Deglaciation chronology of the Donegal Ice Centre, north-west Ireland

: During the Last Glacial Maximum, Donegal in north-west Ireland functioned as an independent centre of ice dispersal that separated and fed into the Donegal Bay Ice Lobe (sourced in the Irish Midlands) to the south and the Hebrides/Malin Sea Ice Stream to the north. We report geochronological data that demonstrate marked contrasts in the timing and rate of deglaciation in northern and southern Donegal. In northern Donegal, which occupied an inter-ice-stream/lobe location, decoupling from the Hebrides/Malin Sea Ice Stream resulted in formation of a marine embayment along the north coast by (cid:1) 22–21ka, and subsequent slow ( (cid:1) 4 (cid:3) 1ma (cid:4) 1 ) climatically driven inland retreat of the ice margin to mountain source areas by (cid:1) 17ka. By contrast, in southern Donegal, which lay near the axis of the Donegal Bay Ice Lobe, deglaciation was delayed until (cid:1) 18 ka following readvance of ice to a moraine in outer Donegal Bay. The ice margin subsequently underwent net retreat, apparently uninterrupted by readvances, at a net rate of (cid:1) 18 (cid:3) 6 m a (cid:4) 1 . A mean terrestrial cosmogenic nuclide age of (cid:1) 15.0 ka obtained for samples from the foothills of the Blue Stack Mountains in south-east Donegal indicates that ice persisted in valley heads and cirques at the beginning of the Lateglacial Interstadial, suggesting that these and nearby mountains supported the last remnants of the Irish Ice Sheet before complete deglaciation of Ireland, and that almost all the shrinkage of the ice sheet in this sector occurred under stadial conditions before the onset of interstadial warming at (cid:1) 14.7 ka. # 2018 The Authors. Journal of Quaternary Science Published by John Wiley & Sons Ltd.


Introduction
During the last (Late Devensian/Late Midlandian) ice-sheet glaciation of Britain and Ireland ($32-15 ka), the mountains of Donegal in north-west Ireland formed an independent centre of ice dispersal within the more extensive British-Irish Ice Sheet (BIIS). Ice radiating from the Donegal Ice Centre fed north and north-west into a major ice stream (the Hebrides/ Malin Sea Ice Stream) on the adjacent Malin Shelf, west and south-west into Donegal Bay, and to the east was confluent with the ice occupying the Irish Midlands. The Donegal ice dome was therefore pivotal in separating ice flows from the Irish Midlands and western Scotland (Fig. 1). Radial ice flow over Donegal for at least part of the last glaciation is demonstrated by the distribution of local erratics, absence of allochthonous erratics, and the alignments of drumlins, moraines, roches moutonn ees, striae and meltwater channels (Charlesworth, 1924;Dury, 1957Dury, , 1958Dury, , 1964Stephen and Synge, 1965;Colhoun, 1973;McCabe et al., 1993;Knight and McCabe, 1997;Ballantyne et al., 2007;Smith and Knight, 2011;Knight, 2012). Flowsets reconstructed by Greenwood and Clark (2009a,b) show that ice moving south from the Donegal Ice Centre was confluent with west-flowing ice from the Irish Midlands in Donegal Bay, forming an ice lobe (the Donegal Bay Ice Lobe; O Cofaigh et al., 2012) that extended north-westwards towards the shelf edge.
Geophysical data obtained for the adjacent offshore shelves indicate that at the global Last Glacial Maximum (gLGM; 26.5-19 ka, Clark et al., 2009a) grounded ice extended as far as the shelf break, $100 km to the west, where it terminated in a marine setting (Benetti et al., 2010;Dunlop et al., 2010;O Cofaigh et al., 2012). These data also indicate that Donegal ice coalesced with ice from western Scotland $60 km north of the present Donegal coastline. Donegal is therefore the key location for determining the timing of decoupling of Irish-and Scottishsourced ice during the last deglaciation, and is also important for establishing the chronology of ice retreat after the ice margin had retreated to the present coastline.
As part of the wider BRITICE-CHRONO project (http:// www.sheffield.ac.uk/geography/research/britice-chrono/ home), designed to establish a detailed deglaciation chronology of the last BIIS, we present 20 new 10 Be and 36 Cl terrestrial cosmogenic nuclide (TCN) surface exposure ages from six sites in Donegal that were selected to complement and extend the existing deglaciation chronology. The aims of this paper are: (i) to establish the timing of the decoupling of Scottish and Irish ice flowing west across the Malin Shelf; (ii) to reconstruct the chronology of ice margin retreat in Donegal Bay; (iii) to determine the net rate of ice margin recession inland from the north coast of Donegal and in Donegal Bay; (iv) to establish for how long ice persisted locally in the Donegal mountains following its retreat from coastal lowlands; and (v) to explore the wider implications of our results for the interpretation of the deglaciation chronology of the western sector of the last BIIS. The chronology of offshore ice margin retreat from the shelf edge towards the present coastline is considered in a separate paper based on new radiocarbon ages obtained from marine microfauna retrieved from sediment cores along a transect from the shelf edge to the outer part of Donegal Bay ( O Cofaigh et al., 2019).

Regional setting and ice dome extent
County Donegal (54˚28 0 -55˚22 0 N, 06˚55 0 -08˚46 0 W) is predominantly underlain by granites, quartzites and schists with a pronounced NE-SW structural grain that has been accentuated by repeated episodes of Quaternary glacial erosion McConnell, 1997, 1999). The north and west of the county are mountainous; many summits exceed 500 m OD with the highest point (Errigal) at 751 m OD. From detailed mapping of erosional and depositional landforms, Charlesworth (1924) proposed that the Donegal mountains had nourished and maintained an independent ice dome during the last glaciation, and placed the former ice divide along a line running approximately north-south from the Derryveagh Mountains to the Blue Stack Mountains, close to the present watershed ( Fig. 1). He showed that ice-flow from this elongated dome was essentially radial with a focus along pre-existing structurally controlled valleys, but argued that during maximum ice extent topography was probably less of a constraining influence on ice-flow directions than during build-up and retreat phases. Subsequent work is generally supportive of this ice dome hypothesis (e.g. Ballantyne et al., 2007;Greenwood and Clark, 2009a, b;Smith and Knight, 2011). In contrast, the thickness attained by the ice dome has been a contentious issue. Charlesworth (1924) claimed that all summits lay beneath the ice, although it is not clear if he was referring to the local Last Glacial Maximum (lLGM), which is placed at $26.3-24.8-ka at the shelf break to the west of Donegal ( O Cofaigh et al., 2019). Complete burial of the mountains by the last ice sheet was favoured by McCabe (1995), while Sellier (1995) maintained that areas above $550 m OD in the Derryveagh Mountains had remained ice free. On the basis of geomorphological evidence and clayfraction mineralogy, Ballantyne et al. (2007) argued for an ice-shed altitude in excess of 700 m OD, but also reported an absence of evidence for glacial modification on six peripheral summits, including Errigal, and regarded these as being either nunataks during the lLGM or buried beneath a cover of nonerosive cold-based ice.
Conflicting interpretations also concern the lateral extent of the Donegal ice dome. Although Charlesworth (1924) envisaged ice extending offshore to the north and west, others have placed the limit onshore in the north of the county (Stephen and Synge, 1965;Bowen et al., 2002), and a limited offshore extent of $10-30 km to the west has been suggested (e.g. McCabe, 1985;Bowen et al., 1986;Knight, 2003;Ballantyne et al., 2007). More recent work utilizing geophysical techniques to image seabed topography has demonstrated that a concentric sequence of nested moraine ridges indicative of deposition by a grounded ice mass extends westwards to the shelf break 90-100 km from the west coast of Donegal (Sejrup et al., 2005;Benetti et al., 2010;Dunlop et al., 2010;O Cofaigh et al., 2012). Radiocarbon dates obtained for marine microfauna in cores retrieved from sediments on the Atlantic shelf north-west of Ireland confirm that these moraines were deposited at the margin of the last ice sheet, and indicate that ice nourished in Donegal began to retreat from the shelf edge in the interval between 26.3 and 24.8 ka cal BP ( O Cofaigh et al., 2019). Recognition that the last ice sheet extended to the edge of the Malin Shelf strongly suggests that the Donegal ice dome was of sufficient thickness to have buried all mountain summits during the lLGM, a proposition also supported by climate-proxy-driven thermomechanical models of ice-sheet build-up and decay (Hubbard et al., 2009). Support for this premise comes from south-west Ireland where Ballantyne et al. (2011) have argued that the Kerry-Cork Ice Cap attained an altitude of at least 1200 m OD, >200 m above the highest summits, and from north-west Scotland where Fabel et al. (2012) have demonstrated that the last ice sheet overtopped all mountain summits. It is therefore extremely unlikely that any of the mountain summits in Donegal formed palaeonunataks during the lLGM (Ballantyne and O Cofaigh, 2017).  Donegal and  parts of Sligo and Mayo with terrestrial  legacy ages (TCN and 14 C), BRITICE-CHRONO 14 C ages for the Donegal Bay Moraine, with core numbers, and sites of new TCN ages reported in this paper. Only those legacy ages of relevance to deglaciation are shown. The first TCN age for each site was calculated using version 3.0 of the online exposure age calculator formerly known as the CRONUS-Earth online exposure age calculator with the LLPR; the age in parentheses was calculated using CRONUScalc 2.0. Ages are mean values of two or more ages for each site except for Glencolumbkille, which is represented by a single age. Terrestrial 14 C ages are minimum ages for deglaciation. Ice divides and generalized ice flow directions are from Greenwood and Clark (2009a). Inset shows location of Donegal and the LGM limit of the BIIS.

Legacy ages, related BRITICE-CHRONO ages and deglaciation
Several previous studies have used either TCN (cosmogenic 10 Be or 36 Cl) surface exposure dating or 14 C dating to establish the timing of ice retreat and/or readvance from sites in Donegal. Collectively these ages provide the foundation of a deglaciation chronology ( Fig. 1; Table 1). Cosmogenic 10 Be exposure ages cited here have been recalculated using the local Loch Lomond production rate (LLPR), and are followed in parentheses by ages obtained from the CRONUScalc online calculator using a global reference production rate; details of these procedures are given in the next section. Insufficient information was available to recalculate the two 36 Cl ages, and we cite them as published by the original authors. The 14 C ages have been (re)calibrated using OxCal 4.2 and, for marine-derived samples, the Marine-13 curve with a marine reservoir correction of 400 years (Bronk Ramsey, 2009;Reimer et al., 2013). All 14 C ages are reported to two decimal places as cal ka BP. TCN ages are reported to one decimal place as ka. Age uncertainties are AE1s. Mean ages reported for two or more TCN ages below and in Tables 1 and 3 are uncertainty-weighted means. Bowen et al. (2002) obtained 36 Cl ages of 25.1 AE 1.1 ka, from glacially smoothed quartzite bedrock at Malin Head, and 31.0 AE 17.0 ka, from either a glacially transported granite boulder or bedrock at Bloody Foreland, but the large uncertainty on the latter age prevents meaningful interpretation, and the former is probably compromised by nuclide inheritance (Ballantyne and O Cofaigh, 2017). For Corvish, at the head of Trawbreaga Bay on the north coast, McCabe and Clark (2003) reported 14 C ages for marine microfaunas within in situ and deformed marine sediments. The basal in situ laminated muds yielded ages of 20.68 AE 0.16 and 18.24 AE 0.13 cal ka BP; the older date implies initial deglaciation before $20.7 cal ka BP. Overlying deformed sands and muds gave ages of 19.50 AE 0.50, 18.32 AE 0.18 and 19.03 AE 0.19 cal ka BP, and were interpreted by McCabe and Clark (2003) as evidence for reworking of the underlying laminated muds by ice readvance at $18 ka. An age of 17.06 AE 0.18 cal ka BP from in situ rhythmically bedded marine muds overlying the deformed muds was regarded as minimal for final deglaciation of the bay.
Seven consistent 10 Be exposure ages from glacially transported boulders on a lateral moraine at Bloody Foreland, the north-westernmost point of Donegal, have given an uncertainty-weighted mean age of 21.6 AE 0.7 ka (21.7 AE 1.8 ka) (Ballantyne et al., 2007;Clark et al., 2009b;Ballantyne and O Cofaigh, 2017;Fig. 2a). Two 10 Be exposure ages from bedrock and a glacially transported boulder on Aran Island, 20 km south-west of Bloody Foreland, have yielded an uncertainty-weighted mean age of 21.7 AE 0.8 ka (21.5 AE 1.8 ka) (Cullen, 2013). The consistency of these two mean ages provides strong support for retreat of the icesheet margin across the Irish sector of the Malin Shelf between $26-25 and $22-21 ka (Clark et al., 2012a;O Cofaigh et al., 2012O Cofaigh et al., , 2019, decoupling of Malin Shelf ice from Donegal Bay ice at $22-21 ka and the beginning of ice retreat at that time from the present coast towards the mountains. At Glencolumbkille in south-west Donegal, 10 Be exposure ages of 17.8 AE 0.6 ka (17.9 AE 1.5 ka) and 19.6 AE 0.7 ka (19.8 AE 1.7 ka) from vein quartz in, respectively, a glacially transported boulder and a roche moutonn ee were reported by Ballantyne et al. (2007). The latter age overlaps within 1s uncertainties with the mean values from Bloody Foreland and Aran Island, but may be compromised by nuclide inheritance (see below).
The timing of deglaciation of the mountains of Donegal is indicated by 10 Be exposure ages for two sites. Glacially plucked bedrock at 405-430 m OD on a col to the east of Errigal in north Donegal has produced three consistent 10 Be exposure ages averaging 18.0 AE 0.6 ka (17.8 AE 1.4 ka) and a minimum age for deglaciation of Slieve League in south-west Donegal is provided by three consistent 10 Be exposure ages averaging 17.3 AE 0.6 ka (17.1 AE 1.5 ka) obtained for samples from rockslide run out debris (Ballantyne et al., 2013b).
A 14 C age of 15.38 AE 0.12 cal ka BP from the basal organic mud of Lough Nadourcan (Watson et al., 2010) provides a minimum age for deglaciation of the low ground along the eastern margin of the Derryveagh Mountains. However, this age is $700 years earlier than the rapid warming identified in the Greenland ice core records and INTIMATE event stratigraphy as marking the onset of the Lateglacial Interstadial at $14.7 ka (Rasmussen et al., 2014), suggesting that the Lough Nadourcan basal 14 C age may be compromised by the incorporation of reworked older carbon. Nevertheless, it is unlikely that ice on low ground survived much beyond the start of interstadial warming even if small glaciers persisted in the mountains.
Legacy ages from sites in north Mayo, along the south side of Donegal Bay, and BRITICE-CHRONO ages from Donegal Bay ( Fig. 1) are relevant to the deglaciation chronology of south Donegal, and therefore are also considered here. McCabe et al. (1986McCabe et al. ( , 2005 reported eight 14 C ages obtained for marine shells and foraminifera within glacimarine sediments at Fiddauntawnanoneen and Belderg Pier on the north coast of Mayo. Seven of these ages range between 20.38 AE 0.31 and 19.16 AE 0.21 cal ka BP; the remaining age (22.09 AE 0.28 cal ka BP) is significantly older and may indicate the reworking of older sediment (Clark et al., 2012b). Deglaciation of these adjacent sites and, by inference, the outer reaches of Donegal Bay therefore appears to have occurred around or slightly before $20 ka (Ballantyne and O Cofaigh, 2017).
To the north-east of these two sites, a distinct ice margin position is represented by the Donegal Bay Moraine (DBM), an offshore moraine that extends for 35 km north-south across outer Donegal Bay (Benetti et al., 2010;O Cofaigh et al., 2012). Deformation of stratified glacimarine deposits indicates that the moraine represents a readvance of the ice margin. Radiocarbon ages for mixed benthic foraminifera within glacimarine sediments in 76-99 m water depth on either side of the moraine ( . Finally, eight cosmogenic 10 Be exposure ages from vein quartz in glacially transported boulders at three sites associated with the Tawnywaddyduff moraine system on the northern slopes of the Ox Mountains south of Sligo Bay ( Fig. 1) returned ages ranging from 20.9 AE 1.5 ka (21.1 AE 2.3 ka) to 15.7 AE 1.5 ka (16.0 AE 2.0 ka). The overall average of these ages ($18 ka) was taken by Clark et al. (2009c) to represent the timing of a readvance of the ice sheet and construction of the moraine. However, Ballantyne and O Cofaigh (2017) questioned this conclusion, noting that the age range spanned >5 ka and that two distinct age groupings are represented, with three older ages [mean 20.2 AE 1.1 ka (20.3 AE 1.9 ka)] and five younger ages [mean 16.6 AE 0.6 ka (16.7 AE 1.5 ka)]. The older sample ages are from a site on the west side of the Ox Mountains and may be compromised by nuclide inheritance as they are inconsistent with the wider dating evidence; four of the younger ages are from the east side and the other age came from the northern slopes.

Field sites and methods
The six sites sampled for TCN surface exposure dating were selected to provide deglaciation-age transects along the north and south coasts of Donegal and an additional deglaciation age for the northern (Derryveagh) mountains ( Fig. 1). For the north coast transect we sampled on the headlands of Rosguill and Malin Head, respectively 29 and 63 km north-east of the Bloody Foreland site dated by Ballantyne et al. (2007) and Clark et al. (2009b). For the southern transect we sampled at Glencolumbkille, close to the western extremity of the Slieve League peninsula, at Kilcar on the south-west coast of Donegal, and on the lower southern slopes of the Blue Stack Mountains. The latter two sites are, respectively, 14 km south-east and 41 km east of Glencolumbkille. In the northern mountains we obtained samples from a prominent valley-floor boulder limit in the Poisoned Glen, Derryveagh Mountains. Samples were collected from the upper surface of large, glacially deposited boulders or ice-scoured bedrock using a hammer and chisel. Twelve boulder samples comprised whole rock (granite, conglomerate sandstone or dolerite), four were from protruding quartz veins in quartzite or schist boulders, two consisted of quartz pebbles embedded in conglomerate boulders, and two samples were from quartzite bedrock ( Fig. 2; Table 2). A compass and clinometer were used to record the geometry of the sampled surfaces and the skyline topography. Locations and altitudes were determined with a handheld GPS unit cross-referenced to a 1: 50 000 topographic map. Sample thickness was measured using callipers, density was determined by the displacement of sub-samples in water, and topographic shielding was calculated using the online calculators formerly known as the CRONUS-Earth online calculators (Table 2; Balco et al., 2008: http://hess. ess.washington.edu/math/). Samples were processed for cosmogenic 10 Be and 36 Cl analysis at the NERC Cosmogenic Isotope Analysis Facility (CIAF). For 10 Be, samples were crushed and sieved to 250-500 mm and quartz was separated in a Frantz isodynamic magnetic mineral separator, before being repeatedly etched with HF (Kohl and Nishiizumi, 1992). Purified quartz was spiked with $0.22 mg of 9 Be and dissolved. Be was extracted and isolated following the methodology described in Child et al. (2000) before being precipitated as Be(OH) 2 and baked to BeO in a quartz crucible. BeO was mixed with Nb and pressed into a copper cathode. For 36 Cl, samples were crushed and sieved to <500 mm, leached in hot HNO 3 (trace metal analysis grade) and then washed thoroughly with ultrapure water to remove meteoric 36 Cl contamination from grain surfaces. Each sample was then split into two fractions: about 2 g for elemental analysis by inductively coupled plasma (ICP) optical emission spectrometry and ICP mass spectrometry, and about 20 g for analysis of 36 Cl by accelerator mass spectrometry (AMS). Chlorine was extracted and purified from the 125-250 mm fraction of leached samples and precipitated as AgCl using a modified version of procedures developed by Stone et al. (1996). Samples were spiked with $1.26 mg of Cl and sample Cl concentrations were determined by AMS isotope dilution (Di Nicola et al., 2009). Samples were processed together with full chemistry blanks. 10 Be/ 9 Be, 36 Cl/ 35 Cl and 36 Cl/ 37 Cl ratios were measured using the 5 MW pelletron at SUERC (Xu et al., 2010;Wilcken et al., 2013) and normalized to NIST SRM4325 with a 10 Be/ 9 Be ratio of 2.79 Â 10 À11 (Nishiizumi et al., 2007), and Z93-0005 (PRIME Lab, Purdue) with a 36 Cl/Cl ratio of 1.2 Â 10 À12 . Cosmogenic nuclide concentrations include a blank correction of 3-14% for 10 Be and 1-5% for 36 Cl (Table 3). The uncertainties in the cosmogenic nuclide concentrations include the AMS counting statistics and scatter uncertainties from sample, procedural blank and standards measurements.

Age calculation and filtering
The cosmogenic 10 Be exposure ages were calculated using two methods. First, ages were determined using version 3.0 of the online calculators formerly known as the CRONUS-Earth exposure age calculators (Balco et al., 2008; http://hess. ess.washington.edu/math/) using the independently-constrained LLPR (Fabel et al., 2012) with time-dependent LM scaling (Lal, 1991;Stone, 2000) and assuming 1 mm ka À1 of post-depositional surface erosion (cf. Andr e, 2002;Nicholson, 2009;Larsen et al., 2012). The value for LLPR generated by version 3.0 of the online calculators is 3.953 AE 0.093 atoms g À1 a À1 . The uncertainty in this value (2.4%) represents the standard deviation of the measurements in the calibration data set and is unlikely to capture the real scaling uncertainty in the production rate estimate, thus leading to a likely underestimate of the computed external uncertainty in exposure ages (for further details see http://sites.google.com/a/bgc. org/v3docs/home/4-ancillar-calculations-and-plots). Second, we used CRONUScalc (http://cronus.cosmogenicnuclides. rocks/2.0/; Marrero et al., 2016a) with the default global 10 Be production rate of 3.92 atoms g À1 a À1 for LM scaling (Borchers et al., 2016), again assuming an erosion rate of 1 mm ka À1 . Both production rates agree within AE1s uncertainties with the range of production rates determined for other high-latitude sites in the northern hemisphere (Phillips et al., 2016). 36 Cl ages were determined using CRONUScalc with LM scaling, an erosion rate of 1 mm ka À1 and production rates of 56 AE 4.1 at 36 Cl (g Ca) À1 a À1 for Ca spallation, 155 AE 11 at 36 Cl (g K) À1 a À1 for K spallation and 759 AE 180 neutrons (g air) À1 a À1 (Marrero et al., 2016b). Table 3 presents the 10 Be and 36 Cl data and exposure ages with associated uncertainties (AE1s). The data files for the online calculators are provided as supplementary data (Table S1). The effect of varying the assumed erosion rates between 0 and 2 mm ka À1 for both 10 Be and 36 Cl calculations results in <2% change for ages up to 18 ka, 3% change for ages 19-25 ka, and up to 13% for 36 Cl ages between 32 and 43 ka. None of these variations affected our conclusions.
Within-site consistency of ages was tested using the reduced chi-square statistic (x 2 R ) (Bevington and Robinson, 2003). Where the x 2 R-value for a sample of ages from a single site exceeds the critical value at the 95% level, it was inferred that geological uncertainty contributed to the observed age scatter. In such cases outlier ages were manually removed until a x 2 R-value less than the critical value was obtained; the remaining ages were regarded as consistent with and representative of a single age population, with age scatter being due to measurement error alone (Balco, 2011;Applegate et al., 2012;Small and Fabel, 2016;Small et al., 2017a). For sites having two or more internally consistent ages, the uncertainty-weighted mean was determined and is regarded as providing the best estimate exposure age for the site. As with the legacy TCN ages discussed above, we cite the 10 Be-weighted mean ages determined with the LLPR first, followed by the equivalent ages calculated with CRONUScalc in parentheses.

Results
The 20 new TCN surface exposure ages and uncertaintyweighted mean values for internally consistent ages for each site are given in Table 3 and Fig. 3. These ages are assessed below in relation to published ages for the region (Table 1). The ages for Rosguill and Malin Head, on the north coast of Donegal, complement published deglacial age estimates for the northern sites of Aran Island, Bloody Foreland and Corvish. A mean value was not calculated for Malin Head because the three samples failed to yield an acceptable x 2 R-value, due to their wide age scatter ($25.7-20.9 and 25.8-20.7 ka). However, sample MH-03 yielded an age of 20.9 AE 0.9 ka (20.7 AE 1.8 ka), reasonably consistent with the TCN mean ages of 21.7 AE 0.8 ka (21.5 AE 1.8 ka) for Aran Island, 21.6 AE 0.7 ka (21.7 AE 1.8 ka) for Bloody Foreland and 19.0 AE 0.7 ka (18.8 AE 1.6 ka) for Rosguill; in addition, the MH-03 age is consistent with the minimum deglaciation 14 C age of 20.68 AE 0.16 cal ka BP for Corvish, and it therefore provides the best fit age of the three Malin Head ages.
Furthermore, Malin Head is the most easterly of our sites and is unlikely to have been deglaciated before the more westerly sites. The other two Malin Head samples are considered compromised by nuclide inheritance.
The mean age of 16.9 AE 0.7 ka (16.7 AE 1.5 ka) for the Poisoned Glen boulder limit in north Donegal is statistically indistinguishable from the mean age of 18.0 AE 0.6 ka (17.8 AE 1.4 ka) obtained from ice-plucked bedrock on Errigal, 2.2 km north and 350 m higher. Together these two sites indicate that the northern mountains were largely deglaciated by $18-17 ka.
In south Donegal, the consistent exposure ages obtained from three boulders from Glencolumbkille yield a mean age of 16.7 AE 0.6 ka (16.6 AE 1.4 ka). Two legacy samples from this location had returned ages of 17.8 AE 0.6 ka (17.9 AE 1.5 ka) and 19.6 AE 0.7 ka (19.8 AE 1.7 ka). The former age is statistically indistinguishable from the three new ages, and collectively all four ages produce an uncertainty-weighted mean age of 17.2 AE 0.6 ka (17.0 AE 1.4 ka) (x 2 R ¼ 1.24 and 1.77, respectively). The latter age possibly reflects the  1.2 ka), and an uncertainty-weighted mean age of 15.0 AE 0.5 ka (14.8 AE 1.2 ka). The fourth sample (BS-01) is younger than the other three. BS-01 was collected from a boulder showing significant differential weathering (Fig. 2f). The assumed erosion rate necessary to bring the age in line with the youngest of the other three Blue Stack Mountain samples is an unlikely 10 mm ka À1 . Alternatively, the weathering may indicate that the sampled surface was covered by peat or soil in the past, resulting in lower 10 Be production. Either scenario decreases confidence in this sample and we do not include it in further discussions.
The results from three Blue Stack Mountain boulders imply deglaciation of all low ground at the head of Donegal Bay before $14.7 ka. They also suggest that ice persisted much later in the Blue Stack Mountains than in the Derryveagh Mountains of northern Donegal, where the available dating evidence suggests deglaciation of the Errigal col at $18.0 ka and ice withdrawal from the Poisoned Glen boulder limit at $16.9 ka.

Discussion
In conjunction with the published legacy ages discussed earlier, the new ages presented here provide spatially consistent constraints on the timing of deglaciation in Donegal indicate progressive eastward retreat of the ice margin along the northern coast of Donegal between $21.7 and $20.7 ka. The Bloody Foreland and Aran Island ages imply that decoupling of ice sourced in Donegal from the Scottishsourced Hebridean Ice Stream commenced within the interval $22-21 ka; this is slightly earlier than previous estimates, which have placed initial disengagement of these two ice masses after $21 ka (Small et al., 2017b). The single Malin Head TCN age and the oldest 14 C age at Corvish indicate that separation of Scottish-sourced ice and Donegal-sourced ice was complete by $20.7 ka, implying that by this time a marine embayment extended eastward along the north coast of Donegal, separating ice flowing north and north-east from the Donegal Ice Centre from the retreating Hebrides/Malin Sea Ice Stream. The timing of ice retreat inland towards the Derryveagh Mountains of northern Donegal is provided by the TCN ages obtained for the two sites in the heart of this range, at $420 m OD on Errigal col [mean ¼ 18.0 AE 0.6 ka (17.8 AE 1.4 ka)] and, 2.2 km to the south, a low-level site ($74 m OD) at the mouth of the Poisoned Glen [mean ¼ 16.9 AE 0.7 ka (16.7 AE 1.5 ka)]. Although these two ages are statistically indistinguishable within uncertainties (Fig. 4), the difference between them may imply exposure of Errigal col by downwasting ice several centuries before retreat of ice in the Poisoned Glen. Irrespective of whether this was the case, the deglaciation ages for both sites imply that $3000-5000 years elapsed between deglaciation of Aran Island and Bloody Foreland and deglaciation of the Derryveagh Mountains (Figs 3 and 4). A further implication is that net ice margin retreat rates were extremely slow. The Errigal col and Poisoned Glen sites lie, respectively, 15 and 17 km south-east of the Bloody Foreland site; if the mean deglaciation ages for these sites are representative, then the net ice-margin retreat rate from Bloody Foreland to both sites was $4 m a À1 ; taking the associated uncertainties into account suggests that net retreat rate is unlikely to have exceeded 5 m a À1 , and may have been as low as 3 m a À1 . By contrast, assuming that the ice margin began to retreat from the shelf edge within the interval 26.3-24.8 ka ( O Cofaigh et al., 2019), the implied net rate of offshore icemargin retreat from the shelf break to Bloody Foreland falls within the range $19.2-33.3 m a À1 .
O Cofaigh et al. (2019) inferred that ice-sheet retreat from the shelf edge was initiated by calving associated with high sea levels induced by glacio-isostatic depression rather than changing climate, and the marked slowing of retreat after the ice margin had become land-based in northern Donegal appears consistent with this interpretation: the inferred slow net retreat rates of Donegal ice in this sector over the period $21-$17 ka suggest that the retreating ice was close to equilibrium with prevailing climate, and experienced only a slight net negative mass balance during this period.
Averaged net retreat rates, however, may obscure oscillations of the ice margin, with periods of retreat alternating with limited readvances. At present, there is dated stratigraphic evidence for only one such readvance, at Corvish, near the head of Trawbreaga Bay (Fig. 3). Readvance occurred over a distance of at least 5 km according to McCabe and Clark (2003). At this site, the youngest radiocarbon age obtained for Elphidium clavatum tests in deformed marine silts (18.32 AE 0.18 cal ka BP) and a single age for E. clavatum tests in overlying undeformed silts (17.06 AE 0.18 cal ka BP) have been interpreted by McCabe and Clark (2003) as bracketing the timing of readvance of the ice margin on the northern coast of Donegal. They placed the timing of this readvance at $18 ka, although the dating evidence appears consistent with readvance of the  Fig. 1. Note that the Malin Head age is a single age and the Glencolumbkille age is the mean of three new ages and the legacy age given in Fig. 1 (2019) ice margin at any time within the interval $18.5-16.9 ka. On the assumption that the Corvish readvance occurred at $18 ka, McCabe et al. (2007) suggested that it correlates with the Clogher Head Readvance (CHR) in north-east Ireland, although reinterpretation of the stratigraphic and dating evidence indicates that the CHR was a short-lived event that peaked rather earlier, at $18.4 ka (Ballantyne and O Cofaigh, 2017). Thus, although the two readvances may be coeval and represent a regional-scale event that occurred in response to climatic forcing (McCabe et al., 2007;Clark et al., 2012b) it is equally feasible that they occurred at different times and represent localized oscillations of the ice margin. Issues associated with recognition and correlation of readvances are discussed by Clark et al. (2012a). The TCN mean age of 19.0 AE 0.7 ka (18.8 AE 1.6 ka) indicative of the timing of deglaciation at Rosguill pre-dates the bracketing ages for the readvance at Corvish ($18.5-16.9 ka), but because of the uncertainties associated with the TCN age we cannot preclude the possibility that the Rosguill site was reoccupied by glacier ice during the same readvance event.
The TCN mean age for the Poisoned Glen [16.9 AE 0.7 ka (16.7 AE 1.5 ka)] implies that ice persisted in the Derryveagh Mountains until $17-16 ka, but the 14 C age of 15.38 AE 0.12 cal ka BP from Lough Nadourcan (Watson et al., 2010) suggests that ice had disappeared from low ground surrounding these mountains before the rapid warming associated with the onset of the Lateglacial Interstadial at $14.7 ka. Valley heads, cirques and plateaus in the Derryveagh Mountains may, however, have retained ice until early in the interstadial, as appears to have been the case for the Blue Stack Mountains of south Donegal (see below).

Deglacial chronology of south Donegal and Donegal Bay
As noted earlier, the DBM that crosses outer Donegal Bay (Fig. 3) represents the limit of a readvance of ice fed from Donegal Bay. Radiocarbon ages of 20.24 AE 0.24 and 17.92 AE 0.16 cal ka BP obtained for foraminifera retrieved, respectively, from the distal and proximal sides of the moraine constrain its age ( O Cofaigh et al., 2019). This broad interval encompasses the timing of both the readvance at Corvish in northern Donegal and that of the CHR in north-east Ireland (McCabe and Clark, 2003;McCabe et al., 2007;Clark et al., 2012b;Ballantyne and O Cofaigh, 2017), but the resolution of the dating evidence is inadequate to establish contemporaneity. The position and alignment of the DBM suggests that the sites at Belderg Pier and Fiddauntawnoneen on the south coast of outer Donegal Bay lay outside the readvance, and the radiocarbon ages of $20-19 cal ka BP obtained by McCabe et al. (1986McCabe et al. ( , 2005 for in situ marine fauna within glacimarine sediments at these sites (Table 1) are consistent with this interpretation (Fig. 3). Conversely, the aggregated TCN mean age for Glencolumbkille in south-west Donegal [17.2 AE 0.6 ka (17.0 AE 1.4 ka)], the single TCN age for Kilcar (18.0 AE 1.7 ka) and the minimum deglaciation age represented by postglacial rockslide debris at nearby Slieve League [17.3 AE 0.6 ka (17.1 AE 1.5 ka)] suggest that southern Donegal lay within the limits of the readvance that produced the DBM. Similarly, the five 'younger' TCN ages [mean ¼ 16.6 AE 0.6 ka (16.7 AE 1.5 ka)] reported by Clark et al. (2009c) for the Twanywaddyduff moraine system of the northern Ox Mountains ( Fig. 3; Table 1) indicate persistence of ice cover along the inner part of Donegal Bay after $17 ka. Collectively, these two sets of ages suggest that much or all of Donegal Bay continued to support ice cover as late as $17 ka, although it is possible that a calving margin along the axis of the bay led to development of an ice-free marine corridor between its northern and southern shores.
The TCN mean age of 15.0 AE 0.5 ka (14.8 AE 1.2 ka) from the southern flanks of the Blue Stack Mountains suggests that Donegal Bay had become ice free by $15 ka but that ice still occupied mountain valleys and cirques near the head of the bay (Fig. 3). The implication of this age is that mountain ice probably persisted for some time following the onset of the Lateglacial Interstadial at $14.7 ka. The Blue Stack ages are the youngest ages for deglaciation hitherto reported for Ireland (cf. Ballantyne and O Cofaigh, 2017), suggesting that these and possibly other mountains in north-west Ireland supported the last remnants of the last Irish Ice Sheet before complete disappearance of glacier ice under the warmer conditions of the Lateglacial Interstadial. The Blue Stack Mountains, along with other mountain areas in Ireland, hosted glaciers during the Younger Dryas Stadial ($12.9-11.7 ka) (Barr et al., 2017;Barth et al., 2018;Tomkins et al., 2018), but it has not yet been demonstrated that these glaciers had persisted throughout the Lateglacial Interstadial.
The sampling sites at Glencolumbkille and the Blue Stack Mountains are separated by a distance of 41 km. The mean TCN ages for these two sites imply that net ice margin retreat between these two sites occurred over $2200 years, implying a net retreat rate of $19 m a À1 ; taking the associated age uncertainties into account implies that net retreat rate of the ice margin along the northern shore of Donegal Bay fell within the range 12-24 m a À1 , markedly faster than the net rate inferred above (3-5 m a À1 ) for ice retreat inland from Bloody Foreland to the Derryveagh Mountains. Although subject to the same caveat (that retreat may have been interrupted by one or more ice margin readvances), there is neither morphological nor seismostratigraphic evidence for later readvances of the ice margin as it retreated eastward from the DBM.  . Equal-area Gaussian probability distributions representing the uncertainty-weighted means and associated uncertainties for the cosmogenic 10 Be exposure ages obtained for samples from the 'coastal' sites on Aran Island (n ¼ 2) and Bloody Foreland (n ¼ 7), and the 'inland mountain' sites of Errigal Col (n ¼ 3) and Poisoned Glen (n ¼ 2). These distributions illustrate the overlap in the ages obtained for the two 'coastal' sites and for the two 'inland mountain' sites, and also the temporal interval of !4 ka that separates the two sets of ages.  (2019) ice-margin retreat from the shelf edge to the DBM is 11.2-14.0 m a À1 , although they considered that this probably incorporated moderately rapid retreat at a minimum rate of 35.7 m a À1 from the shelf edge to mid-shelf, followed by much slower oscillatory retreat at a net rate of 5.5 m a À1 between the mid-shelf and the DBM.

Wider implications
Collectively, the chronological data reported above indicate a marked contrast in both the timing and the rate of net ice margin recession of land-based ice in northern Donegal ($21-17 ka) and retreat of the ice margin in southern Donegal adjacent to Donegal Bay ($17-15 ka). This contrast suggests that the timing of ice-margin retreat was at least partly conditioned by the relationship between ice fed from the Donegal Ice Centre and adjacent ice streams and lobes. Northern Donegal lay in an inter-ice-stream/lobe location, between the Hebrides/Malin Sea Ice Stream to the north and the Donegal Bay Ice Lobe to the south, and here the early ($22-21 ka) decoupling of Donegal ice from the extended Hebrides/Malin Sea Ice Stream appears to have created an ice-free marine embayment along the north coast of Donegal, so that ice flowing from the Donegal Ice Centre was effectively unconstrained, and subsequently retreated gradually in response to a slight net negative mass balance. In contrast, the south coast of Donegal lay near the axis of the Donegal Bay Ice Lobe, which was fed not only by ice from the Donegal Ice Centre, but also by ice from the Irish Midlands. Following initial rapid retreat, the oscillating margin of the Donegal Bay Ice Lobe retreated slowly from mid-shelf to the DBM ( O Cofaigh et al., 2019), so that ice cover persisted over south Donegal until $17 ka, after which it retreated to the footslopes of the Blue Stack Mountains. This contrast in behaviour implies that different dynamics apply to extended marine-based ice streams and lobes, which are sensitive to changes in sea level, confinement and bed slope (Smedley et al., 2017;O Cofaigh et al., 2019;Small et al., 2018) and land-based ice masses in inter-ice-stream/ lobe locations, which respond mainly to changes in climate inputs.
Evidence for marked slowing of ice-margin retreat as the shrinking BIIS stabilized at or near the present coastline is not limited to northern Donegal. TCN ages reported by Small et al. (2017b) for the Sea of the Hebrides to the west of Scotland suggest that termination of ice streaming after $20.6 ka was succeeded by a $3000-4000-year interval during which the ice margin experienced oscillatory net retreat of only 50-70 km as it became progressively land-based among the islands of the Inner Hebrides. The slowing of ice margin retreat in this area coincides closely with the period of very gradual ice-margin recession in northern Donegal.
The mean exposure age of the samples from low ground ($150 m) at the foot the Blue Stack Mountains [15.0 AE 0.5 ka (14.8 AE 1.2 ka)] represents the youngest age for the timing of ice-sheet deglaciation hitherto reported for Ireland and implies that by $15.0 ka the Donegal Ice Centre had shrunk to a small ice cap or transection complex centred on high ground. An analogous situation occurred in south-west Scotland, 240 km to the east, where seven (recalibrated) TCN ages indicate that only fragmented upland remnants of the Galloway Hills Ice Centre remained by $15.1 ka (Ballantyne et al., 2013a). The Galloway Hills TCN ages are statistically indistinguishable from the Blue Stack ages, and both confirm that almost all ice retreat occurred under stadial conditions before $15.0 ka. In both areas it is unlikely that remnant glacier ice survived subsequent rapid warming, when mean July temperatures inferred from subfossil chironomid assemblages rose rapidly by 5-6˚C (Brooks and Birks, 2000;Lang et al., 2010;Watson et al., 2010;Van Asch et al., 2012). A more general implication is that all of Ireland and southern Scotland were probably completely deglaciated early in the Lateglacial Interstadial. For the British Isles as a whole, present evidence suggests that remnants of the BIIS survived the interstadial (if at all) only in the Highlands of Scotland (Finlayson et al., 2011;Ballantyne and Small, 2018 In northern Donegal our TCN data imply very gradual ice margin retreat inland towards mountain source areas at a net rate of 4 AE 1 m a À1 ; by comparison, the inferred net rate of ice margin retreat from south-west Donegal to the foothills of the Blue Stack Mountains near the head of Donegal Bay averaged 18 AE 6 m a À1 . 4. We suggest that the above contrast in timing and rate of ice retreat reflects differences in location relative to those of major ice streams/lobes. Northern Donegal occupied an inter-ice-stream location, and after early decoupling of Donegal-sourced ice from the Hebrides/ Malin Sea Ice Stream the former was unconstrained and retreated mainly in response to changes in climatic inputs. Conversely, southern Donegal lay near to the axis of the Donegal Bay Ice Lobe, which occupied (or reoccupied) much of Donegal Bay as late as $18 ka, delaying deglaciation along the southern coast of Donegal until after $17 ka. 5. A mean TCN age of $15.0 AE 0.5 ka (14.8 AE 1.2 ka) obtained for the footslopes of the Blue Stack Mountains in southern Donegal is the youngest deglacial age hitherto reported for Ireland, and implies that shortly before the onset of rapid warming at the beginning of the Lateglacial Interstadial ($14.7 ka) the Donegal Ice Centre had shrunk to a small ice cap or ice field of very limited extent, and probably disappeared completely during the early part of the interstadial. This date also confirms that virtually all the retreat of the Irish Ice Sheet occurred under stadial conditions before the onset of interstadial warming.

Supporting information
Additional supporting information can be found in the online version of this article.