Late Devensian deglaciation of south-west Wales from luminescence and cosmogenic isotope dating

: The Welsh Ice Cap was a dynamic component of the last British–Irish Ice Sheet at the Last Glacial Maximum, but there are few chronological constraints on the pace and timing of deglaciation. This paper presents new geomorphological and geochronological evidence that constrains the timing of the separation of the Welsh Ice Cap from the Irish Sea Ice Stream and the subsequent deglaciation of south-west Wales; and allow these to be assessed in the context of late Pleistocene climatic events. Luminescence ages from glacial outwash sediments south of Cardigan demonstrate that the Irish Sea Ice Stream was receding by (cid:1) 26.7 ka. The subsequent recession of the Welsh Ice Cap is documented by cosmogenic ages from landforms and sediments in the Aeron and Teiﬁ valleys and upland areas. Deglaciation of the Cambrian Mountains was underway by (cid:1) 19.6 ka. Cross-valley moraines and associated deglaciation deposits show that minor re-advances interrupted the recession of the Aeron Glacier twice, and the Teiﬁ Glacier on at least 12 occasions. By (cid:1) 14.9 ka, the Aeron valley was probably ice-free, but the northwards withdrawal of the Teiﬁ glacier had halted in the Tregaron area. The ﬁnal rapid recession of this glacier into the uplands of central Wales was completed during the Windermere Interstadial (13.5 cal ka BP) when, in common with much of the UK, the whole of Wales is believed to have become ice-free. There is no evidence that the Cambrian Mountains contained ice-free enclaves at the Last Glacial Maximum, as has been previously suggested. The new ages presented here support suggestions that there was rapid change in the conﬁguration of the Welsh Ice Cap between 20 and 17 ka as upland areas became exposed and there was increasing topographic control on patterns of ice discharge. # 2018 The Authors. Journal of Quaternary Science Published by John Wiley & Sons Ltd.


Introduction
Both modelling studies and observations show that the last British-Irish Ice Sheet (BIIS) was a dynamic and rapidly changing ice mass (Hubbard et al., 2009;Chiverrell et al., 2013;Patton et al., 2013a,b,c;Clark et al., 2018). Scourse et al. (2009) also interpreted dynamism, and rapid response to millennial-scale forcing, in their analysis of the continuous ice-rafted detritus record from the adjacent deep sea. This dynamism was largely due to its relatively small size, which was constrained by the continental shelf break, and proximity to the warm North Atlantic Current, part of the Atlantic Meridional Overturning Circulation (Clark et al., 2012). Early numerical simulations of the BIIS (e.g. Boulton et al., 1977Boulton et al., , 1985Boulton et al., , 1991Lambeck, 1993Lambeck, , 1995 suggested that it was an uncomplicated feature, largely centred over the Scottish Highlands, and with gradually decreasing ice-surface elevation across the rest of the country. However, this view differs from that derived from empirical geomorphological and geochronological evidence, which indicates that at the Last Glacial Maximum (LGM), the BIIS was composed of several small, semi-independent ice caps or ice-dispersal centres (McCarroll and Ballantyne, 2000;Golledge et al., 2008;Hubbard et al., 2009;Hughes et al., 2014; Figure 1). The Welsh Ice Cap was one such centre and the timing of recession of this and other sectors of the BIIS are also debated (e.g. Ballantyne, 2010;Chiverrell et al., 2013;Everest et al., 2013;Hughes et al., 2011). We use here the global LGM definition of Peltier and Fairbanks (2006), which places it around 26 ka.
Several workers have attempted to develop a chronology for recession of the Irish Sea Ice Stream and the Welsh Ice Cap (e.g. Ballantyne, 2010;Chiverrell et al., 2013;Patton et al., 2013a,b), but uncertainties remain. Initial work using cosmogenic nuclide ages along the coast implied that decoupling of the two ice masses and deglaciation of the southern Irish Sea Basin was well underway by 23-22 ka (McCarroll et al., 2010) despite some age estimates of 33-100 ka being obtained (Ballantyne, 2010). Although the geomorphology of the landscape produced by the Welsh Ice Cap is reasonably well documented (e.g. Etienne et al., 2005;Jansson and Glasser, 2005), little is known about its recessional history. Recent geochronology in Snowdonia (Glasser et al., 2012;Hughes et al., 2016; Figure 2A) corroborates modelling simulations (e.g. Patton et al., 2013a,b) that show a previously extensive ice cover evolving into a series of valley glaciers. These individual glaciers receded progressively up-valley into accumulation areas in the Cambrian Mountains and Snowdonia. Significantly, the new exposure ages from Snowdonia point to a rapid change in ice-mass configuration between 20 and 17 ka accompanied by transition from an ice cap to a thinning highland ice field with nunataks (Hughes et al., 2016). Before the current study, this apparent rapid change had not been revealed elsewhere in Wales, leaving questions about rates of deglaciation and deglacial dynamics of the Welsh sector of the BIIS. Clark et al. (2012Clark et al. ( , 2018 and Hughes et al. (2011Hughes et al. ( , 2014 provided comprehensive reviews of the geochronological evidence for the glaciation in the British Isles and concluded that the BIIS reached its maximum extent at different times in different sectors. In many parts of Britain, Ireland and the North Sea, ice had reached its maximum extent by 27 ka. For example, the earliest maximum advances recognized by Clark et al. (2012) occurred by 27 ka in the North Sea and off the coast of NW Scotland. More recently, the maximum southern extent of the Irish Sea Ice Stream on the Isles of Scilly has been dated to 26-25 ka (Smedley et al., 2017a). According to Chiverrell et al. (2013), the maximum extent of Devensian ice was delayed until 25.3 and 24.5 ka near our study area in Wales. A large-scale review by Ballantyne (2010) also suggests a glacial maximum in this area between 26 and 21 ka.
During the early stages of the Devensian Glaciation, the Welsh Ice Cap developed as a separate and autonomous feature and the mountainous areas of North Wales provide constraints on ice-surface elevation (Glasser et al., 2012;Hughes et al., 2016). Most evidence points to the Snowdon massif as a major centre of ice dispersal, with either small domes situated above satellite massifs or a continuous ice divide running broadly north to south (Jansson and Glasser, J. Quaternary Sci., Vol. 33(7) 804-818 (2018) 2005) (Fig. 1). The thickest part of the ice cap was positioned over the Arenig and Aran mountains (Fearnsides, 1905;see Fig. 2 for locations named in the text), with ice flowing westwards from these areas towards Cardigan Bay (Foster, 1970a,b) and eastwards towards the border with England (Travis, 1944). To the north and west, the ice cap interacted and coalesced with a southwards-moving ice mass sourced from northern sectors of the BIIS, which occupied the site of the current Irish Sea Basin (Fig. 1). The distinctive deposits of this Irish Sea Ice Stream mark its maximum extent (McCarroll et al., 2010;Chiverrell et al., 2013).
Where it was unimpeded, the Irish Sea Ice Stream spread across the Cheshire Plain into north-east Wales, the northern Vale of Clwyd (Rowlands, 1971;Thomas, 1985Thomas, , 1989Davies et al., 2004), Anglesey and the Llŷn Peninsula (Thomas and Chiverrell, 2007) as part of its southwards progress into Cardigan Bay and across the coastal tract of south-west Wales (Fig. 1). In all these areas, the landward advance of Irish Sea ice was halted by the expanding Welsh Ice Cap so that at the time of the LGM a complex interface existed between the two competing ice masses. The deposits and landforms that locate this zone of interaction have been identified in north Wales (Thomas, 1989;Davies et al., 2004) and in south-west Wales (Davies et al., 2006b). The southward progress of the Irish Sea Ice Stream into south-west Wales was halted locally by the uplands of the Preseli Mountains, which it failed to overtop even at the time of the LGM (Burt et al., 2012). The large ice-dammed glacial Llyn (Lake) Teifi, once thought to have formed during ice recession (e.g. Charlesworth, 1929;Jones, 1965), is now known to have developed between the two converging ice masses during their Late Devensian advance into the Teifi Valley Fletcher and Siddle, 1998;Hambrey et al., 2001;Etienne et al., 2006).
Here we present new luminescence and cosmogenic isotope (surface exposure) ages obtained for the coastal tract and river catchments of south-west Wales and the southern Cambrian Mountains. These new ages, when interpreted in the context of associated glacial landforms and deposits, provide insight into the post-LGM behaviour of the Welsh Ice Cap, notably its separation from the Irish Sea Ice Stream, and subsequent history and pattern of deglaciation.

Study area: geology and geomorphology
This paper focuses on the Teifi, Aeron and Ystwyth river catchments and the flanking upland regions of the southern Cambrian Mountains and Mynydd Bach (Fig. 2). This region was covered by the Welsh Ice Cap and its confluence with the south-eastern margin of the Irish Sea Ice Stream (Fig. 1). South of the Ystwyth valley, the deeply dissected Cambrian Mountains plateau widely achieves elevations above 450 m with a maximum elevation near the headwaters of the Teifi of 594 m. The more westerly Mynydd Bach plateau, on which the River Aeron rises, and which widely exceeds 300 m in elevation, overlooks the lower ground that forms the Cardigan Bay coastal tract. In the far south of the study area, the prominent rock tors of the Preseli Mountains rise to elevations over 460 m. The bedrock of the area principally comprises Figure 2. A. Published cosmogenic isotope ages dates relating to deglaciation events in Wales (Phillips et al., 1994;McCarroll et al., 2010;Glasser et al., 2012;Hughes et al., 2016). Note that four samples referred to in Phillips et al. (1994) are not induced here because no detailed site information is provided that allows us to plot their locations accurately. B. Map of Wales indicating place names mentioned in the text and samples dated in this study. The Teifi, Aeron and Ystwyth valleys are indicated. Background is a hill-shaded NEXTMap DEM provided by Intermap Technologies. Inset box shows location of Fig. 6. [Colour figure can be viewed at wileyonlinelibrary.com] interbedded mudstone, sandstone and localized conglomerate deposited by turbidity currents and other mass-flow processes. Most are of Silurian age, but Ordovician rocks that also include volcanic and intrusive igneous units occur in the south and west (e.g. Davies et al., 1997Davies et al., , 2006aBurt et al., 2012). NE-SW-trending bedrock lineaments reflect the dominant geological structural trend imposed by late Caledonian orogenesis (e.g. Davies et al., 1997). However, the modern drainage systems only locally reflect this structural grain with major ENE-WSW-trending cross faults also influential. Before the Cenozoic capture of its headwaters by the rivers Rheidol and Ystwyth the source of the previously much larger River Teifi was located on the flanks of the Plynlimon (Pumlumon) Dome. Broad glaciated hollows indicate the abandoned former course of this palaeo-River Teifi in the north of the study area. Waters et al. (1997) recognized the current river as a misfit within the sector of the original catchment it still occupies.
The form and distribution of glacially transported and deposited materials present throughout the area have been established by the British Geological Survey's (BGS) mapping programme in the region (e.g. BGS, 1989BGS, , 1993BGS, , 1994BGS, , 2003BGS, , 2006aBGS, ,b, 2007aBGS, ,b, 2008BGS, , 2010. These materials include widespread diamicton of subglacial origin (till) and, principally along valley bottoms, extensive deposits of glaciofluvial and glaciodeltaic sand and gravel, and glaciolacustrine silt and clay. The presence of exotic igneous, metamorphic and sedimentary rock clasts, as well as common shell fragments and Cenozoic lignite particles, allow sediment derived from the catchment region of Irish Sea ice to be distinguished from Welsh ice-sourced materials that contain clasts only of the regional Lower Palaeozoic bedrock (e.g. Davies et al., 2003). Several studies (e.g. Waters et al., 1997;Hambrey et al., 2001;Glasser et al., 2004;Etienne et al., 2006;Davies et al., 2006b;Patton and Hambrey, 2009) have charted the advance of the Irish Sea Ice Stream into western parts of the Teifi and Aeron catchments and the evolution of pro-and sub-glacial drainage features and spillways as the Ice Stream impounded and then displaced glacial Llyn Teifi. Locally extensive sand and gravel deposits, as worked for example at Trefigin, Pantgwyn and Penparc quarries (Fig. 2B), are interpreted as outwash formed during the subsequent recession of Irish Sea ice from this region (see below). Conversely, the distribution of Devensian deposits of Welsh Ice Cap provenance confirms that much of the area away from the coastal tract lay within the reach of the Welsh Ice Cap. The co-incidence of the inland, eastern limit of abundant Irish Sea erratic material and the western extent of Welsh ice-contact deposits indicate that the two ice masses converged. However, Waters et al. (1997; see also Jones, 1965 andDavies et al., 2006b) suggested that, at the time of the LGM, there remained an ice-free enclave along the main Teifi valley; and the reconstructions of Etienne et al. (2006) and Wilby et al. (2007) show this as the northern point of an extensive and widening region to the south that escaped Late Devensian glaciation (see below).
The southern Cambrian Mountains and the valleys of the Teifi (including its major tributary, the River Dulas) and the Aeron were the loci for regional ice discharge (Fig. 2;Jansson and Glasser, 2005;Sahlin et al., 2007;Sahlin, 2010). The deposits and landforms formed during recession in these river catchments were assessed during the BGS mapping of the region (Davies et al., , 2006aWaters et al., 1997;Ross, 2006;Wilby et al., 2007;Burt et al., 2012). Mounds and hummocks of poorly stratified and poorly sorted icecontact sand and gravel record the location of cross-valley and lateral moraines marking the retreat of discrete valley glaciers that occupied these valleys during Devensian deglaciation. Valley floor exposures and boreholes show that, during glacier withdrawal, the recessional moraines in both catchments impounded pro-glacial lakes in which clay, silt and glaciodeltaic deposits accumulated (Merrin, 1999;Davies et al., 2006a,b). In the Teifi valley, the lake dammed by the northernmost of these moraines evolved into Tregaron Bog (Cors Caron in Welsh) À one of the largest raised peat bogs in the UK (Godwin and Mitchell, 1938;Davies et al., 1997). Radiocarbon dating at this site by Hibbert and Switsur (1976) (10 200 AE 220 cal BP) confirms that organic-rich sediments began to accumulate above barren, grey, lacustrine clays soon after the climatic amelioration that signalled the end of the Younger Dryas Stadial and the onset of the Holocene.
In upland and interfluve areas, glacially streamlined bedrock ridges show a strong structural bedrock control. Deposits of head and nivation scree (head gravels) testify to the periglacial conditions that prevailed as Devensian ice withdrew from the region. Landslides, alluvial fans and river terraces record paraglacial adjustment of the landscape that preceded the establishment of the area's modern drainage system and modern alluvial deposition (Davies et al., , 2006a.

Irish Sea Ice Stream deglaciation
Four sediment samples were obtained for luminescence dating from two sand and gravel quarries located to the south of the Teifi valley, SW of Cardigan: two samples from Pantgwyn Quarry (MW05 and MW07; 52.048˚N, 4.735˚E) and two samples from Trefigin Quarry (MW08 and MW09; 52.057˚N, 4.718˚E) (Fig. 3). Sample details are provided in Tables 1 and 2. Both quarries work ridges of poorly sorted gravel and coarse to fine sand, in which abundant crossstratification testifies to deposition by southerly flowing, braided rivers ( Fig. 3) (Owen, 1997). Clast assemblages and abundant shell fragments confirm an Irish Sea Ice Stream source. Extensional faulting (Fig. 3b) seen in the quarry faces has been interpreted by Hambrey et al. (2001) as evidence of collapse of the sediment pile following the decay of flanking ice buttresses. Fining-upwards successions at both sites further support their interpretation as remnants of a once more extensive spread of glacial outwash formed during the withdrawal of the Irish Sea Ice Stream from the area (e.g. Davies et al., 2003). Significantly, both quarries are located several kilometres to the north of the Preseli Mountains, which formed the local southern limit of the Irish Sea Ice Stream advance into the region.

Welsh Ice Cap deglaciation
The absence of glacially transported boulders on the surface of recessional moraines in the Teifi and Aeron valleys precluded dating of these features using cosmogenic isotope techniques. However, 10 samples were collected for cosmogenic nuclide exposure dating from glacially transported boulders and ice-scoured bedrock sites located in the adjacent uplands and interfluves (Table 3; Fig. 4). Following the sampling guidelines of Gosse and Phillips (2001), Dunai (2010) and Balco (2011), samples were collected from the upper surfaces using a hammer and chisel and all samples were <4 cm thick. Detailed site descriptions were made for each sample. Skyline measurements were recorded using an Abney level where the angle to the horizon was >15˚. Sample coordinates were collected using a handheld GPS with a horizontal accuracy of AE4 m. The cosmogenic nuclide  Tables 3 and 4.

Geomorphological mapping
Extensive geomorphological mapping was undertaken by BGS and its collaborators as part of their field investigations in the region (e.g. Waters et al., 1997;Hambrey et al., 2001;Glasser et al., 2004;Etienne et al., 2006;Davies et al., 2006b) and incorporated the results of earlier studies (e.g. Jones, 1965;Watson, 1969). Landforms and sediments were identified according to their size, shape, composition and texture using field observations, including section logging and hand augering, colour stereo aerial photography, borehole records and site investigation reports (see McMillan and Powell, 1999;Hubbard and Glasser, 2006).

Luminescence dating
Samples were taken for luminescence dating by hammering opaque tubes into the sedimentary section to prevent exposure to sunlight during sampling. External beta dose-rates were determined for optically stimulated luminescence (OSL) dating using inductively coupled plasma mass spectrometry (ICP-MS) and inductively coupled plasma atomic emission spectroscopy (ICP-AES), and the external gamma dose-rates were determined using in situ gamma spectrometry (Table 1). To isolate coarse-grained quartz for OSL analysis, each sample was first treated 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 grains 180-212 mm in diameter, and density-separation using sodium polytungstate provided the 2.62-2.70 g cm À3 (quartz-dominated) fractions. The quartz grains were etched for 1 h in 40% hydrofluoric (HF) acid to remove the outer portion of the quartz grains that was affected by alpha irradiation and also remove any contaminating grains of feldspar. After the HF etching, grains were washed in a 10% solution of HCl to remove any fluorides that may have been produced. Grains were finally mounted into 10 Â 10 grids of 300-mm-diameter holes in a 9.8-mm-diameter aluminium single-grain disc for analysis.  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) as described by Smedley et al. (2017a,b) . A preheat plateau test performed on 5-mm aliquots was used to determine the preheat temperature used (200˚C) with a cutheat of 160˚C for the single aliquot regenerative dose (SAR) protocol (Murray and Wintle, 2000). Five screening criteria were applied to the resulting data throughout the analyses unless otherwise specified. Grains were screened based upon whether (i) the test dose response was greater than three sigma above the background, (ii) the test dose uncertainty was <20%, (iii) the recycling and OSL-IR depletion ratios were within the range of ratios 0.8-1.2, (iv) recuperation was <5% of the response from the largest regenerative dose (150 Gy) and (v) the single-grain equivalent dose (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). The associated uncertainties were taken into account when applying the screening criteria. D e values were then calculated for all grains that passed the screening criteria, incorporating the uncertainty from instrument reproducibility of 2.5% (Thomsen et al., 2005). Dose-recovery experiments were performed on Samples MW05 (overdispersion ¼ 16 AE 1%), MW07 (overdispersion ¼ 22 AE 1%) and MW09 (overdispersion ¼ 15 AE 1%) and suggested that the SAR protocol was appropriate for OSL dating. The single-grain D e values determined for each sample are shown (Fig. 5) and included in Appendix 1.

Luminescence dating results
The single-grain D e distributions for Samples MW07 and MW08 were not asymmetrical and so the sediments were deemed to have been well bleached before burial; thus, the central age model (CAM) was used to determine D e values for these samples ( Table 2). The single-grain D e distributions of Samples MW05 and MW09 were asymmetrically distributed and so the sediments were deemed to have been heterogeneously bleached before burial; thus, the minimum age model (MAM) was used to determine D e values for these samples ( Table 2). The overdispersion arising from intrinsic OSL characteristics for each sample was combined in quadrature with the overdispersion arising from external microdosimetry (20%) to determine the values of s b used for the MAM (s b ¼ 0.25). The D e values were then divided by the environmental dose-rates to determine OSL ages ( Table 2). The weighted mean and standard error (eqns 21 and 22 in Aitken and Alldred, 1972) of the two OSL ages at both Pantgwyn and Trefigin were 26.7 AE 1.8 and 26.7 AE 1.6 ka, respectively; consistent with their proximity to one another, these ages appear to confirm that outwash accumulation at both sites was broadly synchronous.

Cosmogenic nuclide dating
Samples were processed in the Natural Environment Research Council (NERC) Cosmogenic Isotope Laboratory Facility, at the Scottish Universities Environmental Research Centre (SUERC). After the mechanical procedures and chemical etching to separate the quartz from other minerals and remove the meteoric 10 Be, the pure quartz was spiked with 9 Be (approx. 0.2 mg 9 Be) and 27 Al (to total up $1500 mg 27 Al) and dissolved. The beryllium and aluminium isotopes were extracted following chemical procedures adapted from Child et al. (2000). Prepared BeO and AlO targets were subsequently mixed with Nb and Ag, respectively, and pressed into copper cathodes. Target measurements were performed using the 5-MV accelerator mass spectrometer at the SUERC AMS Laboratory (Xu et al., 2010). Measured 10 Be/ 9 Be ratios and 26 Al/ 27 Al ratios were converted to nuclide concentrations in quartz and exposure ages were calculated using the Cronus Earth online calculator v.2.3 (hess.ess. washington.edu/; Balco et al., 2008). The calculated age uncertainties are expressed as AE1s (Table 4). Internal uncertainties are quoted when describing our results and the external uncertainties are quoted when comparing our results to previously published data sets.
A range of 10 Be production rates can be used to calculate exposure ages and there remain uncertainties over suitable Table 3. Location and details of samples collected for cosmogenic isotope dating (see also Fig. 4).

Sample
Latitude (   values for Wales. To compare with other published data, we used the same 10 Be production rate for west Wales and the Irish Sea area as Chiverrell et al. (2013) to calculate ages: 4.20 AE 0.15 atoms g À1 a À1 . This production rate value is based on a calibrated dataset from north-west Scotland (Ballantyne and Stone, 2012). The 10 Be ages were calculated using the Lal (1991)/Stone (2000) production rate model. Exposure ages (Table 4) were calculated using a 10 Be decay constant of 4.99 Â 10 À7 a À1 , based on the half-life value of 1.387 AE 0.012 Myr (Chmeleff et al., 2010;Korschinek et al., 2010), and Stone scaling (Stone, 2000) to sampling site for both spallation and muon production. Previously published 10 Be ages from the neighbouring Aran mountains (Glasser et al., 2012) were recalculated using the same approach with the same range of production rates to enable direct comparison.

Cosmogenic nuclide dating results
Both the 10 Be and the 26 Al concentrations were calculated to explore the possibility of multiple exposure histories on the sampled rocks. For most of the samples in this study, the 26 Al/ 10 Be ratios are not depleted with respect to the production rate ratio of 6.75, suggesting a simple continuous exposure with no periods of burial since ice recession (Table  4). However, for two samples (MP12/01 and YST12/01), wide discrepancies in the 10 Be and 26 Al ages may indicate more complex histories of exposure (see below). Apparent 10 Be and 26 Al ages range between 14.3 AE 0.4 and 19.6 AE 0.6 ka. As the individual sample 10 Be and 26 Al ages are not statistically the same for all, we have not included the 26 Al ages in the discussion.

Irish Sea Ice Stream deglaciation
The OSL ages obtained for the Pantgwyn and Trefigin deposits constrain the onset of this deglaciation. Significantly, they indicate that Irish Sea ice had withdrawn from the foothills of the Preseli Mountains as far north as these sites by $26.7 AE 1.6 ka. These findings are consistent with the timing of ice impingement on the Isles of Scilly at 26-25 ka, suggesting that retreat from the maximum in the Celtic Sea occurred rapidly (i.e. within the uncertainties of the ages determined).

Welsh Ice Cap deglaciation in the Teifi Valley and Cambrian Mountains
Most of this sample set came from the eastern side of the Afon Teifi catchment, along the western flanks of the Cambrian Mountains plateau (Figs 2 and 6). These dates provide evidence for the age and pattern of the Welsh Ice Cap withdrawal across this upland region. The ages and locations of Samples CT12/01 (19.3 AE 0.7 ka), BB12/01 (16.1 AE 1.3 ka), BB12/02 (17.7 AE 0.6 ka), TP12/03 (15.7 AE 0.5 ka) and YST12/02 (17.1 AE 0.5 ka) tell a broadly consistent story. They suggest that ice withdrew from some of the most prominent upland ridges at c. 19 ka in the south of the region, exemplified by Criag Tyrch. This was preceded by a more widespread westwards retreat of the ice cap from the southern Cambrian Mountains towards the Teifi Valley, which was largely complete by c. 16-17 ka (Fig. 6). Sample MP12/01 (13.8 AE 0.8 ka) is the most southerly sample of the Teifi Valley/Cambrian Mountains set. This sample, collected at a higher elevation (392 m) than others to the north, implies that the withdrawal of ice from this site was delayed until around 13.8 ka (Fig. 6). Whereas the other samples of this set provide a fairly consistent picture for the timing and pattern of ice retreat from the western flanks of the Cambrian Mountain plateau (see below), the date obtained for Sample MP12/01 appears to be an outlier. The mean exposure age provided by the northernmost sample, YST12/ 01 (14.3 AE 0.4 ka), in being significantly younger ($2.9 kyr) than that obtained for the nearby Sample YST12/02, also appears aberrant. For MP12/01 and YST12/01, the wide discrepancies between the Be and Al exposure data imply that both these samples may record a more complex exposure The ratios ranged between 3.20 Â 10 À14 and 1.88 Â 10 À13 for 10 Be/ 9 Be, and between 1.26 Â 10 À13 and 4.18 Â 10 À13 for 26 Al/ 27 Al. The 10 Be concentrations are corrected for the procedural blank values of 5.1615 Â 10 À13 and 5.8032 Â 10 À15 . Data relative to NIST SRM 4325 ( 10 Be/ 9 Be taken as 2.79 Â 10 À11 ; Nishiizumi et al., 2007).

Welsh Ice Cap deglaciation in the Aeron Valley and Mynydd Bach
The ages obtained for Samples AV12/01 (19.6 AE 0.6 ka) and AV12/02 (19.1 AE 0.6 ka) testify to the early emergence of the highest part of the Mynydd Bach plateau and indicate withdrawal of the Welsh Ice Cap from the whole of this upland region was complete by c. 19-20 ka. Sample TV12/01 (14.9 AE 0.4 ka) implies that, in common with much of the Cambrian Mountains, the interfluve between the Teifi and Aeron catchments was similarly ice-free by c. 15 ka.

I n t e g r a t i n g t h e g e o m o r p h o l o g i c a l a n d geochronological evidence
In Figs 6 and 7 we present a putative chronology and palaeogeographical model for the deglaciation of southwest Wales in the context of late Pleistocene climate events including those identified by the Greenland-based ' INTIMATE' project (e.g. Bj€ orck et al., 1998;Blockley et al., 2014). The regional maximum limits of the Irish Sea Ice Stream and Welsh Ice Cap in the region are based, with only minor modification, on published accounts (e.g.  Waters et al., 1997;Davies et al., 2006b;Etienne et al., 2006;Wilby et al., 2007). The absence of obvious end moraines along the Teifi Valley probably reflects the termination of both ice masses in the surviving remnant of Llyn Teifi. Waters et al. (1997) follow Jones (1965) in suggesting that benches of sand and gravel near Pentre-Cwrt are the remnants of a former delta À the Pentre Cwrt Delta À that built out into the lake close to the Welsh Ice Cap's terminal position (Fig. 6).
The succession of ice-retreat lines shown on Fig. 6 are constrained by the new exposure ages and geomorphological evidence. The latter, including cross-valley moraines (numbered on Fig. 6), kame terraces, glaciolacustrine basins, and remnant deltas and outwash plains, chart the early withdrawal of both ice masses from the coastal tract and show that the subsequent deglaciation of the area was interrupted by episodes of stagnation or re-advance. The new OSL dates for Irish Sea glacial outwash are consistent with the early retreat of this western, maritime component of the BIIS in response to post-glacial maximum climatic amelioration perhaps linked to the INTIMATE GI-3 event. Its withdrawal from the Preseli Mountains and lower Teifi valley areas was evidently well advanced by c. 26 ka. Glacial deposits and landforms preserved to the north provide a record of the decoupling of the Irish Sea Ice Stream from the Welsh Ice Cap and the local impounding of temporary lakes between the separating ice masses (Davies et al., 2006b;Ross, 2006). The eroded remnants of associated glaciofluvial fan deltas supplied by meltwater from the Irish Sea Ice Stream to the north, as for example at Talgarreg, locally allow the position of the retreating ice front to be reconstructed with some precision.
The later deglaciation of Welsh ice from Mynydd Bach and parts of the southern Cambrian Mountains saw the early emergence of bedrock prominences such as Hafod Ithel (Sample AV12/01) and Craig Twrch (Sample CT12/01). This marked the onset of a period of rapid reconfiguration of distal parts of the Welsh Ice Cap at c. 20-19 ka, which resulted in the formation of discrete Teifi and Aeron valley outlet glaciers. By this time, the Irish Sea Ice Stream had fully decoupled from the Welsh Ice Cap and was positioned on the northern Llŷn Peninsula (Smedley et al., 2017b) and both ice masses had completely vacated the coastal tract of southwest Wales. Significant Welsh ice withdrawal from the lower Teifi Valley had probably already occurred (Figs. 6 and 7) and the early retreat moraine T1 had formed and been abandoned. Deglaciation during this interval occurred during the climatic flux associated with the GS-3 to GS-2.1 INTI-MATE events.
The extensive kamiform deposits and associated crossvalley moraines (T2-3) to the east of Llandysul mark the onset of this process in the Teifi Valley. Comparable landforms suggest that withdrawal of the Welsh Ice Cap outlet glaciers that occupied the lower reaches of the Aeron (A1), Ystwyth (Y1) and Rheidol (R1) valleys was also in progress from around this time (Fig. 6). Glacier recession from the Aeron Valley was subsequently interrupted by episodes of stagnation or re-advance on at least two occasions (A2-3), but the succession of landforms seen along the Teifi and Dulas valleys provides evidence of a longer period of ice occupation and of a staggered glacial retreat that halted on at least a further nine occasions (T4-12, D1-4). Although a precise chronology for these events remains difficult to tie down, the new cosmogenic exposure ages provide a basis for a provisional interpretation and comparison with the Greenland-based INTIMATE climate events (Figs 6 and 7). A majority of the Cambrian Mountain dates describe a westwards withdrawal of ice from the upland plateau towards the axis of the Teifi Valley. Sample TV12/01 indicates that by 14.9 ka ice had also withdrawn from the region of the Aeron-Teifi interfluve and it is likely that the whole of the western sector of the study area including the Aeron Valley may then have been ice-free (Fig. 6).
Glacial landforms between Llandysul and Tregaron Bog reveal a staged northwards recession of the Teifi glacier along this 45-km-long reach of the valley (Figs 6 and 7). Recession of the ice from the Teifi valley probably commenced soon after the decoupling of Irish Sea ice from the Welsh Ice Cap, ongoing before 19 ka (see above). The chronology of its subsequent retreat is constrained ostensibly by the MP12/01 exposure age (13.8 AE 0.8 ka) in the south and, in the north, by the onset of early post-glacial deposition at Tregaron Bog (10.2 ka). However, Late Devensian ice is widely assumed to have disappeared from Wales, as throughout much of the UK, during the early part of the Windermere Interstadial c. 13.5 cal ka (Clark et al., 2012). Even the maximum interval these figures allow (c. 1700 years) appears too short to generate the complex suite of recessional landforms and deposits preserved to the south of Tregaron Bog. Such analysis underlines the probability that the cosmogenic isotope values obtained for Sample MP12/01 may not accurately reflect the date of ice withdrawal from this locality (see above).
The exposure age obtained for Sample TV12/01 (c. 14.9 ka) offers an alternative, more realistic chronology arguably more consistent with other data. Its proximity to Tregaron Bog implies that the withdrawal of ice from this sample site can be linked to the recession of the Teifi glacier into the area currently occupied by the bog and to the formation of its associated recessional moraines (T11 and T12; Fig. 6). Hence, and contrary to MP12/01, it suggests that much of the 45-km northwards retreat of the Teifi Valley Glacier into the Tregaron area spanned a period of over 5000 years (Fig. 7). The succession of cross-valley moraines and impounded lake deposits formed during this interval undoubtedly testify in some way to the impact of climatic variations associated with the INTIMATE GS-2.1a-b interval. However, the dynamic nature of the glacier probably meant that its response to these events was neither synchronous nor simplistic.
The absence of recessional moraines to the north of Tregaron Bog appears to signal a marked change in ice behaviour and that the final phase of ice withdrawal from the upper reaches of the modern and palaeo-Teifi valleys was rapid and uninterrupted. This must have occurred between c. 14.9 ka and c. 13.5 ka in response to the climatic amelioration associated with the Windermere Interstadial (¼ INTI-MATE GI-1) (Fig. 7). The undated lacustrine clays present beneath the organic-rich Holocene sediments at Tregaron Bog may include material laid down during this period of warming or the subsequent, short-lived, cold period (Loch Lomond Stadial/Younger Dryas; INTIMATE event GS-1), as at other sites in mid-Wales (e.g. Walker et al., 2001).
The 10 cosmogenic nuclide exposure age samples from these valleys provide the first chronological control on landforms related to the terrestrial deglaciation of south-west Wales. They indicate that deglaciation was well underway by 19 ka and that discrete outlet glaciers subsequently occupied and receded up the Aeron and Teifi valleys. By 15 ka these glaciers had retreated into the uplands of Wales and the lower reaches of these valleys were ice-free. Glasser et al. LGM) and at stages during their retreat, based on published maps of glacial deposits (see text) and the new OSL and cosmogenic isotope ages presented in this paper. Notes: 1. Some geomorphological features have been extrapolated to show their likely extent before later Pleistocene and Holocene deposition and erosion. 2. Only the most seaward of the retreat moraines present in the Ystwyth and Rheidol valleys are shown. 3. Only the largest proglacial lakes are named À many of the others were probably short-lived features that drained rapidly as their moraine dams were breached soon after ice withdrawal. 4. 'Unglaciated' refers only to the Late Devensian À the Preseli Mountains were over-ridden during earlier, more extensive episodes of Pleistocene ice advance. 5. Ice-front positions incorporate and update those proposed in the area by Davies et al. (1997Davies et al. ( , 2006b Figure 7. Timeline for the deglaciation of south-west Wales as suggested by the dating results of this study (see Tables 1-4 and Fig. 6 clear pattern demonstrating that, following a period of rapid thinning, mountain summits in North Wales became exposed as nunataks between 19 and 20 ka. Our samples from the Cambrian Mountain are farther south and at lower elevations than those collected in North Wales. Nevertheless, our 10 Be exposure ages appear to confirm that there was a period of significant thinning around 20 ka, that bedrock prominences became emergent around this time, and that deglaciation of the south-west portion of the Welsh Ice Cap was then dominated by a series of increasingly discrete outlet glaciers that receded progressively up-valley. Numerical ice-sheet models show a similar recessional pattern, with dynamic outlet glaciers occupying valleys during deglaciation as the ice cap receded into the interior of Wales (Patton et al., 2013a,b,c). These new data from the south-west of the Welsh Ice Cap are also relevant to the debate about the extent of the ice cap at the LGM. In a series of papers, Watson (1965Watson ( , 1969 argued that this area of mid-Wales is predominantly a periglacial landscape, lacking evidence of glaciation. Watson argued that, where they exist, glacial deposits are buried beneath periglacial landforms and sediments implying that there was a period of prolonged or intense periglacial activity following glaciation. However, our geomorphological mapping and associated 10 Be exposure ages indicate that upland areas in south-west Wales share a similar deglaciation history to mountainous regions in North Wales, that valley glaciers remained active in the region until c. 15 ka, and that their subsequent recession into upland source areas and final down-wasting pre-dated the Windermere Interstadial. Evidence for a regionally widespread or significantly more protracted Late Devensian periglacial phase is therefore lacking.

Conclusions
i. Using luminescence and cosmogenic isotope dating, we have produced the first detailed chronological framework for Late Devensian glacial events in south-west Wales and the southern Cambrian Mountains. ii. Luminescence ages for ice-contact sand and gravel deposits at Pantgwyn and Trefigin of 26.7 AE 1.8 and 26.7 AE 1.6 ka, respectively, indicate that the Irish Sea Ice Stream was undergoing active recession from the coastal fringes of south-west Wales at that time and, consequently, that the maximum advance of the Irish Sea Ice Stream into south-west Wales occurred before this. iii. Deglaciation of the south-western area of the Welsh Ice Cap was marked by a progression from ice cap conditions to valley glaciation as the ice cap thinned and its outlet glaciers became increasingly constrained by the local topography. iv. Deglaciation of the Welsh Ice Cap probably commenced immediately after the LGM. Bedrock prominences were the first features to emerge from the ice cover. By 19 ka, the adjacent upland regions were ice-free and outlet glaciers occupied, and had begun to recede up, the Aeron and Teifi valleys. By this time, the Irish Sea Ice Stream had fully decoupled from the Welsh Ice Cap and both ice masses had probably withdrawn from the coastal tract of south-west Wales. v. Cross-valley recessional moraines and glacial lake and outwash sediments chart multiple periods of stagnation or re-advance of the Teifi and Aeron glaciers. By c. 15 ka, the whole of the Aeron catchment and much of the Cambrian Mountains plateau were ice-free and the Teifi glacier had receded into the Tregaron area.
vi. A period of rapid deglaciation followed, which saw the Teifi glacier retreat into its upland source areas in the Cambrian Mountains prior to final down-wasting before the onset of the Windermere Interstadial around 13.5 ka. The deglaciation geochronology for south-west and south-central Wales matches that of parts of North Wales, confirming that the Welsh Ice Cap underwent a period of marked thinning and reconfiguration around 17-20 ka.