Trough geometry was a greater influence than climate-ocean forcing in regulating retreat of the marine-based Irish-Sea Ice Stream

Marine terminating ice streams are a major component of contemporary ice sheets and are likely to have a fundamental influence on their future evolution and concomi-tant contribution to sea-level rise. To accurately predict this evolution requires that modern day observations can be placed into a longer-term context and that numerical ice sheet models used for making predictions are validated against known evolution of former ice masses. New geochronological data document a stepped retreat of the paleo– Irish Sea Ice Stream from its Last Glacial Maximum limits, constraining changes in the time-averaged retreat rates between well-defined ice marginal positions. The timing and pace of this retreat is compatible with the sedi-ment-landform record and suggests that ice marginal retreat was primarily conditioned by trough geometry and that its pacing was independent of ocean-climate forcing. We present and integrate new luminescence and cosmo genic exposure ages in a spatial Bayesian sequence model for a north-south (173km) transect of the largest marine-ter-minating ice stream draining the last British–Irish Ice Sheet. From the south and east coasts of Ireland, initial rates of ice margin retreat were as high as 300–600 m a –1 , but retreat slowed to 26 m a –1 as the ice stream became topographically constricted within St George’s Channel, a sea channel between Ireland to the west and Great Britain to the east, and then stabilized (retreating at only 3 m a –1 ) at the narrowest point of the trough during the climatic warming of Greenland Interstadial 2 (GI-2: 23.3–22.9 ka). Later retreat across a normal bed-slope during the cooler conditions of Greenland Stadial 2 was un-expectedly rapid (152 m a –1 ). We demonstrate that trough geometry had a profound influence on ice margin retreat and suggest that the final rapid retreat was conditioned by ice sheet drawdown (dynamic thinning) during stabilization at the trough constriction, which was exacerbated by increased calving due to warmer ocean waters during GI-2.


INTRODUCTION
A significant proportion of ice sheet mass balance is regulated by faster flowing corridors of ice (ice streams), which drain accumulation areas and are often marine-terminating (Stokes and Clark, 2001;Bennett, 2003;Stokes et al., 2016). While climate forcing exerts a fundamental control on the retreat of ice masses, internal factors such as phases of over-extension (i.e., an advance due to a dynamic instability rather than toward an equilibrium position), the bed-slope and trough geometry are also important regulators of ice stream behavior (Jamieson et al., 2012;Joughin et al., 2014;Mosola and Anderson, 2006). Marine-terminating ice streams are also susceptible to oceanic influence, including changes in relative sea level, sea surface temperatures and tidal regime (Payne et al., 2004;Arbic et al., 2008). There is presently substantial concern about anthropogenically forced atmospheric and oceanic warming causing the rapid retreat of marine-terminating ice streams in Greenland and Antarctica (Joughin and Alley, 2011;Rignot et al., 2014). To understand fully and predict how these ice masses will evolve, there is a need to understand the complex interactions between external forcings and internal dynamics in modulating ice marginal retreat (e.g., Benn et al., 2007;Jamieson et al., 2012;Schoof, 2007). Constraining the evolution of former ice streams provides important empirical evidence for testing process understanding of modern-day ice masses and evaluating numerical ice sheet models (Stokes et al., 2015).
The Irish Sea Ice Stream (ISIS; Fig. 1) was the largest marine-terminating ice stream to drain the former British-Irish Ice Sheet (BIIS) (Eyles and McCabe, 1989) and during deglaciation it formed an example of marine-based ice stream retreat. The sediments and landforms along the southern and eastern coastal lowlands of Ireland, with >30 km extent of intermittent coastal exposure between the south coast and Dublin (~170 km; Fig. 1), record the dynamics of the western lateral margin of the paleo-ISIS during the last deglaciation. We propose a conceptual model for the retreat of the ISIS inferred from the sediment-landform assemblages and constrain this using a new data set of 13 opti-cally stimulated luminescence (OSL) and 10 cosmogenic nuclide (CN) exposure ages. OSL dating was applied to glacial sediments (glacial outwash) that can be correlated with the presence of a proximal ice margin. CN dating was applied to glacially transported boulders and bedrock to directly constrain the timing of ice margin retreat. The new chronological data are integrated using a Bayesian sequence model and the pace of ice-marginal retreat is compared with existing data sets to explore the importance of potential driving factors. These factors include North Atlantic oceanic and climatic forcing, changes in the confining geometry of the ISIS in the southern Irish Sea Basin (ISB) (e.g., bed-slope and trough geometry) and the feedbacks in the sequence of ice flow behavior with potential over-extension identified in a rapid advance of the ISIS to maximum limits (Chiverrell et al., 2013;Ó Cofaigh and Evans, 2007;Smedley et al., 2017a).

THE IRISH SEA ICE STREAM
The last ISIS ( Fig. 1: Eyles and McCabe, 1989) drained on-shore ice accumulation areas in Ireland, northern England, and southern Scotland (cf. Roberts et al., 2007) and at its maximum extent ca. 25 ka (Ó Cofaigh and Evans, 2007;Smedley et al., 2017a) extended into the Celtic Sea (Scourse et al., 1990;Scourse and Furze, 2001;Hiemstra et al., 2006), possibly as far as the shelf break (Praeg et al., 2015). The advance to this maximum limit has been hypothesized as a rapid, and perhaps short lived, surge-type event (Scourse and Furze, 2001;Ó Cofaigh and Evans, 2001a, 2001b, 2007. Coastal exposures of glacigenic sediments show ubiquitous diamictons containing erratic clasts of an Irish Sea affinity (Irish Sea Tills: Evans, 2001a, 2001b) and document on-shore flow of ice (Thomas and Summers, 1983;Evans, 2001b, 2007;Evans and Ó Cofaigh, 2003). These sediments have been interpreted as representing: (1) the deglacial transition from subglacial to proximal and then distal glacimarine sedimentation in an isostatically-depressed ISB (Eyles and McCabe, 1989;Clark et al., 2012;McCabe, 1997), or (2) subglacial and ice marginal deposition at the lateral grounded margin of an ice stream Summers, 1983, 1984;Evans, 2001a, 2001b;Evans and Ó Cofaigh, 2003). The glacimarine hypothesis is considered unlikely given the magnitude of isostatic loading required (cf. Lambeck and Purcell, 2001;Bradley et al., 2011) and the lack of un ambig uous evidence for glacimarine sedimentation (Ó Cofaigh and Evans, 2001a;Evans and Ó Cofaigh, 2003;Rijsdijk et al., 2010). The sedi ment-landform assemblages distributed along the southern and eastern coasts of Ireland provide the evi-dence base for a conceptual model for the relative order of retreat events and suggests that the ISIS experienced marked changes in the rate of retreat (Fig. 2).

Rapid Initial Advance and Retreat
Initial advance of the ISIS is recorded along the south Irish coast by subglacial diamictons with erratics of Irish Sea provenance (Ó Cofaigh and Evans, 2001a, b). These diamictons contain abundant reworked marine fauna and the youngest ages from a population of 26 radiocarbon ages indicate that advance occurred after ca. 25-24 ka (Ó Cofaigh and Evans, 2007). The timing of the maximum extent of the ISIS on the Isles of Scilly, UK, is indistinguishable within dating uncertainties with new OSL and CN ages constraining this to 25.5 ± 1.5 ka (Smedley et al., 2017a). The Irish Sea diamictons exposed along the south coast of Ireland are overlain by glacial outwash and localized glacilacustrine deposits that record proglacial deposition along the retreating margin of the ISIS ( Fig. 3; Evans, 2001a, 2001b). Finally, following retreat of the ISIS, ice sourced in the Irish midlands advanced beyond the presentday coastline depositing glacial till that is rich in lithologies of an inland origin and deforming the underlying glacifluvial and glacilacustrine sequence Evans, 2001a, 2001b).   The thin and discontinuous nature of these deposits perhaps reflects a relatively rapid retreat with limited development of ice-marginal landforms. This inference is supported by an earlier Bayesian model using legacy geochronological data that suggests that ISIS advance to-and retreat from-its maximum limit was rapid (<1 ka) (Chiverrell et al., 2013).

Establishment of an Oscillating Grounded Ice Margin
Sediment-landform assemblages related to the ISIS also crop out along the east coast of Ireland and provide evidence for relatively greater complexity in ice marginal dynamics. The Screen Hills (Fig. 4) represent the largest glacigenic sedimentary depocenter on the east coast of Ireland, with >150 km 2 of glacial landforms and ~20 km of coastal exposures that locally reach >50 m in height. The stratigraphy re-flects the interaction of subglacial processes and ice marginal deposition, including thrusting and stacking of glacigenic units, and results from at least eleven minor (< km scale) readvances of the ISIS margin (Evans and Ó Cofaigh, 2003;Thomas and Chiverrell, 2011;Summers, 1984, 1983). The sequence suggests that ice marginal retreat slowed as the ISIS margin retreated from the Celtic Sea into St George's Channel (cf. Evans and Ó Cofaigh, 2003) and that a dynamic and oscillating margin was established Summers, 1983, 1984;Thomas and Kerr, 1987;Evans and Ó Cofaigh, 2003).

Renewed Rapid Ice Marginal Retreat?
Further north exposures of glacial sediments related to the ISIS are more sporadic (Fig. 1), which might reflect more rapid ISIS retreat, but extensive Holocene sand dunes may mask the glacial sequence locally. There are no sediment exposures or geomorphological features indicative of ice margin oscillations or any large scale readvance(s) north of the Screen Hills. Consequently, it is inferred that retreat from the Screen Hills was a quasi-continuous process. The first ice marginal position north of the Screen Hills occurs at Greystones, Co. Wicklow, Ireland, which has been interpreted as a morainal bank complex deposited in an ice proximal subaqueous basin prograding from a bedrock high (McCabe, 2008). The sections display no evidence for oscillation or readvance of the ice margin (McCabe and Ó Cofaigh, 1995). Further north at Killiney, Co. Dublin, Ireland, glacial diamictons (Irish Sea and inland origin) and outwash deposits form a complex sequence of glacitectonically stacked units resulting from overriding by ice and oscillation of the ice margin during ISIS retreat (Rijsdijk et al., 2010).

FIELD SITES
The conceptual model (Fig. 2) for ISIS retreat informed our field sampling strategy with sites distributed to provide geochronological constraints on the timing and pace of deglaciation throughout the region. The 14 OSL samples from eight sites targeted glacial outwash sands associated primarily with known ice marginal positions and locations (Table 1) through the retreat sequence. The CN dating using in situ 10 Be targeted a mixture of glacially modified bedrock outcrops or glacially faceted and transported boulders producing 10 samples from four locations (Table 2).

Kilmore Quay
The exposures at Kilmore Quay, Co. Wexford, Ireland, have previously been described in detail (Ó Cofaigh and Evans, 2001a;Evans and Ó Cofaigh, 2003). Two samples were taken from the coastal cliff where relatively thin glacifluvial sands overlie Irish Sea tills that crop out at the base of the sequence. KQ1 and KQ2 were collected for OSL dating from medium-to fine-sand units ( Fig. 3; Table 2) to constrain the timing of ISIS marginal retreat onto the south coast of Ireland. While the sequences are generally glacitectonised the sands retain primary depositional features such as ripples.

Carnsore Point
Twelve km east of Kilmore Quay, Carnsore Point, Co. Wexford, Ireland, is characterized by a spread of large granite boulders derived from local outcrop of the Carnsore Granite (O'Connor et al., 1988). The boulders occur on the surface and reflect deposition by the retreating ISIS given the lack of sedimentary evidence for re-advances of inland ice in this area (Evans and Ó Cofaigh, 2003). There is no obvious orientation to the spread of boulders and no other glacial landforms in the immediate vicinity. Many boulders exhibit signs of human activity including incorporation into field boundaries ( Fig. 5A and Fig. DR6 in GSA Data Repository 1 ). Some of these however, were inferred in the field to be in situ and three samples (CS1-3) were collected for analysis with 10 Be to provide constraint on the passage of the ice margin from the south coast of Ireland into St George's Channel.

Screen Hills
Thirty km to the northeast of Carnsore Point, the Screen Hills represent the largest accumulation of glacial sediments on the east coast of Ireland (Evans and Ó Cofaigh, 2003;Thomas and Chiverrell, 2011;Summers, 1983, 1984). OSL samples were collected from glacifluvial outwash sands at four sites which were chosen from the ~15 km of continuous coastal exposure studied in detail by Summers (1983, 1984) extending from Blackwater Harbour in the south to Tinnaberna in the north, Co. Wexford, Ireland (Fig. 4). These relate to well-defined ice marginal positions (Thomas and Summers, 1984;Thomas and Chiverrell, 2011) and the sampled units retain primary deposi tional features such as ripples and faint laminations. Progressing from south to north-near Blackwater Harbour-two samples of mediumcoarse sands with faint ripples (BW1) and laminations (BW2) were collected from a >30-mthick wedge of outwash sands and gravels that thin to the south away from a glacitectonised ice marginal position (Limit 2: Fig. 4; Thomas and Chiverrell, 2011). The samples were collected at 5-8 m depth in the 20-m-thick series of alternating outwash sand and gravel sheets that dip gently to the south and overlie a basal diamicton (Fig. 6A). Knocknasillogue, Co. Wexford, Ireland, is ~1.7 km further to the northeast, and two samples separated vertically by ~1 m (KS1 and 1 GSA Data Repository item 2018183, details on optically stimulated luminescence methodology and background information on sites, is available at http:// www .geosociety .org /datarepository /2018 or by request to editing@ geosociety .org.  Note: CN-cosmogenic nuclide; ASL-above sea level; Qtz-quartz. *Topographic shielding correction calculated using online calculator (Balco et al., 2008; available at http:// hess.ess.washington .edu /math /general /skyline _input .php). † Samples were spiked with 9 Be carrier 849 ± 12 μg/g. § Relative to NIST_27900 with 10 Be/ 9 Be taken as 2.79 × 10 -11 . # Ratios not blank corrected. **Corrected for a process blank with 10 Be/ 9 Be ratio of 4.85 ± 0.61 ×10 -15 . † † Corrected for a process blank with 10 Be/ 9 Be ratio of 2.37 ± 0.41 ×10 -15 . ## Corrected for a process blank with 10 Be/ 9 Be ratio of 6.05 ± 0.84 ×10 -15 . KS2) were collected from a depth of ~10 m in medium-fine rippled sands with fine laminations (Fig. 6B). The sampled sequence is immediately up-ice of a pronounced glacitectonised marginal position (Limit 3: Fig. 4; Thomas and Chiverrell 2011). The sampling targeted the upper portion of a ~20-m-thick sequence of sands and gravels that overlie a basal Irish Sea diamicton, with both units thrust forward and upwards in a glacitectonic episode linked to the capping diamicton that completes the vertical succession. At Ballyvaldon, Co. Wexford, Ireland, ~2.3 km to the northeast, a single sample (BV1) of mediumcoarse sands was collected at 10 m depth from a ~20 m thick and laterally extensive sequence of outwash sands fronting an ice marginal position ~700 m to the north (Limit 9: Fig. 4; Fig. 6C; Thomas and Chiverrell, 2011). At Tinnaberna, a further ~2.4 km to the northeast, two samples (TB1 and TB2) were collected from sand layers within alternating outwash sands and gravels fronting a further glacitectonised ice marginal positions (Limit 11: Fig. 4; Thomas and Chiverrell, 2011). The samples were taken from outwash sands ~4 m to 0.5 m above a basal Irish Sea diamicton at burial depths of 10 m and 20 m, respectively, but TB2 did not yield an age determination.

Wicklow Point
Fifty km north of Tinnaberna, there is a distinct channel orientated E-W and cut into schist bedrock at ~15 m above sea level (ASL) on Wicklow Point ( Fig. 5B and Fig. DR6 in GSA Data Repository). The channel is ~250 m long and 10 m wide, widening to 20 m at its southern end and has an undulating thalweg. The onset and end of the channel are abrupt and there is no catchment, a configuration common in subglacial meltwater channels. The feature is therefore interpreted as a likely subglacial melt water channel. The channel would have been exposed during deglaciation and thus two samples were collected for 10 Be analysis from the channel wall, >3 m from the top of the channel (WK1-2). This site was chosen to provide constraints on deglaciation between the sedimentary exposures at the Screen Hills and Greystones.

Ballyhorsey Quarry
Located 5 km inland from coastal sections lies a series of low-level (<120 m ASL) basins separated by bedrock-cored ridges. Drainage channels fret these bedrock ridges (Ravier et al., 2014), and they locally feed substantial proglacial depocenters (see Fig. DR7 in GSA Data Repository [footnote 1]). At Ballyhorsey a former sand and gravel quarry has provided exposures into extensive sub-and proglacial outwash in the "Kilpedder Basin" as described by Ravier et al. (2014) (Fig. 7). The lower sections studied by these authors document basal diamictons that interdigitate with glacifluvial deposits that formed subglacially. These units are buried by prograding subaqueous glacilacustrine fan facies that prograde and dip southwards as wedges and both thin and fine (from gravels to sands) with increasing distance from the ice-contact fan apex. This fan was likely deposited in a lake dammed by Irish Sea ice situated to the north and east (Ravier et al., 2014). In March 2014, the exposures were much degraded, but reasonable exposures were identified at the top of the sequence in the northwest corner of the quarry. These exposures were within 2-3 m of a flat horizontal surface and displayed horizon-tally and planar bedded and rippled fine to medium sands that we interpret as deltaic top-sets formed during the late stages of deposition. Two samples (BH1 and BH2) were collected from fine-medium planar sands with ripples preserved from the uppermost ice proximal delta top-set sands (Fig. 7).

Greystones
On the coast ~22 km north of Wicklow Point and ~7 km northeast of the Ballyhorsey Quarry, extensive (1.7 km) cliff exposures at Greystones document a substantial ice marginal position (Fig. 1). McCabe and Ó Cofaigh (1995) interpreted the sequence as morainal bank deposits that accumulated in a subaqueous setting by subglacial discharge from the ice margin, with the sequence deposited so that it did not evolve into a Gilbert-type delta. In March 2014, the exposures showed thin (2-3 m thickness) sand and gravel delta foresets ( Fig. 8) with associated horizontally bedded sands and gravels interpreted as thin topsets. A single sample (GR1) was taken from this thin (~1 m) rippled sand unit that lies above 10-15 m of subaqueous outwash gravels and diamictons of Irish Sea provenance. Relating the unit sampled to the descriptions of McCabe and Ó Cofaigh (1995), it appears to correlate with a thin (>1 m) horizontally stratified series of ripped sands (Sr) interbedded with planar gravels (Gms) that lie immediately below their uppermost lithofacies (LFA4) which comprised 3.5 m of planar massive and poorly sorted gravels (Gm and Gms).

Bray Head
Five km north of the Greystones site, Bray Head is a hill (240 m ASL) composed of quartzite bedrock that shows extensive signs of glacial modification. Abraded and striated surfaces are present on the summit ridge of Bray Head with striations orientated in a general NNW-SSE direction. Plucked faces also occur on the lee side of flow. A single granite erratic boulder was located on Bray Head, ~250 m north of the summit. Given its altitude and proximity to the Greystones site, Bray Head may have been deglaciated prior to deposition of the sediments from which the optically stimulated luminescence (OSL) sample GR1 was taken. In addition to the granite erratic (BR1), two bedrock samples were collected from 220 m ASL on the summit ridge of Bray Head; one sample was collected from an abraded surface (BR2) with a further sample taken from a plucked surface (BR3) that lay >2.5 m beneath the upper abraded surface (Fig. 5C and Fig. DR6 in GSA Data Repository).

Howth Peninsula
Howth is a peninsula 4 km in length that forms the north coast of Dublin Bay, Co. Dublin, Ireland. The Hill of Howth (171 m ASL) lies in the center of the peninsula and is composed primarily of quartzite. At its summit there is clear evidence of glacial abrasion ( Fig. 5D and Fig. DR8 in GSA Data Repository). Striae indicate ice-movement from the north (Stephens and Synge, 1957) as the ISIS impinged onto the eastern coast of Ireland. Two samples were collected from the summit of the Hill of Howth from abraded quartzite bedrock (HH1-2). On the north side of the peninsula there is a ~0.5 km 2 deposit of stratified sands and gravels ( Fig. 9) extending inland that has a maximum thickness of ~15 m (Lamplugh, 1903). The deposits have been interpreted as deltaic in origin composed mainly of reworked glacial deposits transported along local drainage lines and deposited in an ice marginal water-body (Lamplugh, 1903). The sequence was deposited after retreat of the ice margin north and west of Howth but while the ISIS was still present in the ISB to the east and ponding ice-dammed meltwater in Dublin Bay. Two OSL samples (HD1-2) were taken from fine-medium rippled sands that represent the topsets of the "Howth Delta."

OSL Dating
Samples for OSL dating were collected by hammering opaque tubes into the sedimentary sections. External gamma dose-rates were determined in situ using field gamma spectrometry. Concentrations of U, Th, K, and Rb were determined for each sample using inductively coupled plasma-mass spectrometry (ICP-MS) and atomic emission spectroscopy (ICP-AES) ( Table 3). These concentrations were used to calculate the external beta dose-rates and in situ gamma spectrometry determined the external gamma dose-rate. Sample preparation and analysis followed methods outlined in Smedley et al. (2017a). Single grains of quartz were used to determine equivalent dose (D e ) values (Tables DR1-DR14; Figs. DR1-DR3 in GSA Data Repository [footnote 1]) using a Risø TL/OSL DA-15 automated single-grain system equipped with a 90 Sr/ 90 Y beta source (Bøtter-Jensen et al., 2003). Grain sizes of 210-250 µm were used for OSL analysis of each sample, except sample KQ1 that had a grain size 150-180 µm and so it was likely that up to four grains were present in each hole on the single-grain disc (i.e., microhole measurements). Sample analysis methods are summarized in the GSA Data Repository (see footnote 1). The D e distribution of sample TB2 (Fig. DR3) had scatter at lower doses that suggests the potential for post-depositional mixing (e.g., bioturbation or cryoturbation) thus OSL dating of this sample was considered unreliable. The central age model (CAM; Galbraith et al., 1999) was used to determine an age for sample KQ2 as the symmetrical D e distribution suggests that these grains were not heterogeneously bleached prior to burial. The minimum age model (MAM; Galbraith et al., 1999;Galbraith and Laslett, 1993) was used to determine ages for the remaining samples given their asymmetrical D e distributions (Duller, 2008). The D e values were then divided by the doserate to determine an age. See the GSA Data Repository for full details of the OSL analysis performed in this study, in addition to the D e values determined for each sample.

Cosmogenic Nuclide Exposure Dating
Sampling targeted the uppermost fresh surfaces of boulders and bedrock to minimize the possibility of post-depositional adjustment such as toppling or large scale spallation of bedrock. Topographic shielding was measured in the field using a compass and clinometer and correction factors calculated using the CRONUS-Earth online calculator (Balco et al., 2008). CN sample preparation was undertaken at the University of Glasgow, Scotland, UK, using standard mineral separation techniques (cf. Kohl and Nishiizumi, 1992). Beryllium extraction was carried out at the Cosmogenic Isotope Analysis Facility-Scottish Universities Environmental Research Centre (CIAF-SUERC), using procedures based on Child et al. (2000). The 10 Be/ 9 Be ratios were measured on the 5MW accelerator mass spectrometer (AMS) at SUERC (Xu et al., 2010).
There are a variety of online calculators (Balco et al., 2008;Marrero et al., 2016;Martin et al., 2017), scaling schemes (e.g., Lal, 1991;Lifton et al., 2014;Stone, 2000), and production rate calibrations (e.g., Putnam et al., 2010;Small and Fabel, 2015;Young et al., 2013) available for calculating exposure ages. We present exposure ages calculated using the CRONUS-Earth calculator (Balco et al., 2008) and the CRONUScalc calculator  using a selection of scaling schemes (Lm [Lal, 1991;Stone, 2000] and SA [Lifton et al., 2014]) and production rates Fabel et al., 2012). In practice, choice of calculation method and scaling scheme often make little difference for calculation of 10 Be exposure ages (cf. Table 4). Choice of production  rate, however, can make a significant difference and affect subsequent interpretations. Given an apparent geographical bias on production rate values  it has been argued that ages should be calculated using the most geographically appropriate calibration (Small and Fabel, 2016a). Considering this, and to aid comparison to previously published work from Britain and Ireland, we focus discussion on ages calculated with the CRONUS calculator (http:// hess .ess .washington .edu; Balco et al., 2008), the Lm scaling, and with a reference sea-level high latitude production rate of 4.00 ± 0.17 atoms g -1 quartz (cf. Fabel et al., 2012) calibrated from a site in Scotland, <500 km from our study site.

Bayesian Age Modeling
Integration of the new CN and OSL age control was undertaken using Bayesian age modeling, with the focus on discerning the timing of ice marginal retreat. A sequence model was used which requires the dating information to be arranged in a likely younging order irrespective of the actual age values; this prior model (the hypothetical relative order of events) takes the form of a south to north relative distance reconstruction of retreat of the ISIS margin (Fig. 2). The Bayesian modeling was conducted using OxCal 4.3 (Bronk Ramsey, 2017) and run in an outlier mode (Buck et al., 1991;Bronk Ramsey, 2009) to assess for outliers in time (t) with outlier probabilities assigned a minimum outlier probability of 0.05 which was increased on the basis of quality assurance criteria (see Small et al., 2017). This approach allows all data to be included within the model (suitably downweighted) and removes the need for the somewhat subjective approach of identifying and removing outliers. The likelihood data-or age measurements-were expressed as a Student's t-distribution (more long-tailed than a normal distribution) and attributed an outlier scaling of 10 0 -10 4 years (cf. Bronk Ramsey, 2009). The model (see GSA Data Repository for code) was punctuated by uniform in shape prior boundaries that separate a series of phases, essentially an unordered group of likelihoods-or age measurements-for individual locations. The modeled boundary ages document the retreat of the ISIS and allow rates of retreat between boundaries to be calculated. However, the Bayesian approach to constraining ice margin retreat assumes uni-directional ice retreat (i.e., from south to north in this case). Consequently, any unidentified large scale readvances would cause the reconstructed retreat rates to underestimate the maximum retreat rate experienced by the ice margin between any two given Bayesian boundaries. Similarly, any stillstands or periods of ice margin oscillation will have a similar effect. It therefore must be highlighted that the reconstructed retreat rates presented here represent time-averaged retreat rates between the Bayesian boundaries and that the actual rate likely varied by an unquantifiable amount to be both faster and slower than the time averaged rate. However, given the scale of the transect (~175 km), and the duration of retreat (ca. 7 ka) we consider that time averaged rates provide useful context for modern day retreat rates and for comparison to paleo-ice sheet model output. Additionally, given the lack of evidence for large scale readvances in this sector of the ISB we infer that the time averaged retreat rates presented are a reasonable approximation of the true retreat rate of the ISIS.

Luminescence Ages
The new OSL ages range from 37.1 to 16.9 ka ( Fig. 10; Table 3). There was agreement within uncertainties for the two OSL ages determined for sediments from each site-Kilmore Quay, Knocknasillogue, Ballyhorsey, Howth Deltaexcept for Blackwater. At Blackwater, the OSL age for sample BW1 (37.1 ± 5.0 ka) was older than sample BW2 (24.9 ± 3.3 ka). Both samples were heterogeneously bleached prior to burial, thus the difference in OSL ages is potentially due to the minimum dose population in the single-grain D e distribution of sample BW1 not being as well characterized as sample BW2. It is therefore likely that deglaciation at Blackwater is more accurately constrained by the OSL age for sample BW2 (24.9 ± 3.3 ka). This is supported by the fact that the mean age of BW1 (37.1 ka) pre-dates the timing of the ISIS maximum extent. Additionally, the OSL ages for sites to the south (Kilmore Quay) and north (Knocknasillogue) of Blackwater, which constrain the timing of ice-marginal retreat across this transect are consistent with the younger (BW2) age. The OSL ages determined for individual samples from Ballyvaldon (21.8 ± 4.7 ka), Tinnaberna (24.0 ± 3.7 ka), and Greystones (16.9 ± 2.3 ka) are also-within uncertainties-in agreement with the conceptual model of deglaciation relative to the OSL ages at Kilmore Quay, Knocknasillogue, Bally horsey, and Howth.

Cosmogenic Exposure Ages
The cosmogenic ages obtained span the range of 27.8-11.4 ka ( Table 4; Fig. 10) and show significant variation within individual sites. Only at Bray Head do two samples (BR1 and BR3) agree within their 1σ analytical uncertainties, where the mean age (20.7 ± 1.1 ka) represents the best estimate of the deglaciation age. We note that BR2 was collected from an abraded bedrock surface that is more likely to be influenced by inheritance (Briner and Swanson, 1998) compared to the boulder and plucked bedrock samples (Colgan et al., 2002;Putkonen and Swanson, 2003;Heyman et al., 2011). Of the exposure ages from the other sites some are likely erroneous. CS1 has an apparent exposure age that post-dates the termination of the Younger Dryas at which point Britain and Ireland were ice free (Bromley et al., 2014;Small and Fabel, 2016b). CS2 has an apparent exposure age that pre-dates both initial advance of the ISIS (Ó Cofaigh and Evans, 2007) and the timing of maximum extent on the Isles of Scilly (Smedley et al., 2017a). These ages are down-weighted accordingly in the subsequent Bayesian model (cf. Small et al., 2017).

Bayesian Age Modeling
The geochronological data above do not exist in isolation but are accompanied by spatial information (the "prior model") that places dated sites within a morpho-stratigraphical order representing a series of ice-marginal positions arranged in the order of ice margin retreat (cf. Chiverrell et al., 2013). This independently constructed relative order of events (Fig. 2) constrains the independent age measurements with overlapping age probability distributions providing the basis for using Bayesian age modeling (Buck et al., 1991;Bronk Ramsey, 2009). Bayesian age modeling produced a conformable sequence with an agreement index exceeding the threshold of 60 as defined by Bronk Ramsey (2009) (Fig. 11; Table 5). Assigning increased outlier probabilities for those samples where posterior likelihood > prior likelihood did not result in any significant differences in age model output.   (Balco et al., 2008). See text for details of production rates. Ages assume 1 mm ka -1 erosion, no inheritance, and density of 2.65 g cm -3 . † Calculated using CRONUScalc v.2 calculator . Details of production rates in Borchers et al. (2016). Ages assume 1 mm ka -1 erosion, no inheritance, and density of 2.65 g cm -3 .  The model produces a conformable sequence with modeled boundary ages for all sample sites. These ages range from 26.6 ± 1.9 ka to 19.5 ± 1.3 ka (Table 5). Additionally, the model produces boundary ages that have significantly lower uncertainties than the best estimate deglaciation ages produced by geochronological dating for sites where there was agreement between ages.

C _ D a t e H H 1 C _ D a t e H H 2 C _ D a t e H D 1 C _ D a t e H D 2 B o u n d a r y H o w t h h i g h B o u n d a r y H o w t h D e l t a
Our Bayesian age model indicates that initial ice marginal retreat onto the southern Irish coast occurred at 25.9 ± 1.4 ka (Boundary 2). Retreat of the ISIS from the southern coast of Ireland is constrained by the modeled age (Boundary 3) of 25.1 ± 1.2 ka. Deglaciation to the Wexford coast, and associated deposition of the Screen Hills complex, occurred between 24.2 ± 1.2 ka and 22.1 ± 0.7 ka (Boundaries 4-7). Subsequent deglaciation of the eastern coast of Ireland is constrained by modeled ages from Wicklow Point, Bray Head, Greystones, and the Howth Peninsula and occurred between 21.6 ± 0.6-20.1 ± 1.0 ka (Boundaries 8-12). Final deglaciation of the portion of the ISIS contained within our prior model was complete by 19.5 ± 1.3 ka, when deposition of the Howth Delta occurred (Boundary 13). These modeled age boundaries can be interpreted in terms of the timing of retreat of the lateral margin of the ISIS along the south and east coasts of Ireland and used to test our conceptual model of ISIS deglaciation.

Retreat Rate along the Lateral Margin of the Irish Sea Ice Stream
Retreat rates for the ISIS (Table 6) were derived from the mean modeled boundary ages and evidence major changes in the pace of marginal retreat displayed by the ISIS. We separate ISIS marginal retreat into four distinct stages. Stage I (25.9-24.2 ka) deglaciation proceeds at a near constant rate of ~26 m a -1 with the margin passing from Kilmore Quay (S Irish Coast) to Carnsore Point (Rosslare corner) and then to Blackwater Harbour (Blackwater). During Stage 2 (24.2-22.1 ka; Blackwater-Screen Hills) modeled rates of ISIS axial margin retreat exhibit an order of magnitude slowing at the Screen Hills (Table 6) from ~26 m a -1 to ~3 m a -1 . In this deceleration the ISIS ice margin would have likely experienced numerous stillstands and readvances, which is in accordance with sedimentological and geomorphological evidence for repeated ice margin re-advance in the Screen Hills area Summers, 1983, 1984). Similarly, the relatively rapid rates of retreat onto and along the south coast of Ireland accord well with the sedimentological evi-dence suggesting the first major slowing of ice marginal retreat was in close proximity to the Screen Hills (cf. Summers, 1983, 1984;Thomas and Kerr, 1987;Evans, 2001a, 2001b;Evans and Ó Cofaigh, 2003;Thomas and Chiverrell, 2011). During Stage 3 (22.1-21.6 ka; Screen Hills-Wicklow) retreat of the axial ice margins proceeded rapidly at average rates of 152 m a -1 (Table 6) and there are no reported substantial ice marginal landforms or accumulations of glacial deposits; this is commensurate with rapid ISIS retreat. This retreat phase (Stage 3) accounts for 75 km of the total 173 km of ISIS axial retreat in the ISB. Ice marginal retreat during Stage 4 (21.6-19.5 ka; Wicklow-Howth) is less rapid at ~21 m a -1 , supporting sedimentological evidence for ice margin stillstands at Greystones (McCabe and Ó Cofaigh, 1995), and oscillatory ice mar-ginal retreat at Killiney (Rijsdijk et al., 2010). In summary, there is a strong correspondence between the modeled rates of ISIS marginal retreat and our conceptual model inferred from the geomorphology and stratigraphy (Fig. 2).

Ice stream Behavior: External Forcing or Internal Dynamics?
Our modeled boundary age for ISIS deglaciation of the south Irish coast (24.5 ± 1.5 ka) is indistinguishable from the existing geochronology constraining the advance to, and maximum extent of, the ISIS (Ó Cofaigh and Evans, 2007;Ó Cofaigh et al., 2012;Smedley et al., 2017a). Our data and the existing chronology suggest that the advance to, and retreat from, maximum extent (total distance of ~600 km) occurred within 1-2 ka (i.e., within the resolution of the  available geochronology) and, assuming equal time for both components, this points to axial retreat rates of 300-600 m a -1 . These are an order of magnitude higher than those observed during our Stage 1 (S Irish Coast-Blackwater; 25.9-24.2 ka) and we suggest that this reduction of the retreat rates reflects the continued narrowing of the calving margin width toward and into St George's Channel. Given the short duration but extensive nature of ISIS maximum advance, it likely reflects a dynamic instability within the ice sheet during growth leading to rapid advance and probably over-extension such that the ice stream extended far beyond any position that can be reconciled with a glacial system in which accumulation and ablation were in equilibrium. Rapid or surge-like advances of the ISIS into the Celtic Sea have previously been inferred from marine cores (Scourse et al., 1990). The ISIS at maximum extent, given its axial length, would have had a low profile and been thin at its margin (cf. Scourse et al., 1990), and in advancing from the topographic constriction of St George's Channel it spread out as a piedmontlike lobe. This low gradient, thin-ice and wide calving margin configuration likely rendered an over-extended ISIS vulnerable to rapid retreat perhaps conditioned by accelerated calving but driven by factors such as glaci-eustatic changes, a warmer sea surface, a megatidal regime and/or increasing air temperature leading to hydrofracturing (Haapaniemi et al., 2010;Scourse et al., 2009Scourse et al., , 2018Pollard et al., 2015). Notwithstanding limits imposed by uncertainties of the modeled chronology, the ISIS advance and retreat back to the south coast of Ireland by ca. 24.5 ka appears to pre-date any significant climate/oceanic warming in the North Atlantic (i.e., Greenland Interstadial 2; GI-2; 23.3-22.9 ka b2k; Rasmussen et al., 2014). However, even without external climate forcing at maximum extent, the ISIS may have been intrinsically unstable as thin ice masses near buoyancy are vulnerable to full-thickness tensile failure which would also lead to accelerated calving (Ma et al., 2017). Slowing of ice margin retreat (Stage 2; Blackwater-Screen Hills) in our modeled chronology between 24.2 ± 1.2 ka and 22.1 ± 0.7 ka corresponds with transit of the ISIS margin into the narrowing of the St George's Channel and the onset of minor readvance episodes and stillstands on the lateral ISIS margin at the Screen Hills. The Screen Hills mark a step-change in the ISIS axial trough geometry, with narrowing of the potential calving margin and a shallower trough depth (Fig. 12). These likely changes would have reduced the lateral and vertical extent of the calving margin, thus reducing calving rates leading to stabilization of the ice margin. A similar scenario of ice margin stabilization at trough constrictions was present in numerical modeling of the Marguerite Bay Ice Stream (Jamieson et al., 2012) and has been widely recognized in the literature (e.g., Benn et al., 2007). When an ice stream stabilizes at a trough constriction it ceases to thin at the grounding line although longitudinal stress coupling acts to propagate thinning up stream (Payne et al., 2004;Jamieson et al., 2012). This mechanism can fundamentally limit the duration of stability as the upstream drawdown can only deliver ice to the grounding line for as long as the source areas maintain sufficient ice storage. Such a scenario of drawdown would likely be accompanied by faster flow linked to accelerated ice flux  to the grounding line. Geophysical evidence for ice streaming occurs immediately to the north of our region in the central ISB (Van Landeghem et al., 2009) providing evidence for fast-flow, potentially while the ice margin was located at the Screen Hills. Continued ice sheet drawdown would cause the ISIS to become increasingly sensitive to external perturbations that act to increase mass loss at the grounding line as eventually there would be insufficient upstream mass to sustain fast flow. However, the switch from slower oscillating ice margin behavior (Stage 2) to more rapid marginal retreat (Stage 3) at 22.1 ka appears to post-date GI-2 (23.3-22.9 ka; Fig. 13A), a time of ocean warming in the North Atlantic cited as a potential major driver of BIIS deglaciation (Scourse et al., 2009). Additionally, given that modeled relative sea levels are falling at 21 ka (Bradley et al., 2011;Fig. 13B) it is reasonable to assume that a rise in relative sea-level capable of triggering rapid ice margin retreat was unlikely to have occurred prior to 21 ka. Thus, both the slow-down of marginal retreat and later acceleration of retreat rates display an apparent poor coupling to ocean-climate forcing in the North Atlantic.
More rapid retreat (Stage 3) continues across a normal bed-slope that shallows in the direction of retreat (Fig. 12) with a calving margin width that widens gradually from ~65-90 km. While the widening calving margin may have acted to increase calving and thus exacerbate retreat, the decreasing trough depth (i.e., normal bed slope) could have acted to stabilize the ice margin (cf. Jones et al., 2015) although we note that the changes in depth are small compared to those encountered by contemporary ice streams in Antarctica. Given that these factors would counter act each other, and given that the changes are gradual and relatively small, we suggest that this trough geometry is not conducive to the very rapid retreat (Stage 3) implied by our modeled boundary ages. Additionally, this deglaciation continues during the relatively cooler conditions of Greenland Stadial 2 (GS-2) (Fig. 13). Given that neither trough geom etry nor climate forcing can be directly correlated with the rapid nature of the observed retreat we infer that it is the result of ice sheet reorganization (drawdown) due to increased flux of ice to a stabilized grounding line, causing upstream thinning. The mass loss required to initiate the rapid retreat may well have been exacerbated by external forcing (i.e., GI-2) with increased calving due to warmer ocean waters as indicated by numerous records from ocean cores proximal to the former ISIS (Scourse et al., 2009). In this way the rapid retreat could be interpreted as a delayed response to climate forcing. The modeled rates of ice marginal retreat slow during Stage 4 (Wicklow-Howth) and coincide with morphostratigraphic evidence for stillstands or marginal oscillation (McCabe and Ó Cofaigh, 1995;Rijsdijk et al., 2010). This period of slower retreat occurs across a reverse bedslope (Fig. 12). This seems counterintuitive as this scenario is widely cited as driving rapid grounding line retreat (e.g., Favier et al., 2014;Jamieson et al., 2012;Schoof, 2007), but again we note that the magnitude of depth change is small (<50 m). The more substantial accumulations of ice marginal sediment are both associated with bedrock highs on the lateral margin of the ice stream that may have acted as pinning points (McCabe, 2008), and could account for the slowing of the retreat rates observed during Stage 4. A similar scenario has been proposed for the Llŷn Peninsula, Wales, UK, on the opposite lateral margin of the ISIS (Smedley et al., 2017b).

CONCLUSIONS
The geochronological data presented here allow us to test a conceptual model of ISIS deglaciation in south and east Ireland inferred from the sediment-landform assemblage record. Integration of new geochronological data using Bayesian age modeling produces a conformable age model that supports the conceptual model of deglaciation, with extremely rapid (300-600 m a -1 ) retreat from maximum extent, a slowing of retreat (26 m a -1 ) during the period 25.9-24.2 ka, ice margin stabilization (3 m a -1 ; 24.2-22.1 ka), rapid retreat (152 m a -1 ; 22.1-21.6 ka), and finally a return to slower retreat rates (21 m a -1 ; 21.6-19.5 ka).
This timescale strongly suggests that aspects of ISIS behavior during deglaciation displayed a complex relationship to external climatic forcing. Extremely rapid advance of the ISIS to its maximum extent, and its subsequent retreat at 26-25 ka is not directly correlated with distinct climate forcing in the North Atlantic region. Such behavior may be explained as a dynamic instability in response to overextension of the ice stream to the maximum limit that rendered it vulnerable to rapid retreat. Similarly, the stabilization of the ice margin at the Screen Hills spans a time of distinct warming (GI-2) with the subsequent rapid retreat occurring during colder conditions of GS-2. The stabilization at the Screen Hills is the most distinct change in pace of ISIS retreat evidenced by the data presented here and it occurs where there is a step-change in the confining trough geometry highlighting the important role that this plays in conditioning ice margin retreat. However, contrary to this, the ISIS subsequently underwent rapid retreat without major changes in trough geom-etry. We speculate that this represents a delayed response of the ice margin to the climate forcing of GI-2. Overall, changing trough geometry and internal feedbacks related to the overextension, retreat, and stabilization of the ISIS appear to obscure the role of external drivers such as climatic forcing.
The conceptual model and geochronological data presented here provide evidence for specific ice margin behavior during overall deglaciation that provides a testing ground for numerical models that likely require high resolution representations of grounding line dynamics. As contemporary ice streams in Greenland and Antarctica evolve in response to anthropogenic climate change they will undergo retreat that is conditioned both by climatic forcing and their internal dynamics. Our data highlight the potential for the evolution of rapid ice margin retreat to be highly nonlinear and conditioned strongly by trough geometry.

ACKNOWLEDGMENTS
This work was supported by the Natural Environment Research Council (NERC) consortium grant; BRITICE-CHRONO NE/J009768/1 (NE/J007196/1 to Durham University). The cosmogenic analyses were supported by the NERC Cosmogenic Isotope Analysis Facility allocation 9155.1014. Thanks are due to the staff at the Scottish Universities Environmental Research Centre accelerator mass spectrometry (SUERC AMS) Laboratory, East Kilbride, UK, for 10 Be isotope measurements. H. Wynne is thanked for etching the quartz grains for luminescence dating. We would like to thank the two reviewers, Jon Merritt and Benjamin Laabs, for constructive reviews that have improved this manuscript. Data used in this work is available from the author on reasonable request.