LATE QUATERNARY DEPOSITIONAL ENVIRONMENTS, TIMING AND RECENT DEPOSITION: NARRAGANSETT BAY, RHODE ISLAND AND MASSACHUSETTS

Glacial Lake Narragansett, which occupied much of the southern portion of Narragansett Basin during Late Wisconsinan deglaciation, and was contiguous with glacial lakes in Rhode Island Sound and Block Island Sound, is the focus of this study. This dissertation examines the deposits of glacial lakes to develop a history of deglacial chronology, isostatic rebound, and climate change across the glaciated northeast. All of these issues are addressed in this dissertation. This work synthesizes the Late Wisconsinan deglacial evolution of Glacial Lake Narragansett using digital elevation models, sub-bottom seismic reflection, ground penetrating radar profiles, sediment cores and borehole stratigraphy. The impetus for the research was to fill a perceived gap in the understanding of the Quaternary geology of southern New England. While the spatial extent of the Late Wisconsinan Laurentide Ice Sheet and general timing of deglaciation is known, more detailed analyses of the nature and timing of deglaciation are needed, particularly near the terminal margin of the Laurentide Ice Sheet. Improving the understanding of the timing and nature of deglaciation allows links to be made between the Laurentide Ice Sheet and northern

iv The support of my colleagues at Bryant University has allowed me to expand my teaching skills immensely while accommodating my schedule as a graduate student, giving me invaluable professional experience. I have had the privilege to work with folks from a variety of agencies and department beyond the geosciences department.
In particular, Jim Turenne, (USDA-NRCS), Dan Goulet (RI CRMC), Chris Damon and Mike Bradley (URI-EDC) have all contributed time and effort in the field or lab that contributed to this work. I am happy to count you all as colleagues, and more importantly consider you all good friends.
On a personal level, I am forever indebted to my wife Julie for her support through the long process of graduate school. We transitioned from dating to marriage to parenthood during this ordeal and her support, and at times, subtle pushes have contributed greatly to this dissertation. My parents George and Laura have always supported what I was doing, even if they did not always understand the process. Their support, emotionally, financially, providing childcare and reminders that they were always proud helped immensely during this process. Similarly, I need to thank my inlaws, Jeri and Roland, and Brian and Annick, whose generosity with their time since the birth of Aidan has been immeasurable and allowed us to keep a day-to-day working schedule, knowing Aidan was in good hands. Last, and of course not least, I need to thank my son Aidan, who was born during the ABD phase of my dissertation.
Thank you for always being there with a hug or a smile as well as with welcome (and needed!) distraction during this process. Without knowing it, you continually serve both as I reminder as to why I do what I do, and provide much needed doses of perspective.
v PREFACE During late Wisconsinan deglaciation in New England, a series of glacial lakes formed along the margin of the retreating ice sheet were impounded in topographic lows behind bedrock outcrops, sediment dams or blocks of stagnant ice. The deposits of glacial lakes provide a record of deglacial chronology, isostatic rebound, and climate change across the glaciated northeast. All of these issues are addressed in this dissertation. Glacial Lake Narragansett, which occupied much of the southern portion of Narragansett Basin during deglaciation, and was contiguous with glacial lakes in Rhode Island Sound and Block Island Sound, is the focus of this study. This work synthesizes the late Wisconsinan deglacial evolution of Glacial Lake Narragansett using digital elevation models, sub-bottom seismic reflection profiles, ground penetrating radar profiles, sediment cores and borehole stratigraphy. This dissertation is presented in manuscript format, and comprises a series of papers that address different aspects of the late Quaternary evolution of Narragansett Bay. There is some overlap and differences in formatting between chapters, as two are only slightly modified from manuscripts submitted for publication.
The impetus for the research was to fill a perceived gap in the understanding of the Quaternary geology of southern New England. The spatial extent of the Late Wisconsinan Laurentide Ice Sheet and general timing of deglaciation is well documented, however, more detailed analyses of the nature and timing of deglaciation are needed, particularly near the terminal margin of the Laurentide Ice Sheet.
Improving the understanding of the timing and nature of deglaciation allows links to be made between the Laurentide Ice Sheet and northern hemisphere climate. Debate vi still exists over the timing of the last glacial maximum and beginning of Laurentide Ice Sheet recession (Denton et al., 2010;Peltier and Fairbanks, 2006;Toucanne et al., 2009). Understanding the timing of Late Wisconsinan deglaciation is critical to understanding the climate dynamics and links to ice sheet behavior during that time Denton et al., 2010).
Recent work on the deglaciation of southern New England has focused on glacial lakes in other areas, including Long Island Sound (Lewis and Stone, 1991;, Connecticut, and Massachusetts Ridge et al., 2001;Ridge et al., 2004;Rittenour et al., 2000). Previous work on the subsurface sediment of Block Island and Rhode Island Sounds (Goss, 1995;Needell and Lewis, 1984;Needell et al., 1983) provide insight into the Quaternary evolution of these areas, however, those studies did not extend into Narragansett Bay, and offered a limited view of Quaternary glacial depositional environments.
Glacial lakefloor sediment, interpreted to be varves, was previously identified in several areas within or around Narragansett Bay, including the Providence River (Antevs, 1922;Antevs, 1928;, Bristol (Smith, 1955) and the West Passage . Two of these studies  also proposed estimates of water levels of Glacial Lake Narragansett. Previous studies on the Holocene stratigraphy of Narragansett Bay, McMaster, 1984), did not take a detailed approach to deciphering the late Quaternary evolution of Narragansett Bay. This work is the first to focus on the extent, timing, volume, and water level of Glacial Lake Narragansett, vii and the distribution of Quaternary depositional environments throughout present-day Narragansett Bay.
The first paper (Chapter 1; Accepted for publication in Quaternary Research, 23 January 2012), recreates the topography of southern New England prior to isostatic rebound and evaluates Late Wisconsinan isostatic depression in the region based on the limit of marine inundation in New England. Determining the pre-rebound topography is also an important component to interpreting glacial lake water levels, drainage patterns, spillways and other geomorphic features (Leverington et al., 2002).
Narragansett Bay and adjacent areas, situated close to the terminal margin of the ice sheet and adjacent to the Atlantic Ocean, is an ideal location to examine the total isostatic depression, using relative sea level curves based on different values of total isostatic depression from previous studies and local indicators of ice thickness. The well-mapped limit of Late Pleistocene marine inundation north of the study area provides a constraint for maximum depression. Relative sea level curves created using published values of isostatic depression in southern New England, suggest inundation of the study area well south of the mapped limit of marine inundation. Different models of isostatic rebound are tested, and the maximum isostatic depression in southern New England is constrained. The response of the lithosphere, specifically the amount of isostatic depression under the ice sheet in North America has been studied extensively; primarily using continental scale models based on inferred ice thicknesses and assumed properties of the lithosphere. Reconstructions of isostatic rebound at the terminal margin have previously been a compromise between local indicators of ice thickness near the margin and estimates of ice thickness in the viii central portion of the ice sheet (Braun et al., 2008;Peltier, 2004). Isostatic depression near the margin was poorly constrained.
The second paper (Chapter 2) provides measurements of the elevation of delta plain-delta slope contacts within deltas deposited into Glacial Lakes Block Island, Rhode Island and Narragansett, and examines the water levels of these lakes. The now isostatically uplifted water level of Glacial Lake Narragansett extends above and below present sea level. Previously mapped and formerly unmapped glacial deltas below present sea level were imaged using high-resolution seismic reflection profiles.
Deltas above present sea level were imaged using ground-penetrating radar (GPR).
Determining the elevation of these deltas served two purposes. The first was to determine the former water level(s) of Glacial Lake Narragansett; the second was to compare the isostatic uplift profile recorded by Glacial Lake Narragansett with measured uplift in central New England. Projected water levels reflecting regional isostatic rebound were fit to the present elevation of the deltas, supporting the hypothesis that one large lake occupied much of the southern portion of the Narragansett Basin during deglaciation. The lake levels were projected onto the prerebound topographic model reported in chapter 1 to determine the extent and geometry of Glacial Lake Narragansett as the Laurentide Ice Sheet retreated north through present-day Narragansett Bay.
The third paper (Chapter 3) investigates the timing of Glacial Lake Narragansett based on a new 265-year varve series collected in the Providence River.
The hypothesis of this paper was that varves from Glacial Lake Narragansett could be correlated with the calibrated North American Varve Chronology (NAVC) of Antevs ix (1922;1928), and Ridge (2010, and references contained therein). The varve series from Glacial Lake Narragansett was not correlated with the NAVC, and it is interpreted that Glacial Lake Narragansett is older than both the varve sequences of the NAVC, and the other uncorrelated records in southern New England and eastern New York. The regional context of the Glacial Lake Narragansett varves was examined using these varve records and cosmogenic exposure dates on recessional end moraines, and the age of Glacial Lake Narragansett was constrained between 20,400 and 19,500 yBP.
The fourth paper (Chapter 4) synthesizes the Quaternary evolution of Narragansett Bay based on the interpretation of 800 km of seismic reflection profiles.
Interpreted side-scan sonar records from portions of Narragansett Bay (Chapter 6;  provided additional information on the extent of some of the glacial depositional environments. Previous studies on the stratigraphy and depositional environments of Narragansett Bay focused on the Holocene stratigraphy (McMaster, 1984). Other studies using seismic reflection profiles and sediment cores focused on select areas of Narragansett Bay . This work represents the first comprehensive, detailed mapping of the Quaternary glacial depositional environments throughout Narragansett Bay. Mapping the deglacial depositional environments provides an understanding of the style of deposition and subglacial dynamics of the Laurentide Ice Sheet. This chapter also includes a 1:50,000 scale Quaternary Geologic map of Narragansett Bay.
The extent, continuity, and elevation of lakefloor deposits support the hypothesis of a single-lake occupying the southern portion of the Narragansett Basin x during deglaciation. A key deglacial depositional environment is the lacustrine fans deposited at the margin of the ice sheet beneath the surface of the glacial lake. These mark the position of the ice margin and the location of subglacial tunnels during deglaciation. The presence of lacustrine fans indicates that a considerable amount of meltwater reached the base of the Laurentide Ice Sheet, and the spacing of subglacial tunnels indicated by the distribution of lacustrine fans provides insight into the nature of subglacial drainage of the Laurentide Ice Sheet. The total volume of sediment deposited in the present-day watershed of Narragansett Bay indicates that this drainage was very efficient, and began > 5 km from the southern margin of the ice sheet.
Subglacial drainage has become a controversial topic in recent years, and it has been hypothesized that subglacial meltwater of the Greenland Ice Sheet may be accelerating flow of outlet glaciers (Luthcke et al., 2006;. These similarities point to the importance of understanding the Late Pleistocene dynamics of the Laurentide Ice Sheet, especially in a world of changing climate. The fifth paper (Chapter 5, Accepted for publication in the Journal of Coastal Research 11 December 2011) examines the distribution of modern depositional environments in two shallow embayments within Narragansett Bay using side-scan sonar, surface sediment grab samples, underwater video imagery and digital aerial photography. The concept of benthic geologic habitats is introduced and defined, and the areas were classified using a flexible naming convention that combines information about natural geologic and anthropogenic processes, morphologic form, sediment characteristics and biota. The result of this work contributes to the overall understanding of the distribution of depositional environments in glaciated estuaries, xi where the distribution of facies is controlled by the extent of the glacial depositional environments and the present geologic processes.
The papers presented here provide a scientific foundation that addresses questions regarding the Late Quaternary evolution of Narragansett Bay and southern New England. The effects of climate change on ice sheets, especially the Greenland Ice Sheet, remains highly debated within the scientific literature. A more thorough understanding of the behavior of the Laurentide Ice Sheet, especially during the early stages of deglaciation, can help guide models and predictions regarding the fate of the modern ice sheets. The conclusions of chapters 1 through 4 contribute to the growing evidence for a thinner, older ice sheet in southern New England (Clark, 1992;Peltier and Fairbanks, 2006).
The interpretations of this work extend into research areas beyond the deglacial evolution of southern New England. The conclusions of chapter 1 and 2 contribute to models of isostatic rebound or thickness of the Laurentide Ice Sheet in southern New England. Chapter 3 places the uncorrelated varve sequences in southeastern New England in a regional context, and constrains the age of Glacial Lake Narragansett.
The age (> 19,500 yBP) of varves in the northern terminus of the lake (75 km from the terminal margin of the Laurentide Ice Sheet) indicates deglaciation was fully underway by this time, and rebuts the hypothesis that deglaciation began 18,000 to 20,000 yBP (Denton et al., 2010).

Chapter 4 contributes information on processes and behavior of the Laurentide
Ice Sheet during Late Wisconsinan deglaciation, which can inform models of present ice sheets in a world of changing climate. The Quaternary depositional environments xii mapped can be applied in other regions of the glaciated northeast, and provides managers with an understanding of the subsurface resources of Narragansett Bay, particularly for future dredging projects, disposal of dredged material, marina construction, and siting of offshore wind farms. Chapter 5 introduces the concept and definition of benthic geologic habitats, and builds the understanding between Quaternary (glacial) depositional environments and the present distribution of benthic geologic habitats within two areas of Narragansett Bay. This paper provides a concise method of mapping these areas in a manner that is useful to managers and scientists within other disciplines, and has been applied successfully in lagoon, shoreface, and inner shelf depositional environments.    . Details of digital elevation model discussed in the text. Hatched region on inset map of New England shows extent of glacial marine inundation based on Stone and Peper, (1982) and Thompson and Borns, (1985).  , Goldsmith (1982), Sirkin (1982) and . B -Elevation relative to sea level at  (Clark et al., 1994;Peltier, 2004)    , Ridge, ( , 2004 and . All dates reported as calendar years before present ................ 118 Narragansett Bay varve years 5000 to 5150. A constant offset was added to some of the records to display them on one graph without overlap. EW-1 was originally collected by . B. Correlation of varve records collected in the Providence River for Narragansett Bay varve years 5150 to 5260. A constant offset was added to some of the records to display them on one graph without overlap. C. Correlation of varve records from the Providence River and Seekonk River (Antevs, 1928). Arrows point to two discrepancies, where it appears one varve may be missing from each sequence. The Providence composite curve was offset by 3 cm to display the records on one graph without overlap. D. Correlation of varve records from Pawtuxet Cove  and Gaspee Point Varve thickness of PC-6 was offset . NGRIP ice core chronology and δ18O profile of the NGRIP ice core Rasmussen et al., 2008). Red arrows indicate cosmogenic exposure ages of the Ledyard-Congdon Hill (L-CH) and Old Saybrook-Wolf Rocks (OS-WR) recessional end moraines in southern New England   . Green arrows indicate proposed correlated ages of the    Rayburn and Teller, 2007), and small areas with DEM's created using LiDAR (Light Detection and Ranging) with resolution < 1m (Salcher et al., 2010). Digital elevation models with < 30 m resolution allow for a shift in focus from coarse regional reconstructions to finer scale local reconstructions. Total isostatic rebound and ice sheet thickness remain poorly defined near the terminal margin. This represents a critical region for the deglacial history of the ice sheet, particularly in comparing models and observations of modern ice sheets to the LIS. By projecting relative sea level on to pre-rebound topographic models and comparing the modeled limit of marine inundation to the mapped glacial marine limit (Figure 1), the amount of total isostatic depression in southern New England can be constrained, and the method outlined here can be utilized in other previously glaciated areas.

Glacial isostatic adjustment
Glacial isostatic adjustment refers to the change in elevation of the lithosphere due to the loading and unloading of an ice sheet. Isostatic depression refers to the downward correction of the Earth's lithosphere under the weight of the ice sheet, while isostatic rebound will refer to the upward motion of the Earth's surface after unloading of the ice sheet (Figure 2A, B). The response of the lithosphere under the LIS in North America has been studied extensively; primarily using models based on inferred ice thicknesses, and remains largely unresolved at a local level due to the coarse nature of ice sheet models (Braun et al., 2008;Peltier, 2004). The amount of surface load, and the resultant downward adjustment imparted, is directly dependent on the thickness and density of the overlying ice and properties of the underlying mantle (Peltier, 1999). Reconstructions of ice sheet thickness and resultant isostatic adjustment are typically a compromise between local indicators of ice thickness near the margin and coarse estimates of ice thickness in the central portion of the ice sheet (Braun et al., 2008;Peltier, 2004).
The uplift profile of isostatic rebound in New England is recorded by the formerly horizontal water levels of proglacial lakes. The present elevation of deltas deposited into Glacial Lake Hitchcock in the Connecticut River Valley project on a linear plane, with an uplift profile of 0.89 m km -1 , towards the presumed center of the LIS in Hudson Bay, Canada (339º). The uplift profile in coastal central New England is slightly less, at 0.85 m km -1 , also towards the northwest (331º) (Koteff and Larsen, 1989;Koteff et al., 1993). Based on the linear trend of rebound Koteff and Larsen (1989) inferred that isostatic rebound did not begin until onset of Glacial Lake Hitchcock drainage sometime after 16 ka, thus a significant portion of southern New England was ice-free for 5,000 -8,000 yr. Further to the west, in the Champlain (Rayburn, 2004) and Hudson River (Stanford and Harper, 1991) valleys, isostatic rebound was found to be similar to the observations from central New England, with linear uplift profiles measured between (0.7 -1.0 m). Delayed, linear isostatic rebound implies that in New England, the lithosphere responded as a rigid block, and rebound did not begin until the southern margin of the Laurentide Ice Sheet was in northern New England (Koteff and Larsen, 1989;Koteff et al., 1993). This differs from geophysical models that suggest uplift is contemporaneous with ice retreat (Clark et al., 1994;Peltier, 1982Peltier, , 2004. While the uplift profile of isostatic rebound is well understood in New England, the total isostatic adjustment in New England remains poorly constrained by a few scattered data points or coarse regional models. Based on the elevation and extent of the marine incursion in North America, 150 m of maximum total isostatic rebound was reported at the northern boundary of this study. This work was done at a very coarse scale (1:30,000,000) and was not designed to provide detailed information along the southern extent of the ice sheet (Andrews, 1973). An interpreted glacial marine delta 40 m below present sea level in central Long Island Sound is thought to represent a marine incursion prior to the onset of rebound (Lewis and Stone, 1991;Stone et al., 2005b). Comparing the present elevation of this delta with the eustatic sea level curve of Bard et al. (1990), Lewis and Stone (1991) and Stone et al., (2005b) concluded that the minimum amount of isostatic depression in Central Long Island Sound was 80 m.

Modeled and Observed Isostatic Rebound in the Great Lakes
The margin of the LIS south of the Great Lakes has been extensively studied and the timing of margins is better constrained than any other segment of the ice sheet . Clark et al. (1994) compared the results from five different solid Earth models (Models E1 -E5, Figure 2) with existing shoreline and outlet elevations in the Great Lakes to determine which geophysical and ice thickness model that best describes observed isostatic rebound. The models varied lithosphere thickness and upper mantle viscosity, ranging from a 112 to 212 km thick lithosphere overlying different heterogeneity models of upper mantle viscosity ( Figure 2C). The thick (212 km E2) elastic lithosphere model did not produce a good fit with the Great Lakes, and can be discounted for New England where the lithosphere thickness ranges from 90-110 km (Griffin et al., 2004;Rychert et al., 2005). The other models (E1, E3-E5) used a 112 km elastic lithosphere, with varying upper mantle properties. Clark et al (1994) tested a thick ice sheet model (Boulton et al., 1985;Hughes et al., 1981) and thin ice sheet model ) ( Figure 2D) for each of the solid earth models, at different ice sheet durations, including 10,000 yr, which is similar to the duration of the LIS in southern New England (Clark et al., 1994;. While none of the models completely matched the observed isostatic rebound in the Great Lakes region, an important conclusion was that the thick ice models overestimated the amount of isostatic adjustment, and the thin ice models underestimated it. Thin ice models came closer to producing a better fit, and by increasing the ice thickness by 30%, the thin ice model more closely matched the predicted rebound (Clark et al., 1994). The underestimation in total rebound in the 7 thin ice models may have different results if a model was used that closely matches the lithosphere characteristics in New England. The Clark et al. (1994) E3 Model did not agree with the outlet chronology of the Great Lakes that required a rapid initial isostatic adjustment and rebound, and underestimated the present rate of tilting in the Great Lakes region. Uncertainties in ice sheet thickness and loading history prevented discrimination between lithosphere models; i.e. the same results obtained using a thicker or stiffer lithosphere can be offset by a thinner ice sheet. The lithosphere properties, especially the presence of a stiff, relatively thick lithosphere is the dominant factor controlling the amount and relaxation rate of isostatic rebound (Dyke and Peltier, 2000).

Present terrain model
A seamless grid of the present topography and bathymetry of southern New England provides the base map for recreating the pre-isostatic rebound topography. is well within the vertical error of the NED (Root mean square error of 2.5 m (Gesch, 2007)). Spatial accuracy of hydrographic soundings depends on the age of the data, (1943 to 1991). Surveys after 1965 conform to international hydrographic standards, with vertical accuracy +/-0.3 m in < 20 m water depth, +/-1 m in 20 -100 m water depth and 1% of the water depth > 100 m. Horizontal accuracy also varies depending on age, but is < 30 m, with higher accuracy in more recent surveys (NOS, 2009).

Isobase surfaces
Isostatic adjustment was accounted for by creating isobase surfaces reflecting the total isostatic depression. Using the linear plane of rebound measured in coastal New England of 0.85 m . km -1 uplifted northwest (336º) towards the presumed center of the LIS in Hudson Bay Canada (Koteff and Larsen, 1989;Koteff et al., 1993), raster surfaces were created in a geographic information system (ESRI ArcMap TM 9.3 Surfaces with additional isostatic depression were created by increasing total isostatic depression at the terminal margin, assuming the same linear rebound profile. The points were interpolated into a Triangular-irregular network (TIN), and converted into a raster grid surface (pixel size 15 m) ( Figure 3A). The isobase surfaces were subtracted from the modern DEM in ESRI ArcMap 'Raster Calculator' to produce the pre-rebound isostatic elevation models ( Figure 3B).

Results:
Three pre-isostatic rebound digital elevation models were generated using differing values of total isostatic depression.  (Andrews, 1973;Clark et al., 1994;Stone et al., 2005b). The final model used 50 m of depression at the terminal margin of the LIS and was chosen as an intermediate value.

Relative Sea Level
Relative sea level curves were generated using the algebraic difference between isostatic adjustment and eustatic sea level. The eustatic sea level curve presented here ( Figure 4) is based on previously published sea level data (Donnelly and Bertness, 2001;Oldale and O'Hara, 1980;Peltier and Fairbanks, 2006;van de Plassche et al., 1998). The age, projected sea level, and source material of all of the samples used to generate the selected sea level curves were entered into a database and were converted from radiocarbon years before present to calendar years before present using the calibration of Fairbanks et al., (2005). The resulting points were plotted for comparison with previously published curves, and a best-fit line was drawn through the data. Late Pleistocene and Early Holocene time portion of the curve closely mirrors the curve of Peltier and Fairbanks (2006), and includes two portions of the curve with extremely rapid (> 2 m . yr -1 ) rates of sea level rise, interpreted to be meltwater pulses 1A and 1B (Fairbanks et al., 1992;Peltier, 2005;Peltier and Fairbanks, 2006). The Late Holocene portion of the curve skews sea level in southern New England slightly older than the original Peltier and Fairbanks (2006) model. This local trend is governed by a reasonable best fit between the Peltier and Fairbanks (2006) curve and samples with good stratigraphic control recovered by Oldale and O'Hara (1980). The latest Holocene portion of the curve is a best fit line describing the sea level rise history of a salt marsh in Clinton, CT and other sites in New England (Donnelly and Bertness, 2001;van de Plassche et al., 1998) (Figure 4).
The isostatic rebound curves have a half-life of 1000 yr, and are flat prior to 16 ka with the assumption that isostatic rebound did not begin prior to this time. Based on the projected water level of Glacial Lake Hitchcock, it was originally proposed that rebound was delayed until 14.0 cal ka (Koteff and Larsen, 1989). More recent work suggests the drainage of history of lakes in the Connecticut River Valley was more complicated, and rebound may have begun earlier than 14.0 ka (Ridge, 2004).
Regardless of these differing interpretations, the assumed onset of rebound at 16.0 ka used in this paper appears to be consistent with regional observations at this time.
Differing half-lives of isostatic rebound have been proposed, including a rate of 1750 yr in Long Island Sound . A half-life of 1750 yr would imply isostatic rebound was still occurring < 5 ka, which is not consistent with sea level rise curves in southern New England. A half-life of 1000 yr is favored near the terminal margin if the LIS and is consistent with other studies in New England (Belknap et al., 1987;Dyke and Peltier, 2000).

Constraints on maximum isostatic depression
The relationship between relative sea level and the total downward adjustment of the lithosphere was evaluated by creating three reconstructed terrain models reflecting relative sea level at 26 ka, 21 ka and 16.5 ka for each of the models discussed above. These three time slices represent the last glacial maximum (26 ka) (Peltier and Fairbanks, 2006), the formation of the Charlestown-Point Judith-Buzzards Bay end moraine based on cosmogenic exposure dates (21 ka) , and conditions prior to the onset of isostatic rebound (16.5 ka).

m of depression at Terminal Moraine
Using the 75 m isostatic model, relative sea level at 26 ka, 21 ka, and 16.5 ka was 45 m, 41 m and 29 m below sea level. When the terrain model is adjusted to project these sea levels, it becomes apparent that at 26 ka, marine water would have extended to the margin of the ice sheet ( Figure 5A). Most of Block Island, as well as the southwest shoreline of Rhode Island, Martha's Vineyard, Massachusetts, and eastern Long Island, NY would have been inundated with marine water by 21 ka ( Figure 5B). Marine inundation extends across most of the ice-free Narragansett Basin and southeastern Massachusetts at 16.5 ka ( Figure 6A).

m of depression at Terminal Moraine
Using the 50 m isostatic model, relative sea level at 26 ka, 21 ka, and 16.5 ka was 70 m, 66 m and 53 m below present sea level. Similar results to the 75 m model are obtained when relative sea level is projected onto the terrain model. The southern margin of the LIS would be grounded in marine water at both 26 ka and 21 ka.
Marine waters would have inundated most of Narragansett Basin and much of southeastern Massachusetts south of the glacial marine limit by 16.5 ka ( Figure 6B).

m of depression at Terminal Moraine
When the terrain model reflects 30 m of isostatic adjustment at the terminal margin of the ice sheet, relative sea level at 26 ka, 21 ka and 16.5 ka was 90 m, 86 m, and 73 m below present ( Figure 4). The margin of the ice sheet was not grounded in marine water at either 26 ka or 21 ka in this scenario, and marine water does not inundate any of the study area. By 16.5 ka marine waters may be just south of the study area, extending into Block Channel, but did not extend into Rhode Island or Block Island Sounds ( Figure 7).

Assumptions
A major assumption in this work is that there has not been significant erosion or deposition on the landscape after the LIS retreated from southern New England.
The ubiquitous presence of the eolian mantle over the landscape suggests that other than in the present river valleys, where channel incision and floodplain deposition has occurred, most of the landscape has been largely unchanged. Grading and filling in urban areas has altered the landscape, however most alterations probably fall within the range of uncertainty of the terrain model and this does not change the results of this work. The topography below present sea level was not corrected to reflect postglacial deposition. The thickness of postglacial sediment is less than 6 m across most of the study area, except in deep channels in the East Passage of Narragansett Bay, Rhode Island Sound and Block Island Sound McMaster, 1984;Neddell et al., 1983;Needell and Lewis, 1984;O'Hara and Oldale, 1980). Due primarily to large uncertainties in the amount of sediment eroded during the transgression, there was no attempt to account for sediment lost to erosion since deglaciation, and it is assumed that the post-glacial deposition is in the same order of magnitude of post-glacial erosion.
Isostatic depression resulting from water loading as sea level rose from the Late Wisconsinan low-stand was not considered here. While some isostatic adjustment probably occurred due to water loading since the last glacial maximum, this would have begun after isostatic rebound was complete when relative sea level inundated the inner continental shelf ( Figure 4). Total isostatic displacement at the present shoreline in southern New England from water-loading was likely < 5 m (Bloom, 1967).

Relative Sea Level vs. Glacial Marine Limit
Projecting relative sea levels on the reconstructed terrain models provide some constraints on the downward adjustment of the lithosphere in southern New England.
The limit of glacial marine deposition in New England has been well established to intersect present sea level just south of Boston, Massachusetts and extends north into coastal New Hampshire and Maine ( Figure 1) (Bloom, 1963;Dyke et al., 2005;Koteff et al., 1993;Stone and Peper, 1982;Thompson and Borns, 1985). If a projected relative sea level shows inundation across parts of the study area south of the mapped limit of marine inundation, then another explanation must be examined. Possible explanations include: 1. The marine limit as mapped is incorrect 2. A peripheral forebulge on the continental shelf blocked sea level rise 3. The uplift profile of isostatic rebound was not linear, resulting in a more complicated isobase surface near the ice margin 4. There was less total isostatic adjustment, resulting in a lower relative sea level.
Each of these will be examined and discussed below.

Evidence of the marine limit
The projected inundations in the 75 and 50 m models do not fit with the current extent of glacial marine sediment. Areas that were inundated, i.e. northern Massachusetts and coastal Maine, have a distinct stratigraphy, dominated by the presence of a fossiliferous marine deposit (Bloom, 1963;Hitchcock, 1861;Stone and Peper, 1982). Wave-formed features (Spits, swash bars etc.) are also common (Koteff et al., 1993). The fact that no glacial marine deposits are found above present sea level within the study area supports the idea that the area was not inundated by marine water during deglaciation. Numerous glacial varve sequences from Rhode Island and southeastern Massachusetts (Antevs, 1928) indicate these areas contained glacial lakes and were not inundated with marine water during deglaciation. If the marine limit is correct as mapped, and glacial marine inundation of the landscape did not happen south of Boston, Massachusetts, then any reconstruction that shows marine inundation of the landscape can be considered incorrect and an alternative explanation is warranted.

Peripheral forebulge blocks transgression
The presence of a marginal forebulge, beyond the terminal margin of the LIS resulting from the lateral displacement of the mantle from beneath the ice sheet has been debated for decades. The distance beyond the ice margin and height is very poorly constrained, with distances ranging from 30 -250 km, and total relief of 10 -70 m (Barnhardt et al., 1995;Dillon and Oldale, 1978;Knebel et al., 1979;Pardi and Newman, 1987;Peltier, 1982;Stanford, 2010;Uchupi et al., 2001 Barnhardt et al., (1995). No forebulge migration has yet been recognized in lacustrine water levels in the Connecticut River Valley (Koteff and Larsen, 1989) from 18.2 to 12.5 cal ka (Ridge, 2004(Ridge, , 2011. The lack of a migrating forebulge recorded in southern New England, compared to Maine and Atlantic Canada, may be due to differences in the rigidity of the lithosphere offshore of Atlantic Canada, or lateral changes in mantle temperature, both of which cause significant changes in mantle viscosity (Barnhoorn et al., 2011;Zheng and Arkani-Hamed, 2002). Alternatively, the higher rates of Holocene subsidence in Atlantic Canada and Coastal Maine compared to the rest of New England may suggest that active forebulge collapse is occurring, where in southern New England, the forebulge has already migrated inland (Dyke and Peltier, 2000).
This study assumes that if there was a peripheral forebulge, it was either; a collapsing, non-migrating forebulge (amplitude 20 -40 m (Barnhardt et al., 1995;Dyke and Peltier, 2000;Stanford, 2010)) at a distance > 150 km beyond the terminal margin), or the forebulge migrated through southern New England after the onset of isostatic rebound. A forebulge > 150 km south of the terminal margin of the LIS would have been located beyond the edge of the continental shelf, where it may have impacted southern Atlantic and Caribbean sea levels (Potter and Lambeck, 2004), and would not have had a major impact on relative sea level in southern New England.

Isostatic rebound is not linear
The topographic reconstructions presented here assume that isostatic rebound was linear, based on isostatically uplifted water levels of glacial lakes in New England and eastern New York (Koteff and Larsen, 1989;Rayburn, 2004;Stanford and Harper, 1991). Early models of the Great Lakes proposed a more 'hinged' profile of uplift, with a relatively stable area south of the hinge, and increasing uplift to the north (Goldthwait, 1908). An uplift profile of this shape would increase depression to the north, further inundating the landscape with marine water. Models suggest that over thousands of kilometers, the uplift profile may be slightly exponential (Clark et al., 1994), but for distances of < 300 km (recorded by the uplifted water-levels of glacial lakes in New England and the Hudson River Valley), the plane of rebound is linear (Koteff and Larsen, 1989;Koteff et al., 1993;Rayburn, 2004).

Isostatic rebound was not delayed
The assumption that isostatic rebound was delayed until after ~16 ka, and perhaps as late as 14.2 cal ka (Koteff and Larsen, 1989), is based on the initiation of drainage of Glacial Lake Hitchcock. If isostatic rebound began immediately after ice retreat, the landscape would have had to be even more depressed to meet the required elevation for a marine delta in Long Island Sound. As stated above, increasing the amount of depression further inundates the ice-free portions of the landscape, and produces a grounded marine ice margin in southern New England.

Less isostatic depression
The simplest explanation to align the amount of isostatic rebound to the observed geology in New England is that the total amount of isostatic depression was less than previously assumed. The results presented here suggest that total isostatic important note is that the forebulge height is based on radiocarbon ages of non-in situ marine shells (Ewing et al., 1963;Richards and Werner, 1964), and a slightly lower 20 forebulge (30 m) would bring the isostatic depression in the Hudson Valley in line with the interpretations presented here for southeastern New England.

Implications
This work has several implications. First, previously published values of isostatic rebound in southern New England overestimate total isostatic adjustment by 50 m or more. The glacial marine limit has been well constrained by over 100 years of mapping throughout New England (Bloom, 1963;Stone and Peper, 1982). The glacial marine delta in Long Island Sound remains enigmatic, as it requires 70 to 80 m isostatic depression to intersect eustatic sea level at the onset of isostatic rebound (Lewis and Stone, 1991;Stone et al., 2005b). Reexamination of radiocarbon dates that aided the original interpretation of a Late Pleistocene marine incursion into Long Island Sound, suggests the incursion is closer to 10 cal ka (Varekamp et al., 2004;Varekamp et al., 2006).
The volume of the delta (11.5 x 10 9 m 3 ) requires a significant source of sediment, and is interpreted to record drainage of glacial lakes down the present-day Connecticut River valley (Lewis and DiGiacomo-Cohen, 2000;Lewis and Stone, 1991). The sea level curve presented here ( Figure 4) suggests that deposition of this delta (and drainage of a lake(s) in the Connecticut River Valley) culminated around 11 cal ka, after isostatic rebound had uplifted the landscape, creating enough gradient to incise the present-day Connecticut River through Late Wisconsinan glacial deposits (R. Lewis, personal communication). This is just after meltwater pulse 1B (MWP-1B) (Fairbanks 1989;Fairbanks et al., 1992). While the timing and magnitude of MWP-1B is still debated (Bard et al., 2010), rapid marine incursion into Long Island Sound during MPW-1B could account for the apparently complicated transgression history (Stone et al., 2005b;Varekamp et al., 2004;Varekamp et al., 2006).
The linear profile of rebound in New England, coupled with the maximum depression presented here and the relatively thick lithosphere beneath southern New England supports the ideal of Koteff and Larsen's (1989) 'crustal block' model for delayed rebound across southern New England. The lower value of isostatic adjustment also supports a thinner Laurentide Ice Sheet, (Clark, 1992;Clark et al., 1999), which would produce less isostatic rebound. Rayburn (2004)  Maine and the mid-western United States (Clark, 1992;Shreve, 1984). Maximum marine inundation in Maine approximately 400 km from the terminal margin of the LIS at 15.1 ka requires isostatic depression to be 170 m approximately 1000 yr after the onset of isostatic rebound in New England (Barnhardt et al., 1995). Based on a 1000 yr half-life of isostatic rebound near the terminal margin of the LIS (Belknap et al., 1987;Dyke and Peltier, 2000), total isostatic depression at the Maine coast was 340 m, and would be in line with 30 m of isostatic depression at the terminal margin presented here.
A simple estimation of isostatic depression at equilibrium can be calculated for a given ice thickness using the density ratio of glacial ice (900 kg . m -3 ) to the underlying crust and upper mantle (3300 kg . m -3 ). Using this ratio, the amount of 22 isostatic depression for a given ice sheet model can be approximated by dividing ice sheet thickness by 3.6. Three reconstructions of the LIS (solid lines) and the calculated isostatic depression in southern New England (dashed lines) are shown in Figure 8. The resultant isostatic depressions show that the ice thickness of  overestimates both the total and relative uplift profile observed in southern New England. Ice thickness can be estimated by using the same isostatic relationship and the ice sheet model that best describes the observed isostatic rebound in southern New England is a simple, linear ice sheet, with a profile of 3.3 m . km -1 (Figure 8).
The best fits to this ice sheet profile are the ICE-5G model of Peltier (2004) and the 'thin ice' model of Clark et al., (1994). The ICE-5G ice sheet reconstruction shows a step pattern due to the coarse resolution, however the end points of ICE-5G are almost identical to the proposed ice sheet profile presented here. The resultant isostatic depression under the ICE-5G and 'thin ice' models also closely match the rebound profile observed in southern and central New England.
The isostatic rebound model presented here supports a transgression history in southern New England with inundation of Block Island Sound and Narragansett Bay during latest Pleistocene and Holocene time (McMaster, 1984;Oldale and O'Hara, 1980). This differs from the interpretation of a marine incursion into Rhode Island, Block Island, and Long Island Sounds prior to isostatic rebound (Lewis and Stone, 1991;Stone et al., 2005b). It is possible that marine water inundated parts of Long Island Sound via the western end (Gayes, 1987), although the moraine acting as a dam at the southern end of the Hudson River Valley was probably not breached until at 23 least 15.5 cal ka (Stanford, 2010), and maybe as late as 13.4 cal ka (Donnelly et al., 2005).

Conclusions
High-resolution topographic models of the landscape prior to isostatic rebound can be created in a GIS environment by subtracting interpreted isobase surfaces from modern topographic and bathymetric data. By creating multiple reconstructions with differing values of total isostatic depression, the maximum downward adjustment can be constrained based on the mapped glacial marine limit in New England. The result of this work suggests that previous interpretations of isostatic depression in southern New England would inundate the landscape with marine water south of the mapped limit of glacial marine deposition, suggesting that these published values overestimate isostatic depression in southern New England by 40-50 m or more. A first order estimate of ice-thickness based on total isostatic depression and the observed uplift profile in southern New England is in line with the published ICE-5G model and 'thin ice' models of Clark et al (1994). This produces an ice thickness ranging from 100 m near the southeastern margin of the LIS, to > 1000 m 300 km north of the terminal margin of the LIS in New England. This regional model constrains both continental ice sheet reconstructions and geophysical models that should be considered in future work in both disciplines.    . Details of digital elevation model discussed in the text. Hatched region on inset map of New England shows extent of glacial marine inundation based on Stone and Peper, (1982) and Thompson and Borns, (1985).   , Goldsmith (1982), Sirkin (1982) and . B -Elevation relative to sea level at 16.5 ka assuming 75 m of isostatic depression at the terminal margin. The southern margin of the Laurentide Ice Sheet had retreated north of Rhode Island by this time. C -Elevation relative to sea level at 21 ka, assuming 30 m of depression at the terminal margin. D -Elevation relative to sea level at 16.5 ka assuming 30 m of depression at the terminal margin. Marine water may have begun to inundate the channel south of Block Island by 16.5 ka. The southern margin of the Laurentide Ice Sheet had retreated north of Rhode Island by this time. Figure 6: Ice thickness (solid lines) and isostatic adjustment (dashed lines) for three published ice sheet reconstructions (Clark et al., 1994;Peltier, 2004) and the calculated ice sheet profile based on observed isostatic rebound in New England with 30 m of depression at the terminal margin of the ice sheet. Geographic locations are noted, including SNH (Southern New Hampshire, USA) and BI (Block Island, Rhode Island, USA).

Abstract
The elevation of Late Wisconsinan delta plain-delta slope contacts in and around Narragansett Bay, as well as Block Island and Rhode Island Sounds, were determined using ground-penetrating radar and sub-bottom seismic reflection profiles.
Water levels reflecting regional isostatic rebound were fit to the deltas and are interpreted to record three, lowering water levels. Water levels were projected onto a pre-isostatic rebound terrain model of southern New England to examine the extent of Glacial Lakes Block Island, Rhode Island and Narragansett. The reconstructions of the lakes suggested that drops in water level coincide with lower elevation spillways becoming ice-free as the southern margin of the Laurentide Ice Sheet retreated north.
The projected extent of the glacial lakes shows that Glacial Lake Narragansett was contiguous with Glacial Lakes Block Island and Rhode Island, and one large lake occupied much of the southern portion of Narragansett Basin. The linear trend of the measured deltas further validate the idea of delayed isostatic rebound, and the behavior of New England as a lithospheric block during rebound, as first suggested by Koteff and Larsen (1989).

This paper examines the water levels of Glacial Lakes Block Island, Rhode
Island and Narragansett by measuring the elevation of the delta plain-delta slope contact within deltas deposited into these lakes in and around Narragansett Bay and adjacent waters ( Figure 1). The overall goal was to determine the extent of glacial lakes within Narragansett Bay. A critical component of this is determining the water level of the glacial lake(s). The water levels were projected onto a pre-isostatic rebound topographic model of southern New England to examine the extent of Glacial Lake Narragansett.
The hypothesis is that during Late Wisconsinan deglaciation, one large lake (Glacial Lake Narragansett), occupied most of Narragansett Bay, and extended into adjacent waters. Because the deltas of Glacial Lake Narragansett fall on linear planes reflecting the regional trend of rebound, it is surmised that a single lake occupied the study area during deglaciation. The similar north to south trend (roughly parallel to isostatic rebound) of Glacial Lake Narragansett, and of Glacial Lake Hitchcock, where earlier measurements of rebound were made, is important to consider when making comparisons between these areas. Determining the rebound profile, (the amount of uplift over a given distance), is an important component of reconstructing the deglacial topography of previously glaciated areas.

Water level indicators
The primary indicator of water levels in glacial lakes is the elevation of the delta plain-delta slope contact (often referred to as the topset-foreset contact). The deltas measured in this study are ice marginal or ice contact, Gilbert style deltas (Gilbert, 1890), with a fluvial (braided river) delta plain overlying a delta slope, deposited into, and graded to a standing (lacustrine) body of water ( Figure 2). This style of deposition is common in both modern and Late Wisconsinan deglacial depositional environments .
Within each delta system there may be one or more morphosequences.
Morphosequences are time-equivalent groups of landforms that extend from the collapsed former ice margin at the proximal end of the deposit. There is generally a decrease in grain-size from gravel near the ice margin to sand and silt in more distal portions of the sequence (Koteff and Pessl, 1981;. Where more than one morphosequence was mapped as part of a delta system, attempts were made to obtain elevations of both sequences. The contact between the relatively flat-lying delta plain beds and underlying dipping (10-30º) delta slope beds is readily identifiable in outcrop exposures and geophysical imagery (Smith and Jol, 1997). The erosional surface is used as a proxy for determining glacial lake level, but actually represents the elevation of the bottom of a fluvial channel on the delta plain, and is lower in elevation than the actual water level of the lake (Gustavson et al., 1975). Observations on glacial deltas in Alaska that are similar to those deposited in Glacial Lake Narragansett report channel depths < 1.0 m (Boothroyd and Ashley, 1975), and the delta plain-delta slope contact is assumed to be less than 1 m below the true water level of the lake.

Previous work
The elevation of two deltas was reported in earlier studies of Narragansett Bay . The elevations of the delta plain-delta slope contact of several deltas in southeastern Mass and northeastern Rhode Island were estimated from 1:24,000 and 1:25,000 scale topographic maps (Stone, 2010) or published geologic maps (Hartshorn, 1967;Schafer, 1961a, b;Smith, 1955Smith, , 1956. The extent of the deltas around Narragansett Bay has been mapped at scales of 1:24,000 and 1:100,000 ( Figure 3) McCandless, 2002, 2003;RIGIS, 1989) or were interpreted from published maps of stratified deposits (Allen and Ryan, 1960;Schiner and Gonthier, 1965). The presence of a glacial lake in Block Island Sound (Glacial Lake Block Island) has been known for decades (Bertoni et al., 1977;Frankel and Thomas, 1966). The water level of Glacial Lake Block Island was previously estimated to be 15 -20 m below present mean sea level (MSL) Lewis and Stone, 1991). Sediment deposited in a glacial lake was also mapped in Rhode Island Sound, however the extent of any deltas were not mapped separately from lakefloor deposits in the original work .
Isostatic rebound has uplifted the landscape preferentially to the northwest, indicated by the present elevation of deltas graded to the formerly horizontal water level of glacial lakes (Koteff and Larsen, 1989;Koteff et al., 1993). Anthropogenic development on deltas in northern Narragansett Bay and submergence of deltas beneath present sea level in southern Narragansett Bay, Block Island Sound, and Rhode Island Sound necessitated the use of geophysical methods to determine the elevation of the delta plain-delta slope contact. Determining the elevation of glacial lacustrine and marine deltas with geophysical tools has been done in other areas (Barnhardt et al., 1997;Kostic et al., 2005;Smith and Jol, 1997;Tary et al., 2007).
This present work represents the first attempt to combine these techniques in the context of evaluating the former water level and uplift profile of a glacial lake.

Methods
The delta plain-delta slope contact of 15 deltas ( Figure 3) was imaged using ground-penetrating radar (GPR) or sub-bottom seismic reflection profiles ( Figure 4).
The elevation of the delta plain-delta slope contact estimated from topographic maps or published Quaternary geologic maps was examined in comparison with the geophysical data. The highest elevation delta plain-delta slope contact at any given delta is used to estimate the lake level. Wave action during Holocene transgression and anthropogenic activities on the landscape can modify or remove the upper portions of the delta plain and delta slope, but these processes cannot increase the elevation of the delta plain-delta slope contact (Koteff and Larsen, 1989;Koteff et al., 1993).

Sub-bottom profiling
Glacial deltas were interpreted from sub-bottom seismic reflection profiles collected in Lower Narragansett Bay, Rhode Island and Block Island Sounds, using an EdgeTech (Wareham, Massachusetts, USA) 2-16S Chirp Seismic Reflection profiler, producing a linear pulse of energy with a frequency of 2-10 kHz ( Figure 4). Penetration depth in sandy or gravelly delta plain-delta slope depositional environments was limited to < 20 m, which was sufficient to image the dipping reflectors of the delta slope with occasional penetration into the underlying glacial lakefloor deposits. The elevation of the delta plain-delta slope contact was digitized from the seismic reflection profiles using Chesapeake Technologies SonarWeb TM (Mountain View, California, USA) software, assuming a sound velocity of 1500 m . s -1 in both water and sediment (Hamilton, 1974). The elevation of the towfish was maintained at 1 m beneath the surface of the water, and the elevation of the delta plain-delta slope contact was corrected to present mean sea level (MSL) based on recorded water levels at the Newport, RI tide gauge.

Ground Penetrating Radar (GPR)
GPR was utilized to determine the elevation of the delta plain-delta slope contact of delta surfaces now above present mean sea level (MSL) ( Figure 4). The non-intrusive nature of GPR is ideal for working in urban or developed areas where no exposures are available. GPR works best in unsaturated sand and gravel, making it ideal for imaging glacial deltas Jol, 1995, 1997) and has been used successfully on glacial-marine deltas in northern New England (Tary et al., 2007).
GPR signals will not penetrate sediment saturated with soluble salts, limiting its use in saline and tidally influenced areas near the coast (Buynevich and FitzGerald, 2003).
The system used in this study was a SIR3000 system (manufactured by Geophysical  (Smith and Jol, 1997;Widess, 1973).

d = c . t/2 . Er 1/2 (1)
Where: d = depth in feet. c = velocity of light (1 foot/nanosecond). t = pulse travel time in nanoseconds. Er = relative dielectric permittivity of sediment The ground surface above identified delta plain-delta slope contacts was surveyed with Real-time Kinematic Global Positioning System (RTK -GPS) relative to the North American Vertical Datum, 1988 (NAVD88), accurate to < 0.05 m (Trimble, 2008) and the elevation relative to NAVD88 was converted to MSL based on the datums at the Newport, RI tide gauge (NOS, 2010a). The elevation of the New Meadow Neck, Blackstone River and East Providence Plains deltas was determined from a continuous interpolated grid (cell size 1 m) generated from LiDAR data of the area surrounding the Providence River. The accuracy of this raster surface was validated by comparing the RTK-GPS elevations determined on the other deltas around the Providence River (n = 46); the average difference between the raster surface and RTK-GPS points was < 0.2 m.

Vertical Error
The largest source of error in GPR surveys is the result of uncertainty in the dielectric permittivity of the sediment profile. The elevation of the water table is also important and in most cases, the water table was visible as a distinct, horizontal reflector on the GPR profiles. The vertical scale was calculated using dielectric permittivity above and below the water table corresponding to dry and saturated sediment. The elevation of the ground surface above delta plain-delta slope contacts determined using either RTK -GPS or LiDAR is not a significant source of error, with vertical accuracy of +/-.05 m and 0.2 m respectively.
Sub-bottom seismic reflection profiles were corrected to MSL based on the water level at the Newport, RI tide gauge at the time of data collection. The delta plain-delta slope contact was less than 5 m below the present seafloor on all the seismic reflection profiles, so differences in sound velocity are not a significant source of error. The vertical elevation of the delta plain-delta slope contact is reported here to be +/-2 m, which includes the < 1 m difference between the delta plain-delta slope contact and the actual lake level.

Interpreted water levels
The measured delta plain-delta slope contacts were plotted based on the present elevation and northing (Universal Transverse Mercator,zone 19N). No regression line intersected all of the measured delta plain-delta slope contacts.
Multiple water levels are interpreted based on subsets of the data, selected based on geographic location (i.e. Block Island Sound, lower Narragansett Bay, and upper Narragansett Bay).

Pre-isostatic rebound topographic models
The interpreted water levels were projected onto a pre-isostatic rebound digital terrain model to reconstruct the extent of Glacial Lakes Block Island, Rhode Island and Narragansett. This model used a seamless grid of the present topography and bathymetry of southern New England using the raster National Elevation Database 30 m DEM combined with hydrographic soundings from the National Ocean Service .
Isostatic depression that occurred due to the mass of the ice sheet was accounted for by creating isobase surfaces reflecting the total isostatic adjustment using the linear plane of rebound measured in central New England of 0.85 m . km -1 uplifted towards the northwest (336º). The isobase surfaces were subtracted from the modern DEM in ESRI ArcMap TM 'Raster Calculator' to produce the pre-rebound isostatic elevation models.
The position of the retreating Laurentide Ice Sheet is critical to determining the extent of Glacial Lake Narragansett as the ice retreated north. Ice margins used are a combination of previously published moraine locations Smith, 2010;Stone and Borns, 1986;Stone and Peper, 1982) and proximal lacustrine fan or recessional end moraines interpreted from high resolution sub-bottom seismic reflection profiles.
Ages of four ice margins are tied to regional cosmogenic beryllium (Be 10 ) exposure dates of boulders on moraines . Ages of other ice margins are based on assumed ice margin retreat rate of 75 m . yr -1 . These average retreat rates are estimated from the existing exposure dates, and are consistent with regional retreat rates (Ridge, 2004).
Thickness of post-glacial deposits, primarily Holocene estuarine and shelf sediment, was determined using seismic reflection profiles throughout the study area.
Holocene sediment thickness estimates in Narragansett Bay were generated from the sub-bottom seismic reflection profiles calculated by using Chesapeake Technologies SonarWeb TM (Version 3.3) software. Thickness of post-glacial sediment in Rhode Island Sound and Block Island Sound was digitized from published U.S. Geological Survey maps O'Hara and Oldale, 1980). Interpolated Holocene thickness layers were subtracted from the topobathy grid using Raster Calculator prior to adjusting the grid to account for isostatic rebound. Seismic penetration in the Sakonnet River was limited, and the true estimate of post-glacial sediment thickness is not well constrained.

Results
The elevation of delta plain-delta slope contact of fifteen glacial deltas was determined using either seismic reflection profiles (6) or ground-penetrating radar (9). The results are discussed for each delta below. See table 1 for a summary of the measured delta plain-delta slope contacts.

Block Island Sound
Several large deltas that extend south from the Charlestown Moraine into Block Island Sound are interpreted to have been deposited into Glacial Lake Block Island . Sub-bottom seismic reflection profiles were collected from the present upper shoreface south into the interpreted glacial lake basin. South dipping (5 -15º) reflectors imaged in seismic reflection profiles offshore of the Rhode Island south shore are interpreted to have been deposited in a delta slope depositional environment. The delta plain-delta slope contacts are 17 m and 26 m (Figures 5, 6; Table 1) below present MSL, in two discrete sets 5 km apart. The higher elevation delta is north of the lower elevation set ( Figure 3).

Pettaquamscutt River Delta
The only delta measured in Rhode Island Sound was deposited offshore of Narragansett Beach by meltwater routed down the valley presently occupied by the Pettaquamscutt River ( Figure 3). No estimation of the elevation of the delta plaindelta slope contact had previously been made, although it had been realized that the sequence extended below present sea level Schafer, 1961a, b;Smith, 2010). Seismic reflection profiles show an interpreted delta plaindelta slope contact at 14 m below present MSL (Table 1). Flat lying seismic reflectors overlying the delta slope reflectors are interpreted to be sandy delta plain deposits, and the elevation of the delta plain-delta slope contact here appears unaltered.

Dutch Island
The southernmost deltas directly deposited into present day Narragansett Bay were in the lower West Passage, west of Dutch Island and south of the Jamestown-Verrazano Bridge ( Figure 3). This previously unmapped delta is interpreted to be a small ice-contact delta deposited when the margin of the Laurentide Ice Sheet extended across the lower West Passage of Narragansett Bay. The elevation of the delta plain-delta slope measured in seismic reflection profiles is 9.8 m below MSL ( Figure 7; Table 1).

Annaquatucket delta
The Annaquatucket delta ( Figure 3) was first discussed by Peck (1987), and  who estimated the elevation at 5 m below MSL based on borehole data and a seismic reflection profile collected prior to construction of the Jamestown Verrazano Bridge ( Figure 3). The delta likely formed when the ice margin was near Wickford, RI, 9 km further north (Schafer, 1961b;Smith, 2010). Subbottom lines collected for this study refined the  estimate, placing the delta plain-delta slope contact at an elevation of 5.8 m below present MSL (Table 1).

Mill Creek and Hunt-Quonset deltas
The Mill Creek and Hunt-Quonset deltas were deposited in Glacial Lake Narragansett in the present West Passage of Narragansett Bay ( Figure 3) (Schafer, 1961b;Smith, 2010). The delta plain here has been heavily modified by construction during and after World War II. No previous estimates of lake level were made for deltas, although the thickness of the interpreted delta plain deposits near Wickford Harbor was measured at 1 -2 m, and it was noted that the mapped morphosequence extended below sea level (Schafer, 1961b). The estimated elevation of the delta plaindelta slope contact is just above present MSL, but cannot be accurately determined due to the heavily altered nature of the area.

Potowomut
The Potowomut delta south of Greenwich Bay was deposited as two morphosequences extending southeast into the West Passage of Narragansett Bay ( Figure 3). GPR profiles (3 km total line length) collected near the present shoreline of Narragansett Bay at the southern end of the delta, showed an extremely high (presumably salt) water table. This limited penetration of the radar signal; however, several reflectors interpreted to be dipping delta slope beds were imaged, placing the elevation of the delta plain-delta slope contact at 3.5 m above present MSL (Table 1).

Island Park
The previously unnamed Island Park delta makes up the northeastern end of Aquidneck Island at the head of the Sakonnet River ( Figure 3). Ground penetrating radar profiles were collected 500 m from the northern shoreline of the Sakonnet River, and the highest delta plain-delta slope contact was measured at 1.5 m above MSL (Table 1). Borehole data (USACE, 1957) supports this interpretation based on the transition from sand and gravel (delta plain) to sand (delta slope) at approximately 1 m above present MSL.

Warwick Plains Delta
The Warwick Plains delta extends from the Pawtuxet River into present Greenwich Bay ( Figure 3) . 1.5 km of groundpenetrating radar profiles were collected at two sites < 1 km apart. The elevation of the highest interpreted delta plain-delta slope contact was determined to be 3.3 m above present MSL 0.1 km north of the present Greenwich Bay shoreline and 4.1 m above present MSL 0.7 km north of Greenwich Bay ( Figure 8; Table 1).

Conimicut Point Delta
Southwest of Conimicut point ( Figure 3), dipping seismic reflectors interpreted to be delta slope beds, were mapped at an elevation of 4 m below present MSL (Table 1). This delta extends along the western side of upper Narragansett Bay, as a broad, sandy platform.

Barrington
The Barrington delta system consists of three morphosequences deposited into Glacial Lake Narragansett (Smith, 1955;Smith, 2010). The southern Barrington delta (BTS) is a narrow (0.6 km north to south) ice marginal delta that extends east to west along a portion of the present upper Narragansett Bay shoreline ( Figure 3). The delta has a maximum ground surface elevation of 15 m above MSL. GPR surveys here were inconclusive due to extensive collapse of the glacial beds, and the elevation of the delta plain-delta slope contact can only be estimated at < 12 m above present MSL (Stone, 2010).
The northern delta (BTN) is approximately 1 km north of BTS, and was deposited into Glacial Lake (Smith, 2010). The elevation of the delta plain-delta slope contact of the northern delta was determined to be 10 m above MSL ( Figure 9, Table   1). This elevation is < 2 m lower than unpublished and published estimates of the delta plain-delta slope contact (Smith, 1955;Stone, 2010). Smith (2010) mapped a third delta within the Barrington delta system that was not measured here due to lack of suitable field locations.

Riverside
The Riverside delta extends from the eastern shoreline of the present Providence River to the Barrington River ( Figure 3) McCandless, 2002, 2003;Smith, 1955Smith, , 1956Smith, 2010). The delta plain-delta slope contact was imaged with GPR along the western end of the delta at an elevation of 11.5 m above (Table 1) MSL (Table 1). Unpublished estimates of the delta plain-delta slope contact in the northern end of the delta ( Figure 3) place the elevation at 15 -18 m above MSL (Stone, 2010).

New Meadow Neck
Located between the Barrington and Palmer Rivers, the New Meadow Neck delta is lower than the adjacent Riverside, Barrington and Warren River deltas ( Figure   3) (Smith, 1955). The delta plain-delta slope contact was imaged in two different areas of the delta at 6.3 and 4.8 m above MSL respectively (Table 1).

Providence Plains Delta
The Providence Plains Delta extends along the northwest side of the Providence River   (Figure 3). The elevation of the delta plain-delta slope contact was estimated at 10 m above MSL based on the topography of the area . The surface of the delta plain is heavily developed, and GPR profiles (1.8 km total line length) were limited to side streets, sidewalks and lawns. Beds interpreted to have been deposited in a delta slope depositional environment, with an apparent dip of 15º towards the southeast (towards the Providence River, Narragansett Bay) were imaged in GPR lines close to the present shoreline of Narragansett Bay. RTK elevations agree with the original interpretation, placing the interpreted delta plain-delta slope contact at 10.1 m above MSL (Table 1).

Smith Hill
One of the few areas available to survey on the heavily developed Smith Hill delta surface, was the front lawn of the Rhode Island State House located in the center of Providence, RI. Ground Penetrating Radar profiles (300 m total line length) were collected to determine the elevation of the delta plain-delta slope contact. The ground surface here represents the altered delta plain, and the elevation of the delta plain-delta slope contact is interpreted to be 18 m above MSL ( Figure 10; Table 1).

Blackstone River
The Blackstone River Delta is along the eastern shoreline of the Seekonk River in northeastern Narragansett Bay, deposited into and graded to Glacial Lake Narragansett in the area now occupied by the Seekonk River   (Figure 3). The delta surface is highly developed; however, a cemetery along the western shoreline of the delta where the GPR profiles were collected has been in existence since 1871 suggests the topography has remained mostly unchanged (Ancestry, 2010). Ground surface elevation here is 20 m MSL, and the delta plain-delta slope contact is 18.5 m above MSL (Table 1). A best-fit linear regression using all of the delta plain-delta slope contacts measured had an uplift profile of 0.7 m . km -1 and was a good fit (r 2 = 0.92), but did not intersect all of the deltas measured. A linear regression line intersects all the delta plain-delta slope contacts from the Block Island Sound low delta, north to the Potowomut delta (n = 11) with a very good fit (r 2 = 0.994), but does not intersect the deltas further to the north. The uplift profile of this line is similar to the regional trend of rebound (0.84 m . km -1 ) ( Figure 11). A separate linear regression line from Warwick Plains to the Blackstone River deltas (n = 14), showed a good fit (r 2 = 0.961) with an uplift profile of (0.9 m . km -1 ). The Block Island Sound High, the East Providence Plains and New Meadow Neck deltas are outliers to these water levels, and are discussed separately.

Assumptions
A major assumption in this work is that the elevations of the delta plain-delta slope contacts have not been significantly altered since deposition. The ubiquitous presence of eolian mantle and active development of soils suggests that other than in the present river valleys, where channel incision and floodplain deposition has occurred, most of the landscape has been largely unchanged. Grading and filling in urban areas has altered the landscape, but historic topographic maps (topography mapped 1890 -1940) were examined, and the surveyed areas appear to have undergone little topographic change.
Deltas below MSL represent a minimum elevation, and some sediment may have been eroded and/or redeposited during Holocene transgression. This remains difficult to quantify because a relatively flat erosional unconformity could be mistaken for a delta plain-delta slope contact in geophysical images. Working on glacial deltas in central New England, Koteff et al. (1993) assumed < 2 m of the delta slope beds had been removed based on a detailed examination of borrow pit exposures. Similar alteration of the delta plain-delta slope contact in Glacial Lakes Narragansett, Block Island and Rhode Island would not changes the interpretations presented here.

Isostatic Rebound profile
Throughout this work, the uplift profile is assumed to match the regional trend of isostatic uplift. The uplift profile in central New England was determined by surveying the delta plain-delta slope contacts in Glacial Lake Hitchcock in the Merrimack and Connecticut River Valleys. It was consistent over > 300 km, producing a linear plane of rebound of 0.85 -0.89 m . km -1 (Koteff and Larsen, 1989;Koteff et al., 1993). The plane of rebound tilts towards the assumed center of a singledomed ice sheet over Hudson Bay, Canada, although the geometry and thickness of the Laurentide Ice Sheet is still highly debated (Clark et al., 1996;Dyke and Peltier, 2000;Peltier, 2004). Similar uplift profiles have been measured in the Hudson River Valley in eastern New York (Rayburn, 2004). Figure 12 compares the relative uplift profile of Glacial Lake Narragansett, Glacial Lake Hitchcock and Glacial Lake Merrimack (Koteff and Larsen, 1989;Koteff et al., 1993). The similarity between the regional rebound trend (0.85 m . km -1 ) and the trend of the deltas measured in this study suggest that the relative, linear uplift profile of isostatic rebound is consistent across southern New England.

Spillways:
The spillway for Glacial Lakes Block Island and Connecticut has been assumed to be at Block Channel, located between Block Island, RI and Montauk Point, Long Island NY ( Figure 1) (Goss, 1995;Lewis and Stone, 1991;Stone et al., 2005b;Uchupi et al., 2001).

Water levels
The hypothesis is that one lake occupied Narragansett Bay. To test the hypothesis, a single water level, reflecting the regional trend of isostatic rebound (0.85 m . km -1 towards the northwest) should intersect all of the delta plain-delta slope contacts. No single water level intersects all of the deltas measured, leading to the interpretation that the delta plain-delta slope contacts record a lake with three distinct water levels ( Figure 11). These water levels were projected onto a pre-isostatic rebound topographic model of southern New England and are discussed as lake stages based on the geographic extent of the lake. The elevation of the deltas and continuity of projected water levels suggest the lakes in Rhode Island and Block Island Sounds were contiguous with the lake in Narragansett Bay. The present elevation of spillways at Block Channel and the Mud Hole are shown for comparison with projected water levels ( Figure 11).

Projected extent of glacial lakes on pre-isostatic rebound models
Glacial Lake Block Island The highest projected water level is recorded by the higher Block Island delta at an elevation of 17 m below present MSL. Glacial Lake Block Island began to form as soon as the Laurentide Ice Sheet retreated from the Beacon Hill moraine, as meltwater was trapped between the ice margin and the moraine ( Figure 13A).
Directly behind the end moraine on the southwest side of Block Island, a delta completely filled a portion of the lake when the ice was still within 10 km of the terminal position (Oakley et al., 2010). Sub-bottom profiles did not penetrate any dipping delta slope reflectors, and elevation of the delta plain surface can only be estimated at > 25 m below present MSL, projecting to a similar water level as the higher Block Island Sound delta.
Glacial Lake Block Island continued to expand in size as the margin of the ice retreated north, eventually occupying much of the central portion of Block Island Sound. The spillway at Block Channel initially controlled the water level of Glacial Lake Block Island. Previous interpretations suggest the lowering of the spillway at the eastern entrance to Long Island Sound ( Figure 1) was controlled by an eroding spillway at Block Channel, down to a final elevation of 60 m below MSL (Goss, 1995;Lewis and Stone, 1991;Stone et al., 2005b;Uchupi et al., 2001). This interpretation was supported by the inferred deposition of > 20 m post-glacial sediment in an incised channel at Block Channel . Reexamination of existing U.S.
Geological Survey seismic reflection profiles  and new geophysical data has led to an alternative interpretation. Seismic reflection profiles east of the spillway  extending to the subtidal portion of the Beacon Hill Moraine (Sirkin, 1982), show that the underlying semi-consolidated coastal plain strata are at an elevation of 40 m below MSL. This is > 5m below the

Relationship to Glacial Lake Connecticut
Based on the southwest to northeast orientation of the moraines in and around present day Long Island and Block Island Sounds , Glacial Lake Block Island had to predate Glacial Lake Connecticut. Long Island Sound was not ice-free until the Laurentide Ice Sheet retreated from the Harbor Hill -Charlestown Moraine position, which occurred around 21,000 yBP based on cosmogenic exposure dates . The highest water level for Glacial Lake Connecticut projects to 10 m below MSL at the eastern end of Long Island Sound (Lewis and Stone, 1991;Stone et al., 2005b). The highest delta plain-delta slope contact mapped in Block Island Sound (17 m below MSL) is at least 5 m lower than the highest Glacial Lake Connecticut water level. Lewis and Stone (1991) report a slowly lowering spillway at 'The Race' draining Glacial Lake Connecticut into Glacial Lake Block Island. The two lakes would have been separate until the spillway at The Race lowered to the projected elevation of the Block Channel spillway (25 m below present MSL) perhaps around 19,000 yBP (Stone et al., 2005a). Only then could one large lake have been continuous throughout Long Island, Block Island, and Rhode Island Sounds and Narragansett Bay.

Glacial Lakes Block Island and Rhode Island
Glacial Lake Block Island merged with Glacial Lake Rhode Island that was forming in present-day Rhode Island Sound. The combined lake would have continued to expand in size as the ice retreated to the north, eventually extending through most of Block Island and Rhode Island Sounds. A spillway at the Mud Hole controlled the water level of the eastern portion of Glacial Lake Rhode Island ( Figure   13B). The projected water level for the lower Block Island Sound and Pettaquamscutt River deltas is > 5 m lower than the initial water level of Glacial Lake Block Island.
Two scenarios can explain the drop in water level. The first scenario is that there was > 5 m of erosion through the moraine deposits and Coastal Plain deposits in Block Channel. As discussed previously, the boulders and cobble gravel pavement in Block Channel suggest that significant redeposition has not occurred. The second, favored interpretation is that spillway control for the lake switched to an outlet lower than Block Channel. The geomorphology of moraines in Rhode Island Sound show that while the moraine positions are correlative, the southern margin of the Narragansett Bay lobe of the Laurentide Ice Sheet extended further south than the adjacent Connecticut-Rhode Island lobe Stone and Borns, 1986) and the lobes were diachronous during their retreat (Larson, 1982;Smith, 2010). This behavior of the ice sheet kept the Mud Hole spillway separated from Glacial Lake Block Island until the margin of the ice retreated to the Charlestown-Buzzards Bay position. When the lower outlet became ice-free, the water level of the lake dropped > 5 m (Figure 11), and control switched to the lower Mud Hole spillway ( Figure 13C).

Glacial Lake Narragansett
Glacial Lake Narragansett began to form in present day Narragansett Bay when the Laurentide Ice Sheet retreated north of the Whale Rock End Moraine, at the entrance of the present-day West Passage of Narragansett Bay. Glacial Lake Narragansett was contiguous with Glacial Lake Rhode Island to the south ( Figure   14A) By 19,300 yBP, most of Narragansett Bay was ice-free, and Glacial Lake Narragansett extended through most of present day Narragansett Bay.
Three deltas (East Providence Plains, New Meadow Neck, and Conimicut Point) are outliers to the projected water level of Glacial Lake Narragansett. The East Providence Plains Delta is 3 m higher than the adjacent Blackstone River and Smith Hill deltas. This topographic relationship prompted earlier studies to suggest that the East Providence delta is older than adjacent deposits Smith, 1956). This delta may have been deposited in a smaller lake impounded behind the active margin of the ice sheet or a large block of stagnant ice in the Providence River valley. The former interpretation requires a complicated ice margin ( Figure 14B) (i.e. , that may be related to bedrock valleys in the present-day Seekonk and Barrington/Palmer River valleys . The Conimicut Point delta ( Figure 3) is the lowest projected delta in the study area, 10 m lower than the projected water level for Glacial Lake Narragansett ( Figure   11). This delta is interpreted to have been deposited in a later meteoric lake in upper Narragansett Bay that was significantly lower than Glacial Lake Narragansett. Lakes are considered meteoric when the ice sheet retreats out of the watershed. This lower water level occurred after the onset of isostatic rebound, and the delta probably formed as drainage from the Blackstone River flowed down the present Seekonk and Providence Rivers.

Glacial Lake Barrington
The New Meadow Neck delta (Figures 3) is significantly lower than the surrounding Barrington, Riverside and Warren River deltas ( Figure 11) (Smith, 1955).
Two possible scenarios could explain the topographic relationship of the New Meadow neck delta 1. The delta was deposited into, and graded to a lower water level of Glacial Lake Narragansett, or 2. The delta was deposited into a smaller glacial lake between the present Barrington and Palmer River valleys. The latter is the favored interpretation, and the lake is here called Glacial Lake Barrington. Glacial Lake Barrington would have been separate from both Glacial Lake Narragansett and Glacial Lake Taunton. No spillway for Glacial Lake Narragansett has been identified that could account for a drop in water level of Glacial Lake Narragansett prior to the deposition of the New Meadow Neck Delta.

Lake Drainage
The lack of a spillway at an elevation significantly lower than the water level of the lake (Figure 11), suggests that Glacial Lake Narragansett did not begin draining until the onset of isostatic rebound, which tilted the formerly horizontal water plane of the glacial lakes. Based on the geomorphology of Block Island Sound, Rhode Island Sound and Narragansett Bay, and the elevation of the spillways, later non-glacial lakes would have persisted in many of the deeper closed depressions throughout these areas. Ultimately, the lack of data, specifically high-resolution seismic reflection profiles across potential spillways, limit interpretations of lake drainage scenarios.

CONCLUSIONS
• Sub-bottom seismic reflection profiles and ground penetrating radar images are useful for determining elevation of the delta plain -delta slope contacts within glacial deltas submerged below present sea level and on highly developed delta surfaces above present sea level.
• The highest measured delta plain-delta slope contact was deposited in Glacial Lake Block island in present day Block Island Sound. Glacial Lake Block Island predated, and was at a lower elevation than the highest water level of Glacial Lake Connecticut in Long Island Sound. The water level of Glacial Lake Block Island dropped more than 5 m when the ice retreated north of Block Island, and spillway control shifted from Block Channel to the Mud Hole Spillway.
• Glacial Lake Narragansett began to form as the ice margin retreated north of the present-day entrance of Narragansett Bay; the Mud Hole spillway controlled the water level of the lake at this time. The 3 m drop in the water level of Glacial Lake Narragansett is interpreted to be the result of erosion of the Mud Hole Spillway.
• The water levels of Glacial Lake Narragansett, when projected onto preisostatic rebound topographic models, show that the lake extended throughout much of the basin currently occupied by Narragansett Bay, supporting the hypothesis that one large lake occupied the Bay during Late Wisconsinan deglaciation.
• The New Meadow Neck Delta, between the present Barrington and Palmer River Valleys is significantly lower than the surrounding deltas, and is interpreted to represent deposition in a separate glacial lake, named here Glacial Lake Barrington.
• Smaller, non-glacial lakes existed after the drainage of Glacial Lake Narragansett in present-day Upper Narragansett Bay and the Seekonk River valley, and probably existed in most of the closed depressions within presentday Narragansett Bay, Rhode Island and Block Island Sounds.
• The present elevation of these deltas plot on a linear trend, supporting the idea of delayed isostatic rebound in southern New England. This is in agreement with the regional uplift profiles measured in central New England and eastern New York, suggesting that New England did behave as a lithospheric block during isostatic rebound, as first suggested by Koteff and Larsen (1989     .        Figure 12: Comparison of uplifted water levels of Glacial lake Narragansett, Glacial Lake Hitchcock, Glacial Lake Merrimack (Koteff and Larsen, 1989;Koteff et al., 1993).  18,000 to 20,000 yBP, the constrained age of Glacial Lake Narragansett suggests that at least for the southeastern portion of the Laurentide Ice Sheet, deglaciation was well underway by this time.

INTRODUCTION
The absence of constraining radiocarbon ages and other accurate and precise dating techniques has left the chronology of initial deglaciation from the maximum position of the southeastern Laurentide Ice Sheet in New England only crudely estimated. This is in marked contrast to areas further from the terminal margin that are tied to abundant radiocarbon ages and a well-dated glacial varve chronology.
Previous estimates of the timing of deglaciation of southeastern New England used regional correlation and limited radiocarbon ages that have high uncertainties Stone and Borns, 1986;Uchupi et al., 2001). This paper presents a varve series from Glacial Lake Narragansett, and discusses its relationship with the North American Varve Chronology (NAVC) and other varve series (Antevs, 1922;Antevs, 1928;Ridge, 2011) and constrains the age of the lake in the context of the deglaciation of southeastern New England.
Glacial Lake Narragansett occupied much of the southern portion of the Narragansett Basin during Late Wisconsinan deglaciation . Our hypothesis is that varve records from Glacial Lake Narragansett could be correlated with the calibrated NAVC, providing age control on the northern terminus of the lake. The records from Glacial Lake Narragansett were not correlated with the NAVC, nullifying the hypothesis. However, using these uncorrelated varve sequences, and cosmogenic exposure ages from moraines in southern New England, minimum and maximum ages are placed on Glacial Lake Narragansett.
An important concept in glacial varve chronologies is that while distance from the ice margin is an important component to varve thickness (i.e., the closer to the margin of the ice sheet, the coarser and thicker the sedimentary couplets), regional weather patterns are more important. This is because a warmer, wetter melt season will supply more meltwater and sediment to a glacial lake than a colder, drier melt season. Varve records within separate proglacial lakes affected by the same weather conditions contain similar thickness patterns, and correlation between sequences is based on the pattern of couplet thickness through time. Throughout this work, the NAVC refers to varves deposited in Glacial Lake Hitchcock, which occupied a large portion of the Connecticut River Valley, and Glacial Lake Albany in the Hudson River Valley. The NAVC also includes correlated varve sequences from Glacial Lakes Merrimack, Ashuelot, and Winooski; however, these sequences are from northern New England and younger than Glacial Lake Narragansett Glacial varves sequences are described using terminology that reflects the level of correlation. The term 'varve sequence' is a generic term for any succession of varves. Varve record refers to a sequence of varves from a single outcrop exposure or drill core. Varve series represents a number of varve records correlated within a geographically constrained area (i.e. a single lake). Varve chronologies refer to a correlation of varve series that have a broader regional connection, usually between different lakes (Ridge, 2011).

STUDY AREA
The north to south trending, microtidal (spring range 1.47 m) (NOS, 2010b) Providence River is the uppermost portion of Narragansett Bay (Figure 1). The general geomorphology consists of subtidal flats (< 3 m below mean lower low water (MLLW)) that border a deeper (> 10 m MLLW) dredged navigation channel. The sites discussed in this paper are located 75 km from the Late Wisconsinan terminal margin of the Laurentide Ice Sheet . The Quaternary geology of the Providence River is dominated by glacial deltas along both the eastern and western shorelines Smith, 1956). Coring studies  and seismic reflection profiles  indicate that glacial lake floor deposits underlie most of the subtidal flats. GIS based reconstructions suggest Glacial Lake Narragansett occupied much of the southern portion of the Narragansett Basin, and was contiguous with Glacial Lakes Block Island and Rhode Island to the south (Figure 1) .

Previous work
Ernst Antevs (1922;1928) created the New England Varve Chronology by measuring the thickness of individual couplets at over 100 sites in New England and eastern New York. The thickness of each couplet was averaged using multiple, overlapping sections, and a composite curve spanning 4,152 years of deposition was created (Antevs, 1922;Antevs, 1928) Figure   2). This chronology is older than the NAVC from Lake Hitchcock   (Antevs, 1928). Antevs (1928) also reported three sequences from Rhode Island; Gaspee Point (102 years), Barrington (157 years) and along the Seekonk River (54 years) (Figure 2) . The Newfield section of Antevs has been extended to 171 years with recent drilling (Stone, 2012). Correlation of some of the shorter records, particularly the records from Glacial Lake Middletown in Connecticut may be problematic, and may overlap the NAVC (J.C. Ridge, pers commun).
Recent work has refined and extended Antevs original varve series using additional sites, radiocarbon ages on terrestrial plant macrofossils, and paleomagnetic declination records. Fine-tuning the calibrated age chronology continues today Ridge et al., 2001;Rittenour, et al., 2000). The varve chronology is now referred to as the North American Varve Chronology, covering 5,659 years (American Varve Year (AM) 2,700 to 8,358) extending from 18,200 yBP to 12,500 yBP (15,000 14 C yBP to 10,400 14 C yBP)) . Antevs (1928) was unable to correlate the three varve records near the shoreline of upper Narragansett Bay (Barrington, Gaspee Point and Seekonk) ( Figure   2), with the New England Varve Chronology.  reported rhythmic sediment couplets interpreted to be varves beneath Late Holocene estuarine sediment in the Providence River, but a detailed analysis of the varves was not completed.
Glacial lakefloor sediment has also been identified in surface sediment samples, cores, and interpreted seismic reflection profiles from Block Island and Rhode Island Sounds (Frankel and Thomas, 1966;Goss, 1995;Needell and Lewis, 1984;.

METHODS
Eight continuous sediment cores were recovered from the Providence River between 2006 and 2009. The coring equipment used here is based on the system of Lanesky et al. (1979), consisting of an 8 horsepower gasoline-powered concrete vibrator, connected to a section of aluminum irrigation tubing (7-cm inside diameter, wall thickness 1.8-mm). The vibration creates a low-amplitude standing wave, which temporarily liquefies the sediment in contact with the core barrel, leaving the remainder of the core undisturbed (Lanesky et al., 1979). Coring was accomplished using a specially outfitted 7 m pontoon boat with a 'moon pool' and a deck mounted tripod. Cores collected in 2009 utilized a quadrapod assembly placed directly on the bay floor.
The cores were prepared following the protocol of Ridge (2011). Holes (0.635 cm diameter, spacing 0.3 m) were drilled down the length of the core barrel to allow water to evaporate from the core for approximately two weeks prior to splitting. The cores were then cut into 1.5 m sections and split longitudinally. One half of the core was allowed to partially dry to maximize contrast between layers. The original sediment cores collected by  were not archived; however, photographic slides of the original cores were available. Slides were scanned at 4,000 dots per inch (DPI) using a Nikon slide scanner, and resized to reflect actual core dimensions at 600 DPI.
The Providence River cores were imaged using a GEOTEK core logging system, which generates continuous digital bitmap images in red/green/blue color schemes, and creates a scaled image at 300 DPI. The continuous images were split into overlapping sections (Approximately 600 x 800 pixels), in stratigraphic order from bottom to the top of the core. Varve thickness was determined following the protocol of Ridge (Ridge, 2011). The top of each summer and winter layer was digitized, using script written at Tufts University for Image Tool 3.0. Image Tool is free software designed by the University of Texas Health Science Center in San Antonio for processing and analyzing medical images. The varve analysis script keeps a running total of the thickness and count of each couplet, and exports the data as an ASCII text file for later analysis.

Varve matching
Couplet thickness for each core was plotted in Grapher TM v. 8.0 (Golden software, Golden CO, USA), for visual analysis and matching. Matching was accomplished using a desktop computer with two monitors; one displayed images of the cores, and the other displayed the graphical plots. This allowed for simultaneous visual matching between cores using distinct couplets (specifically, couplets with thick non-melt season layers (>1 cm)), and graphical 'wiggle' matching, based on the pattern of couplet thickness through time. Once correlated, the thickness of each varve was averaged and a composite series was created.
Following the convention of the NAVC and Swedish varve series (i.e. Antevs, 1922;DeGeer, 1975;Wohlfarth et al., 1995), glacial varves are numbered in order of deposition, and older varves have a lower (number) varve year than younger varves. The varves in this study were deposited near the northern terminus of Glacial Lake Narragansett. Contiguous glacial lakes existed in the southern end of Narragansett Basin (Glacial Lake Narragansett), Rhode Island Sound (Glacial Lake Rhode Island), and Block Island Sound (Glacial Lake Block Island) ( Figure 1). The base of the sequence in this study was assigned a varve year of 5,000.
This should provide enough time to span the existence of glacial lakes in Block Island Sound, Rhode Island Sound, and Narragansett Bay if future work can expand the varve series into these lakes.
The composite curve from Glacial Lake Narragansett was compared to varve sequences from Antevs (1928), downloaded from the database maintained by Ridge (Ridge, 2011). Graphical matching was carried in the same manner discussed above.
Photographs of Antevs original outcrops were not available, so visual matching of his records was not possible.

RESULTS
Five of the recovered cores collected in this study contain sediment couplets interpreted to be varves, as do the two cores of . Individual varve records span 27 to 200 years. The basic stratigraphy of the cores is summarized in

Core deformation
Some sections of the varve sequences are deformed. It is interpreted that this deformation was induced during coring although some could be the result of mass movement (i.e. delta slumping (Stone, 1976)). Distortion of couplets often coincides with thicker (>1 cm) non-melt season layers, suggesting some deformation induced during coring could be due to the rigidity of the winter layers. 'Arching' of couplets, where the edges of the sediment are pulled down by the core barrel is common, but the internal stratigraphy and thickness remains intact and does not impede accurate measurement of couplet thickness

Varve correlations
Individual varve records were correlated into a series spanning 265 years The gap between the 166 year and 8-year sequence is estimated to only be two years (GLN years 5,210), and the gap between the 8 and 31-year sequence is estimated to be no more than 9 or 10 years (GLN years 5,223 to 5,233) ( Figure 3B).
A correlation exists between the Seekonk sequence of Antevs (1928), and the composite curve created in the Providence River ( Figure 3C). Slight discrepancies suggest a varve may be missing in each of the sequences. Core PC-6  was not correlated with the cores from further north in the Providence River, but a 79-year correlation exists between core PC-6 and Antevs' record from Gaspee Point ( Figure 3D).

Nature of deposition
Deposition in glacial lakes is dominated by underflows driven by the density contrast between sediment laden river water, and the lake. Measurements at Malaspina Glacier in Alaska indicate that suspended sediment increases the density of streams draining the glacier 1.7 -4.7 times the maximum density of freshwater (Gustavson, 1975). Density underflows are interpreted to have deposited varves in Glacial Lake Hitchcock, and this process is responsible for the laminated, usually graded, melt-season layer of the couplet . More (and coarser) sediment is deposited closer to the source of sediment discharge into the lake, with a decrease in grainsize and couplet thickness in more distal areas. There is a very slight decrease in varve thickness from 2.2 to 1.9 cm (r 2 = .015) over the 265-year record. The decrease in thickness through time is probably more significant, but is skewed because many of the thick, sandy, ice-proximal varves at the base of the cores are too deformed to accurately measure. This decrease is expected, coinciding with increasing distance from the source of sediment discharge.
Basal varves in core PC-6 , and in the Seekonk sequence of Antevs (1928)  This represents an increase in the volume of sediment introduced into the lake during the melt season that increased the suspended sediment available for deposition during the non-melt season. Thickness of non-melt season is generally positively correlated with melt season thickness, and has no direct relation to winter temperature .

Correlation with the NAVC
The major assumption in glacial varve chronology studies is that varve thickness is controlled by regional weather conditions, and varves can be correlated between lakes. The correlation between varve records from within Narragansett basin (i.e. the correlations of the Providence River composite curve and the Seekonk sequence of Antevs, (1928) ( Figure 3C) and between core PC-6  and the varve record from Gaspee Point (Antevs, 1928) ( Figure 3D)), suggest that at least within Glacial Lake Narragansett, these sequences can be correlated to eachother.
Varves from Glacial Lakes Ashuelot and Merrimack, which are similar in size to Glacial Lake Narragansett, have been correlated to the NAVC, as have ice proximal varves (> 40 cm thick) from Glacial Lake Albany (Antevs, 1928;.
The intra-basin correlation within Glacial Lake Narragansett, suggest that the varve records presented here should correlate with records from other lake basins if they overlap temporally.
The varve series generated from the Providence River was not correlated with the NAVC or other varve series from New England and eastern New York. Our interpretation is that the varves measured from Glacial Lake Narragansett are older than both the varve sequences from the NAVC and the uncorrelated sequences in New York, Connecticut and southeastern Massachusetts.

Regional context of uncorrelated varve sequences
The lack of correlation between the varve sequences prevents using the calibrated NAVC timescale to provide age control for Glacial Lake Narragansett.
Uncorrelated varve series represent minimum time of deposition within a lake (i.e. within the same lake, two 100 year, uncorrelated series represent at least 200 years of deposition). 604 uncorrelated varves have been identified from the Providence River, representing a minimum time of glacial lakefloor deposition in the present-day Providence River prior to the retreat of the Laurentide Ice Sheet out of the watershed.
Timing of deglaciation in southern New England historically has been based on a limited number of radiocarbon ages, the calibrated NAVC, and regional correlation. More recently, cosmogenic exposure ages of boulders on recessional moraines have provided additional ages . Initially, exposure ages were consistently younger than terrestrial and marine radiocarbon ages and the calibrated NAVC ). The regional production rate for Beryllium 10 has since been adjusted, bringing the NAVC and exposure ages sets into closer alignment . The offset between the cosmogenic exposure age and the actual deposition of the landform, however, remains difficult to quantify (Applegate et al., 2011;Balco, 2011).

Correlation with Northern Hemisphere climate
The retreat of Laurentide Ice Sheet was linked to hemispheric-scale climate changes, , and some correlation appears to exist between varve thickness in the NAVC and oxygen isotope records from Greenland ice core data after 15,000 yBP . , proposed that recessional moraine formation at the southern margin of the Laurentide can be correlated to colder (more negative δ 18 O) intervals in Greenland ice cores.
Assuming this hypothesis is correct, the exposure ages should intersect cold periods in Northern Hemisphere climate, if the age is synchronous with the deposition of the landform.
Comparing the cosmogenic exposure ages with the δ 18 O record from the synchronized NGRIP, GRIP and GISP2 ice cores (Rasmussen et al., 2008), the exposure ages do not intersect significant cold periods. We interpret the formation of these moraines as correlative with colder periods at 20,550 and 20,400 yBP ( Figure 6).
These ages are 200 -250 years older than the reported cosmogenic exposure ages of Ledyard-Congdon Hill and Old Saybrook -Wolf Rocks moraines ( Figure 6) ), and differ slightly from the original correlation of , based on the improved resolution of the NGRIP chronology. The proposed ages for the moraines are within 1 standard deviation of the reported exposure ages (+/-> 500 years) , and may be closer to the actual exposure ages based on updated cosmogenic production rates for southern New England (G. Balco, pers. commun). Older ages suggested for the Wolf Rocks and Old Saybrook moraines (20,700 (Stone, 2012)) correlate with a relatively warm period record in Greenland, suggesting that deposition at that time would have had to be the result of local ice dynamics. While direct correlation between varve thickness and Greenland temperature prior to 15,000 yBP is weaker than after, some relationship between moraine formation and varve thickness is still evident .

Constraining the age of Glacial Lake Narragansett
Summing the uncorrelated varve sequences older than the base of the NAVC (1,200 years), the minimum age of Glacial Lake Narragansett is 19,400 yBP ( Figure   7). Even if the 150 years of deposition recorded by the short varve records in Connecticut are excluded, a minimum 1,050 years elapsed between the deposition of the oldest varves in the NAVC and the youngest varves in Glacial Lake Narragansett.
Given the uncertainty in the length of time between deposition of individual varve sequences, the estimate of 1,200 years seems conservative. The former ice margin in the Providence River areas is complicated , and it is unclear if the Barrington or Gaspee-Pawtuxet Cove sequence is the oldest measured in Glacial Lake Narragansett. Either way, the oldest varve sequences in the presentday Providence River extend to at least 20,000 yBP.
Glacial Lake Narragansett could not have begun to form until the southern margin of the Laurentide Ice Sheet retreated north from the Whale Rock moraine at the entrance to the present-day West Passage of Narragansett Bay. This moraine is interpreted to be correlative with the Wolf Rocks moraine (Figure 2) Smith, 2010). Cosmogenic exposure ages for a moraine in southeast Connecticut correlative with the Wolf Rocks moraine place the age at 20,300 yBP ).
The ice retreat rate in central New England based on basal varve sequences in the NAVC (Ridge, , 2004) is estimated to be 30 -90 m . yr -1 prior to 14,700 yBP.
Assuming the cosmogenic age is the actual age of the landform, this requires the ice to retreat > 35 km in approximately 300 years, at a rate of > 120 m . yr -1 to reach the oldest measured varve sites in the Providence River by 20,000 yBP. Ages based on the correlation with northern hemisphere climate assign an age of 20,550 yBP on the Wolf Rocks and 20,400 yBP on the Congdon Hill moraine. This allows 500 years between the ice retreating from the moraine and the southern margin of the ice sheet reaching the present-day Providence River, with a retreat rate of 70 m . yr -1 , which falls in the middle of the range for regional ice retreat. The relatively low and uniform δ 18 O prior to 15,000 yBP Rasmussen et al., 2008), supports a systematic retreat rate for the Laurentide Ice Sheet during the proposed age of the lake.
The proposed age range for Glacial Lake Narragansett (20,500 -19,400 yBP) is considerably older than previous estimates of Uchupi et al. (2001) (18,700 -18,100 yBP), and Boothroyd and August (2008) (19,000 -17,600 yBP). This proposed age of Glacial lake Narragansett does not imply that the lake drained at 19,400 yBP, and a non-glacial, meteoric lake likely persisted until the onset of isostatic rebound. While the last glacial maximum for the Laurentide Ice Sheet is often presented as 18,000 to 20,000 yBP (i.e. Denton et al., 2010;Denton and Hughes, 1981), the constrained age of Glacial Lake Narragansett presented here, shows that at least in southeastern New England, deglaciation was well underway by this time.

FUTURE WORK
The potential remains for a correlating the Hudson-Quinnipiac and Glacial Lake Narragansett varve sequences with the NAVC. The oldest sequences at the base of the NAVC and Hudson-Quinnipiac are interpreted to contain more varves in the deeper subsurface. It is interpreted that the base of Glacial Lake Hitchcock varve sequence is less than 800 years older than the present base of the NAVC . Based on regional models of deglaciation, 800 years may long enough to span the current gap between the NAVC and Hudson-Quinnipiac sites. Similarly, the oldest varves measured by Antevs (1928) at both the Hudson and Quinnipiac sites are not the base of these sequences. Extending the base of this sequence would be necessary to correlate with the youngest varves measured in Glacial Lake Narragansett. Additional varve records can be examined from the northern end of Glacial Lake Narragansett, however, Antevs outcrops were at or near the top of the varve section, and it seems unlikely that there are hundreds of varves younger than the sequences measured here.

CONCLUSIONS
• A 265-year varve series from Glacial Lake Narragansett was constructed from eight continuous sediment cores collected from the Providence River, Narragansett Bay, Rhode Island • The Glacial Lake Narragansett varve series was not correlated with either the  Antevs (1922Antevs ( , 1928   Orange areas are the mapped extent of glacial deltas around the Providence River discussed in the text McCandless, 2002, 2003;Smith, 1955Smith, , 1956. Abbreviations: RIV -Riverside, PP -Providence Plains, EPP -East Providence Plains. Inset map of Southern New England shows extent of the study area, the maximum extent of the Laurentide Ice Sheet modified from  and the projected extent of Glacial Lakes Narragansett, Rhode Island and Block Island .  , Goldsmith (1982), Sirkin (1982), , , Ridge, ( , 2004 and Stone et al., (2005b). All ages reported as calendar years before present. 1 120 Figure 3: Correlation of varve records based on close annual matches in total couplet thickness. A. Correlation of varve records collected in the Providence River for Narragansett Bay varve years 5000 to 5150. A constant offset was added to some of the records to display them on one graph without overlap. EW-1 was originally collected by . B. Correlation of varve records collected in the Providence River for Narragansett Bay varve years 5150 to 5260. A constant offset was added to some of the records to display them on one graph without overlap. C. Correlation of varve records from the Providence River and Seekonk River (Antevs, 1928). Arrows point to two discrepancies, where it appears one varve may be missing from each sequence. The Providence composite curve was offset by 3 cm to display the records on one graph without overlap. D. Correlation of varve records from Pawtuxet Cove  and Gaspee Point Varve thickness of PC-6 was offset by 3 cm to display the records on one graph without overlap.    . Dashed black lines refer to potential correlation with Greenland Ice Cores LIS readvance/still stands.  . Green arrows indicate proposed correlated ages of the Congdon Hill and Wolf Rock end moraines. Blue arrow marks the maximum range of Glacial Lake Narragansett.

Introduction
This paper synthesizes the Quaternary depositional environments and deglacial evolution of Narragansett Bay based on ~ 800 km of high-resolution seismic reflection profiles, geotechnical boreholes, and published and unpublished Quaternary maps.
The hypothesis carried throughout this work is that a single lake, Glacial Lake Narragansett, occupied much of southern Narragansett Basin and was contiguous with lakes in adjacent areas during Late Wisconsinan deglaciation. This paper focuses specifically on the elevation and continuity of the lakefloor deposits, the distribution of ice-marginal lacustrine fans and the volume of sediment deposited in Glacial Lake Narragansett. Several previous studies focused on the stratigraphy of Narragansett Bay McMaster, 1984;, but this work represents the first detailed mapping of the Quaternary glacial depositional environments. This mapping provides an understanding of the behavior of the southeastern portion of the Laurentide Ice Sheet during the early stages of deglaciation, and provides a better understanding for managers of subsurface resources of Narragansett Bay, particularly for future dredging, disposal of dredged material, marina construction and siting of offshore wind farms.
Present-day Narragansett Bay lies within Narragansett Basin, a complex of non-marine metamorphosed sedimentary rocks deposited in an intermontaine rift basin and later deformed by the Allegehanian Orogeny in Late Carboniferous-Permian time (Mosher, 1983;Murray et al., 2004). The underlying bedrock is mostly the Rhode Island Formation, a coarse to fine-grained metamorphosed sedimentary rock, with some Cambrian and Precambrian igneous and metamorphic rocks   (Denny, 1982). Glacially transported Cretaceous sediment has been sampled in discrete blocks in terminal moraines at Block Island, Rhode Island (Sirkin, 1976;Stone and Sirkin, 1996), and Coastal Plain sediment has been interpreted in seismic reflection profiles from Rhode Island Sound, south of Narragansett Bay (McMaster and Ashraf, 1973;Needell et al., 1983).
Numerous northern hemisphere glaciations likely covered Narragansett Bay with glacial ice during the Quaternary Period . In-place evidence of all but the most recent (Wisconsinan) glaciation was removed by subsequent glaciations, although deformed, pre-Late Wisconsinan age (Illinoian) beds have been reported at Block Island (Sirkin, 1976) . The Late Wisconsinan Laurentide Ice Sheet reached its terminal position south of New England at the last glacial maximum around 26,000 yBP, before beginning to retreat northward Peltier and Fairbanks, 2006;Stone and Borns, 1986). Meltwater issuing from the ice deposited sediment in a variety of depositional environments as the Laurentide ice sheet retreated (Koteff and Pessl, 1981;. Marine water first inundated Narragansett Bay after 10,200 yBP (sea level 30 m below present), and continued to transgress up Narragansett Bay. By 2,500 yBP, (sea level 2.5 m below present), Narragansett Bay looked similar to the present configuration McMaster, 1984). Present-day Narragansett Bay is microtidal (spring tidal range 1.2 m at Newport, 1.5 m at Providence), mixed-energy estuary according to the classifications of Dalrymple et al., (1992) and Hayes (1979).

Sub-bottom seismic reflection profiling
Subsurface interpretations are based on 785 km (490 mi) of sub-bottom seismic reflection profiles collected in Narragansett Bay and adjacent Rhode Island Sound (Figure 1, inset), using an EdgeTech, SB-216S Full-Spectrum sub-bottom profiler (Figure 2), operated at a frequency sweep of 2-10 kHz (vertical resolution < 15 cm (Edgetech, 1998)). Towfish height was maintained at 1 m below the surface of the water, towed at a speed of < 1.5 m . s -1 . Spatial location was embedded into the sub-bottom files using the serial NEMA output of a Trimble DSM-132 GPS with a reported accuracy of +/-1 m ). Depth to reflectors was calculated using an acoustic velocity of 1,500 m . s -1 . Profiles were post-processed using Chesapeake Technologies (Mountain View, California) Sonar Web v. 3.16. See Appendix 4 for a detailed discussion on seismic reflection profile collection and processing.

Interpretation
Interpreting seismic reflection profiles is done by identifying seismic facies.
Seismic facies are sedimentary packages, distinguishable from adjacent units based on internal characteristics, (i.e. the intensity, spacing, continuity, and internal geometry of seismic reflectors), external geomorphic form, and stratigraphic relationship to other units .

Thickness and Volume of stratified deposits
The thickness of stratified deposits was calculated using Chesapeake Technologies SonarWeb TM software, utilizing the 'Seismic Reflectors Thickness' tool, which calculates the algebraic difference between digitized reflectors. This included lakefloor deposits (varves), distal lacustrine fan, and distal delta slope deposits.
Proximal lacustrine fans and delta plain deposits were not included due to limited seismic penetration.
The volume of deltas around Narragansett Bay was estimating using borehole records (Allen, 1956;Allen and Gorman, 1959;Bierschenk, 1954;Halberg et al., 1961;Johnson, 1962;Johnson and Marks, 1959) and the mapped extent McCandless, 2002, 2003;Schafer, 1961;Smith, 2010). The volume of lakefloor sediment within Narragansett Bay was determined by interpolating surfaces representing the elevation till/bedrock and moraine deposits, and the upper surface of glacial lakefloor deposits in a GIS environment. The volume of sediment contained between the surfaces was determined using the Cut/Fill tool within ESRI ArcMap TM Spatial Analyst extension. See Appendix 5 for details of interpolation and assessment of the surfaces.

Water depths and volume of Glacial Lake Narragansett
A single, continuous raster surface that reflects the isostatically uplifted water level of Glacial Lake Narragansett was created in ESRI ArcMap TM 10.1, based on the elevation of the deltas and minimum lake extent determined in Chapter 2. The interpolated lakefloor surface was subtracted from the water surface to approximate the water depth of Glacial Lake Narragansett, and the volume of the lake was estimated using the ESRI ArcMap TM v. 10.1 Spatial Analyst Cut/Fill tool.

Seismic facies and interpreted depositional environments
This study identified eleven distinct seismic facies in Narragansett Bay, which are discussed below. See table 1 for summary of facies.

Facies T/R: Till/Bedrock
This facies represents the stratigraphically lowest reflector, and is not penetrated by the seismic signal. Differentiating between till and bedrock in the subsurface with the seismic reflection profiler used was not possible in most instances.
The topography of this reflector varies from flat to high relief; the highest relief (up to 30 m) is seen in east to west seismic lines (Figure 2), strongly controlled by the regional trend of the bedrock, with a dominate strike of NNE Murray et al., 2004;Reck and Mosher, 1988). This facies is interpreted to be bedrock, thin till over bedrock or thick till deposits. Reflectors were traced laterally to shoreline exposures of bedrock and till . Boreholes in the Providence River and West Passage sampled till or 'till-like' deposits 3 -10 m thick, overlying metamorphosed sedimentary rocks of the Narragansett Basin (USACE, 1957).

Facies EM: End moraine.
Facies EM, is the stratigraphically lowest reflector when encountered, and has a different geomorphic orientation than facies T/R. The upper reflector is intense, with high relief, hummocky topography in a north to south survey lines. There was little penetration of the seismic signal, but where visible, the internal reflectors were chaotic. This facies was not sampled by any boreholes or cores, but is likely composed of a mixture of till and stratified material, interpreted to have been deposited as recessional end moraines. These deposits mark stillstands or fluctuations of the Laurentide Ice Sheet.

Facies IM: Ice marginal deposits
Facies IM is characterized by an extremely hummocky, intense external seismic reflector. This facies is interpreted to represent ice-marginal deposition, primarily as braided rivers flowing between the ice sheet and valley walls and/or debris flows off of the ice. Terrestrial analogs are composed mostly of stratified sand and gravel . This unit is continuous with ice-marginal deposits in existing Quaternary maps Smith, 2010) adjacent to steep bedrock hills along the margin of present-day Narragansett Bay.

Facies HM: Hummocky moraine
Facies HM is similar in morphology to facies IM, with a hummocky, intense external seismic reflector. Seismic penetration in this facies was limited, but occasional parabolic reflectors are interpreted to be large (> 1 m) boulders. This facies was only mapped in a wide (> 1 km) area in the West Passage, and is continuous with units mapped adjacent to Narragansett Bay (Schafer, 1961;Smith, 2010). This unit is interpreted to be fluvial sand and gravel interbedded with debris flow till, deposited around and/or on debris-covered ice. The hummocky topography is the result of burial and subsequent melting of ice blocks during deglaciation (Schafer, 1961;Smith, 2010). Sand and gravel reported in boreholes southwest of Greenwich Bay (USACE, 1957), are correlative with interpreted proximal lacustrine fans in seismic profiles.

Facies DF: Distal lacustrine fan
Facies DF is continuous with facies PF, and often onlaps PF and T/R ( Figure   3). Internally, reflectors are well defined, and are sub-parallel to parallel. Externally, this facies fills topographic lows, is 3 -5 m thick, and individual sedimentary packages tend to thin progressively to the south. This facies is interpreted to be composed of mostly sand, deposited in the distal portion of a lacustrine fan. This unit was probably sampled in boreholes and cores, however distinguishing this unit from sandy lakefloor deposits (Facies GLF) are not possible.

Facies GLF: Lakefloor
Facies GLF is the most ubiquitous glacial facies in the study area, is This facies was sampled in boreholes and vibracores throughout Narragansett Bay USACE, 1957).

Facies DS: Delta slope
Facies DS is the least common glacial depositional environment mapped.
This facies is typically 5 -10 m thick, is characterized by steeply dipping internal reflectors (10 -20º) (Figure 4), and stratigraphically overlies facies GLF. This unit is interpreted to be composed of sand and gravel, deposited as a glacial delta. The steeply dipping reflectors represent proximal delta slope deposition. Seismic penetration was limited within portions of the deltas now submerged below present sea level along the present-day western shoreline of Narragansett Bay, and the mapped extent is largely based on the geomorphology of the area.

Facies E: Estuarine Channel
Facies E is identified by an intense basal reflector that truncates underlying units as an erosional unconformity ( Figure 5), and is always stratigraphically above the units discussed previously. This unit is interpreted to represent post-glacial fluvial channels modified during Holocene marine transgression. This facies has been sampled in boreholes and cores throughout Narragansett Bay, and is composed of a variety of sediment types, ranging from gravel to silt with shells and other organic matter are common USACE, 1957).

Facies M: Estuarine mud
Facies M is ubiquitous throughout much of Narragansett Bay, occurring as an acoustically transparent layer that drapes the underlying units up to 15 m thick. This facies is interpreted to be estuarine mud deposited in low-energy basins (Figures 2 -5).
Where sampled in boreholes (USACE, 1957), it is composed primarily of organic silt or clay, with marine shells common. Occasionally, in low-energy basins, the lower half of the unit shows a slightly darker seismic return, without a distinct seismic reflector. This could represent changes in the sediment characteristics reflecting increasing water depths during transgression Vinhateiro et al., 2007).

Facies NG: Natural Gas
Facies NG has a distinct seismic signature, with a dark, opaque upper seismic reflector that typically has a convex up reflection that obscures or 'wipes out' the underlying seismic record ( Figure 5). This facies is interpreted to represent gas bubbles in the sediment. This gas is likely in the form of buried methane formed from decayed organic matter. The source of the organic matter is probably a combination of freshwater and saltwater marsh peat, post-glacial lake gyttja, and organic rich estuarine sediment. Natural gas is common in the subsurface of other glaciated estuaries Ussler et al., 2003).

Facies ADM: Anthropogenic dredged material
Facies ADM is rare in Narragansett Bay. Where mapped it has a strong upper reflector, with a convex external form. Where visible, internal reflectors are chaotic.
Facies ADM is stratigraphically the uppermost facies, although a thin drape (< 15 cm) of facies M may be present. This facies is interpreted to be dredged material placed adjacent to navigational channels.

Thickness and volume of stratified deposits
The thickest stratified deposits are located in the deepest portions of the Pre-Wisconsinan bedrock valleys now occupied by the East and West Passages, Sakonnet River and Mount Hope Bay. This study estimated that stratified deposits range from < 1 m to > 50 m ( Figure 6). Stratified deposits in Glacial Lake Narragansett cover > 425 km 2 , with a total volume of 8.7 x 10 9 m 3 ( Table 2). Similar volumes have been calculated elsewhere. The volume of seven glacial marine deltas in eastern Maine of similar size (250 km 2 ) to those in Glacial Lake Narragansett was estimated at 4.7 x 10 9 m 3 (Ashley et al., 1991). This estimate did not include sediment deposited as glacial marine mud, analogous to the lakefloor sediment (> 3.5 x 10 9 m 3 ) deposited within Glacial Lake Narragansett.  Table 2: Sediment volume for deltas and lakefloor deposits in Glacial Lake Narragansett in and around present-day Narragansett Bay

Morphosequence concept
The Quaternary glacial depositional environments of the seismic facies were interpreted, and grouped into morphosequences. Morphosequences are timeequivalent groups of landforms that extend from the collapsed former ice margin at the proximal end of the deposit, with a general decrease in grain-size towards more distal portions of the sequence (Koteff and Pessl, 1981;. Traditionally, in terrestrial Quaternary mapping, morphosequences are identified using were not linked to an individual morphosequence. Lakefloor deposits (Qlf), which overlie older deposits, were not associated with any individual morphosequence, as deposition occurred during the entire deglaciation of Narragansett Bay. Eskers, which are a common feature on terrestrial Quaternary maps, could not be differentiated from other deposits (i.e. proximal fans) due to their limited spatial extent.

Ice margin positions and ages
Ice margins in the southern portion of Narragansett Basin were identified using a combination of proximal lacustrine fan and recessional end moraine deposits interpreted from high resolution sub-bottom seismic reflection profiles. The ice-margins were placed along the up-ice side of recessional moraine deposits, and behind the crest of proximal lacustrine fans. Margins were correlated with previously published ice margins from adjacent areas Smith, 2010;Stone and Borns, 1986;Stone and Peper, 1982). The age of the Wolf Rocks -Whale Rock and Congdon Hills-Bonnet Point ice margins are correlated to cosmogenic beryllium (Be 10 ) exposure ages of moraines in eastern Connecticut  and with cold intervals in Greenland ice core records Svensson et al., 2006).

Deglaciation of Narragansett Bay
Deglaciation of what became present-day Narragansett Bay began when the indicate that the moraine was deposited between 20,300 yBP , and 20,550 yBP . Drainage down the present-day Pettaquamscutt River Valley began at this time, depositing a delta (QdPR) (Plate 1) that was graded to Glacial Lake Rhode Island.
The margin of the ice continued to retreat north into the present-day West Passage of Narragansett Bay, and a stillstand or fluctuation of the active ice margin formed the Bonnet Point end moraine (QemBPT) (Figure 8, Plate 1). This topographic feature was first proposed as a moraine by , and , and is interpreted to correlate with the Ledyard -Congdon Hills moraine. This moraine was deposited between 20,200 yBP , and 20,400 yBP . Subglacial drainage in the center of the present-day West Passage deposited the Dutch Island Delta (QdDI) (Plate 1).
Lakefloor deposits seen in seismic reflection profiles below the delta (Figure 4) suggest this landform started as a lacustrine fan and formed a progradational delta in the manner presented by . The present-day East Passage and Sakonnet River were both still occupied by ice at this time.
The Annaquatucket sequence (QimAN, QfAN, QdAN) began to form as the ice margin retreated from the till hill along the west side of the present-day lower West Passage, and drainage shifted from the Pettaquamscutt River Valley to the Hunt-Annaquatucket valley (Schafer, 1961;Smith, 2010). Meltwater from the Hunt-Annaquatucket flowed down the lower West Passage of Narragansett Bay, between the valley wall to the west and a tongue of ice in the deeper portion of the valley (Schafer, 1961). The sequence ends in a delta deposited into Glacial Lake Narragansett in the West Passage (QdAN) (Plate 1). The hummocky moraine deposits at the head of the Annaquatucket sequence (QhmAQ) extend south and east from Fox Island. These deposits are correlative with ice marginal deposits in the center of the Sakonnet River. No correlative ice-marginal or moraine deposits were mapped in the East Passage.
After retreating from the Annaquatucket sequence, drainage down the pre-Wisconsinan bedrock valley west of Narragansett Bay  began to deposit the first in a series of very large (20 -47 km 2 ) deltas into Glacial Lake Narragansett. The Mill Creek Delta (QdMC) surrounds Wickford Harbor. North of the Mill Creek sequence, a stillstand or fluctuation of the ice margin is marked by the Quonset Point end moraine (QemQP) adjacent to Narragansett Bay (Schafer, 1961;Smith, 2010). Moraine deposits offshore of Quonset Point are correlative with this position, as are lacustrine fans north of present-day Conanicut Island, and ice marginal deposits around Dyer Island in the East Passage and in the Sakonnet River valley (Figure 9; Plate 1).
The Hunt-Quonset delta that extends from Wickford into Narragansett Bay was deposited when the ice retreated from the Quonset Point to the Quidnessent (QemQUI) moraine positions ( Figure 9) (Schafer, 1961;Smith, 2010). Two segments of the Quidnessent moraine were mapped between Hope Island and the western shoreline of Narragansett Bay, that are now completely buried by lakefloor deposition.
This ice margin is correlative with lacustrine fans east of Hope Island, ice marginal deposits around Dyer Island in the East Passage, and with the Sakonnet end moraine (QemSR) in the center of the Sakonnet River valley. This also marks the onset of deposition of a prominent set of lacustrine fans that extends from this ice margin north to the deposits at Prudence Island (QimPI) (Plate 1).
The Potowomut delta along the west side of Narragansett Bay was deposited as the ice margin retreated north from the Quidnessent position Schafer, 1961;Smith, 2010 (USACE, 2001), and these deposits are interpreted to be proximal lacustrine fans overlain by glacial lakefloor deposits. The southern margin of the ice sheet continued to retreat north, depositing the Smith Hill delta northwest of the Providence River, and the Blackstone delta, which was deposited into an arm of Glacial Lake Narragansett in the present-day Seekonk River. Assuming an ice retreat rate of 70 m yr -1 (Ridge, 2004), Narragansett Bay was ice-free before 19,500 yBP, although the ice remained in the watershed until at least 19,300 yBP .

Late Pleistocene non-glacial deposits
The Conimicut Point deposits (QfCP, QdCP) (Plate 1) are markedly lower (10 m) than the projected water level recorded by the surrounding deltas (QdWP, BT, RIV, etc) (Plate 1). This is interpreted to represent deposition into a smaller, nonglacial lake in upper Narragansett Bay, probably after the onset of isostatic rebound and initial drainage of Glacial Lake Narragansett as discussed in Chapter  gravel ), but boreholes in the narrow area between Aquidneck Island and the adjacent upland east of the Island Park delta penetrated >200 feet of sand (USACE, 1957). The extent of these fluvial deposits appears to be limited to the sandy, central channel of the present-day Sakonnet River.

Extent of lakefloor deposits
The hypothesis carried throughout this work is that one large lake, Glacial Lake Narragansett, occupied much of the southern portion of Narragansett Basin and was contiguous with lakes in adjacent waters. Paramount to this is determining the continuity of lakefloor deposits in the different geographic areas within Narragansett Bay. If glacial lakefloor deposits are restricted to the deeper basins, then it could be argued that separate, smaller glacial lakes existed. Glacial lakefloor deposits are ubiquitous throughout Narragansett Bay (Figure 9), supporting the hypothesis that one large lake occupied much of southern Narragansett Basin. Natural gas and estuarine deposits (Facies NG and E) obscure the seismic reflection for short distances (e.g. across the dredged channel at Quonset Point and in portions of the Providence River), but these gaps are typically less than 500 m. The seismic record is obscured in parts of the East Passage but silt and clay reported in boreholes from this area (USACE, 1957) are interpreted to be glacial lakefloor deposits.
The only portions of the bay where lakefloor deposits are not seen continuously on seismic reflection profiles is in the channels connecting the East Passage with Mount Hope Bay, and Mount Hope Bay with the Sakonnet River ( Figure   9). The gap between the mapped lakefloor deposits in both areas is < 5 km. Mount Hope Bay and the East Passage are connected by a narrow (< 1 km), deep (> 25 m) incised channel, between two till uplands (Plate 1). Seismic penetration in this area was very limited due to coarse grained surface sediment (gravel) .
No geomorphic evidence exists to suggest that Glacial Lake Narragansett did not extend into present-day Mount Hope Bay via the East Passage.
It is unclear if the glacial lake in the present-day Sakonnet River was contiguous with the both the lake to the south in Rhode Island Sound (Glacial Lake Rhode Island), and Glacial Lake Narragansett in Mount Hope Bay. A 3 km long, 0.5 km wide channel connects the present-day Mount Hope Bay and the Sakonnet River.
The favored interpretation is that an arm of Glacial Lake Rhode Island extended into the Sakonnet River valley, but was not connected with Glacial Lake Narragansett in Mount Hope Bay. The lakes would have been separated by the Island Park delta (QdIP) that originally extended across the head of the present Sakonnet River. The present channel connecting Mount Hope Bay with the Sakonnet River was incised by the post-glacial Taunton River.

Elevation of glacial lakefloor sediment, minimum water-depths, and volumes of Glacial Lake Narragansett
If any of the lakefloor deposits are higher than the interpreted water level of Glacial Lake Narragansett, more than one lake may have existed in Narragansett Bay, and an alternative hypothesis would have to be discussed. The upper surface of lakefloor deposits digitized from seismic reflection profiles were plotted as a point cloud based on their location and elevation. The upper surface of the lakefloor deposits in each geomorphic area of the bay (i.e. East Passage, Providence River etc).
These surfaces were compared to the elevation of delta plain-delta slope contacts around present-day Narragansett Bay, and the interpreted water levels of Glacial Lake Narragansett ( Figure 10) (BAO Water-level chapter)). No lakefloor deposits were mapped higher than the projected water levels of Glacial Lake Narragansett, supporting the hypothesis that one large lake occupied most of present-day Narragansett Bay during deglaciation.

Lacustrine fans and subglacial drainage
The systematic northward retreat of the ice margin through Narragansett Bay is recorded by the deposition of lacustrine fans at the grounded margin of the ice sheet.
The presence of lacustrine fans indicates that a significant amount of meltwater (and sediment) issuing from the ice sheet was subglacial (Shreve, 1972) (Figure 12), similar to the present drainage of the Greenland Ice Sheet . The lacustrine fans often onlap a preceding fan, indicating that the subglacial tunnels stayed open, and in the same position as the ice margin retreated north (Banerjee and McDonald, 1975), and did not reorganize each melt season in the manner presented for the Greenland Ice Sheet (Bartholomew et al., 2010) and Arctic valley glaciers (Binghamn et al., 2005).
Lacustrine fans were most common in the deeper portions of the valleys, suggesting that drainage was focused by a network of subglacial channels into a subglacial tunnel ( Figure 13). This agrees with hydrologic models of the Laurentide Ice Sheet, which suggest subglacial drainage was a distributed drainage system that flowed towards the bottom of valleys, forming a larger single conduit for subglacial drainage (Hooke and Fastook, 2007). This differs from other regions in the glaciated northeast. Ice tunnels were close together (spacing 100's of m) and formed coalesced fans (stratified end moraines) parallel to the ice margin prior to 14,700 yBP in coastal Maine (Dorion et al., 2001;Thompson and Borns, 1985), before switching to larger, more centralized drainages (Ashley et al., 1991). While the spacing of the seismic profiles (< 1km) could be biasing the results, the seismic coverage was sufficient to map fans that were laterally continuous (parallel to the margin ice sheet). Lacustrine fan distribution similar to Glacial Lake Narragansett has been mapped in Long Island Sound (Lewis and Stone, 1991;. The implication of this is that larger, less frequent subglacial tunnels are indicative of a warm based ice sheet with a wide ice marginal area where surface water reached the bed (Hooke and Fastook, 2007). Hooke and Fastook (2007), however, limit subglacial drainage and tunnel formation to within 5 km of the margin of the ice, which differs than the interpretations discussed below.

Volume of sediment
To account for the volume of sediment deposited in Glacial Lake Narragansett Based on the location of morphosequences and uncorrelated lacustrine fans throughout Glacial Lake Narragansett, multiple drainage systems were active as the ice sheet retreated. Major drainage systems were probably located in the paleo-Blackstone and paleo-Taunton River valleys, with numerous smaller drainages .  interpret deep bedrock valleys extending down the Providence River, Mount Hope Bay/Sakonnet, East Passage and West Passage (Figure 13). The interpreted course of the paleo-Blackstone River down a pre-Wisconsinan west of upper Narragansett Bay and through the entrance of Greenwich Bay , contradicts boreholes (USACE, 1957), seismic reflection profiles and the regional trend of bedrock in Narragansett Basin Mosher, 1983). An alternative interpretation is that the paleo-Blackstone flowed down the west side of present-day Narragansett Bay, and down the present-day Pettaquamscutt River valley ( Figure 13).
The paleo-Taunton River probably flowed down the bedrock valley in the Mount Hope Bay and the Sakonnet River . Obviously these drainages would have been active at different times as the southern margin of the Laurentide retreated north up present-day Narragansett Bay.
An important note is that these valleys are coincident with the large deltas along the western shoreline of the present-day West Passage, in Upper Narragansett Bay and the Providence River.
Similar depositional processes would have been active during pre-Wisconsinan glaciations, as well as during the advance of the Laurentide during Late Wisconsinan time (Ridge, 2004). To account for the volume of sediment deposited during Late Wisconsinan deglaciation, these deposits had to have contributed sediment, although how this process would work remains enigmatic. All in place evidence of these deposits have been removed, but stratified deposits have been identified as thrust sheets in the recessional moraine at Block Island (Sirkin, 1976).
Regardless of some localized sediment sources, the immense volume of sediment deposited during deglaciation suggests that the drainage systems beneath the Laurentide Ice Sheet had to be very efficient and certainly began more than 5 km from the ice margin, contradicting some hydrologic models of the ice sheet (Hooke and Fastook, 2007). Previous work in Maine and Scandinavia suggest that throughflowing, subglacial drainage systems can begin 150 -170 km from the margin of the ice sheet (Arnold and Sharp, 2002;Ashley et al., 1991). The present-day Blackstone River begins 120 km north of Narragansett Bay, extending up a pre-glacial bedrock valley , and the northern extent of this valley may have been continuous with other river valleys further to the north prior to isostatic rebound.  report total sediment discharge for the Malaspina Glacier, which is similar in size to the individual lobes of the Laurentide Ice Sheet, to be 15 to 150 million m 3 . yr -1 , of which, half may have been bedload (Ashley et al., 1991;Hooke et al., 1985). Cowan and Powell (1991), measured the sediment yield of McBride Glacier, which is drained by a single subglacial tunnel, to be 1.7 -3.6 x 10 6 m 3 . yr -1 . Using this rate, the largest delta (Warwick Plains 1.2 x 10 9 m 3 ) would have been deposited in 300 -700 years. Based on estimated rates of ice-margin retreat this deposition occurred within < 100 years, suggesting individual drainages were at least three times larger than the present McBride drainage, and/or multiple drainages were active in larger deltas.
The average river discharge of all the temperate valley glaciers (8) in Glacier Bay National Park (including McBride) is 8.8 x 10 6 m 3 . yr -1 (Powell, 1991). This rate of discharge would deposit the Warwick Plains delta in 135 years. The low-end estimate of  for sediment discharge of the Malaspina would deposit the Warwick Plains delta in 160 years. Small deltas (i.e. Dutch Island, that the sediment discharge of the temperate valley glaciers in southeastern Alaska are within the same order of magnitude of individual drainages within Glacial Lake Narragansett.

Climate
The role of subglacial meltwater indicated by the prevalence of lacustrine fans, is directly related to the climate during deglaciation. Climatically, subglacial fans indicate a warm melt season with abundant melt water in sub-glacial tunnels. Icewedge casts and other features interpreted to be the result of permafrost suggest that the mean annual temperatures had to be below 0° C Stone and Ashley, 1992). Modeling of summer climate along the southern margin of the Laurentide Ice Sheet at the last glacial maximum suggests that mean July temperatures were below freezing under the influence of northerly winds (blowing off the ice), and above freezing under the influence of southerly winds. Mean summer temperature was modeled to be 3 -6° C (Bromwich et al., 2005).
The relatively warm temperatures of ice-free surfaces (i.e. the exposed continental shelf) adjacent to the ice sheet produced a strong thermal gradient along the margin of the ice sheet. Episodic low-pressure systems tracked along the southern margin of the ice sheet, producing extended periods of southerly wind and precipitation (rain), and the precipitation that fell on the ice would quickly reach the bed through moulins and drain via subglacial tunnels. These climatic conditions are not dissimilar from present-day coastal Greenland, which has average summer temperature of 6°C and mean annual temps 2 -4° C below zero (Vinher et al., 2006).
The similarities between the Laurentide and parts of the Greenland Ice Sheet are important. Future predictions of modern ice sheet behavior and response to climate warming requires an understanding of previous ice sheets. Subglacial drainage of the Greenland Ice Sheet, specifically the fact that surface meltwater transported to the base of the ice-sheet through moulins may be accelerating flow of outlet glaciers, has become a controversial topic (Bartholomew et al., 2010;Luthcke et al., 2006;Sundal et al., 2011;. A secondary conclusion of those studies is that as climate continues to warm, the Greenland is shifting from a polythermal to a warm-based ice sheet. As a result, the southern margin of the Late Wisconsinan Laurentide Ice Sheet may have been similar to the present-day characteristics of the Greenland Ice Sheet. It has been proposed that freshwater draining from glacial lakes can have an impact on thermohaline circulation in the North Atlantic (Broecker, 1997;Broecker et al., 1989). Freshwater flux from draining glacial lakes had to exceed 0.1 Sv (1 Sv = 10 6 m 3 . s) to affect thermohaline circulation (and climate) in the North Atlantic (Ganopolski and Rahmstorf, 2002;Rahmstorf, 1995Rahmstorf, , 2000. The calculated volume of Glacial Lake Narragansett (40 x 10 9 m 3 ) represents a minimum value. Delta progradation and lakefloor deposition was time-transgressive as the ice sheet retreated north, making it difficult to calculate an instantaneous depth and volume of the lake.
Glacial Lake Narragansett would have had to drain completely in 3 x 10 5 s (107 hr) to have exceeded this flux. While probably not exceeding this threshold, Glacial Lake Narragansett does represent a significant volume of freshwater not accounted for in existing models of freshwater flux (Licciardi et al., 1999). The volume of one lake (i.e. Glacial Lake Narragansett) in conjunction with other lakes in the region (Glacial Lakes Connecticut, Taunton, Cape Cod Bay) may become significant if they drained at approximately the same time (i.e. at the onset of isostatic rebound). If future work can constrain drainage of the lakes, it may be possible to link these events with climate records in the North Atlantic.

Patterns in the seismic record
Individual reflectors in the seismic records of lakefloor deposits show distinct patterns, with dark reflectors (more intense) at intervals of ~0.75, 1.5, and 3 m. This layering pattern may reflect climate fluctuations. Possible causes could be decadal changes in sea surface temperatures due to the Atlantic Multidecadal Oscillation (AMO) (Delworth and Mann, 2000), or increased storminess in the northeast due to the El Nino-Southern Oscillation or North Atlantic Oscillation (Hoerling and Kumar, 2000;Hubeny, 2006;Rittenour et al., 2000;Rogers, 1984). Based on limited core samples collected in the Providence River, the average thickness of individual couplets was approximately 2 cm . If this average is representative throughout the lake, each meter of lakefloor sediment represents approximately 50 years of deposition. The dominant intervals at 0.75 m and 1.5 m would represent ~35 years and 75 years of deposition respectively. These intervals coincide with the signal of the AMO, which has a periodicity ranging from 30 -100 yr or more (Delworth and Mann, 2000). This is thought to be the result of variability of thermohaline circulation in the Atlantic Ocean, and could point to the link between the climate in the North Atlantic and the deglacial evolution of the Laurentide Ice sheet.
Attempts to link the thickness of varves in central New England, with the climate records of Greenland suggest that there is a link between the rate of melting of the Laurentide Ice Sheet and Western Hemisphere climate . Work by Rittenour et al., (2000) in Glacial Lake Hitchcock showed a strong periodicity in the varve record at an interval of 2 -5 years, interpreted to represent variability induced by El Nino -Southern Oscillation. Rittenour et al., (2000) also report a signal with > 40 year variability that could represent the variability induced by the AMO, but was not discussed in detail. Additionally, no mention was made of the North Atlantic Oscillation, which has been recognized from Holocene sediment records in southern New England .

CONCLUSIONS
• Quaternary (glacial) depositional environments were interpreted from seismic facies identified from 800 km of high-resolution seismic reflection profiles collected throughout Narragansett Bay. These deposits were grouped into morphosequences, and mapped in a similar manner to traditional (terrestrial) Quaternary mapping.
• The retreat of the ice-sheet up present-day Narragansett Bay was systematic with five minor stillstands or fluctuations recorded by small recessional moraines. While lakefloor deposits are ubiquitous throughout, large deltas are limited to the western and upper portions of present-day Narragansett Bay, coincident with pre-glacial bedrock valleys.
• The near continuity of glacial lakefloor deposits throughout present-day Narragansett Bay and the elevation of the deposits relative to the projected water level of Glacial Lake Narragansett support the hypothesis that one large lake occupied present-day Narragansett Bay and extended into adjacent areas during Late Wisconsinan deglaciation.
• The prevalence of lacustrine fans deposited at the margin of the ice sheet throughout the study area indicates that subglacial meltwater was dominant throughout the lake basin. The distribution of the fans suggests that subglacial tunnels remained active in the same locations during retreat, and were mostly localized to the deepest portions of pre-glacial bedrock valleys.
• The spacing of the lacustrine fans and nature of subglacial drainage indicates that the Laurentide Ice Sheet was warm-based, and that drainage was focused into a distributed drainage system that flowed from the sides towards the bottom of valleys, forming larger single conduits for subglacial drainage.
• The volume of sediment deposited around and within present-day Narragansett Bay suggests that a very efficient drainage system existed under the Laurentide Ice Sheet during deglaciation, and multiple drainages were active during deposition of larger morphosequences. The amount of excavated sediment needed to account for this volume suggests either a pre-glacial (Illinoian) source of stratified sediment, and/or that sediment was transported from well beyond the present-day watershed of Narragansett Bay.     Schafer, 1961;Smith, 2010) are shown on the left; uncorrelated deposits are shown on the right side.   . B: Same area after the ice margin had retreated north. The 'stacked' fans suggest that the ice tunnels are staying in approximately the same location an extended period (years). Glacial lakefloor deposits have buried the entire sequence as the ice sheet retreated north. Figure 13: Location of proximal lacustrine fans in Narragansett Bay. Dashed black lines represent the thalweg of the bedrock valleys identified by . Fans likely exist in the upper East Passage and Sakonnet River, but seismic penetration was limited in these areas.

ABSTRACT
An integrated mapping approach using high-resolution side-scan sonar, surface sediment grab samples, digital aerial and orthophotography and underwater video imagery was used to map Holocene sediment cover and Late Wisconsinan glacial outcrop in two shallow embayments in Narragansett Bay, Rhode Island, USA. The use of side-scan sonar to characterize the seafloor has become common in a variety of different marine environments. Challenges remain in classifying side-scan or other acoustic data into a naming convention that is useful to scientists and managers. We characterize the benthic geologic habitats of these areas utilizing a flexible naming convention that combines information about geologic processes, morphologic form, particle size, biota, and anthropogenic impacts. Benthic geologic habitats were separated into three habitat groups (depositional environments) (Estuarine bayfloor, estuarine cove and estuarine marginal habitats), and further divided based on morphologic form, surface sediment texture, geologic features, biologic characteristics, and anthropogenic impacts. There is a general trend of decreasing grain size with increasing distance from the open water of Narragansett Bay, however, the types and distribution of facies is complicated, and this work adds to the developing sedimentary models of estuaries. The methods outlined in this paper has been successfully applied in other estuarine, lagoon and shoreface environments, providing a concise method of imaging and characterizing benthic geologic habitats on the seabed.

INTRODUCTION
The objective of this work was to understand the benthic geologic habitats and geologic processes within two important areas of Narragansett Bay and describe them with a naming convention understandable by managers and non-geologists at a scale useful in a wide variety of applications. The high-resolution of the side-scan records and aerial photographs allow for detailed delineation of map units, and identification of geologic features and anthropogenic impacts. The results of this work contribute to the overall understanding of the distribution of depositional environments in microtidal estuaries.

BENTHIC GEOLOGIC HABITATS
A habitat is a spatially recognizable area with physical, chemical and biological characteristics that are distinctly different from surrounding areas (Valentine et al., 2005). We propose that a benthic geologic habitat is a spatially recognizable area with geologic characteristics that are distinctly different from surrounding areas. Identification of these habitats is done using side-scan records in conjunction with ground-truth data and existing or estimated tidal current, wind, and wave information. The definition we propose is analogous to depositional environments, a term often used in geologic mapping. Because many non-geologists are unfamiliar with the concept of depositional environments, we adopted the term benthic geologic habitat . It has been pointed out that referring to interpretive maps of the seafloor as 'habitat' maps is problematic without specific details on the association between species or populations and different areas of the seafloor, and a better term would be 'potential' habitats (Greene et al., 2007).
We feel that our definition of benthic geologic habitats are analogous to the 'potential habitats' defined by Greene et al. (2005), without any connotation of possible correlation with species or populations beyond those visible in the side-scan sonar imagery.
Challenges remain in classifying side-scan or other acoustic data into a naming convention that is useful to scientists and managers, and numerous classification schemes have been proposed (Barnhardt et al., 1998;Goff et al., 2000;Greene et al., 1999;Kennish et al., 2004;Nitsche et al., 2004;Valentine et al., 2005). These existing schemes are not applicable in complex estuarine environments, which we map at very large scales (1:10,000 or 1:5,000). Cowardin et al. (1979), which is commonly applied to estuarine and lagoon environments, uses a hierarchical approach that would classify both study sites simply as 'Estuarine, Subtidal, Unconsolidated Bottom', with further delineation based only on sediment texture.
Recent attempts at creating a national standard for ecological mapping (CMECS; Madden et al., 2010) were examined. We feel that the segmented nature of this largely ecologically based classification scheme, which separates the morphologic form (Geoform), surface sediment characteristics (Surficial Geology) and biological attributes (Benthic Biota) into different components is overly complicated. Our classification scheme provides a concise way to describe the morphologic form, surface sediment texture and other characteristics identifiable in side-scan sonar imagery, using a naming convention that combines information from particle size , geologic processes, depositional environment, and also biota and anthropogenic impacts Oakley et al., 2007).
The interdisciplinary nature of marine mapping makes it important to report data in a manner useful to managers and scientists in fields other than geology. Shumchenia and King (2011) simplified the benthic geologic habitats from Greenwich Bay into silty (low-energy basin and bay channel habitats) and sandy (depositional platforms and bayfloor sand sheets) geologic habitats. This grouping coincided with the broad scale assemblages of benthic macrofauna, and showed a good example of combining the spatial coverage of 'top down' acoustic based mapping with biologically based 'bottom up' information.

STUDY SITES
Narragansett Bay is a microtidal estuary located in eastern Rhode Island and southeastern Massachusetts that is an important resource for fishing, maritime transportation, tourism, recreation, shipbuilding, defense and manufacturing (Harrington, 2000). Present-day Narragansett Bay lies within Narragansett Basin. The underlying bedrock is the Rhode Island Formation, a coarse to fine-grained metamorphosed sedimentary rock . Narragansett Bay is located within the Late Wisconsinan glacial limit, and both study sites are underlain by a variety of sediment types, including glacial fluvial sand and gravel, glacial lakefloor silt and clay, and till.
Marine water first inundated Narragansett Bay after 10,200 yBP (sea level 30 m below present) (all ages reported in calendar ages before present) McMaster, 1984). Greenwich Bay was flooded between 6,500 and 5,000 yBP (sea level 5 m below present), Wickford Harbor between 4,000 and 2,500 yBP (sea level 2.5 to 4 m below present) McMaster, 1984).
Present-day Narragansett Bay is microtidal (spring tidal range 1.2 m at Newport, 1.5 m at Providence), and a mixed-energy estuary according to the classifications of Dalrymple et al. (1992) and Hayes (1979).
Greenwich Bay is bracketed by glacial deltas to the north and south . The eastern shoreline of Greenwich Bay is comprised of till, containing a mixture of gravel, sand, silt and clay, while the western shoreline is a mixture of coarse grained, ice-marginal sand and gravel . Wickford Harbor is located entirely within a glacial delta, and the deeper areas of the harbor likely represent ice-block basins, now inundated with marine water. The north to south trending coves in both sites likely formed by spring sapping, similar to those on southeast shore of Massachusetts (FitzGerald et al., 2002;, with the exception of Greenwich Cove, which may occupy the location of a collapsed ice-margin of the retreating Laurentide Ice Sheet . Both sites support commercially important shellfish, finfish, migratory birds and other aquatic wildlife essential to the ecosystem of Narragansett Bay, and are popular recreational areas for boating, kayaking, fishing, and swimming (Joubert and Lucht, 2000;Dalton et al., 2010). These factors, along with evidence of declining water quality (fish kills, disappearance of eelgrass beds and shellfish closures) led to the creation of a process by the Rhode Island Coastal Resources Management Council to create a Special Area Management Plan (SAMP) for Greenwich Bay and surrounding watershed. The types and extent of benthic geologic habitats were mapped in Greenwich Bay to assist with the creation of this plan. Mapping in Wickford Harbor was completed as part of the ongoing MapCoast project, which has a goal to map, inventory, describe, and classify sediment and subaqueous soils in Rhode Island waters shallower than 5 m (http://www.mapcoast.org).

Side-scan sonar
Between 2003 and 2006, 210 km of high-resolution (500 kHz, 100 m swath width) side-scan sonar data were collected in Greenwich Bay and Wickford Harbor using an towed EdgeTech 272TD side-scan sonar system (Average vessel speed 1. 8 m · s -1 ), spatially located using a Trimble Differential GPS with a reported accuracy < 1 m . Side-scan records were processed using Chesapeake Technologies SonarWeb TM software, and individual data files were manually bottom tracked, adjusted for variations in contrast and time-varied gain, and a slant range correction was applied to correct for the elevation of the towfish above the bay bottom (Fish and Carr, 1990). Side-scan files were combined into a single mosaic of each study site, with a pixel size of 0.3 m (ground distance). The mosaics were exported as a GeoTiff for analysis in a Geographic Information System (GIS), displayed using an inverse medium yellow-orange known as a 'Klein' color scheme.

Side-scan data interpretation
Side-scan records are interpreted based on the texture and intensity of the returning acoustic energy, and spatially recognizable areas with different backscatter patterns represent side-scan sonar facies. Geologic facies are sediment or rock with certain readily identifiable characteristics such as color, particle size, sorting, structure, biologic content, plus others, discernable in either the field or laboratory (Walker, 1990). Side-scan sonar facies are the geologic facies interpreted using the strength and texture of the returning sonar signal; in general, the harder (or denser) the bottom, the stronger the return signal and the darker the side-scan sonar record (using an inverse color scheme) (Goff et al., 2000). This relationship is complicated by bed roughness, vegetation and bioturbation (Nitsche et al., 2004), and surface sediment samples and underwater video imagery were collected to aid interpretation.
Facies boundaries were manually digitized directly onto the digital mosaic at a scale of 1:1,000 (the limit of pixilation) using MapInfo™ GIS software. Identification of features too small to be resolved on the mosaics was aided using digital, boulders, sunken boats) were assigned a point feature.

Surface sediment samples
Surface sediment grab samples (n = 48), were collected using a Petite Ponar TM sampler. Samples were spatially located using the same DGPS utilized in the sidescan survey, and were photographed and described in the field. Selected samples were analyzed using standard sieve and pipette techniques outlined by (Folk, 1980).
Additional sediment samples (24), collected by  were downloaded from the United States Geologic Survey (USGS) East Coast sediment texture database (Hastings et al., 2000). Particle sizes are based on Udden (1914) andWentworth (1922). Sediment texture is classified using a modified version of the Shepard (1954) naming convention ( Figure 2).

Underwater imagery
Limited underwater video imagery was collected in both Greenwich Bay (C.

Deacutis, personal communication) and Wickford Harbor. Video imagery in
Wickford was collected using a SeaViewer Sea-Drop TM color digital video camera. A similar, analog system was used in Greenwich Bay. Files were played back in the lab, and qualitative observations were made regarding sediment texture, geologic features, and biologic characteristics. Visibility in both sites was poor, which reduced the field of view, and limited the usefulness of the video imagery, particularly in muddy environments. Diver collected photographs and interpreted aerial photographs (Bradley et al., 2007), provided additional ground-truth in Wickford Harbor.

Aerial photography
Vertical aerial photographs were used to delineate geologic habitats in shallow (< 2 m) subtidal areas. Polygons created by Boothroyd and Galagan (1992)  to-year persistence of macroalgae seen on side-scan records in Wickford Harbor.

Naming convention of benthic geologic habitats
Benthic geologic habitats were named based on the interpreted depositional environment, abbreviated with a one to three letter acronym (Table 1), sediment texture, and a descriptor if applicable. The names and abbreviations of depositional environments were modified from those presented in Boothroyd et al. (1985) for coastal lagoons along the Rhode Island south shore, which in turn were based on units from Fisher et al., (1972) (Table 1). Grainsize is abbreviated as Gravel (g), sand (sa) and silt (si), with additional classes of fine (f) or coarse (c). Sediment containing more than 5% organic matter (dry weight) was classified as organic (o). Habitats where sediment texture was not determined were mapped as undifferentiated (u) ( Table 2).
Descriptors are based on characteristics seen on the side-scan imagery, and can be verified in the underwater video imagery and aerial photography. These descriptors are frequently biological characteristics (i.e. presence of macroalgae or submerged aquatic vegetation, indicated by a (v) for macroalgae, or (e) for eelgrass, but could also refer to anthropogenic or geological features. A summary of the benthic geologic habitats is shown in table 3. The key to this naming convention is the flexibility, and new depositional environments, descriptors, and additional sediment textures can be added as needed in new study areas.

Benthic Geologic Habitats
Benthic geologic habitats were grouped into three main geologic habitat groups (depositional environments) in both study sites, (estuarine bayfloor, estuarine cove, and estuarine marginal environments), and further divided based on morphologic form (i.e. basins, flats, tidal deltas etc.). The distribution of these habitats can be seen in figures 3 and 4. The general color scheme for is red for gravel, yellows for sand, browns for mud and brown-green or green for macroalgae or submerged aquatic vegetation. Inlet channels, dredged channels, marinas, and distributary deltas are in blue.

Boulder gravel pavement (Pvbg)
A distinct habitat composed primarily of pebbles, cobbles, and large (up to 6 m diameter) boulders was mapped at the entrance of Greenwich Bay. The boulders in this habitat are not transported, and the relatively high tidal energy (compared to the adjacent basins) inhibits the deposition of finer grained sediment.

Shell reef (Shr)
This habitat exists only in a small area of western Wickford Harbor, where it was visible in side-scan sonar records and digital orthophotographs. Field observations show it to be comprised of intact shells and fragments of eastern oyster (Crassostrea virginica) with some slipper shells (Crepidula fornicata). This is likely not a drowned oyster reef (portions are intertidal), which would have been 1-2 m below mean lower low water (McCormick-Ray, 2005). No shell middens have been reported for this location. This probably represents the placement of dredged materials adjacent to a small dredged navigation channel.

Bayfloor sand sheets (Sssa)
Adjacent to the depositional platforms, areas of the bayfloor identified by a moderate to dark side-scan return are interpreted as bayfloor sand sheets. Sediment samples from these habitats were > 90% sand. These habitats often have tidal or wave-generated bedforms visible on side-scan or video imagery.

Low-energy basins (Leb)
The areally most extensive habitats in both Greenwich Bay and Wickford Harbor are identified by a low-backscatter, featureless side-scan return ( Figure

Bay channels (Bc)
The bay channel geologic habitats encompass a variety of energy levels and grain-sizes, primarily related to variations in tidal current velocity. Bay channels connect adjacent low-energy basins, and incised channels are common in many estuarine environments (Dalrymple et al., 1992). Channel habitats range from sandy  Figure 5B). The origin of these features is enigmatic. The larger features (3-5 m) were first mapped as pockmarks caused by methane gas escape from the sediment , which are common in other estuaries in the northeast . Sub-bottom seismic reflection profiles through this habitat show none of the characteristics of pockmarks or pockmark fields (e.g. depressions, gas wipeouts) . It is now thought these represent aggradations of shelled bivalves, however these were not seen in the sediment samples or limited video from this area. Given the nature of these ground-truth methods and relatively small size of the features, a sample needs to be collected directly on an aggradation to confirm this interpretation.

Estuarine Cove
North to south trending coves extend off the central basins in both study sites.
These coves are narrow (generally less than 300 m across at the entrance); extend off the main basins < 2200 m with water depths generally less than 2 m, except dredged channels up to 4 m deep.

Low-energy basins (Leb)
Low-energy basins occupy the deeper areas of coves and have a very light sidescan return that indicates fine-grained sediment. Surface sediment samples range from sandy silt to silt or clayey silt containing 50 -95% silt and were mapped as low-energy basins. These areas tend to be finer-grained (silt to clayey silt) with higher organic content (5 ->10%) than low-energy basins in the less protected basins of Greenwich Bay and Wickford Harbor, and were mapped as low-energy basin silt (Lebsio).

Inlet Channels (Ci, Cf, Ce, Cd)
Inlet channels at the mouth of the coves in both study sites contain natural (Ci, Cf, Ce) and dredged channels (Cd). Natural inlet channels are generally sandy (Cisa), and were further subdivided into ebb (Cesa) or flood channels (Cfsa) depending on the horizontal segregation of tidal currents around tidal deltas (Boothroyd, 1985).
Representative samples from dredged channels contained 85-90 % silt and were mapped as dredged channel silt (Cdsi).

Tidal deltas (Df, De)
A flood-tidal delta and some features of an ebb-tidal delta were mapped at the entrance of Brushneck Cove in central Greenwich Bay. The flood-tidal delta was mapped as Dfsa. Individual units within the flood-tidal delta (i.e. flood ramp) were not mapped due to the small size of the delta and the map scale (1:10,000).
The only obvious features of an ebb-tidal delta were channel marginal linear bars (Decm) that flank the inlet channel. The geomorphic features of these tidal deltas, including the associated ebb and flood-channels are mapped using the terminology described by Hayes, (1975).

Estuarine marginal habitats
Estuarine marginal environments are those that are immediately adjacent to, and often continuous with similar upland environments (i.e. a sandy depositional platform extends landward to a sandy intertidal beach). This work focused on subtidal geologic habitats, although many of these features extend into the intertidal areas. These habitats were broadly classified as erosional (terraces) or depositional (platforms, distributary deltas, tidal deltas and tidal flats).

Erosional habitats Terraces (Te)
Wave-eroded terraces have formed narrow bands of coarse-grained sediment (sand to boulders) in areas of high wave action, and are mapped as gravel erosional terraces

Wave-formed bars
Large bedforms were found on the depositional platform and on the bayfloor sand sheet in Greenwich Bay ( Figure 5D), identified by a moderate acoustic return with distinct bar forms (>100 m in length, crest to crest spacing 10 -300 m). The bars do not appear to be shore connected. Most of the bars on the depositional platform in northern Greenwich Bay are sub-parallel to the shoreline although a few are oriented perpendicular to the shoreline. Wave-formed bars were not present in Wickford Harbor.

Tidal bedforms
Tidal bedforms ( Figure 5C) were identified on the side-scan record in the southwest corner of Greenwich Bay (Habitats Sssa and Dpsa), in an area influenced by strong tidal-current flow. These bedforms are medium to large (crest to crest spacing 5 -15 m) 2-D and 3-D dunes using the terminology of Ashley et al. (1990). The current velocity necessary to form the 3-D dunes is at least 80 cm . s -1 (Boothroyd and Hubbard, 1975). Modeled and observational data suggests a height of approximately 0.6 m (Boothroyd, 1985). They appear to be ebb oriented; however, determining orientation of low-amplitude bedforms from side-scan records can be problematic.

Bedrock
Bedrock outcrops are not common in or adjacent to either study site. One small outcrop of Rhode Island Formation  in the northern portion of Mill Creek in Wickford occurs as a small island, mapped as (Bdrx) (Figure 4).

Isolated boulders (within other habitats)
Isolated large boulders up to 5 m diameter crop out within the depositional platform and bayfloor sand sheet habitat throughout Greenwich Bay ( Figure 5E).

Boulders in Wickford
Harbor were limited to some scattered boulders in the southeast portion of the harbor.

Submerged Aquatic Vegetation and Macroalgae
Drift or attached macroalgae, and rooted submerged aquatic vegetation are identifiable in the side-scan records in both study areas ( Figure 6). Macroalgae was also visible in aerial photography in both sites.  -Bohnsack et al., 1988;Thornber et al., 2008). The extent of eelgrass (Zostera marina), which is an important habitat for juvenile finfish and invertebrates, has declined dramatically in Narragansett Bay due to a deterioration in water quality (Kopp et al., 1995). Eelgrass was mapped in the low-energy basin in Wickford Harbor (Lebsie) (Figures 4, 6A, B). The presence of eelgrass interpreted from side-scan records was confirmed by direct diver observations. No eelgrass was mapped in Greenwich Bay.

Quahog (hard-shell clam) harvesting (rake) trails
Bottom trails resulting from harvesting quahogs (Mercenaria mercenaria) are clearly visible in the side-scan imagery ( Figure 7A). These features are extant in a large area of the western basin and along the northern margin of the midbay channel system within Greenwich Bay ( Figure 3). Quahogs are harvested using a long handled metal rake (approximately 60 cm across) known locally as a bull rake, dragged through the sediment, forming distinct furrows in the bayfloor. Some areas mapped showed no portions of the bayfloor untouched by a rake ( Figure 7B). Wickford Harbor is typically closed to shell fishing, and these features were not seen.

Mooring drags
Distinct circular bottom features, 5-20 m across, appear on the sonar record throughout the mooring fields of both Greenwich Bay ( Figure 7C) and to a lesser extent Wickford Harbor. These features represent bottom disturbance from the ground chain of boat moorings, which disturbs the bottom as the boat pivots around the mooring anchor. This feature is particularly prevalent in a very dense mooring field in southwest Greenwich Bay.

Marina/shoreline structure debris
Debris litters the bay floor along parts of the developed western shoreline of Greenwich Bay, and numerous pilings lay on the bayfloor, possibly the result of Hurricane Carol in 1954. Boulders mapped in the entrance channel to Wickford Harbor appear to have been part of the jetties, but could not be differentiated from glacial boulders cropping out within the surrounding bayfloor sandsheet (Sssa).

Sunken Boats
There are a number of sunken boats scattered around the study sites. Most are < 10 m in length. One unique sunken boat located in Greenwich Bay, in approximately 1 meter of water (MLLW) ( Figure 7D). This large > 30 m wreck shows up in detail on the side-scan imagery. Individual wooden planks can be identified, and it appears that the sides of the wreck have collapsed outward and lay on the depositional platform.

DISCUSSION
The present distribution of benthic geologic habitats is controlled largely by the Late Wisconsinan glacial landforms and sediment characteristics. The gravel erosional terraces (Teg) in eastern Greenwich Bay fringe a headland comprised of till . These terraces form as storm waves erode of the shoreline. Boulders (up to 6 m diameter) were left behind during retreat of the headland bluffs by wave action, and likely are not transported, except perhaps during severe storm events (hurricanes). These processes would have been active during Holocene transgression, and the boulder-gravel pavement (Pvbg) adjacent to the erosional terraces in southeast Greenwich Bay formed as an erosional terrace at a lower relative sea level. Wave erosion of the till could not begin until relative sea level was < 6m below present. Applying the relative sea-level curves proposed for Narragansett Bay McMaster, 1984), this process began around 6,000 yBP. The boulder gravel pavement is presently impacted by breaking waves only during hurricanes and severe extra-tropical cyclones. Relatively strong tidal currents inhibit deposition of fine-grained sediment during fair-weather periods.
This habitat probably marks the extent of the till headland prior to Holocene transgression, and similar habitats have been mapped around till or coarse-grained stratified uplands along the Rhode Island south shore (Klinger, 1996;Zitello, 2002;Oakley et al., 2009).
The sandy depositional platforms (Dpsa) in both study sites are interpreted to have formed from the deposition of sand eroded from the stratified glacial headlands, as the shoreline retreated due to wave action during storm events. The shoreline in northern Greenwich Bay has retreated 20 to 30 m since 1939; the shoreline Wickford Harbor has eroded 5 to 10 m in the same period . The wide depositional platforms (< 500 m) and bayfloor sand sheets (< 1,000 m) in eastern Greenwich Bay are forming on the now submerged sandy delta slope and eroded portions of the sandy and gravelly delta plains of glacial deltas north and south of Greenwich Bay . These deltas provide a ready source of sediment, and the original depositional slope of the delta (2 -3 m . km based on the configuration of modern glacial deltas in Alaska) (Boothroyd and Ashley, 1975) is conducive to forming wide depositional platforms.
The bayfloor sandsheet (Sssa) (Water depth 2 -4 m) adjacent to the modern depositional platforms would have been an active depositional platforms when relative sea level was > 2.5 m below present (prior to 2,500 yBP McMaster, 1984) Low-energy basins are areally the most extensive habitats in both study areas (Figures 3,4), occupying topographic lows in the glacial topography or channels incised into the glacial deposits. These areas are sinks for fine-grained (silt to clay) mineral and organic sediment. Fine-grained sediment has four possible sources in estuaries; fluvial input, biological production, offshore sources and shoreline erosion (Cronin, 2007). Direct fluvial input is low in both of the embayments studied here, and throughout Narragansett Bay as a whole. Major rivers draining into Narragansett Bay have been extensively dammed, and while not well quantified, models show decreasing sediment load in the Blackstone River closer to Narragansett Bay (Ji et al., 2002), and much of the river is at or close to bedrock .
One exception to this, are the small distributary deltas at the mouth of some of the tidal creeks, although exactly how much sediment is presently transported down these creeks is unclear.
Primary production is the most significant source of organic, silt-sized sediment in Narragansett Bay , and the remains of decayed macroalgae and phytoplankton are deposited into low-energy basins (Granger et al., 2000). Offshore sediment was identified as a source of sediment in Chesapeake Bay; however, models of sediment transport in Rhode Island Sound suggest that this area is not a significant source of sediment for deposition in Narragansett Bay (Grilli et al., 2010). The embayments studied here are 15 km (Wickford) and 25 km (Greenwich Bay) from the southern entrance of Narragansett Bay and active exchange of sediment between these areas and Rhode Island Sound is probably negligible.
Shoreline erosion is a significant source for silt and minor amounts of clay within Narragansett Bay as well as other estuaries Cronin, 2007;Marcus and Kearney, 1991). Wave action erodes the till shorelines throughout Narragansett Bay, selectively removing silt and clay from these areas, where it can be transported to, and deposited in, deeper, low-energy areas. Modern channel incision into underlying glacial lakefloor sediment was reported as an additional source for fine-grained sediment in eastern Long Island Sound (Knebel et al., 1999). Seismic reflection profiles in the channel at the entrance to Greenwich Bay shows that 1 -3 m of estuarine mud overlies the glacial lake floor deposits suggesting that active channel incision is not occurring here. Wind-blown silt from adjacent areas likely contribute small amounts of sediment, but this process would have been more active in the recent past, when tilling of fields around Narragansett Bay was more common.
High-resolution chirp sub-bottom seismic reflection profiles indicate the lowenergy basin in western Greenwich Bay contains 7 -8 m of fine-grained Holocene sediment; in Wickford Harbor, the basins contain 2.5 -5 m of fine-grained sediment.
Published sedimentation rates for Greenwich Bay (0.35 -0.5 cm yr -1 ) (Latimer and Quinn, 1996) suggest deposition took 1,600 -2,500 years. This age range correlates with the Holocene inundation model of Narragansett Bay presented by  and McMaster (1984), that suggest by 2,500 yBP, most of Narragansett Bay was at roughly the present extent and configuration . Circulation models of Narragansett Bay based on acoustic Doppler current profiles suggest that southwest 'sea-breeze' winds may create a circulation pattern that significantly increases the residence time for water in Greenwich Bay. This may reduce transport of fine-grained material out of central Greenwich Bay (Rogers, 2008), and could produce higher sedimentation rates.
There is a general trend of decreasing grain size from east to west in both Greenwich Bay and Wickford Harbor, ranging from sand (and some gravel) along the shallow shelf areas in the eastern portions of the study sites, to fine-grained silt and silty sand in the deeper depositional basins. This trend is expected, given the general decrease in wave-energy from the open water of Narragansett Bay and towards the more protected areas. A decrease in grain size in sheltered central basins within an estuary is common in mixed energy environments (Dalrymple et al., 1992). This relationship is complicated in areas where wave action and tidal currents have produced sandy or gravelly deposits in estuarine cove and marginal depositional environments. Similar depositional environments have been mapped in other modern, mixed-energy microtidal estuaries (Biggs, 1967;Kerhin et al., 1988;Knebel, 1986Knebel, , 1989Nitsche et al., 2004). Facies distributions in these estuaries are more complicated than the models of tide-dominated or wave-dominated estuaries illustrated by Dalrymple et al., (1992) and Boyd et al. (2006). This is further complicated in glaciated estuaries, where sediment characteristics exhibit great lateral variability (Barnhardt et al., 1998;Knebel et al., 1991).
The complex lateral distribution of facies interpreted to have been deposited in estuarine environments have begun to be recognized in the rock record, by comparing modern estuarine facies with detailed outcrop analysis (Mack et al., 2003). Few facies models and assemblages of estuarine environments have been proposed that are directly applicable to the rock record (Reinson, 1992). Future work on rocks interpreted to have been deposited in an estuarine environment should consider sandy depositional platforms, which may comprise some of the sandstones in interpreted estuarine sedimentary sequences (Mack et al., 2003), and the low-energy central basins of estuaries and coastal lagoons have been recognized as potential petroleum source rocks (Putnam, 1989;El Hariri, 2008). Transgressive estuaries have a high preservation potential in incised valleys (Belknap and Kraft, 1985). Detailed mapping of the extent and distribution of facies and depositional environments within modern estuarine environments allows for a better understanding of the rock record (Reinson, 1992).
The results of this work also provided the basis for subaqueous soil mapping in both study sites, and other estuarine and coastal lagoons in Rhode Island (i.e. Stolt et al., 2011;Payne, 2007

CONCLUSIONS
• Greenwich Bay and Wickford Harbor, two embayments of western Narragansett Bay were mapped using side-scan sonar, surface sediment samples and vertical and oblique aerial photographs. The combination of techniques allowed for full coverage mapping in these shallow estuarine environments.
• These areas were mapped into benthic geologic habitats utilizing a flexible naming convention that combines information about geologic processes, morphologic form, sediment characteristics, biota and anthropogenic impacts.
• Human activities in Narragansett Bay have a considerable impact on benthic habitats.
Significant bottom disturbances from shell fishing and mooring fields were identified, and debris from marina construction/destruction litters the bay floor adjacent to most of the developed shoreline.
• The Late Wisconsinan glacial landforms and sediment distribution control the distribution of benthic geologic habitats throughout the studies sites, especially in estuarine marginal environments.
• There is a general decrease in grain size from east to west correlating with a decrease of wave-energy in more protected portions of the estuary in both study sites. This is similar to other microtidal estuaries, and published estuarine models; however, the distribution of facies within these embayments is complicated, especially in estuarine cove and marginal environments. The dominant process (wave or tidal) varies spatially and temporally, especially when considering processes active during storms.
• The methods outlined in this paper has been successfully applied in other estuarine, lagoon and shoreface environments, and provides a relatively low-cost, concise method of imaging and characterizing benthic geologic habitats on the seabed.          Ocean Service hydrographic survey points in the study area used to interpolate the digital terrain model. The accuracy of the interpolated hydrographic grid was assessed by randomly withholding 10% of the data points and recreating the terrain model. The difference between the elevations in the grid created using the subset data and the withheld data points was checked using the 'Point Intersect' feature in ESRI ArcMAP TM 'Hawth's Tools' extension. The mean difference was 0.1 m, +/-0.6 m. The high standard deviation was due to outliers located near areas of high relief, specifically in the East Passage of Narragansett Bay, and between Block Island and Long Island Sound (Figure 1). Less than 0.5% of the withheld data points differed by more than 1 m from the grid. The final terrain utilized all of the data points.

Location of the cores
Seven cores were recovered from the upper Providence River (PVD-1 -PVD-8). PVD-4 was recovered, but the core was lost out of the bottom of the barrel during recovery. Cores were located in a general area 300 m south of the Port Edgewood Marina and approximately 50 to 100 m east of the Edgewood shoreline ( Figure 1A).
Water depths were less than 1 m MLLW. Two cores from a precious study  were also used. The location of these cores are shown on figures 1A and 1B.

Post-glacial depositional environments
The interpretations of the post-glacial depositional environments follow those of . The sand capping the cores was eroded and transported from the adjacent glacial deltas during storm events and deposited on a sandy tidal flat. The dark brown interbedded silt and sand in cores PVD-1 to PVD-5 is interpreted to have been deposited in a shallow estuarine environment, evidenced by the presence of estuarine and marine macrofossils . The dark brown ~ 1 cm thick peat capping the varves was interpreted to have been deposited in a freshwater marsh , although a detailed examination of these layers for identifiable macrofossils was not conducted. The thick, organic-rich, silt in core PVD-6 may represent wetland deposition in small kettle hole, however the layer was not examined for macrofossils to confirm this interpretation.    : Image from core PVD-3. Red arrow marks the boundary between tan varves and gray sandy varves discussed in text. Blue arrow marks the boundary between the varves and the overlying medium sand with shell fragments. Core depths represent depth within the recovered core sample. Deformed sections, where accurate varve measurements were not possible are noted.

Figure 5:
Image from core PVD-5. Red arrow marks the boundary between tan varves and gray sandy varves discussed in text. Blue arrow marks the boundary between the varves and the overlying medium sand with shell fragments. Core depths represent depth within the recovered core sample. Yellow dots are plastic disks placed 5 cm apart during imagery collection to aid in scaling images during varve analysis. Deformed sections, where accurate varve measurements were not possible are noted. Figure 6: Image from core PVD-6. Core depths represent depth within the recovered core sample. Red arrow marks the boundary between undeformed and deformed varves. Blue arrow marks the boundary between mottled olive-gray and dark brown organic silt. Brown arrow marks boundary between dark brown organic silt and mottled olive-brown silt. Purple arrow marks the boundary between mottled silt and medium sand with shell fragments. Deformed sections, where accurate varve measurements were not possible are noted.

Figure 7:
Images from core EW-1 from . Blue arrow marks the brown peat layer overlying the varve sequence. Red arrow marks the base of medium sand with shell fragments that caps the core. Core depths represent depth within the recovered core sample.

Figure 8:
Images from core PC-6 from . IRD = Ice rafted debris seen in the first four basal varves. Red arrow marks the boundary between deformed varves and the overlying estuarine silt. Green arrows refer to thick (13 and 25 cm), sandy couplets discussed in the text. Orange arrow marks the boundary between varves and ice-marginal sand and gravel. Core depths represent depth within the recovered core sample. Deformed sections, where accurate varve measurements were not possible are noted.

Introduction
The paleomagnetic properties of glacial lakefloor sediment have been used to support regional correlation between varve sequences. Workers studying changes to the Earth's magnetic field realized that the annual nature of varves provided an ideal platform for constructing a record of paleomagnetic remanent declination and inclination (Johnson et al., 1948;Verosub, 1979aVerosub, , 1979b. Declination is more useful as a correlation tool than inclination, which can be altered during deposition or by post-depositional process (compaction) . It was hoped that paleomagnetic information, particularly remanent declination would support a potential correlation with the North American Varve Chronology (NAVC). Ultimately, results were inconclusive and were not included in the manuscript version of chapter 3.

Methods
Paleomagnetic directional information was measured on continuous longitudinal subsamples of cores PVD-3 and PVD-5 following the procedures outlined by King and Peck (2001). Measurements of inclination, declination, and magnetization were conducted with a 2-G Enterprise TM (WSG Inc., Sand City, CA), pass through cryogenic magnetometer at one-centimeter intervals. Progressive demagnetization with a diminishing alternating field removed magnetic overprints and subsampling disturbances. Declination and inclination were averaged over the depth range of each couplet, and converted to varve years.

Results
The remanent declination and inclination curves produced for cores PVD-3 and PVD-5 show inconsistencies somewhat expected given the level of deformation in the cores. While highly scattered, the remnant declination records in undeformed sections is 30º -32º. Remanent inclination is less clear; values range between -60º and 120º, with clusters at 30º and 90º, and little consistency between cores (Figure 1).

Paleomagnetic Properties
The variability seen in the remanent inclination and declination records from cores PVD-3 and PVD-5 is presumably the result of coring deformation. While depositional processes can reorient grains, and cause declination to vary up to 15º (Verosub, 1979b), the consistency of declination in undeformed sections of the cores suggests that this is not a significant issue here. Previous studies have shown that regional correlation is possible using paleomagnetic records of sediment deposited in Late Wisconsinan and Holocene time (King and Peck, 2001;. The base of the paleomagnetic record in the NAVC are from varves deposited around 17,800 yBP, which is considerably younger than the interpreted age of Glacial Lake Narragansett (discussed below). Other paleomagnetic records exist from the northeastern United States and southeastern Canada for the Holocene and latest Pleistocene, but do not extend back to Late Wisconsinan time (King and Peck, 2001;St-Onge et al., 2003). Ocean Drilling Program (ODP) cores from site 1063 on the Bermuda Rise (1,500 km southwest of the study area) show a 20° westward excursion in relative declination at approximately 19,000 yBP (Lund et al., 2006), but the distance between the sites and coarse temporal resolution of the ODP cores make any correlation with Glacial Lake Narragansett varves speculative at best. The lack of a calibrated, high-resolution, regional paleomagnetic record older than the NAVC limits the usefulness of remanent declination and inclination as a tool for constraining the age of varves deposited in Glacial Lake Narragansett. developed by the Society of Exploration Geophysicists (Norris and Faichney, 2002).

5.
Individual data files were imported into Chesapeake Technologies SonarWeb (v3.16) software for post processing. The full seismic reflection profiles were used (i.e. the records were not clipped at a depth) and the images produced for the profiles had a pixel size of 0.05 cm. Profiles were displayed using an inverse medium yellow-orange to brown known as a 'Klein' color scheme, named for the color of analog paper records produced by that company's wetpaper recordings in the 1970s. We believe the inverse Klein scheme allows us to see more detail on the digital records and provide more contrast between adjacent side-scan facies than traditional grey-scale images.

6.
Profiles were imported into Chesapeake Technologies SonarWeb (v3.16) as standard SEG-Y. The antenna of the DGPS used during surveying was located within 3 m of the towfish, no layback or other spatial transformation was applied.

7.
Files were reoriented in SonarWeb so that the read north to south or west to east (from left to right) on the seismic profiles. Profiles were annotated with the position (Rhode Island State Plane Feel, NAD 1983), date/time and course every 1,000 pings (approximately 200 m at a survey speed of 1.5 m . s -1 ).

8.
The course made good during data collection was calculated using a smoothing function that fits a curve to the GPS reading every 50 pings (Approximately 10 m at a survey speed of 1.5 m . s -1 ) . This removes some of the of the smallscale boat motion without a significant change in spatial accuracy.

9.
Individual profiles were adjusted for both contrast and time-varied gain to maximize contrast between seismic reflectors. Typical values for time-varied gain ranged from 5 -20 dB down the signal with no offset. Differences in seafloor geometry and surface sediment characteristics can change the seismic penetration, even on adjacent seismic profiles, requiring file by file processing.

10.
Occasional files with excess noise, typically a combination of water column interference and multiple reflections of the seafloor, were run through a digital band-cut filter between 600 and 6000 Hz. All other settings in SonarWeb were set to the default settings 11. The upper surface of identified seismic facies were digitized on individual profiles in the 'Seismic Reflectors' section of SonarWeb. Seismic facies are sedimentary packages, distinguishable from adjacent units, based on internal characteristics, (i.e. the intensity, spacing, continuity, and internal geometry of seismic reflectors), external geomorphic form and stratigraphic relationship to other units . These reflectors were