport. Specifically, we tested the hypotheses that: (i) anomalous salinity patterns along the northern Michigan Basin margin cannot be explained by topographically driven recharge of precipitation, instead glacial loading of continental ice sheets must be invoked; and (ii) relatively permeable Devonian carbonate aquifers were areas of focused recharge for subglacial meltwaters into the Michigan Basin. Results from this study have implications for the forces involved in generating large changes in salinity profiles within basins, response times of major aquifer systems to climate change, and distribution of groundwater and energy resources. HYDROGEOLOGY OF MICHIGAN BASIN Hydrostratigraphy As a result of prolific hydrocarbon exploration in the Michigan Basin, the hydrostratigraphic architecture of the basin, rock properties, and fluid chemistries are well known. The intracratonic basin is relatively tectonically undeformed and contains up to approximately 5 km of Paleozoic and Mesozoic sedimentary strata, which are underlain by Precambrian igneous and metamorphic basement rocks (Dorr & Eschman 1970). The dominant lithologies of basin formations are described below from oldest to youngest age, and rock properties are listed in Table 1. The basal Mt. Simon Sandstone is a regional aquifer, containing high permeability and porosity sandstones (Vugrinovich 1986). The Munising Formation is a regional confining unit, composed of interbedded sandstones, dolomite, and mudstones (Swezey 2008). The Prairie du Chien Group and Jordan Sandstone are regional aquifers, composed of dolomites and sandstones with minor amounts of shale and anhydrite. The Trenton-Black River Group is a regional aquifer, dominantly composed of limestones and dolomite. The overlying Utica Shale is an organic-rich fracture shale confining unit and one of the major petroleum source rocks in the basin (Swezey 2008). The Niagara and Clinton Groups at the base of the Silurian section are regional aquifers, composed of pinnacle reef limestone and dolomite. The Salina Group contains limestone, dolomite, and bedded evaporite deposits (over 1 km in thickness) of halite, gypsum and anhydrite. The Salina Group salts are regional confining units. The Bass Island-Burnt Bluff Formations are regional aquifers, composed of limestone, dolomite, and shale, with minor amounts of evaporites. The Detroit River Group, a regional confining unit, is composed of dolomite, bedded halite, anhydrite, limestone and sandstone. The Traverse and Dundee formations are regional aquifers, composed primarily of vuggy limestone with minor dolomite, shale, anhydrite and halite (Vugrinovich 1986).

3 20 J. C. MCINTOSH et al. Fig. 2. Hydrogeologic framework of the study area. (A) Bedrock geology map of Paleozoic basins in the Great Lakes region of the United States, highlighting regional aquifers and shale confining units in relation to Pleistocene glacial limits. (B) Cross-sectional view of the Michigan Basin. Shale and evaporite confining units are shown in gray, and the Devonian organic-rich Antrim Shale is shown in black. (C) Comparison of salinity gradients between the Michigan, Illinois and Appalachian basins [adapted from Hanor (1979)]. Lines represent the maximum total dissolved solids (TDS) concentrations measured with depth. Table 1 Hydrogeologic properties of Michigan Basin aquifer systems and confining units used in the mathematical model. Geologic age Hydrogeologic unit Mesh layer Porosity (%)* K x values (Myr )1 ) K z values (Myr )1 ) Thermal conductivity (Wm )1 C )1 ) Pleistocene Glacial Drift Pennsylvanian Saginaw-Michigan Formation ) )4 1.5 Mississippian Marshall Sandstone Mississippian Coldwater Shale ) )4 1.5 Upper Devonian Antrim Shale ) )4 1.5 Devonian Traverse-Dundee Carbonates Devonian Detroit River Group ) )5 4.5 Silurian Bass Island-Burnt Bluff Carbonates Silurian Salina Group (Bedded Evaporites) ) )5 4.5 Ordovician Utica Shale ) )4 1.5 Ordovician Trenton-Black River Group Cambrian Jordan Sandstone Cambrian Prairie du Chien Group Cambrian Munising Formation ) )4 1.5 Cambrian Mount Simon Sandstone Precambrian Basement ) )7 2.5 *Hydrogeologic property data compiled from Vugrinovich (1986), Garven et al. (1993), Senger (1993), Gupta & Bair (1997), Appold & Garven (1999), Eberts & George (2000), Beauheim & Roberts (2002), and Breemer et al. (2002).

4 Glacial impacts on basinal-scale fluid and solute transport 21 The Upper Devonian Antrim Shale is an organic-rich fractured shale, which is one of the major petroleum source rocks in the basin, although the vast majority of natural gas in the shale is microbial in origin (Martini et al. 1998). The shale matrix porosity and permeability is low, however, the shale fracture systems produce copious amounts of formation water and natural gas. The Coldwater Shale, a regional confining unit, overlies the Antrim Shale. There is a thin (<30 m in thickness) Mississippian sandstone aquifer (Berea Sandstone) between the Coldwater and Antrim shales, although it is not represented in our model hydrostratigraphy. The Marshall Sandstone is a regional aquifer with high measured permeabilities and porosities (Vugrinovich 1986). The Michigan Formation is a regional confining unit, consisting of shales interbedded with minor amounts of limestone, dolomite, anhydrite and gypsum. The overlying Saginaw Group contains a series of interbedded shales, siltstones, sandstones, and minor limestones and coalbeds (Vugrinovich 1986). Paleozoic formations contain saline fluids derived from evaporation of Paleozoic seawater, dissolution of evaporite deposits, and water-rock reactions (Wilson & Long 1993a,b; Hanor & McIntosh 2006). In general, formation water salinity increases exponentially from <0.5 g l )1 total dissolved solids (TDS) near the surface to >350 g l )1 TDS at over 800 m depth (Fig. 2C). Subsidence and sedimentation in the basin stopped in the Jurassic, and since then erosion has not significantly removed bedrock thickness; thus, the present-day bedrock topography has probably been constant since the Jurassic (Vugrinovich 1989). Glacial drift deposits (up to approximately 300 m thick) overlie the Michigan Basin, forming much of the modern surface topographic relief. Glacial history The Michigan Basin was entirely overrun by ice sheets during at least the last three glacial advances (Wisconsinan, Illinoian, and Kansan), as evidenced by remnant glacial deposits (Dorr & Eschman 1970). Figure 2A shows the maximum southern extent of the Laurentide Ice Sheet, and the Last Glacial Maximum (LGM, approximately 18 ka). The Michigan Basin also likely experienced multiple previous continental glaciations over the last 2 Ma (Shackleton 1987). The Laurentide Ice Sheet was wetbased near its southern margin, which covered the Michigan Basin, and meltwater infiltration fluxes were as high as 6 mm year )1 (Lemieux et al. 2008a). Glacial landforms and the distribution of gravels and tills in south-central Michigan point to subglacial flooding during retreat of the ice sheet (Fisher & Taylor 2002; Fisher et al. 2003). In general, the ice sheet advanced from north to south across the basin. Modern hydrology Modern groundwater in the Michigan Basin flows from high elevation recharge areas (glacial moraines) to low elevation discharge regions (e.g. Saginaw Bay lowlands) and eventually to the Great Lakes (Vugrinovich 1986). The presence of Late Pleistocene age waters in low permeability glacial tills and underlying confined aquifers (Kolak et al. 1999; McIntosh & Walter 2006) suggests that modern groundwater circulation is relatively shallow and does not physically interact with the deeper older saline formation waters. EVIDENCE FOR GLACIAL RECHARGE INTO MICHIGAN BASIN Multiple lines of evidence exist from geochemical, isotopic, and hydrologic studies that support infiltration of Laurentide Ice Sheet subglacial meltwater and Late Pleistocene precipitation into the margins of the Michigan Basin. Remnant saline brines in the Upper Devonian Antrim Shale across the northern basin margin have been significantly diluted by freshwater, as shown by the decrease (up to 80%) in bromide concentrations (Fig. 3A) to 300 m depth (McIntosh & Walter 2005). The source of the freshwater recharge appears to be subglacial meltwater, because the 14 C ages correspond to periods of ice cover, and the d 18 O values (as low as )16.4&) show mixing of ice sheetderived waters with saline brines and modern precipitation (Martini et al. 1998; McIntosh et al. 2004; Fig. 3B). Late Pleistocene brines in the Antrim Shale are enriched in Na and Cl relative to Br (Fig. 3A), indicative of halite dissolution. Mass-balance calculations reveal that these NaCl brines obtained >60% of their salinity from halite dissolution (McIntosh & Walter 2005). Interestingly, there are no halite deposits in the Antrim Shale. Therefore, recharge into the fractured shale must have been routed through underlying Devonian carbonate aquifers, which contain localized and bedded evaporites. There are several studies that provide evidence for recharge and storage of glacial meltwaters in Devonian carbonate aquifers in the Michigan Basin. McIntosh & Walter (2006) investigated groundwater in confined Devonian carbonate aquifers along the northern and southern margins of the basin. 14 C ages ranged from modern to >50 ka. Oxygen isotope values ranged from )18 to )13&, and Na Cl Br relations pointed to dilution and displacement of saline brines by meteoric recharge. Weaver et al. (1995) and Husain et al. (2004) report similar results from Devonian carbonate aquifers and overlying glacial tills along the eastern basin margin in southwestern Ontario. One of the important consequences of subglacial recharge into Devonian carbonate aquifers was the generation of economic microbial methane accumulations in the

5 22 J. C. MCINTOSH et al. Fig. 3. Extent and sources of freshwater dilution of Michigan Basin brines. (A) Chloride versus bromide concentrations of formation waters from the Upper Devonian Antrim Shale and underlying Devonian carbonates (data from McIntosh et al and Wilson & Long 1993b). Brines in Devonian carbonates at depth (>375 m) are enriched in Cl and Br, plotting along the seawater evaporation trend (McCaffrey et al. 1987). Brines in the Antrim Shale, along the shallow basin margin (<500 m depth) have been significantly diluted by freshwater and are enriched in Cl from halite dissolution in underlying carbonate aquifers (McIntosh & Walter 2005). (B) Oxygen isotope values of formation waters versus bromide show dilution of basinal brines by two (or more) freshwater endmembers: modern precipitation and Late Pleistocene glacial meltwater (McIntosh & Walter 2005). overlying fractured organic-rich Upper Devonian Antrim Shale. There are over natural gas wells producing methane from the Antrim Shale in northern Michigan at present, and microbial methane has been detected along the shallow eastern basin margin (McIntosh et al. 2009). Microbial methane was generated in situ at the time of glacial recharge, as shown by the correlation of low dd values of CH 4 and co-produced formation waters (Martini et al. 1996). Formolo et al. (2008) hypothesized that 75 88% of microbial methane generated in the Antrim Shale was released to the atmosphere during periods of glacial retreat. Methanogenesis associated with Pleistocene meltwaters drove calcite precipitation in Antrim Shale fractures, which may provide evidence for hydrofracturing of shales by glacial loading (Budai et al. 2002). Several researchers have also reported glacial recharge and flushing of brines in the Pennsylvanian and Mississippian age sandstone aquifers, overlying the Upper Devonian Antrim Shale, in the central Michigan Basin. Kolak et al. (1999) detected dilute formation waters (Cl as low as 600 mg l )1 ) with anomalously low d 18 O values ()18.5&) at depth in the Saginaw Formation; they concluded melting of the Port Huron lobe of the Laurentide Ice Sheet drove subglacial meltwaters into basinal aquifers, reversing current hydrologic gradients. These meltwaters are now refluxing into the Saginaw Bay lowlands and Lake Huron. Ma et al. (2004) detected Late Pleistocene age waters in the underlying Mississippian Marshall Sandstone aquifer that have paleorecharge temperatures of approximately 1 C and oxygen isotope values of )9.8 to )8.6& (within the range of modern precipitation). They suggested that these waters were recharged beneath the Laurentide Ice Sheet based on the 14 C ages and near freezing recharge temperatures, with a significant portion of paleoprecipitation due to the elevated d 18 O values. In addition, based on high He fluxes and salinities in Marshall Sandstone formation waters, they suggest that gases and solutes migrated upward into the shallow aquifer from deeper geologic formations (Ma et al. 2005). Klump et al. (2008) measured noble gases, stable isotopes, and 14 C of groundwaters along the southwestern margin of the Michigan Basin. They report that soil temperatures were approximately 3 C in southeastern Wisconsin and d 18 O values of recharge ranged from )10.3 to )9.9& prior to glaciation (>28 ka). During periods of glaciation, recharge temperatures decreased to approximately 1.4 C, d 18 O values decreased slightly ()12.4&), and excess air concentrations increased, likely due to greater fluctuations in the water table. Similar to Ma et al. (2004), they suggested that subglacial recharge must have contained a significant component of paleoprecipitation to explain the relatively positive d 18 O values that are significantly higher than has been reported for glacial meltwaters at the end of the Late Pleistocene (e.g. )25 to )24& for Glacial Lake Agassiz waters; Remenda et al. 1994). Evaporated surface waters of proglacial lakes may have had more positive d 18 O values (up to )15.8&) (Buhay & Betcher 1998). Bahr et al. (1994) provided hydrologic evidence for glacial loading of the Michigan Basin. Overpressures were observed in the deep Cambrian St. Peter Sandstone, which could not be explained by either ongoing sediment compaction or oil and gas generation. Instead, they suggested that these anomalous pressures were generated during glaciation. These natural tracer studies show deep penetration of freshwater into Paleozoic aquifers along the margins of

6 Glacial impacts on basinal-scale fluid and solute transport 23 the Michigan Basin and flushing of saline brines, most likely driven by Late Pleistocene glaciation. These reservoirs of Late Pleistocene dilute waters are a critical groundwater resource in the Great Lakes region. Our study further investigates the source, mechanisms, and pathways of recharge to the Michigan Basin via numerical modeling. PREVIOUS ICE SHEET MODELING EFFORTS IN GREAT LAKES REGION Several hydrologic modeling studies have been conducted in the Great Lakes region of the United States to explore the effects of past climate change on groundwater recharge to sedimentary basins; results from the most relevant studies are summarized below. Person et al. (in press) provides a more extensive review of ice sheet-aquifer modeling efforts across North America and Europe. Hoaglund et al. (2004) investigated the hydraulic connection between the Saginaw Lobe of the Laurentide Ice Sheet and underlying groundwater systems in the Saginaw Bay Lowlands region (central Michigan Basin) during the Port Huron ice advance using a three-dimensional MOD- FLOW model. They focused on the Mississippian clastic aquifers, and ignored variable density flow and solute transport. They modeled glaciation by increasing the hydraulic head values of surface nodes to equal 91% of the ice sheet thickness (to account for the density of ice, 0.91 g cm )3 ), assuming a thin ice scenario (Clark et al. 1994). They also used a loading efficiency of 1.0, which corresponds to 100% transmission of the ice load into hydraulic load. Model results show that groundwater flow beneath the ice lobe was primarily vertical, and flow directions during periods of ice advance were reversed compared to present-day conditions. Bea Jofre et al. (2009) recently developed a hydrogeochemical model of the eastern margin of the Michigan Basin which includes variable density fluid flow and reactive solute transport. Their preliminary results show that glacial recharge was unable to displace deep saline brines; most of the groundwater circulation and chemical reactions were limited to the shallow basin margin, near the leading edge of the ice sheet. These results are consistent with Sykes et al. (2009) who developed a three-dimensional hydrologic model of the eastern margin of the Michigan Basin, which was focused on the proposed Deep Geologic Repository site for low to intermediate level radioactive waste located in the Bruce Peninsula in southwestern Ontario. Breemer et al. (2002) constructed a cross-sectional finite difference model of the Illinois Basin west of the Michigan Basin (location shown in Fig. 2) to investigate meltwater fluxes and drainage patterns beneath the Lake Michigan Ice Lobe. They concluded that groundwater flow patterns were reversed and groundwater velocities were significantly higher during periods of glaciation compared to modern ice-free conditions. The presence of confining units along the northwestern margin of the basin, that truncate regional aquifers, limited meltwater infiltration into the Illinois Basin. As a result, most of the meltwater discharged through tunnel channels. The model of Breemer et al. (2002) did not consider variable density fluid flow, which is likely important considering the high salinities of brines in the Illinois and Michigan basins (up to >350 g l )1 TDS). Lemieux et al. (2008a c) constructed a three-dimensional model of the Canadian and northern United States landscape to investigate the impacts of the Wisconsinian glaciation (approximately 120 ka to present) on groundwater flow and brine transport in Paleozoic sedimentary basins, including the Michigan Basin, and fractured bedrock (Canadian Shield). Coupled processes, such as variable density fluid flow, hydromechanical loading, subglacial infiltration, permafrost development, and flexure of the Earth s lithosphere by loading and unloading of ice sheets were incorporated into the model. Results show profound effects of glaciation on groundwater flow: hydraulic head values increased up to 3000 m down to a depth of 1.5 km, subglacial meltwaters recharged basins during glacial advances, meltwaters discharged from basins during glacial regressions, and there were hiatuses in recharge and build up of high hydraulic heads during periods of permafrost cover. They estimate that 15 70% of Laurentide Ice Sheet meltwaters infiltrated into the subsurface as recharge, which represents a vast reservoir of freshwater resources stored on the continents. The continental scale of the model, although impressive, required oversimplification of the hydrogeology; the model included only four hydrostratigraphic units (Canadian Shield, sedimentary basins, orogenic belts, and oceanic crust) and only 10 vertical nodes for a 10-km column of rock. Because of this coarse discretization, it was impossible to evaluate the impacts of glaciation on individual aquifers and confining units within sedimentary basins. Bense & Person (2008) developed a basin-scale numerical model of a generic intracratonic sedimentary basin to evaluate the effects of variable density fluids, permafrost development, mechanical loading and unloading, and lithospheric flexure during glaciation. Results of transient ice sheet simulations show infiltration of dilute, young, isotopically depleted meltwaters into aquifers and confining units along the basin margin, and preservation of anomalous pore pressures in aquifers and confining units at depth that are still responding to past glacial events (more than 10 ka). These coupled hydrologic processes were also likely important for the Michigan Basin, although beyond the scope of this project, and should be considered in future modeling efforts.

7 24 J. C. MCINTOSH et al. Despite multiple numerical models constructed for Paleozoic basins in the Great Lakes region, there are no adequate models published to date that can be used to evaluate the impacts of Pleistocene glaciation on groundwater flow patterns and salinity gradients across the northern margin of the Michigan Basin, where economic accumulations of microbial methane associated with low salinity, isotopically depleted Late Pleistocene waters have been identified (Martini et al. 1998; McIntosh et al. 2004). None of the previous modeling studies included the northern margin of the Michigan Basin, with the exception of Lemieux et al. (2008a c); however, their model grid is too coarse the Michigan Basin is represented as one hydrogeologic unit to investigate hydrologic processes in specific geologic formations (e.g. Devonian carbonate aquifers and overlying Antrim Shale). NUMERICAL MODEL OF MICHIGAN BASIN A two-dimensional finite-element model was constructed for the northern half of the Michigan Basin, using the program CPFLOW developed by Raffensperger & Garven (1995) to simulate variable density fluid flow, heat and solute transport under modern and Late Pleistocene hydrologic conditions. The model was then used to test the hypothesis that continental glaciation is required to explain the presence of low salinity formation waters observed at depth in the basin (McIntosh & Walter 2005), and that Devonian carbonates acted as subglacial drains for meltwater recharge (Grasby & Chen 2005). Our conceptual model envisioned the complex physical interaction between two regional flow regimes, a shallow freshwater flow system driven by differences in hydraulic head and a deeper hypersaline flow system driven largely by differences in fluid density (Fig. 4). The position of the transition zone between these systems should depend on the magnitude of the gradient in hydraulic head and on the background density of the brines. Increased hydraulic heads and meltwater fluxes resulting from glaciation should significantly increase groundwater flow beneath ice sheets (e.g. Boulton & Caban 1995; Breemer et al. 2002; Person et al. 2007; Lemieux et al. 2008a c). We anticipated that glacial meltwater would penetrate more deeply into basinal aquifers during ice cover, versus modern topographic and climate conditions, and drive fluids to migrate further basinward. The model parameters, boundary conditions, and governing equations are outlined below. Mesh and rock properties A geologic cross-section of the basin modified from Lilienthal (1978) and Passero et al. (1981) was used to construct the finite-element mesh shown in Fig. 5. The mesh has 28 rows, 159 columns, 4452 nodes, and 8532 Fig. 4. Conceptual model of Late Pleistocene versus modern hydrology for the northern Michigan Basin. Continental glaciation likely enhanced freshwater recharge into Paleozoic aquifers and confining units, diluting and flushing basinal brines and stimulating generation of natural gas (microbial methane) in organic-rich shales. triangular finite elements; the cross-sectional grid is 300 km wide and 4.6 km deep. The largest element size is 2500 m in the y-direction, and 323 m in the x-direction. Rock properties for each geologic formation were compiled from the literature (Vugrinovich 1986; Garven et al. 1993, 1999; Senger 1993; Gupta & Bair 1997; Appold & Garven 1999; Everham & Huntoon 1999; Eberts & George 2000; Beauheim & Roberts 2002; Breemer et al. 2002). Representative values used in our model simulations are reported in Table 1. Hydraulic properties of bedded evaporites in the Michigan Basin were approximated from data from Permian salts in the Delaware Basin of New Mexico (Beauheim & Roberts 2002). Salinity gradient and fluid density An initial salinity gradient was imposed to approximate the distribution of brines in the basin prior to Late Pleistocene glaciation; the salinity gradient was assumed to be equal to the maximum observed salinity with depth (Hanor 1979; Fig. 6A). The concentration at the surface boundary was set equal to 0 g L )1 TDS to provide a constant source of freshwater. The density of brines (Fig. 7) was calculated

8 Glacial impacts on basinal-scale fluid and solute transport 25 Fig. 5. Two-dimensional hydrologic model of the northern margin of the Michigan Basin with the finite-element mesh and hydrostratigraphic units (see corresponding numbers in Table 1 for unit properties) highlighted. Model boundary conditions are discussed in the text. using an algorithm from Kemp et al. (1989), which is applicable for salinities up to 600 g L )1, temperatures <174 C, and pressures <100 MPa. The Kemp et al. equation of state also takes into account the elemental composition of fluids. Michigan Basin brines are highly enriched in Ca 2+, in addition to Na + and Cl ) (Wilson & Long 1993a,b). Other equations of state typically used in basin modeling studies (e.g. Rowe & Chou 1970; Phillips et al. 1981; McCain 1991; Batzle & Wang 1992) assume brines are NaCl-type, thus underestimating the density of Michigan Basin fluids, and many are only applicable up to approximately 350 g L )1 TDS (Fig. 7). Heat and pressure The present-day geothermal gradient for the Michigan Basin is C per km, as calculated from bottom hole temperatures (Cercone 1984; Speece et al. 1985; Vugrinovich 1989). The Michigan Basin has been geologically stable since the Jurassic, and there is no geological evidence for volcanic or igneous activity in the underlying crust since the Paleozoic (Dorr & Eschman 1970). Thus, it is reasonable to assume that the present geothermal gradient is a good approximation of the geothermal gradient prior to Late Pleistocene glaciation. A surface temperature of 10 C was applied across the upper boundary of the model to approximate the average annual temperature for Michigan. A uniform basal heat flux of 45 mw m )2 was applied across the bottom boundary, which is consistent with observed heat flow values for the Michigan Basin (Speece et al. 1985). Figure 6B shows the initial temperature gradient for model simulations. Initial boundary conditions It was assumed that the hydraulic conductivities of the Precambrian basement rocks underlying the sedimentary package were low enough that flow is minimal, therefore the bottom of the model and the right hand side of the model where Paleozoic sediments onlap onto the Canadian Shield were set as no flow boundaries. The left-hand side of the model corresponds to the center of the basin and a natural hydrologic divide, where regional groundwater discharges in the Saginaw Lowlands to the Great Lakes (Vugrinovich 1989; Hoaglund et al. 2002) (Fig. 5). We chose to only model the northern half of the basin to focus our efforts in areas where we had good geochemical and isotopic constraints on the timing and source of recharge into Devonian carbonate and shale aquifers. This also enabled us to increase grid resolution and decrease computation time. Extending the model to incorporate the full basin would likely result in different salinity patterns in the central basin (left-hand side of model), but would probably not significantly modify the salinity gradients along the northern basin margin (focus of this study). The elevation of the water table under current hydrologic conditions was assumed to be equal to the land surface, a reasonable assumption given the vertical and horizontal scale of the

9 26 J. C. MCINTOSH et al. Fig. 7. Calculated density versus total dissolved solids (TDS) relations from several published equations of state, compared to measured values of Michigan Basin brines (Wilson & Long 1993a,b; McIntosh et al. 2004). Adams & Bachu (2002) provide a detailed discussion of these equations of state. Densities calculated via the Kemp et al. (1989) equation, which accounts for fluid composition and is applicable up to 650 g l )1 TDS, most closely match Michigan Basin brines. The other equations of state assume a NaCl solution, which underestimates the density of CaCl 2 -rich Michigan Basin brines, and have lower salinity limits (as indicated by the extent of modeled lines). Fig. 6. Initial salinity and temperature conditions applied to the Michigan Basin hydrologic model. (A) The maximum salinity of Michigan Basin brines with depth (shown in inset) was used as the initial salinity gradient. (B) The initial temperature gradient was established by applying a basal heat flux of 45 mw m )2 and a surface temperature of 10 C. model. We used dispersivity values of 1000 m in the longitudinal (x) direction and 10 m in the transverse (y) direction, and time steps of 10 years, given the large scale of the sedimentary basin model [e.g. de Marsily (1986), Domenico & Schwartz (1990) and Neuman (1995)]. CPFLOW Modeling theory and methods The software code CPFLOW developed by Raffensperger & Garven (1995) was used to model coupled fluid (variable density), heat and solute transport in the Michigan Basin. For transient flow in slightly deformable media, CPFLOW solves equations representing conservation of fluid mass and thermal energy (per unit volume of bulk porous medium): where t is time, x i are the spatial coordinates, / porosity, P fluid pressure, T temperature, q fluid density, l dynamic viscosity, g gravitational acceleration, C the rate of internal fluid production or depletion (due to chemical reactions, sediment compaction, external loading, etc.), k intrinsic permeability, k bulk thermal conductivity-dispersion, q Darcy velocity, and (qc) the heat capacity (with subscripts for the fluid phase f and for the bulk porous medium b). In CPFLOW, fluid density and viscosity in Eqns 1 and 2 are assumed to depend on pore pressure, temperature, and salinity according to equations of state from Kemp et al. (1989) and McCain (1991). For a nonreactive groundwater brine system, the concentration of salt (salinity, C) can be obtained by solving the mass-balance equation representing advection, diffusion, and dispersion processes: In Eqn 3, the coefficient D is the hydrodynamic dispersion tensor (Bear 1972), and V is the average linear velocity of the pore fluid as defined by Darcy s Law, and therefore the fluid velocity and the medium s dispersion

10 Glacial impacts on basinal-scale fluid and solute transport 27 coefficients must be calculated before the concentration salinity can be computed. Laboratory and field measurements of D are well known for many siliciclastic and carbonate formations which serve as shallow aquifers, but few measurements exist for deep basin strata (Domenico & Schwartz 1990). In homogeneous laboratory cores, D is approximately equal to a(v), and the dispersivity coefficient (a) is approximately 10 )3 to 10 )1 m, but it could scale up to approximately 10 1 to 10 3 m for long-distance flow in highly heterogeneous and fractured aquifers in sedimentary basins (de Marsily 1986). A limitation of CPFLOW is the two-dimensionality of the code. While some finite-element hydrothermal flow programs have full three-dimensional capability (e.g. López & Smith 1996; Garven et al. 1999; Coumou et al. 2008), the sparseness of three-dimensional parameter data and the cubic increase in computing memory and CPU time make a three-dimensional analysis difficult on any serial computer for large-scale basin models involving coupled flow. We chose to limit our analysis to two-dimensional profiles, and felt this was a good first-approximation to understanding the large spatial-scale effects and long time-scale effects of groundwater brine mixing in the Michigan Basin. Threedimensional studies are computationally possible, especially with the availability of cluster parallel computing, but we did not feel three-dimensional models were well justified or needed in this case. RESULTS OF MODEL SIMULATIONS AND DISCUSSION Recharge under modern hydrologic conditions Current hydrologic and boundary conditions were applied across the top of the model to simulate modern meteoric recharge and observe the effects of freshwater influx on fluid migration, salinity, and groundwater flow directions (Fig. 8). The model was run for 1 million years. Under modern water table configurations, the hydraulic head gradients are minimal (Fig. 8A), and there is very limited penetration of freshwaters into the basin (Fig. 8B,C). Dilute water (<20 g L )1 TDS) penetrated approximately 27 km laterally into the Devonian carbonate aquifers and Upper Devonian Antrim Shale (Fig. 8C). Flow patterns, as indicated by the stream functions (Fig. 8B), show meteoric water circulation is limited to the shallow subsurface, driven by the hummocky topography of glacial deposits. Topographic highs, such as glacial moraines, serve as major recharge areas, while shallow groundwater discharges into surface drainages in topographic lows and eventually out to the Great Lakes. These results are consistent with previous reports of modern groundwater flow in the basin from hydraulic head measurements, MODFLOW modeling, and age tracer studies (Vugrinovich 1986; Mandle & Westjohn Fig. 8. Hydrologic model simulations of steady-state topographically driven recharge along the northern margin of the Michigan Basin for 1 million years. Results for hydraulic head (A), stream function (B), and total dissolved solids (C) are shown. The black arrows in (B) highlight fluid flow directions. 1989; Hoaglund et al. 2002; McIntosh & Walter 2006). Maximum groundwater fluxes (mass flowing between stream lines, Fig. 8C) are kg Myr )1, and maximum horizontal and vertical velocities are 1.5 and 0.07 m year )1, respectively. Topographically driven recharge of meteoric waters cannot explain the presence of low salinity, Na and Cl enriched, Late Pleistocene age waters at depth in the Michigan Basin. Steady-state glacial conditions To investigate the effects of Pleistocene glaciation on subsurface hydrology, the top boundary condition was altered to imitate hydraulic heads and differences in surface boundary conditions during the LGM. Several studies on the Laurentide Ice Sheet provide good constraints on the ice thickness and surface topography, dynamics of ice movement, initiation development of the ice sheet, meltwater drainage patterns and mechanisms of basal melting (e.g. Clark & Walder 1994; Clark et al. 1994; Licciardi et al. 1998; Marshall & Clarke 1999). The thickness of the

11 28 J. C. MCINTOSH et al. Laurentide Ice Sheet was estimated using the following quadratic equation for the ice profile from Nye (1952): where x is the distance from the ice margin, s is the basal shear stress, q is density, g is gravity, and A is an ice-profiling coefficient. The value for the ice-profiling coefficient (A) for the Laurentide Ice Sheet varied regionally from 0.32 to 4.1 (Matthews 1974); we used a value of 1.8 for our modeling efforts, which is consistent with a thin ice reconstruction (Clark et al. 1994). For ice-covered surface nodes, the hydraulic head was calculated by multiplying the ice sheet thickness by the density of ice (0.91 g cm )3 ) and adding this to the average modern land surface elevation. The loading efficiency of the ice sheet was assumed to be 100%; this is the same approach employed by Person et al. (2003, 2007), Hoaglund et al. (2004), and Bense & Person (2008) to model ice sheet-aquifer interactions, and assumes that liquid water on the surface of the ice sheet periodically drains to the bottom of the ice sheet, as is observed today in the Antarctic and Greenland ice sheets (Dowdeswell & Siegert 1999; Fricker et al. 2007; Jansson et al. 2007; McMillian et al. 2007). The model was run for 1 million years under modern hydrologic conditions to reach a steady-state salinity profile prior to any glacial loading. The northern margin of the basin was then instantaneously loaded with an ice sheet (Fig. 9) for years to investigate the alteration of flow and salinity patterns in response to glaciation. The hydraulic head values increased from 4200 to 5500 m beneath the ice sheet (Fig. 9A). Groundwater flow directions, as indicated by the stream functions (Fig. 9B), completely reversed compared to simulations of modern topographically driven groundwater flow (Fig. 8B). Maximum groundwater fluxes were > kg Myr )1, and maximum horizontal and vertical velocities were 1.3 and 0.07 m year )1, respectively. A plume of dilute waters penetrated to approximately 80 km laterally into the Devonian carbonate aquifers and Upper Devonian shale (Fig. 9C). Subglacial recharge at the northern basin margin also caused brine upwelling in the central basin; this may be an artifact of the no flow boundary we imposed on the lefthand side of the model. However, this may also be a realistic hydrologic phenomenon, as there is geochemical evidence of upward vertical brine migration and discharge in the central basin (Kolak et al. 1999; Ma et al. 2005). Transient glacial conditions The Laurentide Ice Sheet advanced and retreated across the Michigan Basin multiple times over the last 2 Ma; therefore we constructed a transient model to investigate Fig. 9. Hydrologic model simulations of stagnant ice cover at the northern margin of the Michigan Basin for years. Results for hydraulic head (A), stream function (B), and total dissolved solids (C) are shown. The black arrows in (B) highlight fluid flow directions. Note the reversal of groundwater flow compared to Fig. 8B. the effects of ice advance and retreat on subsurface hydrology. Simulations were run under modern topographic and hydrologic conditions until an approximate steady state was reached (by 1 million years) to generate an initial salinity profile prior to glacial loading. A periodic surface boundary was then applied to represent the last glacial advance and retreat. One full glacial cycle ( years) was simulated (Fig. 10). The hydraulic head of each icecovered node was determined using Eqn 4, as explained above. Hydraulic heads of surface nodes not covered by ice were set equal to the land surface elevation. Because only one glacial cycle was modeled, our results underestimate the likely influx of glacial meltwaters into the basin. As the ice sheet advanced across the Michigan Basin (from north to south), pore water hydraulic head values increased from 4200 to 6000 m at the maximum glacial extent (8500 years; Fig. 10A). Hydraulic head values were highest in the glacial drift deposits and bedrock aquifers, including the Mississippian Marshall Sandstone and Devonian Traverse and Dundee formations. Confining units, such as the Devonian Detroit River Group and Silurian

12 Glacial impacts on basinal-scale fluid and solute transport 29 Fig. 10. Transient hydrologic model simulations of the northern margin of the Michigan Basin for one full glacial cycle (approximately 0 17 kyr). The ice sheet advanced from the north (right hand side of model) to the south. Hydraulic head (A) and total dissolved solids (B) results are shown for five time intervals; the 8500 years time interval represents the maximum glacial advance.

13 30 J. C. MCINTOSH et al. Salina Group, maintained low hydraulic head values, in some cases lower than original values prior to glaciation. As the ice sheet retreated, hydraulic head values decreased in the glacial drift and bedrock aquifers. However, by years (end of one full glacial cycle) hydraulic head values of regional aquifers were still elevated compared to preglacial conditions. These results are consistent with previous studies that suggest groundwater systems in glaciated sedimentary basins are still responding to hydrologic conditions from the end of the LGM (Person et al. 2007). Groundwater velocities were as high as 26 m year )1 in the horizontal direction and 2.6 m year )1 in the vertical direction. These values are approximately 20 times higher than velocities under modern and stagnant ice hydrologic conditions. Model simulations of TDS clearly show infiltration of freshwaters into the basin during ice sheet advance, with preferential recharge into regional aquifers (Fig. 10B), although brines in confining units also appear to have been displaced by freshwater recharge. As seen in the stagnant ice simulations, brine upwelling has occurred along the left-hand side of the model (central Michigan Basin). At the maximum glacial extent (8500 years), dilute waters (<20 g L )1 TDS) penetrated up to approximately 60 km laterally in the Devonian Traverse and Dundee formations, and up to approximately 48 km in the Upper Devonian Antrim Shale. Radiocarbon ages and d 18 O values of groundwaters in Devonian carbonates and overlying shales along the basin margins are consistent with recharge beneath the Laurentide Ice Sheet (14 50 ka) (McIntosh & Walter 2006). Dilute waters also invaded the basal Cambrian and Ordovician aquifers to significant depths along the northern margin of the basin (<20 g L )1 TDS at approximately km lateral distance from the subcrop). Comparing the TDS profile at the end of the modeled glacial cycle ( years; Fig. 10B) to the glacial maximum (8500 years; Fig. 10B) there is a slight rebound in the freshwater-saline water interface. This may suggest that the freshwater-saline water interface in the Michigan Basin is still responding (rebounding) from the LGM. It is informative to compare modeled salinity gradients with actual measured values in the basin, although it is important to note that our intent was not to try to recreate the actual glacial history and salinity gradients of the Michigan Basin. Modeled results show lower TDS concentrations than measured values in regional aquifers and confining units (Fig. 11). Formation waters with TDS values <100 g L )1 have been observed approximately 70 km into the Devonian Traverse and Dundee formations, whereas model results show 100 g L )1 TDS waters up to approximately 110 km from the subcrop. The overestimation of freshwater infiltration and brine flushing in our model simulations may be due to: (i) poor estimation of rock hydraulic properties, or (ii) lack of halite dissolution in the Fig. 11. Comparison of observed versus modeled salinity gradients. The total dissolved solids (TDS) result from the transient ice sheet model at years (Fig. 10B) is shown in the background. Black numbers highlight TDS values from the ice sheet model. White numbers display observed TDS values for Michigan Basin brines from Passero et al. (1981), Dannemiller & Baltusis (1990), Wilson & Long (1993a,b), and McIntosh et al. (2004). Model simulations show deeper penetration of meltwaters (lower salinity values) than are observed in the basin today, likely due to the exclusion of halite dissolution in the model. model. It is unlikely that we greatly overestimated hydraulic conductivities and porosities of geologic formations. In addition, only one glacial cycle was simulated. Multiple glaciations would have increased infiltration of freshwater into the basin. More likely, halite dissolution by freshwater recharge would have regenerated salinity and maintained high TDS values at shallow depths, as shown in the formation water geochemistry (McIntosh & Walter 2005). Future work Our model results provide a first order approximation of ice sheet-subsurface hydrologic processes in the Michigan Basin. Future research needs to be conducted to evaluate the importance of reactive solute transport (e.g. halite dissolution), permafrost development, hydromechanical loading and lithospheric flexure on fluid flow and brine migration in glaciated sedimentary basins. In addition, our model only considered the northern margin of the basin. Sharper salinity gradients along the southern basin margin (McIntosh & Walter 2006) may point to differences in the infiltration depth of Pleistocene meltwaters that should be investigated. CONCLUSIONS Numerical models were constructed of the northern half of the Michigan Basin to evaluate important drivers and pathways for freshwater infiltration and displacement of saline fluids. Model simulations show that the presence of low salinity waters at depth across the northern basin margin cannot be explained by topographically driven recharge of meteoric waters over the past 1 million years; rather, increased hydraulic heads from continental ice sheets must