Origin and Extent of Fresh Paleowaters on the Atlantic Continental Shelf, USA by Denis Cohen1,Markperson2,Pengwang3, Carl W

Case Study/ Origin and Extent of Fresh Paleowaters on the Atlantic Continental Shelf, USA by Denis Cohen1,MarkPerson2,PengWang3, Carl W. Gable4, Deborah Hutchinson5,AndeeMarksamer6, Brandon Dugan7, Henk Kooi8, Koos Groen8, Daniel Lizarralde9, Robert L. Evans9, Frederick D. Day-Lewis10,and John W. Lane Jr.10

Abstract While the existence of relatively fresh groundwater sequestered within permeable, porous sediments beneath the Atlantic continental shelf of North and South America has been known for some time, these waters have never been assessed as a potential resource. This fresh water was likely emplaced during Pleistocene sea-level low stands when the shelf was exposed to meteoric recharge and by elevated recharge in areas overrun by the Laurentide ice sheet at high latitudes. To test this hypothesis, we present results from a high-resolution paleohydrologic model of groundwater ﬂow, heat and solute transport, ice sheet loading, and sea level ﬂuctuations for the continental shelf from New Jersey to Maine over the last 2 million years. Our analysis suggests that the presence of fresh to brackish water within shallow Miocene sands more than 100 km offshore of New Jersey was facilitated by discharge of submarine springs along Baltimore and Hudson Canyons where these shallow aquifers crop out. Recharge rates four times modern levels were computed for portions of New England’s continental shelf that were overrun by the Laurentide ice sheet during the last glacial maximum. We estimate the volume of emplaced Pleistocene continental shelf fresh water (less than 1 ppt) to be 1300 km3 in New England. We also present estimates of continental shelf fresh water resources for the U.S. Atlantic eastern seaboard (104 km3) and passive margins globally (3 × 105 km3). The simulation results support the hypothesis that offshore fresh water is a potentially valuable, albeit nonrenewable resource for coastal megacities faced with growing water shortages.

1Department of Geological & Atmospheric Sciences, Iowa State Introduction University, Ames, IA 50011 Hydrologists have long regarded the marine environ- 2 Corresponding author: Hydrology Program, New Mexico ment as the domain from which salt water invades coastal Tech, NM 87801; (575) 835-6506; fax (575) 835-6436; [email protected] aquifers in response to groundwater pumping (e.g., Henry 3University Information Technology Services, Indiana University, 1959; Gill 1962; Schaefer and Walker 1981). However, Bloomington, IN 87801 data from well bores indicate that relatively fresh ground- 4Los Alamos National Laboratory, Los Alamos, NM 87545 water exists beneath the Atlantic continental shelf far 5U.S. Geological Survey, Woods Hole, MA 02543 offshore North and South America (Figure 1). One of the 6 Indiana University, Department of Geological Sciences, most striking examples of fresh groundwater offshore is Bloomington, IN 87801 found on the Atlantic continental shelf of New England, 7Department of Earth Science, Rice University, Houston, TX 77005 where a series of well bores (Hathaway et al. 1979; Schlee 8VU University Amsterdam, Amsterdam and Fritsch 1982; Mountain et al. 1994; Austin et al. 9Department of Geology and Geophysics, Woods Hole 1998; Malone et al. 2002) reveal nearly fresh water (less Oceanographic Institution, Woods Hole, MA 02543 than 5 parts per thousand total dissolved solids, or ppt tds) 10 U.S. Geological Survey, Office of Groundwater, Branch of 200 m beneath the seafloor more than 100 km offshore Geophysics, Storrs, CT 06269 of New Jersey and Long Island (Figure 1, section 3). Correction added after online publication September 15, 2009: Emplacement mechanisms that have been proposed to on page 7, in Equation 9, the value in the square root should have account for this fresh water include (1) ocean-directed, read: ‘‘1 − (ξ/L)2’’. We apologize for this error. Received April 2009, accepted August 2009. lateral inflow of fresh water during Pleistocene sea-level Copyright © 2009 The Author(s) low stands from confined aquifers whose recharge areas Journalcompilation©2009NationalGroundWaterAssociation. were, and often still are, above sea level today (right-hand doi: 10.1111/j.1745-6584.2009.00627.x side of Figure 2A; Meisler et al. 1984; Pope and Gordon NGWA.org GROUND WATER 1 Figure 1. Salinity distribution (contours in ppt) for four cross-sectional transects (1 to 4) across the Atlantic continental shelf of North and South America (from Hathaway et al. 1979; Johnston 1983; Kooi and Groen 2001; Marksamer et al. 2007). The location of cross sections 1 to 4 is shown in the inset (numbered red section lines). The fresh water beneath these shelves was likely emplaced via several mechanisms, as discussed in the text. Orange and yellow shaded patterns represent crystalline and continental shelf sedimentary rocks, respectively.

1999); (2) localized, vertical paleorecharge on subaerially units. Along the New England coast, however, onshore exposed continental shelf during Pleistocene sea-level low hydraulic heads are too low to drive meteoric water far stands (central part of Figure 2A; Kooi and Groen 2001; offshore (Kooi and Groen 2001). In addition, confined Kooi et al. 2000) facilitated by gaps in confining units aquifers north of Martha’s Vineyard and Nantucket out- due to either nondeposition or erosion and local topo- crop or subcrop below sea level (not shown). Therefore, graphic variations in the water table; or (3) sub-ice-sheet onshore recharge cannot be happening today. The fresh recharge along portions of the continental shelf overrun and brackish subseafloor pore waters along the north- by the Laurentide ice sheet (Siegel and Mandle 1984; eastern U.S. Atlantic coast thus can be considered prime Person et al. 2003, 2007a) between 21 and 14 KyBP examples of paleogroundwater emplaced during Pleis- (Figure 2B; Uchupi et al. 2001). Proglacial lakes that tocene sea-level low stands that escaped salinization dur- form in front of ice sheets could also contribute recharge ing Holocene sea level rise (e.g., Hathaway et al. 1979; (Marksamer et al. 2007; Lemieux et al. 2008a, 2008b, Meisler et al. 1984; Kooi and Groen 2001; Person et al. 2008c; Figure 2B). To date, no study has estimated the 2003). These conditions have also been reported along the volume of emplaced fresh water on the continental shelf Atlantic continental margin in Europe (Edmunds 2001). in New England or for passive margins globally. What is Because pore water data are extremely limited, there is a clear from inspection of Figure 1 is that the distribution dearth of information regarding the distribution and ori- of present-day fresh and salt water cannot be explained gin of these offshore paleowaters. Paleohydrologic models by assuming hydrodynamic equilibrium conditions with represent a powerful tool to assess offshore groundwater modern sea level (Figure 2C). resources. The presence of large volumes of offshore fresh To estimate the amount of fresh water present on water along the southeastern United States (Figure 1, continental shelf environments, we analyzed four salinity section 2) can be explained, in part, by present-day mete- cross-sectional profiles from Suriname (Kooi and Groen oric recharge to confined aquifers that outcrop on land and 2001), Florida (Johnston 1983), New Jersey (Hathaway the presence of high-permeability subseafloor karstified et al. 1979), and Massachusetts (Marksamer et al. 2007) 2 D. Cohen et al. GROUND WATER NGWA.org would control the degree of subaerial exposure of the shelf during sea water low stands. However, mean water depth is only one factor controlling fresh water occurrence. Other factors, such as the presence/absence of glacial loading, the elevation and areal extent of onshore recharge areas for confined aquifers, and recharge rates, are also important. The average volume of fresh water (salinity less than 1 ppt) per kilometer width for these four cross sections is estimated to be about 3.8 km3.This estimate assumes an effective sediment porosity of 0.2 and includes fresh water in both coarse and fine-grained sediments. Assuming that the length of the U.S. Atlantic continental shelf is 3310 km and assuming an average of 3.8 km3 of available fresh water per kilometer of coastline (i.e., using all four section lines), then the volume of available fresh water along the eastern seaboard of the United States is about 12,000 km3. Globally, continental shelf width varies considerably from almost zero near sub- duction zones off of South America to more than 900 km off of Siberia. If we restrict our analysis to passive margins, then the total global length of the continental shelf is about 80,000 km. Together with a fresh water volume of 3.8 km3/km, this yields a global estimate of sequestered fresh water along passive margins of about 300,000 km3. Figure 2. Conceptual models for fresh water emplacement. The order-of-magnitude calculation suggests that passive (A) Meteoric recharge under modern sea level conditions. margins may represent an important unconventional fresh Sea water–fresh water interface is assumed to be in equi- water resource. However, there is considerable uncertainty librium with modern sea level conditions; (B) fresh water in these estimates. driven offshore by sub-ice-sheet recharge from the Lauren- The purpose of this study is to evaluate the mech- tide ice sheet during the last glacial maximum, 21,000 years before present; and (C) vertical infiltration of meteoric water anisms of emplacement, distribution, and volume of this through confining units induced by local flow cells on the unconventional, nonrenewable water resource using high- continental shelf during sea-level lowstands associated with resolution numerical modeling. Obviously, this work rep- glaciation. Salt water is shaded yellow, fresh water is white. resents only a first step toward estimating the extent of Gray and orange shaded patterns represent confining units this potential resource, and additional hydrologic and geo- and bedrock, respectively. physical exploration is required to verify the simulation results. We present order-of-magnitude estimates of pas- (presented in Figure 1 and Table 1). The lithologic com- sive margin fresh water volumes globally, but focus our position of these sections varies from siliclastic (Suri- analysis on New England, where unusually fresh pore name, New Jersey, Nantucket) to carbonate dominated fluids are found in boreholes located up to 100 km off (Florida). The limit of the 1 ppt isochlor varies consid- the coast (wells 6001, 6009, 6011, 6020; Figure 3A). erably from about 10 km off New Jersey to 120 km off These wells have salinity below 3 ppt within Pleistocene, Florida. Inspection of Table 1 suggests that the volume Pliocene, and Miocene sand units from 100 to 500 m of continental shelf fresh water appears to be inversely below seafloor (Kohout et al. 1977; Folger et al. 1978). correlated with mean continental shelf water depth. The Because the response time for solute transport on con- average depth (and hence slope) of the continental shelf tinental shelf aquifer systems is long (at least 106 years

Table 1 Atlantic Continental Shelf Fresh Water Estimate

Distance Fresh Water Lens Vertical Fresh Water Lens Mean Continental Volume Fresh Cross Section Extends Offshore (km) Thickness (km) Shelf Depth (m) Water1 (km3/km)

Suriname 40 0.4 45 3.2 Florida 120 0.5 32 12 New Jersey 5 0.2 65 0.2 Nantucket 50 0.3 37 3

1Assuming a porosity of 0.2.

NGWA.org D. Cohen et al. GROUND WATER 3 Study Area Sediments of the Atlantic continental margin of New England have been studied for their petroleum and min- eral resources (Perlmutter and Geraghty 1963; Hathaway et al. 1979; Mattick and Hennessy 1980; Poag 1982; Mal- one et al. 2002) and probed to assess water resources and chemical fluxes (Kohout et al. 1977; Meisler et al. 1984; Pope and Gordon 1999; Buxton and Modica 1992; Person et al. 1998, 2003). A series of boreholes were completed along the Atlantic continental shelf during the 1970s and 1980s as part of the Atlantic Margin Cor- ing (AMCOR) project (Hathaway et al. 1979), as continental offshore stratigraphic test (COST) wells (Schlee and Fritsch 1982), and most recently through the Ocean Drilling Program (ODP) Legs 150 and 174A (Mountain et al. 1994; Austin et al. 1998; Malone et al. 2002). Sedi- ments that accumulated on the Atlantic continental margin following the Mesozoic break-up of Africa and North America locally exceed 10 km in thickness. Tertiary and Cretaceous unconsolidated sands extend tens of kilome- ters offshore before grading into finer-grained facies. The Tertiary and Cretaceous shelf deposits of New England are capped by a roughly 100- to 200-m-thick layer of marine sediments and Pleistocene glacial tills and outwash sands. Multiple glacial/interglacial stages are documented on Nantucket and Martha’s Vineyard (Kaye 1964; Oldale et al., 1982). The permeability of medium to fine-grained sedimentary units along the continental shelf is estimated to range between about 10−10 to 10−18 m2 (Table 2). These have been measured through aquifer tests or estimated through model calibration. Most of the direct measurements were made in the near-shore environment where sediment burial depths are minimal; thus, offshore values may be overestimated. Hydrogeologic model calibration (e.g., Pope and Gordon 1999; Person et al. 2003) has been used to assess hydrogeologic properties of sediments far offshore. Sediment compressibility for the Pliocene-Oligocene sands has been estimated to be 4.4 × 10−7 Pa−1, which would yield a specific storage −2.3 −1 coefficient (Ss) of about 10 m (Dugan and Flem- Figure 3. (A) Areal footprint of three-dimensional paleohy- ings 2002). Pope and Gordon (1999) estimated the spe- drologic model (dashed purple line) along Atlantic conti- cific storage coefficient for deeper New Jersey confined nental shelf in New England and location of cross sections aquifer units to be about 10−6 m−1 based on calibration and wells presented in Figures 8 and 9. (B) Cross-sectional of their sharp interface model. Measured porosity of the transects depicting hydrostratigraphic units represented in three-dimensional model: silt/clay (blue), fine sand (red), and Pleistocene and Miocene sands ranged between 0.30 and medium to coarse sand (green) units. Note that the shal- 0.48 (Person et al. 1998). Off New Jersey, Miocene sands low sands crop out in submarine canyons. Imposed surface appear to be significantly overpressured and porosity has hydraulic heads (C) and topography/bathymetry of conti- been estimated to be as high as 0.7 (Dugan and Flemings nental shelf (D) during LGM, 21,000 years before present. 2002). Scholle (1977, 1980) reported porosity variations Areas of high imposed heads (greater than 250 m) delin- eate the approximate location of the Laurentide ice sheet in between 0.6 and 0.15 within COST wells B-2 and B-3 Figure 3C. for the mid-Atlantic region. Gill (1962) reported porosity variations within the Cohansey Sand off Cape May, New Jersey, to range from 0.41 to 0.28. − assuming a solute diffusion coefficient of 3 × 10 10 m2/s Estimates of modern recharge rates (∼0.5m/a)for and a confining unit thickness of 100 m), we must recon- New England using environmental isotopes (Knott and struct hydrogeologic conditions during the Pleistocene in Olimpio 1986), potential evapotranspiration estimates order to arrive at a representative account of present-day (Steenhuis et al. 1985), or regional groundwater flow salinity conditions. models (Guswa and LeBlanc 1981; Person et al. 1998; 4 D. Cohen et al. GROUND WATER NGWA.org Table 2 Aquifer and Confining Unit Properties from In Situ and Core Tests

Unit Permeability Range (m2) Hydraulic Conductivity Range (m/d)

Pleistocene glacial outwash sands 10−11.5 –10−10.1 22–70 Pliocene/Oligocene silt/clayey silt 10−18.4 –10−14.8 0.000001–0.0012 Cretaceous sandstones (Magothy and Raritian)1 10−12.7 –10−11.1 0.15–22

1Sources: Perlmutter and Geraghty 1963; Guswa and Le Blanc 1981; Buxton and Modica 1992; Person et al. 1998; Pope and Gordon 1999.

Pope and Gordon 1999) indicate that about 40% to 50% of three-dimensional transport processes, an extensive of the annual precipitation goes to recharge. On Nantucket regional scale (125,000 km2), and a relatively high spatial Island, Cape Cod, and Martha’s Vineyard, Pleistocene discretization (more than 5 million tetrahedral elements) unconsolidated outwash sands are the only source of fresh of the sediments that control the flow patterns in the water (Person et al. 1998). On Long Island and New Jer- subsurface. sey, deeper Cretaceous aquifers are also exploited for fresh water, because shallow aquifers have become contaminated by point and nonpoint source pollutants (Katz Mathematical Model et al. 1980) and shallow salt water intrusion (Schaefer We developed a high-resolution, three-dimensional, and Walker 1981). Water table elevations vary from 2 m variable-density groundwater flow model for this study above sea level on Nantucket Island (Person et al. 1998) (GW_ICE) to reconstruct the extent of fresh water on to 24 m on Long Island (Buxton and Modica 1992) and the Atlantic continental shelf in New England during the more than 38 m in New Jersey (Pope and Gordon 1999). Pleistocene. GW_ICE is a modified (parallelized) ver- As indicated earlier, confined offshore aquifers in Maine sion of the serial finite-element based paleohydrogeo- and Massachusetts receive no modern meteoric recharge logic model GEOFE (Person et al. 2007b). Documen- because they outcrop or subcrop below sea level. tation describing the governing transport equations for Radiocarbon ages for glacial tills of the continental groundwater flow, heat, solute transport, numerical imple- shelf off New England indicate that the Laurentide ice mentation, and partial model validation exercises for the sheet reached its southern most extent across Long Island serial version of this code (GEOFE) are available at and Nantucket about 21 ka before retreating onto the http:\\ww.ees.nmt.edu/person. mainland by approximately 14 ka (Uchupi et al. 2001) A continental shelf paleohydrologic model needs to (Figure 4). Ice sheet thickness was 1000 m near Cape re-create hydrologic conditions through time by incor- Cod and 500 m on Long Island (Denton and Hughes porating, as realistically as possible, the temporal evo- 1981). Several proglacial lakes formed following the last lution of sea level, temperature, subsea topography and, glacial maximum (LGM) as the Laurentide ice sheet where appropriate, glaciations. This requires representa- retreated (Uchupi et al. 2001; Mulligan and Uchupi 2003) tion of changes in land surface temperature, sea level (Figure 4). Seepage from proglacial lakes in Nantucket variations, glaciations, and salinity conditions on geologic Sound, Cape Cod Bay, Block Island Sound, and Long time scales. Island Sound could have provided extensive recharge to confined aquifers of the New England continental Groundwater Flow shelf (Marksamer et al. 2007). When these lakes burst, The groundwater flow equation solved in this study rapid sedimentation on the continental shelf would have represents the effects of variable-density flow, sea level changed the land surface morphology, induced rapid change, and ice sheet loading through time following the sediment loading, and perhaps induced infiltration of fresh approach of Provost et al. (1998): water. ∂h ρi ∂η A series of prior hydrologic models constructed ∇x · [Kμ ∇x (h + ρ z)] = S − (1) r r s ∂t ρ ∂t for the continental shelves off New Jersey (Meisler o et al. 1984; Pope and Gordon 1999), Long Island, and where ∇x is the gradient operator, K is the hydraulic Massachusetts (Person et al. 2003) concluded that both conductivity tensor, h is hydraulic head, η is ice sheet onshore and sub-ice-sheet recharge played important roles thickness, z is elevation, ρ is the density of water ◦ o in explaining the occurrence of sequestered fresh water. at the standard state (10 C, salinity of 0.0 mg/L, and Because of their restrictive simplifying assumptions (e.g., atmospheric pressure), ρr is the relative density (defined sharp interface theory, cross-sectional geometry), these below), ρi is the ice density, μr is the relative viscosity models lacked the ability to provide an accounting of (defined below), and Ss is the specific storage. Equation 1 the total amount of sequestered fresh water on the is similar to other variable-density flow models except that New England’s continental shelf. Our analysis overcomes it includes an ice sheet loading term (second term on right- this limitation by combining a detailed representation hand side of Equation 1). Equation 1 assumes a loading NGWA.org D. Cohen et al. GROUND WATER 5 Figure 4. Distribution of proglacial lakes (black) and position of Laurentide ice sheet (thick red lines) during Late Quaternary on the Atlantic continental shelf (after Uchupi et al. 2001). The following abbreviations are included: L.A. is Lake Amsterdam; L.At is Lake Attlebury; L.F. is Lake Fishkill; L.B. is Lake Bascom; L.BIS is Lake Block Island Sound; L.C. is Lake Connecticut; CH.S. is Champlain Sea; L.H. is Lake Hudson; L.Hi is Lake Hitchcock; L.K. is Lake Kiskaton; L.M. is Lake Marrimack; L.Mi is Lake Middletown; L.N. is Lake Narragansett; L.P. is Lake Passaic; L.S. is Lake Schohaire; L.Sc. is Lake Sacandaga; T. is Lake Tomhannock; L.T. is Lake Tilson; L.Ta. is Lake Tauton; L.W. is Lake Warrensberg; L.Wa. is Lake Warwarsing; and L.V. is Lake Vermont. efficiency of 1.0 (i.e., all of the load of the ice sheet permeability heterogeneity within individual lithologic is transferred to the fluid; Lemieux et al. 2008a). This units (Gelhar 1993). may overestimate induced pore pressure increases by this We solved a variable-density form of Darcy’s law mechanism at shallow depths (less than 300 m). However, (Garven and Freeze 1984) for the specific discharge at greater depths, a loading efficiency of 1 is probably vector: reasonable. Support for this assumption at shallow depths =− μ ∇ + q K r x [h ρrz](2) comes from preconsolidation stress data collected across North America and Europe, which indicate that tills where the relative density (ρr) and relative viscosity (μr) overrun by ice sheets are less consolidated than would terms are given by: be expected based on the reconstructed ice sheet loads. − = ρf ρo Estimated pore water pressures within glacial till based ρr (3) ρo on consolidation data are between 78% and 98% of = μo the ice-overburden pressure for maximum and minimum μr (4) μf ice thickness estimates (Piotrowski and Kraus 1997; ρ μ Hooyer and Iverson 2002). In other words, most of the where f is the density of groundwater, o is the viscosity of water at standard state, and μ is the viscosity of ice sheet weight was supported by elevated pore fluid f the fluid at elevated temperature, pressure, and salinity pressures. Our analysis neglects the effects of Pleistocene conditions. Because Atlantic shelf sediments have a sedimentation on simulated fluid pressures (see Dugan gentle slope, we neglected off-diagonal terms of the and Flemings 2000; Marksamer et al. 2007). Although hydraulic conductivity tensor (i.e., the principal directions this prevents us from accurately reconstructing anomalous of anisotropy are aligned with vertical and horizontal heads in fine-grained sediments, it probably has little directions). effect on the salinity distribution in relatively coarse- grained, near-shore sands. Our analysis also neglects Solute Transport subgrid scale density instabilities induced by buoyancy We used a conventional advective/dispersive equation effects (Simmons et al. 2001; Post and Kooi 2003) and to represent time-dependent transport of a conservative 6 D. Cohen et al. GROUND WATER NGWA.org = − solute: polynomials and P is pressure (P [h z]ρfg). These polynomial expressions are valid for temperatures between ∂C ◦ ◦ φ =∇x [φD∇x C] −q∇xC (5) 10 C and 150 C and salinities between 0 and 6 molal ∂t NaCl. Fluid density is less sensitive to changes in fluid where D is the standard three-dimensional hydrodynamic pressure than it is to changes in temperature and salinity dispersion-diffusion tensor (Konikow and Grove 1977) for the range of conditions encountered in sedimentary ◦ ◦ and C is solute concentration (solute mass fraction; kg basins. Between −3 C and 10 C, fluid density remains solute/kg solution). D is a function of solute diffusivity fixed at standard state conditions although permeability ◦ as well as longitudinal and transverse dispersivity. Our reduction prevents fluid flow from occurring below −2 C. approach neglects the potential effects of salt exclusion on salinity changes due to the effects of freezing (Starinsky Boundary Conditions and Katz 2003). Prior studies have used different strategies to represent the hydraulic boundary condition beneath the Heat Transport ice sheet, including specified flux (Provost et al. 1998; The thickness of continental shelf sediments (up to Breemer et al. 2002), specified head (Bense and Person 3 km in this study) requires that we represent the tempera- 2008), or a mixed boundary (Lemieux et al. 2008a). For ture effects on fluid density. GW_ICE solves a conductive this study, we specified a hydraulic head equivalent to and convective-dispersive heat transfer equation: 90% of the ice sheet height beneath the glacier (Boul- ∂T ton et al. 1995). Subglacial heads of this magnitude have [c ρ φ + c ρ (1 − φ)] =∇ [λ∇ T ] −qρ c ∇ T f f s s ∂t x x f f x been observed beneath ice streams (Engelhard and Kamb (6) 1997). We assume that recharge flux in excess of local melt rates is supplied along a network of sub-ice-sheet where λ is the thermal dispersion-conduction tensor; T channels, which were supplied by melt water from the is temperature; c and c are the specific heat capacities s f northeast along an esker network (Shreve 1985). Direct of the solid and liquid phases, respectively; and ρ s influx of glacial surface melt water can also be a source is the density of the solid phase. Permafrost during of sub-ice-sheet groundwater recharge. In Greenland, sur- Pleistocene glaciations can cause important reductions in face melt water lakes have been observed to drain rapidly permeability near the land surface (Kleinberg and Griffin through fracture networks in areas where the ice sheet 2005; Bense and Person 2008). In the model, the thermal thickness is less than 1000 m (van der Veen 2007; Zwally dispersion-conduction tensor is a function of the solid et al. 2002). The thickness of the ice sheet (η) was esti- thermal conductivity, the fluid thermal conductivity, the mated using a polynomial expression presented by van porosity, the longitudinal and transverse dispersivities, the der Veen (1998): fluid density, the fluid heat capacity, and the magnitude and direction of the Darcy flux (de Marsily 1986). For η = H 1 − (ξ/L)2 (9) all time steps where the atmospheric temperature was − ◦ below 2 C, surface elements in our model beyond the where H is the maximum ice sheet thickness at a ice sheet toe were assigned a permeability 1000 times particular time step, L is the ice sheet length at that time lower than their Holocene values (Kleinberg and Griffin step, ξ is the distance from the margin of the domain 2005) to represent near-surface freezing. Permeability margin in an upgradient direction, and η is ice sheet was reassigned to normal levels for these elements if thickness at distance ξ from the margin of the basin. temperatures rose above freezing in subsequent time steps. Both H and L were scaled using the Greenland Ice Core Because the surface mesh thickness (z) varied between Project (GRIP) sea level data (Figure 5) such that during 0.6 and 15 m, our permafrost algorithm would not allow the LGM, L, and H were at their maximum values. The for the development of a thick (greater than 100 m) maximum thickness of the ice sheet during the LGM is permafrost zone. shown in Figure 3C. The weight of the ice can cause lithosphere deflection (Peltier 1998). This resulted in a Equation of State reduction of the hydraulic heads assigned to nodes beneath Thermodynamic equations of state are used to com- the ice sheet on the continental shelf surface by as much pute the density and viscosity of groundwater at elevated as 100 m and diminished the hydraulic gradients beneath temperature, pressure, and salinity conditions. GW_ICE the ice sheet. While the full isostatic compensation of uses the polynomial expressions of Kestin et al. (1981): ice sheet loading is 28% (Marshall and Clarke 2002; Tarasov and Peltier 2004, 2007), the ice sheet was out on 1 = + + 2 + + 2 a(T ) b(T )P c(T )P Cd(T) C e(T ) the continental shelf for only a few thousand years and ρf probably did not attain isostatic equilibrium (Redfield and 2 h(T ) 2 −PCf(T)− C Pg(T)− P (7) Rubin 1962; Barnhardt et al. 1995). The deflection of the 2 = + land surface due to ice sheet loading at the LGM is shown μf μo[1 B(T,C)P](8) in Figure 3D. We used the ice sheet thickness to deflect where a(T ), b(T ), . . . , h(T ),andB(T,C) are third- and the lithosphere isostatically by 10%. Sea level in New fourth-order temperature- and concentration-dependent England was varied by 0 to −120 m relative to modern NGWA.org D. Cohen et al. GROUND WATER 7 finite element method. Advective solute transport was solved using a Lagrangian-based modified method of characteristics (MMOC; Zheng and Bennett 1994). A time step size of 10 years was used. We used the ParMETIS parallel mesh partitioning library (Karypis and Kumar 1996) and Aztec (Tuminaro et al. 1999) toolkits to iteratively solve the matrices formed in GW_ICE.To verify the performance of our parallel code, we performed Figure 5. Isotopic reconstruction of Late Pleistocene sea a scalability analysis using two test grids comprised of level (bold black line) and atmospheric temperature (red 104,758 and 1,277,068 nodes, respectively. The matrices line) vs. time in years before present (data source: Imbrie solved in the serial version used the SPLIB iterative et al. 1984 and Johnsen et al. 1995). solver (Bramley and Wang 1995). We enforced consistent parameters and solution schemes in both the serial and parallel versions. For the relatively coarse mesh, the conditions (Figure 5). Subaerial nodes not overrun by the speedup plateaus around 128 processors (Figure 6A). For ice sheet were assigned a hydraulic head equal to the land the more refined mesh (Figure 6B), we found that linear surface elevation. Nodes below sea level were assigned a speedup prevailed, even using 256 processors. For the head consistent with sea level elevation. A temperature of ◦ New England continental shelf models, we ran GW_ICE 4 C was specified for all nodes below sea level. This was on 256 processors using the NSF Teragrid (BigRed at based on modern records of deep (greater than 2000 m) Indiana University) and Encanto at the New Mexico ocean temperatures and is likely lower than modern Super Computer Application Center. The New England thermal conditions on the continental shelf. Land surface study described in the following paragraph required about temperatures for unglaciated regions were prescribed 5 days of central processing unit (CPU) time. using the continental temperature reconstruction estimated The convergence may have been faster and the from the SPECMAP and GRIP ice core record (Figure 5). ◦ total simulation time reduced had we solved all three Beneath the ice sheet, temperatures were fixed at 0 C. A equations simultaneously as it is done in some codes basal heat flux typical of the North American craton (Sass (e.g., Hayba and Ingebritsen 2001). However, this would et al. 1989) of 0.06 W/m2 was imposed along the base of have required replacing the MMOC method used in the model. Insulated boundary conditions were prescribed GW_ICE, which typically uses a smaller Courant-limited for all sides for heat transfer. Boundary conditions for time step size, with an upwinding scheme (e.g., Noorishad solute transport equations were a constant concentration of 0 or 35 ppt on the sea floor, depending on whether the local land surface elevation was above or below sea level. For submarine canyons, a spring or advective type (zero lateral concentration gradient) boundary condition was imposed.

Initial Conditions We assigned hydrostatic initial conditions for groundwater flow using the local elevation of the water table. Initial temperatures varied linearly with depth using a ◦ gradient of 30 C/km. Assigning initial salinity conditions was problematic because of the long response time (∼106 years) associated with solute diffusion. Salt water was ini- tially assigned to be present within all sediments. We then ran the model for 2 million years holding sea level constant at Late Pliocene levels (−53 m), thereby adjusting to the imposed pre-Pleistocene sea level conditions. At the start of the Pleistocene, we varied sea level by repeat- edly recycling the GRIP sea level record (Figure 5). We assume that the Laurentide ice sheet extended out onto the continental shelf in New England three times (Kaye 1964). We used the Wisconsinan ice sheet history on the Figure 6. Comparison of serial and parallel model performance. Parallel version of GW_ICE used Aztec solver pack- shelf as our guide to represent the Nebraskan and Illinoian age. The serial version (GEOFE) used SPLIB (Bramley and glacial cycles. Wang 1995). Both solver packages used the GMRES iterative scheme for this test. Model performance using a coarse Numerical Implementation (A, 104,758 nodes) and more refined (B, 1,277,068 nodes) grid is presented. Simulations were performed on NCSA I64 We solved variable-density groundwater flow, heat, cluster. The dashed line indicates optimal speedup. and solute transport equations sequentially using the 8 D. Cohen et al. GROUND WATER NGWA.org et al. 1992) to minimize numerical instabilities associated Results with advection-dominated solute transport. Upwinding A full model calibration (Hill 1998) to estimate all schemes tend to smear solute concentration fronts which parameters is impractical given the model’s complexity, we wanted to avoid. long simulation period, and regional scale, as well as the inaccessibility of most of the area represented by New England Grid the model domain to field measurements. Addressing the Our finite element model represents paleohydrologic problem as a hypothesis test, we conducted a limited conditions on the continental shelf (Figure 3A) from model calibration exercise in which permeability condi- New Jersey to Maine in both glaciated and unglaciated tions were varied for the three lithologic units over their areas. We divided the Quaternary, Tertiary, and Creta- known range of uncertainty (±102 m2). We did not vary ceous sediments into three lithologic units comprised of specific storage in our sensitivity study. We found that medium-coarse sands, fine sands, and silt/clays (Table 3; only a narrow range of permeability was able to reproduce Figure 3B). Our hydrogeologic framework model was observed salinity profiles. Confining unit permeability was based on existing onshore and offshore lithologic well found to be one of the most important factors controlling logs and aerial structure maps of Pleistocene-Upper Cre- offshore salinity distribution. If confining unit permeabil- taceous strata from Klitgord and Schneider (1994) and ity was too low, fresh water was restricted to the near- Klitgord et al. (1994). The model domain extends to a shore environment. Parameters resulting in the best fits to depth of about 3 km near the continental slope. The model salinity profiles are listed in Table 3. Our best-fit param- domain extends about 250 km offshore and about 750 km eters fall within the range of values reported from prior parallel to the coastline. Many of the deeper fine- and studies (Table 2) for aquifers and confining units. Param- medium-grained sand units are encased in finer-grained eters held constant for all three units include permeability silt and clay layers (e.g., cross section A-A’ in Figure 3B). anisotropy (ratio of vertical to horizontal permeability; The notable exceptions are in submarine canyons where 100), longitudinal dispersivity (100 m), transverse disper- − Miocene sands are exposed (Hampson and Robb 1984). sivity (10 m), and solute diffusivity (3 × 10 10 m2/s). The Tertiary and Cretaceous shelf deposits are capped Once the model was calibrated, we ran one additional by about 100- to 200-m-thick units of marine and Pleis- simulation in which the glacial loading effects were tocene glacial tills, outwash sands, and silt/clays. The removed. This was done to assess whether or not glacial stratigraphy used in our model was based on coastal loading influenced the distribution of fresh water on the borings (Kohout et al. 1977), AMCOR wells (Hathaway continental shelf in the vicinity of the ice sheet. This was et al. 1979), and COST wells (Mountain et al. 1994). accomplished by modifying the specified head boundary We include details of the continental shelf morphology condition, removing the effects of the ice sheet, and and stratigraphy at the 1- to 100-m scale. Using grid- setting the glacial loading term (∂η/∂t) in Equation 1 to generation algorithms developed at Los Alamos National zero. Labs (LaGriT©; Gable et al. 1996), we constructed a Computed heads for the model that included the high-resolution tetrahedral finite element mesh of the effects of glacial loading varied considerably laterally and Atlantic continental shelf using 5.8 million elements and with depth (Figure 7). Figure 7 presents computed head 987,995 nodes. The grid spacing in the x–y directions patterns across the top, southern, and seaward edges of was about 4 km per element. In the z-direction, grid spac- model domain. Prior to the LGM (37 ka in Figure 7), sea ing varied between 3 and 45 m. We chose this relatively level was about 80 m lower than today and heads were refined vertical discretization to ensure that there was a highest (30 m above modern sea level) in New Jersey and high enough grid resolution near sand–silt interfaces to in sand/silt units far offshore. The highest head conditions avoid numerical dispersion. We chose a minimum element in the offshore environment are found at depth in isolated width of 3 m for representing strata pinch outs in order sand units, which do not outcrop and which are within to avoid high element aspect ratios. overlying relatively thick, silt units. This is because of both density effects and remnant heads associated with higher sea level boundary conditions and prior (Illinoian) glacial loading. This same pattern is seen today (0 ka in Table 3 Figure 7). During the LGM (21 ka in Figure 7), glacial Hydrogeologic Properties Assigned in loading resulted in highest heads beneath the ice sheet Paleohydrologic Model (up to 700 m). High heads (up to about 30 m above 2 1 2 −1 3 sea level) also occurred in both aquifers and confining Unit kmax (m ) φ Ss (m ) units alike as the effects of glacial loading propagated Medium sand 10−12 0.2 10−3 out across the continental shelf. This was facilitated by − − Fine sand 10 14 0.2 10 4 low permeability conditions associated with permafrost − − ◦ Silt/clay 10 17 0.3 10 4 formation (temperatures below −2 C) beyond the toe of the ice sheet in periglacial areas. At 37 ka, sea level 1Permeability. 2Porosity. was 80 m lower than present, exposing a large portion of 3Specific storage coefficient. the continental shelf to meteoric recharge and enhancing the shore-normal hydraulic gradient. The topographic NGWA.org D. Cohen et al. GROUND WATER 9 Figure 8. Simulated concentrations (in parts per thousand) along cross sections extracted from the three-dimensional finite element model of the Atlantic continental shelf. Cylin- ders depict concentration of offshore AMCOR, COST, and ODP wells (well radii not to scale). The wells were raised 500 m above the sea floor in this image, so that the cross sections would not obscure them. The locations of the cross sections and wells are presented in Figure 3A.

brackish water is predicted to be present in the basal sand unit across eastern Long Island (section C-C’, Figure 8). Stumm (2001) documents that salt water intrusion has Figure 7. Changes in computed heads (in meters) during occurred in response to pumping during the 1970s within Late Pleistocene and Holocene along the Atlantic continental the Lloyd aquifer (basal Cretaceous sand, section B-B’) shelf in New England. The locations of New Jersey (NJ), along Long Island’s northwestern shore on the Great Neck Long Island (LI), and Nantucket Island (NT) are listed to Peninsula. The more significant discrepancy between the provide a geographical frame of reference. The dashed white line represents the position of the coastline through time. model and field observations occurs along the southern shore of western Long Island, where Perlmutter et al. (1959) reported the presence of sea water within the lowlands to the north and east of Cape Cod drove upper Pleistocene outwash sands, brackish water within groundwater northeastward from Massachusetts toward the upper Cretaceous Magothy aquifer, and fresh water Maine. Flow directions were reversed as the Laurentide within the basal Cretaceous sands (Lloyd aquifer) along ice sheet overran the continental shelf at 21 ka. As sea southwestern Long Island. Theareawheresaltwateris level continued to rise and the ice sheet retreated, meteoric observed in the Magothy-Jameco aquifer has complex recharge became restricted to New Jersey and Long hydrogeology with local erosion features that may play Island. Heads in excess of modern sea level (+30 m) a role in the presence of salt water (Buxton and Shernoff are indicated at depth today (Figure 7), due in part to 1999). It appears that our model overpredicts the extent of density effects and to fossil heads associated with ice sheet fresh water beneath southwestern Long Island extending loading (Marksamer et al. 2007). under the ocean, although predevelopment salinity data Fresh water is sequestered in sediments from the are not available. On the other hand, predevelopment present-day coastline to about 50 km offshore. Within salinity data of Pope and Gordon (1999) suggest that this region, fresh water occurs within both aquifers and our model appears to underpredict the amount of fresh confining units (Figure 8), matching observations from water within the Upper Cretaceous sands beneath New Martha’s Vineyard (well ENW-50 in Figure 9). Between Jersey. Within the shallow Miocene sand unit off New about 50 to 100 km offshore, a complex transition zone Jersey (section A-A’, Figure 8), fresh to brackish water exists, characterized by brackish to fresh water within is transported to 100 km offshore. This is because these sand units and brackish to near sea water in confining shallow sands outcrop in Baltimore and Hudson Canyons, units. This transition zone is observed beneath Nantucket creating submarine springs which discharged brackish Island (Figure 9, well 6001) wells. Our model predicts water to the ocean during sea-level low stands. these patterns (section D-D’, Figure 8) in Massachusetts, Models for which the effect of glaciation was but places them farther offshore. The position of this removed predicted higher salinity in regions proximal to transition zone is variable from south to north along the the ice sheet toe (e.g., blue dashed lines, Figure 9). For shelf. Our model predicts fresh water conditions beneath several wells (6014, 6011, 6001), the computed salini- western Long Island (section B-B’, Figure 8). Some ties in the upper 200 m of the sea floor were about 5 ppt 10 D. Cohen et al. GROUND WATER NGWA.org Figure 9. Comparison of simulated (lines) and observed (red squares) concentrations (in parts per thousand) for offshore AMCOR, COST, and ODP wells on Atlantic continental shelf. Simulations including (ice sheet, solid black line) and removing (sea level, blue dashed line) the effects of glacial loading are presented. The locations of the wells are presented in Figure 3A. higher than for the model run, where the effects of glacia- rates occur due to enhanced shore-normal hydraulic gra- tions were removed. In one well (6016), the computed dients associated with sea-level low stands (Figure 10B). salinities were 15 ppt saltier when glaciations effects were Along the continental slope, paleo submarine springs removed. This suggests that glaciations had a pronounced discharge groundwater (about 1 cm/year) along Hudson effect on continental shelf salinities proximal to the ice and Baltimore Canyons (Figure 10C), where Miocene sheet toe. However, some wells far from the ice sheet sands outcrop (Hampson and Robb 1984). Flow direc- (wells 6009, 6010, 6015, 6018) were not impacted by tions fluctuate through geologic time within these canyons glaciation. We calculated the volume of sequestered fresh (Figure 10C–E). During periods of sea-level low stands water in New England to be about 1300 km3 by inte- and glaciation, brackish (16 ppt) groundwater discharge grating over the sand-dominated elements in our model occurs (Figure 10C). Today, sea water invades these sands containing fresh water of less than 1 ppt. along the canyon (Figure 10D) and in the near-shore environment (red areas of along the sea floor in Figure 10D), During the modeled LGM, predicted recharge/ where fresh water sands outcrop at the sediment–water discharge rates in the vicinity of the ice sheet reached interface, density-driven convection cells (Simmons et al. about 2 m/year, about four times higher than modern esti- 2001; Post and Kooi 2003) are seen in the model (areas mates for Nantucket Island (Knott and Olimpio 1986). below sea level with alternating fresh and sea water pat- High hydraulic heads predicted near the toe of the ice terns in Figure 10D). Calculated temporal (Figure 10E) sheet are consistent with glaciotectonic features observed variations in submarine canyon discharge during sea- within tills on Nantucket Island and Martha’s Vineyard level low stands of the LGM help explain why fresh (Oldale 1985; Oldale and O’Hara 1984; Boulton et al. water extends farther off New Jersey than in areas far- 1995). There is a consistent pattern of recharge beneath ther to the north (Johnston 1983). Computed discharge the ice sheet and focused discharge just beyond the ice rates (1 cm/year) are consistent with groundwater sapping sheet toe (Figure 10A). Permeable sand units helped prop- features observed during submersible dives along Balti- agate the effects of high imposed heads beneath the more Canyon off New Jersey (Robb 1984). It appears ice sheet outward over a wide region beyond the ice that submarine canyons influence the fresh water distri- sheet toe, thereby increasing flow rates. In unglaciated bution only in relatively shallow, permeable deposits that regions during the LGM, higher than modern recharge outcrop (section A-A’, Figure 3). NGWA.org D. Cohen et al. GROUND WATER 11 Stumm (2001) reports salt water intrusion occurred following about two decades of heavy (more than 3 billion gallons/year) pumping within the Magothy and Lloyd aquifers on the Great Neck Peninsula, Long Island. It should be noted, however, that the Lloyd aquifer termi- nates (and crops out) along Long Island Sound proximal (2 to 3 km) to these well fields. The more problematic area is along the southern shore of western Long Island, where saline water has been observed in the Magothy-Jameco aquifer since measurements have been made (Buxton and Shernoff 1999). Possible erosional breaks in the confining unit overlying the Magothy aquifer (not represented in our regional scale model) could be the cause of this observed salt water intrusion. However, it is important to note that our hypothesis testing model is regional in scale and it is not possible to perfectly match local salinity variations everywhere without excess parameter tuning, which we elected not to pursue. Our model results indicated that glacial loading reduced the salinity of groundwater in wells situated near the ice sheet margin between about 5 to 15 ppt. Wells far from the ice sheet margin in New Jersey were not significantly impacted by glaciations. Some of these New Jersey wells (e.g., well 6009, Figure 9) had remarkably Figure 10. Computed magnitude of surface groundwater low salinity within shallow Miocene sands and were flux (in log m/year) during the last glacial maximum (LGM; located far (greater than 100 km) offshore. Our model A) and present day (B). Computed groundwater salinity results indicate that this was caused by enhanced offshore (total dissolved solids in parts per thousand) concentration flow in these shallow sands which cropped out in Hudson during the LGM (C) and present day (D). Computed temporal variations in vertical groundwater flux along Hudson Canyon. Our finite element model results suggest that both Canyon (E) during the last glacial cycle (positive for dis- ice sheet loading and sea-level low stands have played a charge, negative for salt water reflux). The location of the significant role in the emplacement of the offshore fresh discharge point is shown in Figure 10D (black circle). The groundwater on New England’s continental shelf. We locations of New Jersey (NJ), Long Island (LI), and Nan- believe that the occurrence of distal offshore fresh water is tucket Island (NT) are listed in Figure 10D to provide a geographical frame of reference. primarily controlled by sea level fluctuations and proximal fresh water is most strongly ice influenced by ice sheets at high latitudes. This suggests that data from different parts of the continental shelf are important to disentangle Discussion and Conclusions and constrain fresh water emplacement mechanisms. Our back-of-the-envelope and model-based estimates While our paleohydrologic model represents one of provide only a first-cut assessment of the potential for the largest (spatially) and highest resolution regional offshore fresh water. These calculations require field ver- groundwater models ever developed, it is idealized; ification. Porosity on the Atlantic continental shelf varies nonetheless, it facilitates useful insight on the emplace- between about 0.6 and 0.1. If we vary the porosity ment mechanisms, distribution, and volume of offshore between 0.1 and 0.4, the resulting estimated fresh water fresh water. We did not account for surface water drainage volume ranges from 500 to 2000 km3 for New England. patterns (Lemieux et al. 2008a) or the effects of Plio- Uncertainty regarding the thickness of fresh water satu- Pleistocene sedimentation on anomalous pore pressures rated sediments and the seaward extent of the fresh water (Dugan and Flemings 2000). We did not account for lens could also change our estimates of fresh water vol- local sea level fluctuations associated with the effects ume, probably by a factor of ±2. In addition, these esti- of a proglacial forebulge (Barnhardt et al. 1995). It is mates do not consider how much water can be extracted. likely that during periods of glaciation, high hydraulic As in petroleum extraction, practical recovery is likely to heads resulted in enhanced permeabilities due to low be substantially less than 100%. Finally, there is consid- effective stress (Boulton et al. 1995), as evidenced erable variability in offshore fresh water estimates among by glaciotectonic features observed on Nantucket and these four sections (12 to 0.2 km3/km). Hydrologic and Martha’s Vineyard (Oldale and O’Hara 1984). (or) geophysical field data collected in the future could be Overall, our model did a reasonable job of predict- used to refine the model, resolve more heterogeneity in ing offshore well salinity profiles (Figure 9). As noted hydraulic properties, and reduce prediction uncertainty. earlier, our model may have overpredicted the extent of With demands for fresh water increasing in coastal fresh water beneath Long Island’s western shorelines. areas, sequestered continental shelf fresh water represents 12 D. Cohen et al. GROUND WATER NGWA.org a large, valuable, untapped, albeit nonrenewable resource. M. Katz, H. Krawinkel, C. Major, F. McCarthy, C. Mc- Many coastal megacities (e.g., Karachi, Pakistan; Rahman Hugh, G. Mountain, H. Oda, H. Olson, C. Pirmez, et al. 1997) are facing acute water shortages, and local C. Savrda, C. Smart, L. Sohl, P. Vanderaveroet, W. Wei, groundwater resources are becoming contaminated. These and B. Whiting. 1998. Continuing the New Jersey Mid- problems will become worse under conditions of climate Atlantic sea-level transect; covering Leg 174A of the cruises of the drilling vessel JOIDES Resolution, Halifax, change (Nicholls 2004). To put these fresh water estimates Nova Scotia, to New York, New York. Proceedings of the into perspective, during the 20th century the cumulative Ocean Drilling Program, Part A: Initial Reports 174A: volume of fresh water withdrawn from the High Plains 1071–1073. Reston, Virginia: USGS. Aquifer in the United States for agricultural and other Barnhardt, W.A., W.R. Gehrels, D.F. Belknap, and J.T. Kelley. 3 purposes is estimated to be about 270 km (Konikow 1995. Late Quaternary relative sea-level change in the 2002). The volume of useable offshore water could be western Gulf of Maine: Evidence for a migrating glacial actually much greater than stated earlier if reverse osmosis forebulge. Geology 23, no. 4: 317–320. facilities (Schiermeier 2008; Shanon et al. 2008) were Bense, V.F., and M.A. Person. 2008. Transient hydrodynam- used to process brackish (less than 10 ppt) offshore ics within intercratonic sedimentary basins during glacial groundwater. It is important to note, however, that in cycles. Journal of Geophysical Research 113, 1–17. many coastal settings, this fresh water is not being Boulton, G.S., P.E. Caban, and K. van Gijssel. 1995. Ground- replenished today. Pumping of offshore aquifers will water flow beneath ice sheets. Part 1, large scale patterns. Quaternary Science Reviews 14, no. 6: 545–562. deplete the resources and will eventually lead to salt water Bramley, R., and X. Wang. 1995. A User’s Guide for the intrusion. CADYF-SPLIB Project. Bloomington: Department of Com- Development of these resources would be expensive puter Science, Indiana University. ftp://ftp.cs.indiana.edu/ and would require offshore drilling and construction of pub/techreports/TR453.ps.Z. pipelines to transfer water inland, and/or the use of Breemer, C.W., P.U. Clark, and R. Haggerty. 2002. Modeling modern land-based petroleum drill rigs capable of drilling the subglacial hydrology of the late Pleistocene Lake large-scale (greater than 10 km) horizontal wells to access Michigan Loge, Laurentide Ice Sheet. Geological Society the offshore resource. Marine seismic surveys would of America Bulletin 114, no. 6: 665–674. need to be conducted to ensure that drilling hazards Buxton, H.T., and P.K. Shernoff. 1999. Ground-water resources are avoided, such as natural gas deposits hosted in of Kings and Queens Counties, Long Island, New York. structural traps. In addition, extensive hydrologic and U.S. Geological Survey Water-Supply Paper 2498. Reston, geophysical characterization would be necessary to map Virginia: USGS. Buxton, H., and E. Modica. 1992. Patterns and rates of ground- the distribution of groundwater salinity offshore and water flow on Long Island, New York. GroundWater 30, identify viable drilling targets. Electromagnetic methods no. 6: 857–866. (e.g., transient and frequency-domain controlled source Denton, G.H., and T.J. Hughes. 1981. The Last Great Ice Sheets. techniques) are well suited to the first task (e.g., Hoefel New York: John Wiley & Sons. and Evans 2001), whereas marine multichannel seismic de Marsily, G. 1986. Quantitative Hydrogeology, Groundwater methods are well suited to the second. Hydrology for Engineers. New York: Academic Press. Our study may have implications for estimating sub- Dugan, B., and P.B. Flemings. 2002. Fluid flow and stability marine groundwater discharge to oceans because cur- of the US continental slope offshore New Jersey from the rent levels reported in the literature (e.g., Li et al. 1999; Pleistocene to the present. Geofluids 2, no. 2: 137–146. Taniguchi 2002; Michael et al. 2005) may not be represen- Dugan, B., and P.B. Flemings. 2000. Overpressure and fluid tative of long-term (i.e., Pleistocene) average conditions flow in the New Jersey continental slope: implications for slope failure and cold seeps. Science 289, 288–291. (Figure 10). Edmunds, W.M. 2001. Paleowaters in European coastal aquifers: The goals and main conclusions of the PALAEAUX Project. Geological Society Special Publica- Acknowledgments tions 189, 1–16. This work was supported by the U.S. National Engelhard, H., and B. Kamb. 1997. Basal hydraulic system of Science Foundation and the Malcolm & Sylvia Boyce a west Antarctic ice stream: Constraints from borehole Endowment at Indiana University. We gratefully acknowl- observations. Journal of Glaciology 43, 207–230. edge computational support from the NSF Teragrid Folger, D.W., R. Hathaway, R.A. Christopher, P. Valentine, and Program at Indiana University and the New Mexico C. Poag. 1978. Stratigraphic test well. Nantucket Island, NMCAC supercomputer center, and support from the Massachusetts. U.S. Geological Survey Circular C 773. USGS Groundwater Resources Program. We thank Jean- Reston, Virginia: USGS. Gable, C.W., H.E. Trease, and T.A. Cherry. 1996. Geological Michel Lemieux and Steve Ingebritsen for their thorough applications of automatic grid generation tools for finite reviews of this paper. elements applied to porous flow modeling. Numerical Grid Generation in Computational Fluid Dynamics and Related Fields: 1007-1015. References Garven, G., and R.A. Freeze. 1984. Theoretical analysis of the Austin, J.A., N. Christie-Blick, M. Malone, S. Berne,´ M.K. role of groundwater flow in the genesis of stratabound ore Borre, G. Claypool, J. Damuth, H. Delius, G. Dickens, deposits. 1. Mathematical and numerical model. American P. Flemings, C. Fulthorpe, S. Hesselbo, K. Hoyanagi, Journal of Science 284, 1085–1124. NGWA.org D. Cohen et al. GROUND WATER 13 Gelhar, L.W. 1993. Subsurface Stochastic Hydrology. Engle- Kestin, J., H.E. Khalifa, and R. Corriea. 1981. Tables of the wood Cliffs, New Jersey: Prentice Hall. dynamics and kinematic viscosity of aqueous NaCl solu- ◦ Gill, H.E. 1962. Ground-water resources of Cape May County, tions in the temperature range of 20–150 Candthepres- New Jersey: Salt-water invasion of principal aquifers. sure range of 0.1–35 MPa. Journal of Physical Chemistry New Jersey Department of Conservation and Economic Reference Data 10, 71–87. Development Special Report 18. Kleinberg, R., and D. Griffin. 2005. NMR measurements of Guswa, J.H., and D.R. Leblanc. 1981. Digital models of ground- permafrost: Unfrozen water assay, pore-scale distribution of water flow in the Cape Cod aquifer system, Massachusetts. ice, and hydraulic permeability of sediments. Cold Regions Water Resources Investigation Open-File Report: 80-67. Science and Technology 42, 63–77. Hampson, J., and J. Robb. 1984. A geologic map of the conti- Klitgord, K.D., and C.M. Schneider. 1994. Geophysical nental slope off New Jersey: Lindenkohl Canyon to Toms database of the East Coast of the United States, northern Canyon. U.S. Geological Survey Miscellaneous Investiga- Atlantic margin: Velocity analyses. U.S. Geological Survey tions Map I-1608. Reston, Virginia: USGS. Open-File Report: 94-192. Reston, Virginia: USGS. Hathaway, J.C., C.W. Poag, P.C. Valentine, R.E. Millerk, D.M. Klitgord, K.D., C.W. Poag, C.M. Schneider, and L. North. Schultz, F.T. Manheim, F.A. Kohout, M.H. Bothner, and 1994. Geophysical database of the East Coast of the United D.A. Sangrey. 1979. U.S. Geological Survey core drilling States, northern Atlantic margin: Cross sections and gridded on the Atlantic Shelf, 1979. Science 206, no. 4418: database (Georges Bank basin, Long Island platform, and 515–527. Baltimore Canyon trough). U.S. Geological Survey Open- Hayba, D.O., and S.E. Ingebritsen. 2001. The computer model File Report: 94-637. Reston, Virginia: USGS. HYDROTHERM, a three-dimensional finite-difference Knott, J.F., and J.C. Olimpio. 1986. Estimation of recharge rates model to simulate ground-water flow and heat transport in to sand and gravel aquifer using environmental tritium. the temperature range of 0 to 1,200 degrees C. U.S. Geo- U.S. Geological Survey Water-Supply Paper 2297. Reston, logical Survey Water-Resources Investigations 94-4045. Virginia: USGS. Reston, Virginia: USGS. Kohout, F.A., J.C. Hathaway, D.W. Folger, M.H. Bothner, E.H. Walker, D.F. Delaney, M.H. Frimpter, E.G.A. Weed, and Henry, H. 1959. Saltwater intrusion into fresh-water aquifer. E.V.C. Rhodehamel. 1977. Fresh groundwater stored in Journal of Geophysical Research 64, no. 11: 1911–1919. aquifers under the continental shelf, Implications from Hill, M.C. 1998. Methods and guidelines for effective model cal- a deep test, Nantucket Island, Massachusetts. Water ibration. U.S Geological Survey Water- Resources. Inves- Resources Bulletin 13, no. 2: 373–386. tigations Report 98-4005. Reston, Virginia: USGS. Konikow, L.F. 2002. Ground-water depletion and overexploita- Hoefel, F, and R. Evans. 2001. Impact of low-salinity porewater tion; a global problem. Geological Society of America on seafloor electromagnetic data: A means of detecting Abstracts with Programs 34, 229. submarine groundwater discharge? Estuarine Coastal and Konikow, L.F., and D.B. Grove. 1977. Derivation of equations Shelf Science 52, 179–189. describing solute transport in groundwater. U.S. Geologi- Hooyer, T.S., and N.R. Iverson. 2002. Flow mechanism of the cal Survey Water-Resources Investigations 77-19. Reston, Des Moines lobe of the Laurentide ice sheet. Journal of Virginia: USGS. Glaciology 48, 575–587. Kooi, H., and K. Groen. 2001. Offshore continuation of coastal Imbrie, J., J. Hays, D. Martinson, A. McIntyre, A.C. Mix, J. groundwater systems; predictions using sharp-interface Morley, N.G. Pisias, W.L. Prell, and N.J. Shackleton. approximations and variable-density flow modelling. Jour- 1984. The orbital theory of Pleistocene climate: Support nal of Hydrology 246, 19–35. 18 from a revised chronology of the marine δ O record. Kooi, H J., K. Groen, and A. Leijnse. 2000. Modes of seawater In Milankovitch and Climate, 269–305. Boston, Mas- intrusion during transgressions. Water Resources Research sachusetts: D. Reidel Publishing Company. 36, no. 12: 3581–3589. Johnsen, S.J., D. Dahl-Jensen, W. Dansgaard, and N. Gunde- Lemieux, J.-M., E. Sudicky, W. Peltier, and L. Tarasov. 2008a. strup. 1995. Greenland paleotemperatures derived from Simulating the impact of glaciations on continental ground- GRIP borehole temperature and ice core isotope profiles. water flow systems. I. Relevant processes and model for- Tellus 47B, 624–629. mulation. Journal of Geophysical Research 113, F03017. Johnston, R. 1983. The saltwater-freshwater interface in the Ter- Lemieux, J.-M., E. Sudicky, W. Peltier, and L. Tarasov. 2008b. tiary limestone aquifer, Southeast Atlantic outer-continental Simulating the impact of glaciations on continental ground- shelf of the U.S.A. Journal of Hydrology 61, no. 1–3: water flow systems. II. Model application to the Wiscon- 239–249. sinian glaciation over the Canadian landscape. Journal of Karypis, G., and V. Kumar. 1996. A parallel algorithm for mul- Geophysical Research 113, F03018. tilevel mesh partitioning and sparse matrix ordering. 10th Lemieux, J.M., E.A. Sudicky, W.R. Peltier, L. Tarasov. 2008c. International Parallel Processing Symposium, Honolulu, Dynamics of groundwater recharge and seepage over the Hawaii, 314–319. Canadian landscape during the Wisconsinian glaciation. Katz, B.G., J.B. Lindner, and S.E. Ragone. 1980. An evaluation Journal of Geophysical Research 113, F01011. of sources of nitrogen as causes of ground-water contam- Li, L., D.A. Barry, F. Stagnitti, and J.Y. Parlange. 1999. Subma- ination in Nassau County, Long Island. GroundWater 18, rine groundwater discharge and associated chemical input 607–616. to a coastal sea. Water Resources Research 35, no. 11: Kaye, C.A. 1964. Outline of pleistocene geology of Martha’s 3253–3259. Vineyard, Massachusetts. U.S. Geological Survey Pro- Malone, M., M. Claypool, G. Jonathan, and G.R. Dickens. fessional Paper P0501-C. C134–C139. Reston, Virginia: 2002. Variable methane fluxes in shallow marine systems USGS. over geologic time: The composition and origin of pore 14 D. Cohen et al. GROUND WATER NGWA.org waters and authigenic carbonates on the New Jersey shelf. Perlmutter, N.M., and J.J. Geraghty. 1963. Geology and ground- Marine Geology 189, no. 3–4: 175–196. water conditions in southern Nassau and southeastern Marksamer, A.J., M. Person, F. Day-Lewis, J.W. Lane, D. Queens Counties, Long Island, New York. U.S. Geological Cohen, B. Dugan, H. Kooi, and M. Willett. 2007. Integrat- Survey Water Supply Paper 1613-A. Reston, Virginia: ing geophysical, hydrochemical, and hydrologic data to USGS. understand the freshwater resources on Nantucket Island, Perlmutter, N.M., J.J. Geraghty, and J.E. Upson. 1959. The Massachusetts. In: Subsurface Hydrology Data Integration relation between fresh and salty groundwater in southern in Subsurface Hydrology, 1–54. Washington, DC: AGU Nassau and southeastern Queens Counties, Long Island, Water Resources Monograph. New York. Economic Geology 54, 416–435. Marshall, S.J., and P. Clark. 2002. Basal temperature evolution Person M., J. McIntosh, V. Bense, V.H. Remenda. 2007a. Pleis- of North American ice sheets and implications for the tocene hydrology of North America: The role of ice sheets 100-kyr cycle. Geophysical Research Letters 29, 769–793. in reorganizing groundwater flow systems. Reviews of Geo- Mattick, R.E., and J.L. Hennessy. 1980. Structural framework, physics 45, 1–28. stratigraphy, and petroleum geology of the area of oil and Person, M., P. Roy, H. Wright, E. Ito, T. Winter, D. Rosen- gas lease sale No. 49 on the U.S. Atlantic continental shelf berry, W. Gutowski, D. Cohen, and P. Morin. 2007b. and slope. U.S. Geological Survey Circular C812. Reston, Hydrologic response of the Crow Wing watershed, Min- Virginia: USGS. nesota to mid-Holocene climate change. Geological Society Meisler, H., P.P. Leahy, and L.L. Knobel. 1984. Effect of of America Bulletin 119, no. 3–4: 363–376. Eustatic sea-level changes on saltwater-freshwater in North Person, M., B. Dugan, J. Swenson, L. Urbano, C. Stott, J. Tay- Atlantic Coastal Plain. U.S. Geological Survey Water- lor, and M. Willett. 2003. Pleistocene hydrogeology of the Supply Paper 2255. Reston, Virginia: USGS. Atlantic continental shelf, New England. Geological Soci- Michael, H.A., A.E. Mulligan, and C.F. Harvey. 2005. Seasonal ety of America Bulletin 115, 1324–1343. oscillations in water exchange between aquifers and the Person, M., J. Taylor, and S.L. Dingman. 1998. Sharp interface coastal ocean. Nature 436, no. 7054: 1145–1148. models of salt water intrusion and well head delineation Mountain, G.S., K.G. Miller, P. Blum, P.G. Alm, M.P. on Nantucket Island, Massachusetts. GroundWater 36, Aubry, L.H. Burckle, B.A. Christensen, J. Compton, J.E. 731–742. Damuth, J.F. Deconinck, L. de Verteuil, C.S. Fulthorpe, S. Piotrowski, J.A., and A.M. Kraus. 1997. Response of sediments Gartner, G. Guerin, S.P. Hesselbo, B. Hoppie, M.E. to ice-sheet loading in northwestern Germany: Effective Katz, N. Kotake, J.M. Lorenzzo, S. McCracken, C.M. stress and glacier-bed stability. Journal of Glaciology 145, McHugh, W.C. Quayle, Y. Saito, S.W. Snyder, W.G. ten 495–502. Kate, M. Urbat, M.C. Van Vossen, and A. Vecsei. 1994. Poag, C.W. 1982. Stratigraphic reference section for Georges New Jersey continental slope and rise; covering Leg 150 Bank Basin-depositional model for New England Passive of the cruises of the drilling vessel JOIDES Resolution, Margin. American Association of Petroleum Geologists Lisbon Harbor, Portugal, to St. John’s Newfoundland. In Bulletin 66, no. 8: 1021–1041. Proceedings of the Ocean Drilling Program, Part A: Initial Pope D.A., and A.D. Gordon. 1999. Simulation of ground- Reports, 902-906. College Station, Texas, Ocean Drilling water flow and movement of the freshwater-saltwater Program. interface in the New Jersey Plain. U.S. Geological Survey Mulligan, A., and E. Uchupi. 2003. New interpretation of glacial Investigations Report 98-4216. Reston, Virginia: USGS. history of Cape Code may have important implications for Post, V.E.A., and H. Kooi. 2003. On rates of salinization by free groundwater contaminant transport. EOS Transactions 84, convection in high-permeability sediments; insights from no. 19: 177–183. numerical modeling and application to the Dutch coastal Nicholls, R. 2004. Coastal megacities and climate change. area. Hydrogeology Journal 11, 549–559. GeoJournal 37, 369–379. Provost, A.M., C.I. Voss, and C.E. Neuzil. 1998. SKI SITE- Noorishad, J., C.F. Tsang, P. Perrochet, and A. Musy. 1992. 94 Glaciation and regional ground-water flow in the A perspective on the numerical solution of convection- Fennoscandian Shield. SKI Report 96. Stockholm: Swedish dominated transport problems: A price to pay for the easy Nuclear Power Inspectorate. way out. Water Resources Research 28, no. 2: 551–561. Rahman, A., H. Lee, and M. Khan. 1997. Domestic water Oldale, R.N. 1985. Geologic map of Nantucket and nearby contamination in rapidly growing megacities of Asia: Case islands, Massachusetts, Miscellaneous Investigations of Karachi, Pakistan. GeoJournal 44, 339–360. Series. U.S. Geological Survey Report I-1580. Reston, Redfield, A.C., and M. Rubin. 1962. The age of salt marsh Virginia: USGS. peat and its relation to recent changes in sea level in Oldale, R.N., and C.J. O’Hara. 1984. Glaciotectonic origin of Barnstable, Massachusetts. Proceedings of the National the Massachusetts coastal end moraines and a fluctuating Academy of Sciences of the United States of America 48, late Wisconsinan ice margin. Geological Society of America no. 10: 1728–1735. Bulletin 95, 61–74. Robb, J. 1984. Spring sapping on the lower continental slope, Oldale, R.N., P.C. Valentine, T.M. Cronin, E.C. Spiker, B.W. offshore New Jersey. Geology 12, no. 5: 278–282. Blackwelder, D.F. Belnap, J.F. Wehmiller, and B.J. Szabo. Sass, J.H., D.D. Blackwell, D.S. Chapman, J.K. Costain, E.R. 1982. Stratigraphy, structure, absolute age, and pale- Decker, L.A. Lawver, C.A. Swanberg, D. Blackstone, C. ontology of the upper Pleistocene deposits at Sankaty Brott, H. Heasler, A. Lachenbruch, B.V. Marshall, P. Mor- Head, Nantucket Island, Massachusetts. Geology 10, no. 5: gan, R. Munroe, and J. Steele. 1989. Heat flow from the 246–252. crust of the United States. Physical Properties of Rocks Peltier, W.R. 1998. Postglacial variations in the level of the and Minerals 2, no. 2: 503–548. sea: Implications for climate dynamics and solid-earth Schiermeier, Q. 2008. Purification with a pinch of salt. Nature geophysics. Reviews of Geophysics 36, 603–689. 452, no. 7185: 260–261. NGWA.org D. Cohen et al. GROUND WATER 15 Schaefer, F.L., and R.L. Walker. 1981. Saltwater intrusion into Island, New York, U.S.A. Journal of Hydrology 79, the Old Bridge Aquifer in the Keyport-Union Beach area no. 1–2: 145–169. of Monmouth County, New Jersey. U.S. Geological Sur- Stumm, F. 2001. Hydrogeology and extent of saltwater intrusion vey Water-Supply Paper Report W2184. Reston, Virginia: of the Great Neck Peninsula, Great Neck, Long Island, New USGS. York. U.S. Geological Survey Water-Resources Investiga- Schlee, J.S., and J. Fritsch. 1982. Seismic stratigraphy of the tions Report 99-4280. Reston, Virginia: USGS. Georges Bank Basin complex offshore New England. Tarasov, L., and W.R. Peltier. 2007. Coevolution of continental American Association of Petroleum Geologists Memoir 43, ice cover and permafrost extent over the last glacial- 223–251. interglacial cycle in North America. Journal of Geophysical Scholle, P.A. 1980. Geologic studies on the COST No. B-3 Research 112, F02S08. Well, U.S. Mid-Atlantic Outer Continental Shelf Area. U.S. Tarasov, L., and W.R. Peltier. 2004. A geophysically con- Geological Survey Circular 833. Reston, Virginia: USGS. strained large ensemble analysis of the deglacial history of Scholle, P.A. 1977. Geologic studies on the COST No. B-2 the North American ice-sheet complex. Quaternary Science Well, U.S. Mid-Atlantic Outer Continental Shelf Area. U.S. Reviews 23, 359–388. Geological Survey Circular 750. Reston, Virginia: USGS. Taniguchi, M. 2002. Tidal effects of submarine groundwater Shanon, M.A., P.W. Bohn, M. Elimelech, J.G. Georgiadis, B.J. discharge into the ocean. Geophysical Research Letters 29, Marinas,˜ and A.M. Mayes. 2008. Science and technology no. 12: 2.1–2.3. for water purification in coming decades. Nature 452, Tuminaro, R.S., M. Heroux, S.A. Hutchinson, and J.N. Shadid. 301–310. 1999. Official Aztec User’s Guide: Version 2.1. Sandia Shreve, R.L. 1985. Late Wisconsin ice-surface profile calculated National Laboratories Technical Report, SAND99-8801J. from esker paths and types, Katahdin esker system, Maine. http://www.cs.sandia.gov/CRF/aztec1.html. Quaternary Research 23, 27–37. Uchupi, E., N. Driscoll, R.D. Ballard, and S.T. Bolmer. 2001. Siegel, D.I., and R.J. Mandle. 1984. Isotopic evidence for Drainage of late Wisconsin glacial lakes and the mor- glacial melt water recharge to the Cambrian-Ordovician phology and late quaternary stratigraphy of the New Jer- Aquifer, north-central United States. Quaternary Research sey–southern New England continental shelf and slope. 22, 328–335. Marine Geology 172, 1170–145. Simmons, C.T., T.R. Fenstemaker, and J.M. Sharp. 2001. van der Veen, C.J. 2007. Fracture propagation as means of Variable-density groundwater flow and solute transport in rapidly transferring surface meltwater to the base of heterogeneous porous media: Approaches, resolutions and glaciers. Geophysical Research Letters 34, L01501. future challenges. Journal of Contaminant Hydrology 52, van der Veen, C.J. 1998. Fundamentals of Glacier Dynamics. 245–275. Rotterdam, Netherlands: A.A. Balkema. Starinsky A., and A. Katz. 2003. The formation of natural Zheng, C., and G.D. Bennett. 1994. Applied Contaminant Trans- cryogenic brines. Geochimica et Cosmochimica Acta 67, port Modeling. New York: Van Rostrand Reinhold. 1475–1484. Zwally, H.J., W. Abdalati, T. Herring, K. Larson, J. Saba, and Steenhuis, T., C. Jackson, S. Kung, and W. Brutsaert. 1985. K. Steffen. 2002. Surface melt induced acceleration of Measurement of groundwater recharge on eastern Long Greenland ice-sheet flow. Science 297, 218–222.

16 D. Cohen et al. GROUND WATER NGWA.org