Maximum extent and readvance dynamics of the Irish Sea Ice Stream and Irish Sea Glacier since the Last Glacial Maximum

: The BRITICE ‐ CHRONO Project has generated a suite of recently published radiocarbon ages from deglacial sequences offshore in the Celtic and Irish seas and terrestrial cosmogenic nuclide and optically stimulated luminescence ages from adjacent onshore sites. All published data are integrated here with new geochronological data from Wales in a revised Bayesian analysis that enables reconstruction of ice retreat dynamics across the basin. Patterns and changes in the pace of deglaciation are conditioned more by topographic constraints and internal ice dynamics than by external controls. The data indicate a major but rapid and very short ‐ lived extensive thin ice advance of the Irish Sea Ice Stream (ISIS) more than 300 km south of St George's Channel to a marine calving margin at the shelf break at 25.5 ka; this may have been preceded by extensive ice accumulation plugging the constriction of St George's Channel. The release event between 25 and 26 ka is interpreted to have stimulated fast ice streaming and diverted ice to the west in the northern Irish Sea into the main axis of the marine ISIS away from terrestrial ice terminating in the English Midlands, a process initiating ice stagnation and the formation of an extensive dead ice landscape in the Midlands. © 2021 The Authors Journal of Quaternary Science Published by John Wiley & Sons, Ltd.


Introduction
The Irish Sea Ice Stream (ISIS) was the largest ice stream draining the former British-Irish Ice Sheet (BIIS; Fig. 1A).Extending for some 750 km from feeder catchment uplands in southwestern Scotland, eastern Ireland, the English Lake District, Snowdonia and west Wales, it flowed southwards through the Irish Sea Basin (ISB).During its maximum extent, it penetrated and crossed the Celtic Sea to a marine terminus south of St George's Channel (Fig. 1A).ISIS was, however, unique in the BIIS context in that it was a compound ice stream with a secondary terrestrial non-streaming terminal zone in the English Midlands, here defined as the Irish Sea Glacier (ISG; Fig. 1A; Thomas, 1985Thomas, , 1989Thomas, , 2005;;Chiverrell et al., 2020) with a bifurcation zone north of the current North Wales coast; please note that the term Irish Sea Glacier has previously been used to denote ice more generally within the ISB but is used in this more restricted sense here (cf.Lewis and Richards, 2005).Unlike some of the other ice streams draining the former BIIS, the significance of the ISIS has been recognized for over 100 years (Wright, 1914) based on erratic trains, notably of Ailsa Craig microriebeckite and Shap granite (Fig. 1A), till matrix lithologies and landform distribution; 'Irish Sea Till', though spatially variable, has been a mappable lithostratigraphic unit along the margins of the Irish and Celtic seas readily distinguishable from more local tills of inland ice source on the basis of colour (dark blue-grey in the south, red in the north), lithology, high matrix to clast ratio and, above all, the incorporation of marine carbonate macro-and microfauna (e.g.Austin and McCarroll, 1992) giving rise to the designation 'shelly till'.In parallel with the description and process understanding of ice streams within contemporary ice sheets (e.g.Alley et al., 1986), the identification of marginal fast flow sectors in the form of drumlin fields (Clark et al., 2012) and clear evidence of ice streaming on the floor of the Irish Sea (Van Landeghem et al., 2009;Van Landeghem and Chiverrell, 2020) confirmed the recognition that the Irish Sea system was indeed a former ice stream (Roberts et al., 2007).
Despite earlier interpretations that the Irish Sea Till might pre-date the last glacial stage, based purely on lithostratigraphy (Mitchell, 1972;Bowen, 1973), the gradual accumulation of geochronological evidence over the past 30 years has forced the realization that the Irish Sea Till is overwhelmingly of last glacial stage age.Indeed, the richness of this legacy of geochronological data constraining the advance, maximum extent, retreat and readvance phases of the BIIS enabled the first application of Bayesian analysis to any retreating ice stream (Chiverrell et al., 2013).These geochronological data included optically stimulated luminescence (OSL) ages on proximal glacifluvial sequences, terrestrial cosmogenic nuclide (TCN) ages on glacially transported boulders and glacially scoured bedrock surfaces, and radiocarbon ages ( 14 C) on materials from below, within and above glacigenic units.This analysis enabled the identification of outliers and reduction in the uncertainty of individual determinations.The reconstruction emerging from this analysis was of advance of the ice stream between 30 and 24.3 ka to a maximum limit in the central Celtic Sea (Scourse et al., 1990) to the southwest of the Isles of Scilly (Scourse, 1991) between 25.3 and 24.5 ka, before rapid but stepped northwards retreat with readvance phases out of the Celtic Sea into the northern Irish Sea.The timing of the maximum extent of this advance was corroborated by the independent dating of ice-rafted detritus (IRD) sourced from the Celtic Sea in a deep-sea core from the Goban Spur adjacent to the ISIS terminus in the Celtic Sea to between 24 and 25 ka (Scourse et al., 2009;Haapaniemi et al., 2010).
The demonstration that Bayesian analysis could be successfully applied to geochronological series constraining ice stream dynamics, and the quantification of IRD flux from the BIIS in the adjacent deep-sea sequences, were key motivations behind the establishment in 2012 of the BRITICE-CHRONO Consortium Project (Small et al., 2017).Even though, at the onset of the project, the ISIS was the best constrained ice stream draining the former BIIS, there were significant uncertainties concerning the dynamics of the ice stream.These included defining the maximum extent of the ISIS in the Celtic Sea sector, establishing the retreat dynamics of the ice stream as it backstepped towards St George's Channel, constructing a chronology for the multiple readvance sequences identified in southeast Ireland (Screen Hills; Thomas & Summers, 1983, 1984) and the Llŷn Peninsula in north Wales (Thomas and Chiverrell, 2007) and across the Isle of Man (Thomas et al., 2004), establishing a retreat chronology for the dynamics of the English Midlands lobe (ISG; Thomas, 1989;Chiverrell et al., 2016) and defining the nature and timing of the bifurcation between the ISIS and English Midlands ice.It was not known whether the retreating terminus within the Celtic and Irish  Finlayson et al. (2010), major bathymetric contours, and the inner, mid-and outer locations for modelled relative sea level (RSL) water depths from a glacio-isostatic adjustment model (Bradley et al., 2011).(B) The retreat patterns of the ISIS and ISG (Fig. 8) identifying the Boundary limits (BL), typically well-defined ice margin limit positions that punctuated the Prior models used in the Bayesian modelling (Fig. 6).ISIS limits are BL1 to BL22, ISG limits in the English Midlands E-BL1 to E-BL10 and in northeast Ireland NI-BL2, 4 and 5. Sample location names and the sampling track of the RRS James Cook cruise (JC106) are also shown.[Color figure can be viewed at wileyonlinelibrary.com].
seas was everywhere marine, or whether it incorporated a component of terrestrial retreat on an emergent seabed (cf.Furze et al., 2014), how the retreat sequence related to IRD flux on the adjacent margin, or how the changes in the rate of retreat related to the primary external forcings and controls such as atmospheric and marine climate, relative sea level, palaeotidal amplitudes, bed slope and topographic constrictions.To address these questions, constraining the retreat dynamics of the BIIS in the Irish Sea sector was defined as a core component of BRITICE-CHRONO, and two of the project transects, Transect 3 (T3; English Midlands lobe, ISG) and Transect 4 (T4; Celtic and Irish sea axis, ISIS) were established for detailed investigation (Fig. 1).Both transects involved the application of OSL and TCN techniques to terrestrial sequences, in marginal settings in the Isles of Scilly, east Ireland, west and north Wales (T4), and in the English Midlands lowlands, northwest England, southwest Scotland and the Isle of Man (T3).Sediment core material yielding samples for 14 C dating was acquired during investigation of the marine sequence for both transects during research cruise JC106 of the RRS James Cook during the summer of 2014 (Fig. 1B).
A series of sub-transect papers have been published reporting the detail of the new geochronological data and their stratigraphic context.These include terrestrial data from the Isles of Scilly (Smedley et al., 2017a), southeast and east Ireland (Small et al., 2018), the Llŷn Peninsula (Smedley et al., 2017b), English Midlands (Chiverrell et al., 2020), and offshore for the Celtic Sea (Lockhart et al., 2018;Scourse et al., 2019) and northern Irish Sea (Chiverrell et al., 2018).The purpose of this paper is to provide a synthesis of the legacy and new BRITICE-CHRONO data for T3 and T4 presented in these subtransect contributions, and to present a revised Bayesian analysis of these data.We present new geochronological data for three sites in west Wales, Abermawr, Banc-y-Warren and Tonfanau (Fig. 1B), and integrate these into this synthesis, which includes further independent data from southwest Wales published recently (Pantgwyn-Trefigin; Fig. 1B; Glasser et al., 2018).The patterns, timing and rates of retreat of both the ISIS (T4) and the ISG (T3) are then compared with independent data on potential drivers of retreat to highlight the major forcings responsible for deglaciation.The contrast between the largely marine ISIS and largely terrestrial ISG sectors (Fig. 1) is particularly valuable in this context as a natural experiment because marine controls early in deglaciation (relative sea-level change, palaeotidal amplitude change, ocean heat supply) only apply for the ISIS.

Isles of Scilly
The interpretation that the well-defined ice limit across the northern Isles of Scilly (Scourse, 1991;Hiemstra et al., 2006) dated to the Late Devensian, and not an earlier glacial stage (cf.Mitchell and Orme, 1967;Bowen, 1973), was first proposed by Scourse (1991).The evidence for this was on the basis of conventional 14 C dating of organic deposits lying stratigraphically beneath the Scilly Till (Scourse, 1991) and some experimental thermoluminescence (TL; Wintle, 1981) and OSL ages (Smith et al., 1990) on loessic material associated genetically with the Scilly Till.These data were later supplemented by OSL data from glacifluvial sands from north Tresco (Battery site; Scourse and Rhodes, 2006) and TCN ages on the Shipman Head Boulder Moraine and other bedrock surfaces by McCarroll et al. (2010); these data were also consistent with a glacial event during Marine Isotope Stage 2. None of these determinations were considered sufficiently robust to be included in the legacy dataset screened within BRITICE-CHRONO (Small et al., 2017) so a further dating campaign was undertaken (Smedley et al., 2017a).A series of OSL ages on glacifluvial outwash sand lenses at the Battery site (Tresco) indicated emplacement at 25.5 ± 1.5 ka and, together with TCN ages on boulders overlying till on Scilly Rock of 25.9 ± 1.6 ka, confirmed that ISIS impinged on the northern Isles of Scilly, but did not override the archipelago, at ~26 ka (Smedley et al., 2017a).Scilly therefore constitutes a tightly dated icemarginal position for the maximum extent of the eastern flank of ISIS within the Celtic Sea (Fig. 1).

Celtic Sea
Undated sporadic patches of till and glacimarine sediments found by the British Geological Survey to the west and southwest of Scilly in the Celtic Sea, constituting the Melville Till and Melville laminated Clay, respectively (Pantin and Evans, 1984;Scourse et al., 1990) have long been considered a correlative of the Scilly Till (Scourse, 1991;Scourse and Furze, 2001).These remained undated until new cores of glacimarine sediment were recovered from close to the shelf break, one containing an Arctic mollusc yielding a 14 C age of 24 265 ± 195 cal a BP (Praeg et al., 2015).BRITICE-CHRONO further recovered a further series of nine cores from the Celtic Sea containing glacimarine sequences and, in one case, till from close to the shelf break (JC106-003VC; Fig. 1; Scourse et al., 2019).The glacimarine facies have yielded a suite of five 14 C ages, ranging from 24 280 ± 230 to 26 991 ± 350 cal a BP (ΔR = 0).It is not possible at present to identify whether this spread of ages incorporates multiple readvance phases or only a single advance-retreat phase, but the mean of the six ages available is 25.2 cal ka BP.This is within the error of the combined OSL and TCN ages from Scilly and indicates the extension of the ISIS as far as the continental shelf break some time prior to ~25.2 cal ka BP and initiation of deglaciation at that time.

Celtic Deep
Cores taken through sequences within the Celtic Sea have typically recovered thick Holocene sandy mud to mud sequences overlying a coarse, shelf gravel lag above an erosional unconformity cut through glacigenic mud sequences (Austin and Scourse, 1997;Scourse et al., 2002;Furze et al., 2014).BRITICE-CHRONO recovered two piston cores (JC106-051/052PC) supported by acoustic data from the Celtic Deep (Fig. 1; Lockhart et al., 2018;Scourse et al., 2019); these data indicate very similar sequences to those described previously.Furze et al. (2014) attribute the basal glacigenic muds to probable glacilacustrine rather than glacimarine conditions on the basis of their contained microflora and fauna, with the implication that deglaciation across the northern Celtic Sea might have been terrestrial rather than marine, a conclusion consistent with Tóth et al. (2020).

South and East Ireland
Before the BRITICE-CHRONO project, exposures of Irish Sea Till in coastal sections in southern Ireland had yielded 14 C and  OSL data (Ó Cofaigh and Evans, 2001a,b, 2007; Ó Cofaigh  et al., 2012) indicating the flow of the ISIS after 26 cal ka BP westwards into the Celtic Sea and then onshore on exiting the constriction of St George's Channel, lending support to the notion of a piedmont lobe across the Celtic shelf.Ó Cofaigh et al. (2012) collected OSL ages for glacilacustrine and glacifluvial sediments overlying Irish Sea Till at Whiting Bay and Ballycroneen Strand constraining Irish Sea ice withdrawal from the coastline of southern Ireland to ~24 ka, and thus affirming a short-lived and rapid advance of the ISIS to maximum limits in the Celtic Sea.Further north and east, sedimentological and structural interpretation of the complex but undated glacitectonized sequences of the Screen Hills in southeast Ireland indicated deposition and thrusting during a series of readvance phases of the right lateral terrestrial margin of the marine-terminating ISIS within the ISB (Thomas andSummers, 1983, 1984;Thomas and Chiverrell, 2011).The structural framework of the Screen Hills complex provided a chronological template of readvance phases tested by a programme of OSL and TCN dating by BRITICE-CHRONO, in addition to the application of these techniques at other sites in southeastern and eastern Ireland.
A total of 13 OSL and 10 TCN determinations have been generated by BRITICE-CHRONO (Small et al., 2018) that document the initially rapid retreat of the ISIS out of the Celtic Sea.Initial rates may have been as high as 300-600 m a −1 .It has been inferred that such rapid retreat here might be related to grounding line retreat across the nearby reverse slope of the Celtic Deep.The deepest part of the Celtic Deep is between −140 and −150 m and the adjacent Celtic Deep Southern Sill less than −80 m (Furze et al., 2014).The data support the interpretation for a deceleration in retreat rate (26 m a −1 ) followed by stabilization (3 m a −1 ), indicated by the series of readvance phases between 23.3 and 22.9 ka in the Screen Hills sequence that are located in the topographic constriction of St George's Channel.Subsequent retreat northwards of this constriction after 22.9 ka along the normal slope of the ISB was then faster (~125-150 m a −1 ) (Smedley et al., 2017a;Small et al., 2018).

Wales
Luminescence ages from deglacial sequences in Cardigan, southwest Wales, indicate the ISIS retreat potentially as early as ~26.7 ka (Glasser et al., 2018).The Llŷn Peninsula in northwest Wales contains numerous ice-marginal morphosedimentary indicators of stepped retreat and readvance phases, notably morainal banks with thrust, stacked and tectonized sequences delineating inter-morainal basins (Thomas and Chiverrell, 2007).The Thomas and Chiverrell interpretation (2007) of this previously largely undated retreat sequence provided a morpho-stratigraphic template for tightly targeted OSL sampling of proglacial sequences and reevaluation of existing TCN data within the BRITICE-CHRONO project (Smedley et al., 2017b).This sector of the ISIS is characterized by topographic constriction, a reverse bed slope and changing widths of the former calving margin.The new data from Abermawr, Banc-y-Warren and Tonfanau have been integrated into a regional Bayesian model that indicates centennial-scale oscillatory behaviour of the ice front between 24 and 20 ka BP across a relatively short axial distance (~120 km) consistent with a slowing in the overall rate of ISIS retreat (Smedley et al., 2017b).

English Midlands
Luminescence measurements on cobbles and sediments from glacifluvial sequences deposited by the terrestrial-terminating ISG, which extended into the English Midlands, indicate initial substantial advance reaching south Lancashire >30 ka and a maximum ice limit established ~26.5 ka in the vicinity of Bridgwalton and Seisdon (Fig. 1B; Chiverrell et al., 2021).Extensive geomorphological mapping (Thomas, 1985(Thomas, , 1989;;Thomas et al., 2004;Thomas and Chiverrell, 2007) has demonstrated that this advance was sourced by ice from the Lake District, the northern Irish Sea and north Wales.Retreat ice margins facing towards a series of reverse bedrock slopes generated numerous ephemeral proglacial lakes that strongly conditioned the style of deglaciation which commenced at ~25.3 ka in the south; the ice margins pulled-back from the Cheshire lowlands into the eastern Irish Sea and Lancashire after 22.1 ± 1.1 ka.

Northern Irish Sea
A synthesis of offshore geological and geophysical data combined with TCN and OSL dating of glacigenic landforms and sediments in the northern Irish Sea, north Wales, Isle of Man and western Cumbria provides constraints on the timing and retreat pattern of the ISIS northwards once the discrete west and east components of the ISIS and ISG had deglaciated (Chiverrell et al., 2018).This synthesis incorporated previously published 14 C ages on marine microfauna (McCabe et al., 1998;McCabe et al., 2007;McCabe, 2008;Ballantyne and Ó Cofaigh, 2017).As the retreating ice front escaped the topographic constriction of the central Irish Sea -Llŷn Peninsula, ice retreat accelerated at ~20 ka in response to this wider calving margin but slowed on reaching the Isle of Man as the ISIS stabilized and oscillated with a series of readvance phases.These generated the glacitectonized thrust stratigraphy characteristic of the Isle of Man sequence (Thomas and Summers, 1984;Thomas et al., 2004) including components produced during the Scottish Readvance of ice margins.The new ages demonstrate that the Scottish Readvance occurred between 19.2 and 18.2 ka, earlier than previously suggested and not coincident with Heinrich Event 1 (Chiverrell et al., 2018).The ISIS then retreated northwards towards the Southern Uplands of Scotland, with the Bayesian analysis showing deglaciation of the southern Solway Lowlands at ~15.5 ± 0.6 ka (Chiverrell et al., 2018).

Materials and methods
The materials and methods used in the contributory investigations are included in detail in the individual papers reporting the findings, as listed above.New OSL ages developed for three sites in west Wales are presented here.

OSL geochronology
OSL dating determines the time elapsed since a mineral grain (typically quartz or K-feldspar) was last exposed to sunlight before burial (see Rhodes, 2011;Smedley, 2018).Because of exposure of the mineral grains to naturally occurring ionizing radiation, charge accumulates at defects within the crystal structure and builds up through time.The charge can be released in the laboratory by stimulating with light at a particular wavelength, and the energy liberated from this process produces light which can be detected at a different wavelength.This emission of light is called luminescence, and is the OSL signal measured from the sample.The OSL signal can be used to calculate the equivalent dose (D e ), a measure of the total ionizing radiation received by the sample during its period of burial.To determine an OSL age, D e is divided by the environmental dose-rate (a measure of how much radiation the sample experiences per year) which combines the contributions from alpha, beta and gamma radiation, along with the cosmic dose-rate.
The new OSL ages reported in this paper were determined using measurements of the OSL signal from single grains of quartz.To calculate the environmental dose rate to the quartz grains, U, Th, K and Rb concentrations were measured on milled and homogenized bulk sediment samples using inductively coupled plasma mass spectrometry (ICP-MS) and atomic emission spectroscopy (ICP-AES) (Table 1).The beta dose-rates were calculated from the concentration of these radionuclides in each sample using the conversion factors of Guérin et al. (2011) and corrected for grain size and HF etching using the beta dose-rate attenuation factors of Guérin et al. (2012).Gamma dose-rates were determined in the field using in situ gamma spectrometry with a 2-inch-diameter NaI:Tl scintillator.Water contents were estimated considering the field and saturated water contents, and the environmental history for each sample.The dose-rate due to cosmic rays was calculated after Prescott and Hutton (1994).All dose rate calculations were undertaken using the Dose Rate and Age Calculator (DRAC; Durcan et al., 2015).
Grains of quartz used for OSL measurements to determine equivalent doses (D e ) were isolated by treating each sample with a 10% (v/v) dilution of 37% HCl and with 20% (v/v) of H 2 O 2 to remove carbonates and organics, respectively.Dry sieving then isolated the 212-250 µm diameter grains for all samples.Density separation using sodium polytungstate then provided the 2.62-2.70− g cm −3 (quartz-dominated) fractions, which were etched for 1 h in 40% hydrofluoric acid (HF) to remove the outer portion of the quartz grains that was affected by alpha irradiation and any contaminating feldspar.Grains were then washed in a 10% solution of HCl to remove any fluorides that may have been produced during the HF etch.Finally, grains of quartz were mounted on a 9.8--mm-diameter aluminium single-grain disc for analysis, which contained a 10 by 10 grid of 300-µm-diameter holes.
All luminescence measurements were performed 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).Optical stimulation was performed for 1 s using a green laser at 125 °C and the OSL signal was detected through a 2.5-mm-thick U-340 filter and convex quartz lens placed in front of the photomultiplier tube.The first 0.1 s and final 0.2 s of stimulation were summed to calculate the initial and background OSL signals, respectively.A preheat plateau test performed on sample T4BANC02 was used to determine the preheat temperature (200 °C for 10 s) applied throughout the single aliquot regenerative dose (SAR) protocol (Murray and Wintle, 2000).The grains were accepted after applying the following screening criteria and accounting for the associated uncertainties: (i) whether the test dose response was greater than three sigma above the background, (ii) whether the test dose uncertainty was <20%, (iii) whether the recycling and OSL-IR depletion ratios were within the range of ratios 0.8-1.2,(iv) whether recuperation was <5% of the response from the largest regenerative dose and (v) whether the singlegrain D e values were not from a population of very low doses that were identified by the finite mixture model (FMM) to be inconsistent with the geological context of the sample (i.e.<1 ka).To assess the performance of the SAR protocol, dose-recovery experiments were performed on all the samples with the OSL signals bleached using blue light emitting diodes (LEDs).The samples gave dose recovery ratios within 10% of unity, showing that the SAR protocol used for analysis could accurately recover a known laboratory radiation dose and was therefore appropriate for these samples.

Bayesian methodology
Each of the geochronological methods employed to constrain ice retreat generates an age probability distribution, with each technique affected by method-specific limitations in their application (discussed above).Bayesian age modelling (Buck et al., 1996;Bronk Ramsey, 2008) describes a group of approaches applied widely to integrated dating control for sediment sequences.The process refines the probability distributions for individual ages and relies on a series of ages that are presented as an order of events discerned independently of the chronology.The approach is routinely applied to sets of age measurements related by stratigraphy and governed by the principle of superposition, for example radiocarbon ages from lake or peat sequences (Bronk Ramsey, 2008).Bayesian modelling has been applied increasingly to spatially distributed geomorphological datasets where there is a single actor responsible for the relative order of events, for example the progression of fluvial terraces (Chiverrell et al., 2009) and the retreat of glacial margins (e.g.Chiverrell et al., 2013Chiverrell et al., , 2018Chiverrell et al., , 2020;;Bradwell et al., 2019).The deglaciation sequence for the ISIS and ISG as evidenced in the onshore and offshore geomorphology provides this hypothetical 'relative-order' of events for Bayesian modelling of the geochronological determinations (e.g.Chiverrell et al., 2013).In the terminology of Bayesian chronological modelling this is the Prior model (e.g.Chiverrell et al., 2013).
The Prior model for the ISIS extends with ice marginal retreat from maximum limits near the continental shelf break in the southern Celtic Sea through well-defined ice margin configurations identified on the sea floor and in the coastal zone of adjacent land masses through the ISB across the Isle of Man and stepping-back on to land in southwest Scotland (Ballantyne et al., 2013;Praeg et al., 2015;Chiverrell et al., 2016;Smedley et al., 2017a,b;Chiverrell et al., 2018Chiverrell et al., , 2020;;Jenkins et al., 2018;Small et al., 2018;Wilson et al., 2018;Ballantyne and Small, 2019;Scourse et al., 2019).Two further published Bayesian models are included here; the Prior models for these were developed for (i) the largely terrestrial retreat sequence of the ISG from maximum limits in the English Midlands to the Lancashire coast (Chiverrell et al., 2020), and (ii) the step-back of the ISIS from the western ISB into the northeast of Ireland (Chiverrell et al., 2018).Together, these three Prior models describe the retreat dynamics for this sector Table 1.Environmental dose-rates to grains of quartz, determined using ICP-MS and ICP-AES analysis and field gamma spectrometry.The chemical concentrations are presented to the appropriate decimal places according to the associated detection limit.The grain size for all samples was 212-250 µm.The water contents are expressed as a percentage of the mass of dry sediment.Dose-rates were calculated using the Dose Rate and Age Calculator (DRAC; Durcan et al., 2015).Total doserate (Gy ka −1 )

Sample
of the former BIIS with the exception of the retreat of ice into the mountainous interior of Wales which operated as a separate ice cap (McCarroll and Ballantyne, 2000;Patton et al., 2013a,b,c;Hughes et al., 2016).Each Prior model was developed independently of the age information and included all geochronological samples according to procedures set out in Bronk Ramsey (2008Ramsey ( , 2009a,b) ,b) and Bronk Ramsey and Lee (2013).Markov chain Monte Carlo (MCMC) sampling in the Bayesian analysis uses the Prior models and the likelihood (measured ages) probabilities for all samples to build distributions of possible solutions, thereby generating modelled probabilities termed posterior density estimates (1σ age ranges).The Bayesian models were coded in OxCal 4.3 (Bronk Ramsey and Lee, 2013) as Sequence models, and the Prior models were structured as a series of Phases that contain dating information grouped for retreat stage zones, each including sites that share relationships with the adjacent zones.Phases, unordered group of events/parameters in the OxCal nomenclature, are delimited by the Boundary command located in the code typically at well-defined ice margin limit positions and are termed Boundary Limits (BLs).The Boundary command in Sequence models generates modelled age probability distributions that here reflect the ages of those BLs, for the ISIS labelled BL1 to BL20, for the ISG in the English Midlands labelled E-BL1 to E-BL10 and for the step-back of ice into northeast Ireland labelled NI-BL1 to NI-BL7 (Fig. 1A).In summary, the Bayesian modelling uses uniform Phase Sequence models punctuated by Boundaries, and run to assess outliers in time using a scaling of 10 0 to 10 4 years using a Student's t-distribution to describe the outliers (Bronk Ramsey, 2009b).
Numerous iterations of the modelling were undertaken varying the outlier probabilities for individual age determinations to achieve overall model agreement indices exceeding the >60% threshold advocated by Bronk Ramsey (2009a).Outliers were given ultimately a probability scaling based on their fit within the model, thereby allocating probabilities of p < 0.2, p < 0.5, p < 0.75 and p = 1 (100%) on a scale of increasing outlier severity.Individual ages were assessed based on their fit both within an individual Phase and in the progression between Phases of the Sequence model.Bottlenecks in the modelling were handled by increasing iteratively the outlier probability for all ages in selected Phases until the model produced overall agreement, at which point agreement indices are produced for all ages individually.The statistical outliers identified by that process were not arbitrarily excluded but scrutinized for reasons explaining outlier behaviour either in the Prior model (e.g. the sample context) or in the measurement data (e.g.nuclide inheritance).Once outliers (p = 1; 100%) had been identified, the outlier probabilities were decreased for subsequent iterations, with this process-cycling continuing until the overall model agreement was >60%.Samples handled as outliers for the ISIS are detailed in later sections.Outlier handing for the English Midlands and northeast Ireland Sequence models are described in Chiverrell et al. (2018Chiverrell et al. ( , 2021)).

Site descriptions and stratigraphy
Abermawr [51.969813, −5.0831921] Abermawr is located on the north coast of Pembrokeshire (Fig. 1B), where Irish Sea ice has impinged on a north-draining catchment 4-5 km west of Fishguard.The sequence at Abermawr (McCarroll and Rijsdijk, 2003) shows Irish Sea glacial sediments overlying locally derived breccia and gravels.The glacigenic sequence is deformed and comprises Irish Sea till overlain by offlapping outwash sands and gravels.The offlapping upper flows represent melt-out, flow tills and paraglacial slope wash redistributing the glacial sediments.The Abermawr coastal exposures in July 2013 showed Irish Sea diamicton, incorporating thin channel fill outwash sands, which were sampled for T4ABMW01 (Fig. 2A).The OSL sample was taken from a ~0.1-0.15-m-thick unit composed of horizontally laminated fine to medium sand that appeared to form an ice proximal outwash channel fill (Fig. 2B).
Banc-y-warren [52.106176, −4.624386] Banc-y-Warren (Fig. 1B) is an important site in the Quaternary geology of SW Wales and a listed Geological Conservation Review locality.Geomorphologically the site is one of a series of steep-sided conical low hills that rise ~50 m above the northern flanks of the lower Teifi valley.Various models have been proposed to explain the sedimentology through these mounds exposed for many years in extensive sand and gravel workings.These models vary from end-moraines (Charlesworth, 1929), to coalescing ice front deltas feeding into proglacial Lake Teifi with bottom-set sands and crossed-bedded gravels a foreset component (Jones, 1965;Helm and Roberts, 1975).Alternatively, the deposits have been interpreted as subaerial rather than subaqueous outwash spread on top of a down-wasting Irish Sea glacier (Allen, 1982).Hambrey et al. (2001) attribute the mounds of sand and gravel to recession of the Late Devensian Irish Sea ice margin with localized ice-bounded bodies of glaciofluvial sediment that accumulated on relatively high areas becoming ice-free on flanks of the lower Teifi valley and between remaining lobes of ice.In July 2013, the quarry at Banc-y-warren exposed the deposits of an ice-contact sandur system or kame terrace (Hambrey et al., 2001); these yielded samples T4BANC01 and T4BANC02 (Fig. 3).OSL sample T4BANC01 was taken from high in the sequence in a unit of rippled fine and medium sand planar cross-sets reflecting relatively shallow back-bar environments, but adjacent to larger (metre-scale) trough cross-sets that probably indicate deeper water deltaic or channel systems (Fig. 3A).OSL sample T4BANC02 was taken from lower in the sequence from a unit of rippled fine to medium sand that also forms a larger (metre-scale) trough cross-sets probably deposited in deep channel systems (Fig. 3B).
Tonfanau [52.619176, −4.1253003]The coastal cliffs at Tonfanau (central west Wales; Fig. 1B) comprise 2-km-long exposures of glacigenic sediments on the eastern margin of the ISB.Patton and Hambrey (2009) described a sequence recording the dynamics and deposition of sediments by both Welsh and Irish Sea ice, which included six lithofacies: diamicton, sandy gravel, well-rounded gravel, well-sorted sand, muddy sand and laminites.The lowermost units comprise a grey clast-poor muddy diamicton with sand and gravel in close association recording Welsh ice extending into Cardigan Bay.Higher in the sequence, sediments with a northern provenance bury the basal units, as a thick grounding-line fan complex developed fed by the ISIS flowing south and interacting with an ice-contact lake (Fig. 4A).Proximity of the ice margin is reflected in evidence for later overriding by the ISIS.These sediments are capped by a few metres of outwash (OLF-I to OLF-IV; Fig. 4B).OSL sample T4TONF01 was taken from high in the sequence (OLF-IV) in a unit of fine and medium sand in the upper sections of the grounding line fan complex (Fig. 4B,C).The sampled unit is part of a thicker succession of sand (OLF-III to OLF-IV) interpreted as subaerial outwash plain deposits by Patton and Hambrey (2009).

OSL data from west Wales sites
The OSL dating results for the four samples described above are presented in Tables 1 and 2, and each of the D e distributions used to calculate ages are presented in Fig. 5.The D e distributions determined for samples T4ABMW01 (Fig. 5a), T4BANC01 (Fig. 5b) and T4BANC02 (Fig. 5c) were asymmetrical, which is typical of sediments that were heterogeneously bleached before burial as is commonly observed for glacial sediments (e.g.Duller, 2008;Smedley et al., 2017b).The minimum age model (MAM; Galbraith and Laslett, 1993;Galbraith et al., 1999) is used to identify the well-bleached part of a heterogeneously bleached D e distribution to determine an age, and it was therefore used to determine ages for these samples.A symmetrical D e distribution was determined for sample T4TONF01 (Fig. 5d) and suggests that these grains were homogeneously bleached before burial; thus the central age model (CAM; Galbraith et al., 1999) was used to determine this age and broadly calculates the weighted mean of a D e distribution, accounting for the individual uncertainties of each datapoint.This difference in the scatter of the D e data from these samples can be expressed numerically by calculating the overdispersion of each set of D e values.Overdispersion is a measure of the scatter in the data beyond what would be expected from the uncertainties on individual D e values, and much higher values are seen for the incompletely bleached samples (with values of 54-78%), than for the bleached sample with an overdispersion of 37% (Table 2).
To apply the MAM it is necessary to define the overdispersion that would be expected if the sample had been completely bleached at deposition.One approach to defining this overdispersion is to measure a well-bleached correlative of the sediments in question, but this is not possible for these sites.Instead, the approach used previously by Smedley et al. (2017b) was used.In this approach the overdispersion determined from the single-grain D e dose recovery experiments, which quantified the amount of scatter caused by intrinsic sources of uncertainty, was added in quadrature to extrinsic overdispersion arising from external microdosimetry (estimated at 20%) to determine the value of σ b used in application of the MAM.These σ b values varied from 0.20 to 0.35 for the three samples where the MAM was used (Table 2).
The distribution of D e values for the two samples from Bancy-Warren are particularly scattered, and a large proportion of the grains do not appear to have been bleached at deposition, perhaps a result of short transport subaerially and deposition in deep water (as suggested by the sedimentology) where light exposure would be expected to be limited.However, in both cases it was possible to identify a population of grains for age determination (Fig. 5b,c, and ages in Table 2).
These four new OSL ages presented here constrain potentially three locations on the eastern margin of ISIS at a critical juncture during the passage northwards of ice margins between Ireland and Wales.They were selected to provide Table 2. Results from OSL dating, including the total number of grains analysed and the number of grains (n) that passed the screening criteria.The overdispersion (OD) from intrinsic sources of uncertainty was determined from dose-recovery (DR) experiments and combined in quadrature with OD from extrinsic sources (20%).The overdispersion values determined from intrinsic and extrinsic sources were rounded to the nearest 0.05 for consistency and used to determine the respective σ b values for the three-parameter minimum age model (MAM).Small et al., 2018).Located broadly midway between Pembroke and the Llŷn Peninsula, the grounding-line fan complex at Tonfanau developed in an ice-contact lake between Welsh and Irish Sea ice (Patton and Hambrey, 2009).That interplay of ice masses, coupled with evidence for later overriding by Irish Sea ice, suggest a setting ideal for recording retreat of ISIS.Notwithstanding the homogeneous bleaching, the age of the 17.2 ± 1.6 ka (T4TONF01) is young compared with the wellconstrained ice margin retreat chronology across the Llŷn Peninsula (Smedley et al., 2017b).Constrained by the limited availability of suitable sand units, the sampling targeted sand units in the uppermost sandur deposit above the graveldominated proglacial fan wedges.It is possible that the unit sampled reflects either a late-stage subaerial sandur fed by Welsh ice or fluvial remobilization of the deposits under the periglacial conditions that Patton and Hambrey (2009) attribute to the uppermost muddy sand units at the location.In either case, the OSL age provides a minimum age on the passage northwards of Irish Sea ice margins.

Bayesian analysis modelling of ice retreat
Bayesian age modelling (Bronk Ramsey, 2009a) of the dating control for the ISIS provides a basis for calculating the timing for the advance and then retreat northeast and north through the Celtic and the ISB to the mountains of SW Scotland.The length and complexity of the prior model for the ISIS, coupled with an inherent scatter of using many age determinations, rendered obtaining a conformable agreement challenging.The model has therefore been broken into two sub-models, describing the south and north sectors respectively and including a zone of overlap where the Phases are included in both models (Fig. 6A; Table 3).The models have overall agreement indices of 120% for the south and 98% for the north, both exceeding the >60% threshold advocated by Bronk Ramsey (2009a).The iterative approach to assessing outliers, as described previously, resulted in a total of 28 out of the 93 age determinations being identified as absolute outliers (Table 3), with the remaining 65 receiving a variable outlier scaling of p < 0.2, p < 0.5 and p < 0.75 based on increasing outlier severity as prescribed by the model.Out of the 28 total outliers, six TCN ages on the Isle of Man and in western Cumbria are unconformable as they constrain ice sheet lowering (thinning) rather than ice marginal retreat (Chiverrell et al., 2018).Nine further TCN ages were treated as outliers (p = 1); seven of these ages are unconformable on the basis of being 'too old'.TCN ages that are older than the true age of deglaciation are interpreted as reflecting nuclide inheritance.In some cases (e.g.TRE01, T4SCI02) this is obvious as the apparent exposure ages pre-date build-up of the BIIS (e.g. Brown et al., 2007).Other samples (e.g.T4CSP02) have apparent ages that are only slightly older than the modelled deglaciation age but are still unconformable within the broader context of the prior model.Similarly, two 'too young' ages probably reflect post-depositional modification of boulders; T4CSP01, for example, is incorporated into a field boundary and has thus potentially been moved.Five 14 C ages are older than other members in a Phase of ages indicating marine fauna reworked into younger sediments, whereas the younger members of the Phase provide better constraint on the timing of deglaciation.Eight OSL ages were treated as outliers (p = 1), with two reflecting the sampling of units younger than deglaciation (e.g. at Battery, Isles of Scilly; Smedley et al., 2017a) and six where other samples suggest poor resetting of the OSL signal.Throughout, all outlier samples were identified as outliers in the iterations of the Bayesian modelling and were not excluded subjectively.The Boundary ages constraining the retreat of ice margins were derived from the modelling (Fig. 6), with the zone of overlap using an average of the south and north modelled ages and uncertainties.In subsequent sections, italics denote the posterior density estimates or modelled ages derived from the Bayesian modelling to distinguish them from the unmodelled ages obtained directly for individual samples.Details on the Bayesian age modelling for the ISG and stepback of ice from the western ISB into the northeast of Ireland are provided in Chiverrell et al. (2020) and Chiverrell et al. (2018) respectively.
Basal constraint on ice advance in the Irish-Celtic Sea sector is provided by dated evidence for ice-free conditions in several locations.Four Ashes in the English Midlands (Staffordshire) yielded 14 C ages of around 34.6 ± 0.8 cal ka BP (Morgan, 1973) with additional OSL ages of around 27.8 ± 2.6 ka reported from further north at Dowkabottom (Lancashire; Telfer et al., 2009).The Four Ashes ages are similar to equivalent dating of ice-free conditions at around 34.4 ± 1.8 cal ka BP based on 14 C ages from Sourlie on the Ayrshire coast in the feeder zone to ice in the North Channel (Finlayson et al., 2010(Finlayson et al., , 2014)).Further east in central Scotland equivalent organic-rich sediments spanning 39.8-32.8ka BP have been identified at Balglass Burn (north of Glasgow; Brown et al., 2007); these sites are in the central sector of the former BIIS and provide important constraint on ice sheet build-up.Further constraint on the build-up of ice in the Irish Sea sector is provided by a cluster of OSL ages around 29.9 ± 1.2 ka from Lydiate Lane Quarry, which constrains the expansion of ice in the north ISB and extension into south Lancashire (Chiverrell et al., 2021).The Four Ashes and Lydiate Lane ages have been incorporated into the Bayesian model for the Irish Sea Glacier (Fig. 6C), but they apply in a wider Irish Sea context recording a substantial build up and expansion of ice into the central Irish Sea by 28.3 ± 1.4 ka.The retreat stages indicated by the Bayesian modelling are shown in Fig. 7.

Celtic Sea and Celtic Deep
The maximum extension of the ISIS to the shelf break in the south Celtic Sea is well dated by TCN and OSL from Scilly and 14 C-dated fauna from glacimarine facies overlying subglacial diamict constraining the initial pull-back in the south Celtic Sea, all overlapping within uncertainties.These ages constrain the timing for the maximum extension of the ISIS (Fig. 7) near the continental shelf break to 25.6 ± 0.5 ka (BL1) and the initial pullback of ice margins to 25.3 ± 0.2 ka (BL2).Although probably integrating across pulses of ice margin advance, still-stand and retreat, the expansion phase amounts to a net axial advance of the ice stream of ~150 m a −1 .The Bayesian modelling has calculated modelled age probability distributions for the ISIS limits that show substantial changes in the pace of retreat, vacating the Celtic Sea rapidly between 25.3 ± 0.2 ka (BL2) and 24.7 ± 0.2 ka (BL4) at rates between 230 and 650 m a −1 .The rapid advance of the ISIS and high discharge of icebergs associated during the rapid retreat probably required high velocities in the ice stream, which is consistent with sea floor geomorphological indicators of fast flow in the central ISB, where a field of megascale glacial lineations are aligned NNE-SSW and axial parallel to ice flow towards the Celtic Sea evidence ice streaming (Van Landeghem and Chiverrell, 2020).
The glacimarine sequences of the Celtic Sea are in many cases, though not all, highly deformed and contorted, and are characterized by high values of undrained shear strength as measured by shear vane (Lockhart et al., 2018;Scourse et al., 2019).The geotechnical properties of these sequences are difficult to explain without subsequent overriding and glacitectonism as a result of ice readvance (Lockhart, 2019) and their distribution identifies the specific sector of the Celtic  (Chiverrell et al., 2021) and retreat of ice margins into the north of Ireland (Chiverrell et al., 2018).All boundary ages are expressed as ±1σ.
Ages marked with an asterisk are identified as outliers that did not influence the modelled outputs.Model structures show the named Phases in the Bayesian age models and groups of dating information for the three models.Boundary ages are spaced vertically to highlight correlation of timing between sectors.

Irish Sea Ice Stream
Irish Sea Glacier   Sea interpreted to have been subject to readvance (Fig. 1).Whether this was a major readvance, or merely slight adjustments to the grounded margin during overall retreat, is uncertain.It is notable that there is a peak of Celtic Sea sourced IRD in the Goban Spur at 26 ka before a larger peak coincident with Heinrich event 2, and two smaller peaks afterwards (Haapaniemi et al., 2010;Fig. 8).It is not yet clear whether the spread of 14 C ages, or the interpretation of readvance, relate to these temporally separated IRD peaks.Lockhart et al. (2018) have generated a seismic-stratigraphic synthesis of the sub-bottom acoustic profile and core data collected during the James Cook (JC106) cruise.This indicates that the glacigenic facies of the Celtic Sea form part of the upper Little Sole Formation (cf.Pantin and Evans, 1984) and that the northwestern series of linear sediment megaridges of the Celtic Sea represent tidal bedforms anchored on an eroded subglacial landsystem.A clear distinction in morphology is apparent between the 'western' ridges and those in the southeast which Lockhart et al. (2018) attribute to the influence of this subglacial basement on subsequent ridge evolution.The implication of this is that the western ridges are coincident with the limit of the ISIS advance, whereas the eastern ridges lie outside this ice advance.This interpretation is supported by a clear morphological distinction between the ridges in the two sectors, with the western ridges being larger/longer and orientated NE-SW contrasting with smaller/shorter eastern ridges orientated NNE-SSW.This distribution is consistent with the boundary between glacitectonized and undeformed glacimarine sequences (Fig. 1).
There are no marine geological or geophysical data available currently to resolve the ISIS eastern flank limit or how the ISIS advance relates to the earlier ice advance postulated to have overridden Lundy in the outer Bristol Channel (Fig. 1; cf.Rolfe et al., 2012) although the nature of any direct evidence of glaciation and associated timings in the outer Bristol Channel remain equivocal (Carr et al., 2017;Gibbard et al., 2017;Hiemstra et al., 2019).However, the distribution of glacigenic sediments and landforms on Scilly (Scourse, 1991;Hiemstra et al., 2006;Smedley et al., 2017a) is difficult to explain without a major portion of the ISIS east flank overriding at least the western part of the Lundy Platform (Lockhart, 2019).
The youngest 14 C ages derived from shells reworked into Irish Sea Till along the south coast of Ireland (Ó Cofaigh andEvans, 2001a,b, 2007) post-date the shelf break maximum of 25.6 ± 0.5 ka (BL1).In the Bayesian model they are included as a retreat stage cluster of ages, suggesting that they represent retreat-stage diamicts deposited during oscillatory retreat of the ISIS margins onshore across south and southeast Ireland.Furze et al.'s (2014) interpretation of possible glacilacustrine deglaciation in the Celtic Deep supports early glacial isostatic adjustment (GIA) simulations of a landbridge between Britain and Ireland during deglaciation (Lambeck, 1995).Tóth et al. (2020) imply from geomorphological and seismostratigraphic evidence onshore and offshore the Irish southeast coast that a regional palaeolake developed following ice recession of the ISIS and they reconstruct a southerly advance of Irish Ice Sheet ice going at least 15 km offshore onto the shelf across County Waterford.No micropalaeontology has yet been undertaken on the JC106 piston cores so it has not been possible to further test whether deglaciation was marine or lacustrine.However, the glacigenic mud sequences at the base of the two cores are undeformed, with high water content, and are very soft, indicating that they have not been subsequently overridden by ice.This suggests that the terminus just to the south of St George's Channel retreated without the readvance phases characteristic of the Celtic Sea; this supports the interpretation of faster retreat across the reverse slope of the Celtic Deep (Small et al., 2018).

West Wales and the central Irish Sea
The geochronology collected for southwest Wales (Pembrokeshire) and from the southern coast of Ireland together constrain the pullback of ice to the pinning point between southeast Ireland and southwest Wales after the advance to maximum limits (Fig. 7).OSL ages obtained for proglacial sediments on the Cork and Waterford coasts of southern Ireland have constrained the pull-back of ice margins from the coast to around 24.4 ± 1.8, 24.2 ± 2.3 (Whiting Bay) and 23.8 ± 2.1 ka (Ballycroneen Strand) (Ó Cofaigh et al., 2012).Glasser et al. (2018) presented pairs of OSL ages from outwash deposits in Pembrokeshire (southwest Wales) exposed in the Pantgwyn and Trefigin Quarries.Those single grain OSL measurements targeted poorly sorted gravel and coarse to fine sand attributed to deposition by southerly flowing braided rivers at 29.1 ± 4.2, 26.2 ± 2.0 (both Pantgwyn), 28.1 ± 1.9 and 22.8 ± 3.2 ka (Trefigin).From this and considering that the ages overlapped with admittedly wide uncertainties, Glasser et al. (2018) calculated weighted mean ages both centred on ~26.7 ka for sediments attributed to a post-maximum ice extent stage of the ISIS retreat sequence.This is problematic, as they pre-date the timing for maximum positions obtained for the south Celtic Sea (Praeg et al., 2015;Smedley et al., 2017a;Scourse et al., 2019).
As part of BRITICE-CHRONO further OSL ages were determined from Kilmore Quay (25.7 ± 4.7 and 26.2 ± 2.2 ka) and TCN ages from Carnsore Point (27.8 ± 1.5 and 24.0 ± 1.3 ka), both in southeast Ireland (Small et al., 2018).Here, the OSL age determined for sample T4ABMW01 at Abermawr (Fig. 5a; 25.2 ± 5.0 ka) is consistent with the chronology from Pembroke and southern Ireland.The OSL ages from Banc-y-Warren, T4BANC01 (Fig. 5b; 22.1 ± 3.6 ka) and T4BANC02 (Fig. 5c; 39.6 ± 5.7 ka), are in stratigraphical order within the sedimentary sequence.It is possible that the large difference between the two ages is due to the minimum dose population being characterized better for sample T4BANC01 (Fig. 5b) than for sample T4BANC02 (Fig. 5c).In the Bayesian model (Fig. 6) all the ages mentioned have been incorporated as a single Phase containing the Waterford-Kilmore-Carnsore-Pembroke retreat stage geochronology.This model structure produces a good fit, achieved by handling the older ages from Banc-y-Warren, Pantgwyn, Trefigin (all OSL) and Carnsore Point (TCN) as outliers.The boundary ages produced by the Bayesian model suggest that ice had retreated from this north Pembrokesouth Wexford -Waterford alignment between 24.3 ± 0.2 ka (BL5) and 24.0 ± 0.4 ka (BL6; Fig. 7).
The timing of ice margin retreat north from St George's Channel, interpreted based on the Bayesian analysis of the BRITICE-CHRONO data, suggests significant deglaciation by ~23.3 ka.The OSL age determined for sample T4TONF01 (17.2 ± 1.6 ka) is younger than the adjacent OSL chronology from the Llŷn Peninsula (Smedley et al., 2017b), and even the southern portion of the Isle of Man (Chiverrell et al., 2018).Given the stratigraphical position of the OSL sample at Tonfanau, this age may instead date late-stage reworking of the deposit either as a subaerial sandur sourced from Welsh ice or periglacial fluvial remobilization (Patton and Hambrey, 2009).In either case, the Tonfanau OSL age provides limited constraint on passage northwards of Irish Sea ice margins and was excluded from the Bayesian model on the grounds of context.
The pace of ice margin retreat slowed as the ice stream narrows between SE Ireland and SW Wales, and there was probably a minor readvance and stabilization of the ice marginal position between 24.7 ± 0.2 ka (BL4) and 24.2 ± 0.2 ka (BL5).The pace of net axial ice margin retreat remains slower, 20-40 m a −1 , moving northwards from BL5 (24.2 ± 0.2 ka) through the Llŷn Peninsula to Wicklow alignment by 22.1 ± 0.6 ka (BL9), overlapping with a narrower constriction of the palaeo-ice stream (Fig. 7).The northern end of the Wicklow Trough may mark this transition from rapid to slow retreat (Coughlan et al., 2020).There is evidence for a slight acceleration of retreat through the wider portion of Cardigan Bay, north of the extensive Screen Hills moraine complex in County Wexford (BL7) (Thomas andSummers, 1983, 1984;Small et al., 2018).Across the Wicklow -Llŷn Peninsula configurations of the ice margin (BL8 to BL12; Fig. 7), between 22.6 ± 0.6 and 21.4 ± 0.4 ka, the pace of net axial margin retreat was slow initially (~25 m a −1 ) and with an oscillatory ice front (Thomas and Chiverrell, 2007), but then accelerated (~125 m a −1 ) as the ice front widened and pulled back from a reverse bed slope into the deeper water of the northern St George's Channel (Smedley et al., 2017a).

Northern Irish Sea Basin
The temporal correlations of EBL5 with the ISIS margins at a Wicklow -Llŷn Peninsula position provides a mechanism for ponding the extensive proglacial lakes that developed in the Manchester embayment, the rivers Weaver (Glacial Lakes Delamere and Congleton) and Dee (Glacial Lake Bangor) basins (Crofts, 2005;Chiverrell et al., 2021).In addition, this configuration of the ice margins would have created circumstances conducive for ice-dammed lakes in northeast Wales (Peake, 1961;Thomas, 2005).Retreat of ice margins through the central Irish Sea displays marked asymmetry with more rapid retreat (100-250 m a −1 ) across the western deeper waters (Fig. 7).With >130 m typical water depths at present and extending to join the North Channel between northeast Ireland and southwest Scotland, ice margins reached the coastline of northeast Ireland between 21 ± 0.4 ka (NE-BL1) and 19.5 ± 0.5 ka (NE-BL2).In the east the ISIS margins retreated across the shallower (>65 m) eastern Irish Sea and across the Isle of Man 20.6 ± 0.3 ka (BL15) to 20 ± 0.4 (BL18) ka.The sea floor geomorphology between the Isle of Man and Wales reflects these changing glacial dynamics, with ribbed moraine and latterly drumlinized subglacial bedforms evidencing a more southwest to westerly flow direction and renewed ice streaming conditioned by the topography and increasing proglacial space in the western Irish Sea (Van Landeghem et al., 2009;Van Landeghem and Chiverrell, 2020).
Readvances of the ice margins characterize the glacial dynamics during the step back across the northern ISB and into northeast Ireland (Fig. 7).These include the Scottish Readvance (Trotter et al., 1937) constrained in the Bayesian model to between 20 ± 0.4 and 18.9 ± 1.1 ka (BL18-19), but predating advance limits (possibly linked with H1) in the north of Ireland.Geomorphological evidence for >1-2-km-scale readvances of the ice margin is unequivocal on the Isle of Man and western Cumbria (Huddart, 1977;Thomas et al., 2004;Livingstone et al., 2010;Chiverrell et al., 2018).Readvances of the ice margins in northeast Ireland occurred 18.7 ± 0.2 ka (NI-BL3) and at Killard Point 17.4 ± 0.3 to 16.5 ± 0.2 ka overlapping with Heinrich event 1 (McCabe, 2008;Ballantyne and Ó Cofaigh, 2017;Chiverrell et al., 2018), but there is a lack of a dated limit position equivalent to Killard Point in the northeast Irish Sea; it may be situated in the Solway Firth.Ultimately, the Bayesian model shows similar ages for retreat and deglaciation of mountain regions in southwest Scotland and the English Lake District ~16-15 ka (cf.Ballantyne et al., 2013;Wilson et al., 2018).

Controls on ice advance, maximum extent and deglaciation
The available terrestrial and shallow marine geochronological evidence indicates that significant ice sheet growth in this sector of the BIIS initiated before 26 ka and probably around 30 ka (Fig. 8) although the deep ocean IRD record indicates clearly BIIS calving margins in Marine Isotope Stage (MIS) 3 and initiation during the MIS5/4 transition (Scourse et al., 2009).The IRD record indicates that initial ice sheet growth and readvance dynamics in MIS3 appear tightly coupled to millennial-scale external climate forcing (Scourse et al., 2009).Ice in the ISB had expanded as far south as central Lancashire by ~30-29 ka (Fig. 8) (Chiverrell et al., 2020).The expansion to maximum extent of the ISIS revealed here is early in the global Last Glacial Maximum (LGM) at 25.6 ± 0.5 ka (Clark et al., 2009).The ISG reached maximum limits slightly earlier than the maximum extent of the ISIS, at 26.4 ± 1.1 ka, and with a slower net axial advance rate of ~35 m a −1 (Fig. 8).In 4000 years the marine terminating ISIS had accomplished 565 km of axial ice margin retreat (BL1 to BL10), which contrasts markedly with the terrestrial-terminating ISG reaching eBL5 by 22.1 ± 1.1 ka, some 115 km at only 26 m a −1 (Figs. 7 and 8).
The data are consistent with the build-up of ice within the ISB ~30 ka through to 26 ka and then rapid advance across the shallow southern Celtic Sea soon after; whether this advance was associated with ice streaming or not is unclear, but rapid splaying of the ice front into the Celtic Sea, more than 300 km south of the constriction of St George's Channel suggests rapid advance as a piedmont lobe across the full width of the ice front.Some support for ice streaming during this advance is provided by the extensive mega-scale glacial lineations north of Anglesey where these bedforms are >60 km in length and aligned NNE-SSW parallel to ice flow towards the Celtic Sea (Van Landeghem and Chiverrell, 2020).This advance can be conceived of as a response and temporary over-extension of the ice front preceded and conditioned by significant ice accumulation behind the St George's Channel topographic barrier which steepened the ice surface profile to beyond its yield stress.Once released, the ISIS is interpreted to have captured ice that formerly fed into Cheshire-Shropshire from the ISG from accumulation in the hinterland of the northern ISB and redirected this ice into the main axis of the Irish Sea.This ice piracy starved the terrestrial ISG of ice and initiated slow and early deglaciation, evidenced by the dead ice land system ubiquitous in the Cheshire-Shropshire lowlands (Chiverrell et al., 2020).Deglaciation of the ISIS from its maximum extent at the Celtic shelf break followed soon afterwards because of ice starvation following the overextension event accentuated along a very wide marine margin.
The plausibility of our reconstructed behaviour of ice stream over-extension and ice catchment piracy could be usefully tested by ice sheet modelling experiments.However, in the absence of this it is clear that the timing of this rapid deglaciation event must surely be a function primarily of internal ice dynamic effects as it clearly precedes the warming of Greenland interstadial 2 that followed Heinrich Event 2 and it is not associated with any significant increase in relative sea level (Fig. 8).Thereafter, the sequence of deglaciation of the ISIS and readvance dynamics of the Ice Stream were strongly conditioned by topographic factors, notably trough geometry, rather than simply external climatic or sea level controls (Fig. 8; Smedley et al., 2017a;Small et al., 2018).External climatic controls, including marine and atmospheric temperature regimes, and precipitation, seem to have played a significant role in early ice growth in this sector of the BIIS.Thereafter, at maximum extent and during deglaciation, ice dynamics and topographic controls exerted a stronger role (Fig. 8).

Conclusions
The Irish Sea sector of the former BIIS differs from all other sectors in that it was a compound (bifurcating) system with two outlets, one a marine-terminating ice stream flowing southwards through the ISB into the Celtic Sea, the ISIS, and a nonstreaming terrestrial terminus flowing southwards through Cheshire-Shropshire lowlands into the English Midlands, denoted the ISG.This contrast constitutes a natural laboratory for identifying major drivers of deglaciation since marine forcing was absent in the Cheshire-Shropshire sector until late in deglaciation.The timing and readvance dynamics of both the ISIS and the ISG have been reconstructed via a Bayesian analysis of the geochronological data from both on-and offshore sequences.
The regional chronology (Fig. 8) shows that ice in the Irish Sea sector built-up and expanded substantially from 30 ka through to ~26 ka (Chiverrell et al., 2020).The ISIS extended as far as the continental shelf break to the SW of Britain and Ireland at 25.6 ± 0.5 ka; independently dated IRD is registered from the ISIS in adjacent deep-sea cores synchronous with this advance.The ISIS then retreated rapidly northwards through the Celtic Sea, with evidence for readvance phases reaching St George's Channel by 24.3 ka and the Llŷn Peninsula by 23.9 ka.The complex readvance sequences identified on the Llŷn Peninsula (24-20 ka) and in eastern Ireland have now been tightly constrained to register centennial-scale oscillations of the ice front driven by internal ice dynamics over topographic pinning points and constrictions of the ice-stream.The ISG in the Cheshire-Shropshire lowlands displayed differing dynamics with little evidence of streaming but reached a slightly earlier maximum at 26.4 ± 1.1 ka, with confining topography and the interaction with ice-dammed lakes contributing to regulate the terrestrial retreat dynamics, eventually pulling back offshore by ~21 ka.Retreat northwards into the northern Irish Sea then accelerated, evacuating the deeper waters of the western Irish Sea first, developing pronounced ice margins across the northern Isle of Man by 19.1 ka.The final retreat phase, with ice margins pulling back onto terrestrial settings in the north of Ireland and SW Scotland around 17 ka, was ultimately deglaciation accomplished in a fully marine context as evidenced by the preservation on the seabed of subglacial landforms and by increasing influence of local ice sources with flow realignment during drawn-down and retreat.
The timing and patterns of retreat of the ISIS and ISG revealed by this analysis suggest a 'plugging' of ISIS behind the St George's Channel topographic barrier steepening the ice surface profile, which once released produced a rapid overextension into the Celtic Sea.This initiated piracy of the ISG through the capturing of ice accumulation sources in the northern ISB ultimately leading to slow deglaciation characterized by a dead ice landsystem.Initial deglaciation of the ISIS from the shelf-break is therefore attributed to ice starvation following its temporary over-extension, and its overall retreat behaviour was largely marine in character and strongly modulated by topographic controls.
The data that support the findings of this study are included in the text and Supporting Information, or previously published sources as cited.All relevant data will also be made available in the forthcoming BRITICE-CHRONO online data repository, or upon reasonable request from the lead author.

Figure 1 .
Figure 1.Location maps plotted over the CGIAR-CSI SRTM (Version 4) 30-m-resolution topography (http://srtm.csi.cgiar.org/srtmdata/)and EMODnet bathymetric data (https://www.emodnet-bathymetry.eu/).(A) The Irish Sea Ice Stream (ISIS) and Irish Sea Glacier (ISG), place names, ice maximum extent, ice flow lines, ice divides (as maximum glaciation) based on Clark et al. (2012) apart from the divide in the Galloway Hills based on Finlayson et al. (2010), major bathymetric contours, and the inner, mid-and outer locations for modelled relative sea level (RSL) water depths from a glacio-isostatic adjustment model (Bradley et al., 2011).(B) The retreat patterns of the ISIS and ISG (Fig. 8) identifying the Boundary limits (BL), typically well-defined ice margin limit positions that punctuated the Prior models used in the Bayesian modelling (Fig. 6).ISIS limits are BL1 to BL22, ISG limits in the English Midlands E-BL1 to E-BL10 and in northeast Ireland NI-BL2, 4 and 5. Sample location names and the sampling track of the RRS James Cook cruise (JC106) are also shown.[Color figure can be viewed at wileyonlinelibrary.com].

Figure 2 .
Figure 2. Coastal sections at the north end of Abermawr beach in July 2013 showing (A) the sequence of channelized horizontally laminated sands draped over an over-consolidated Irish Sea-type diamict and buried by more stratified diamictons interpreted as flow/melt-out tills giving way to paraglacial slope deposits.Inset (B) shows a close-up of OSL sample T4ABMW01.For full description and interpretation of these sections see McCarroll and Rijsdijk (2003).[Color figure can be viewed at wileyonlinelibrary.com].

Figure 3 .
Figure 3.The quarry sections at Banc-y-warren visited and sampled in July 2013 showing exposures in the upper 5-4 m of the sequence (A) and the lowest in sequence deposits then visible ~20 m below the surface (B).(A) T4BANC01 was sampled from horizontally (Sh) and planar cross-stratified (Sp) beds of rippled medium-fine sand, but adjacent to 1-m-scale trough cross-stratified sand (St) channel fill deposits.Inset (left) shows a close-up of OSL sample T4BANC01.(B) T4BANC02 was sampled from rippled fine to medium sand in stacked >1-m-scale trough cross-stratified sand (St) mostly likely deep-water channel fill deposits.Inset (right) shows a close-up of OSL sample T4BANC01.[Color figure can be viewed at wileyonlinelibrary.com].

Figure 4 .
Figure 4. (A) Coastal section at Tonfanau sampled in December 2013 located 10 m north of the Log profile TON8 described by Patton and Hambrey (2009).The sequence comprises a steeply dipping basal package of ice-contact fan gravels with minor sand beds, buried by more horizontally bedded outwash lithofacies (OLF) OLF-I and OLF-II.Progressing up sequence, sand beds are increasingly dominant and OLF-III and IV dip more steeply to the south to south-west.(B) OLF-IV south-dipping beds of rippled coarse-to fine sand and the two samples collected.(C) Close-up of the sample T4TONF01 measured for OSL.For full description and interpretation of the Tonfanau sections see Patton and Hambrey (2009).[Color figure can be viewed at wileyonlinelibrary.com].
important constraints mirroring developing chronology in eastern Ireland(Small et al., 2018), and for the eastern flank of ISIS to punctuate data from Scilly(Smedley et al., 2017a) and the Llŷn Peninsula (northwest Wales)(Smedley et al., 2017b).Furthest south of the three, a thin ice proximal outwash deposit at Abermawr (west Pembroke) yielded at age of 25.2 ± 5.0 ka, which considering the heterogeneous bleaching and the large age uncertainty is a good fit to the retreat sequence.Further north, Banc-y-Warren (north Pembroke) yielded a more equivocal chronology.The OSL age determined for T4BANC02 of 39.6 ± 5.79 ka came from low in the sequence and relatively deep-water channel or deltaic sands.This age pre-dates the build-up of ice in the Irish Sea sector(Scourse et al., 2019;Chiverrell et al., 2020) and thus is considered an inaccurate estimate of the time of deposition reflecting poor resetting of the OSL signal.Higher in the Bancy-Warren sequence, though still heterogeneously bleached, these rippled fine and medium sands within trough cross-sets yielded an age of 22.1 ± 3.6 ka (T4BANC02).Both the Abermawr and the younger of the Banc-y-Warren ages postdate the chronology from Scilly, are similar in timing to ice margin retreat to the south coast of Ireland and pre-date ice margin retreat to the Llŷn Peninsula(Smedley et al., 2017a,b;

Figure 5 .
Figure 5. Abanico plots showing the single-grain D e values determined from quartz.Data are presented in two forms to aid data interpretation, a radial plot and a probability density function.Each datapoint represents the D e value determined from a single grain.The position on the x-axis indicates the precision of the individual D e (more precisely known values are plotted to the right).The value of D e for any given point can be read off by drawing a straight line from 0 on the primary y-axis (standardized estimate) through the datapoint to the secondary y-axis (D e ).Typically, samples well bleached before burial will have more symmetrical D e distributions, in comparison to partially bleached sediments, which have asymmetrical D e distributions.The line and grey shading represent either the CAM or MAM D e value for that sample.

Figure 7 .
Figure 7. Chronology for the major retreat limits calculated in the Bayesian age modelling for the wider Irish Sea sector plotted over topographic and bathymetric data [the CGIAR-CSI SRTM (Version 4) 30-m-resolution (http://srtm.csi.cgiar.org/srtmdata/)]and EMODnet data (https://www.emodnet-bathymetry.eu/).The locations sampled for geochronology ( 14 C, TCN and OSL) are identified.The retreat limits show the boundary ages ±1σ for ISIS, retreat into the northeast of Ireland (Chiverrell et al., 2018) and the ISG (English Midlands) (Chiverrell et al., 2020).The ice divides shown represent ice maximum positions after which their position was dynamic; these are based on Clark et al. (2012) apart from the divide in the Galloway Hills based on Finlayson et al. (2010).[Color figure can be viewed at wileyonlinelibrary.com].

©Figure 8 .
Figure 8. (A) Boundary ages (circle ±1σ whisker plots) from both the ISIS and ISG Bayesian models against the respective net axial distance (bottom axis); also plotted are net integrated axial advance/retreat rates in m a −1 (solid bars, top axis) calculated from the Bayesian models.(B) Total (outline) and detrital carbonate (orange) ice-rafted debris (IRD) flux records from the OMEX 2 K core (49°5′N, 13°26′W)(Haapaniemi et al., 2010).Heinrich Events H2 and H1 are highlighted(Bond et al., 1992).(C) δ 18 O concentrations, Greenland Stadials (GS) and Interstadials (GI) from the GISP2 and NGRIP Greenland ice cores(Rasmussen et al., 2014), and modelled surface-air temperatures (black line) relative to present for land masses north of ~45°N(Bintanja et al., 2005).(D) Sea surface temperature (SST) records determined for the North Atlantic using SST (°C) calculated using planktonic foraminifera for core SO82-02 at 59°N, 31°W (red line)(Van Kreveld et al., 2000;Rasmussen et al., 2016) plotted using theWaelbroeck et al. (2019) age model and the MD01-2461 site from the Porcupine Seabight at 51.7°N, 12.9°W (blue line)(Peck et al., 2006;Peck et al., 2007).(E) Ice volume equivalent sea level including uncertainties (black/solid)(Lambeck et al., 2014) plotted alongside summer insolation (red/pecked) for 60°N(Berger and Loutre, 1991), and modelled relative sea level water depths for the inner, mid-and outer ISIS derived from a glacio-isostatic adjustment model(Bradley et al., 2011) updated to include the latest BRITICE-CHRONO ice sheet reconstruction and accounting for global ice sheet variations.Palaeo-water depths reproduced here are calculated by subtracting the predicted relative sea level from the present-day bathymetry.(F) Calving front width, and the minimum and median ice front elevations estimated from the NEXTMap elevation and EMODnet bathymetry (www.emodnethydrography.eu/) datasets for the ISIS plotted against the modelled boundary age chronology (Figs. 6 and 7).[Color figure can be viewed at wileyonlinelibrary.com].

Boundary BL13 -Anglesey -Howth Hill
(Bronk Ramsey, 2009b)020)equence age-model output of dating constraints using OxCal 4.3 for (A) ISIS, (B) retreat into the northeast of Ireland(Chiverrell et al., 2018)and (C) ISG (English Midlands)(Chiverrell et al., 2020).The model structure shown uses OxCal brackets (left) and keywords that define the relative order of events(Bronk Ramsey, 2009a).Each original distribution (hollow) represents the relative probability of each age estimate with posterior density estimate (solid) generated by the modelling.Shown are 14 C ages (R_Date, green), OSL ages (C_Date, orange), cosmogenic nuclide ages (C_Date, blue) and modelled boundary ages (red).Outliers are denoted by '?' and their probability (p) of being an outlier indicated by low values, e.g.<5 (95% confidence).Model agreement indices for individual ages show their fit to the model with >60% the widely used threshold for 'good' fit(Bronk Ramsey, 2009b).[Colorfigure can be viewed at wileyonlinelibrary.com].Boundary NI-

Table 3 .
The modelled boundary limit ages for the ISIS, alongside Bayesian model boundary ages for the ISG