Glacial–interglacial cycles in the south-central and southeastern Pyrenees since ~180 ka (NE Spain–Andorra–S France)

Abstract This study uses luminescence and 14C accelerator mass spectrometry procedures to date relevant glaciofluvial and glacial deposits from the south-central and southeastern Pyrenees (Andorra–France–Spain). We distinguish two types of end-moraine complexes: (1) those in which at least a far-flung moraine exists beyond a frequently nested end-moraine complex (the most common) and (2) those in which a close-nested end moraine encompasses at least two glacial cycles. Both types formed within six distinctive glacial intervals: (1) A penultimate glacial cycle during Marine Oxygen Isotope Stage (MIS) 6 and older glaciofluvial terraces occurred beyond the range of the luminescence dating method. (2) An early glacial advance in MIS 5d (~97 −15/+19 ka) was followed by glacial retreat during MIS 5c (< 91 ± 9 ka). (3) The last maximum ice extent (LMIE) was in early MIS 4 (~74 ± 4.5 ka). (4) Unexpectedly, glaciers thinned during the second half of MIS 3 (~39 −6/+11 ka). (5) During the MIS 3–2 transition, glaciers subsequently fluctuated behind the LMIE limits. (6) The global last glacial maximum (LGM) started as early as ~26.6 ± 0.365 ka b2k, and the corresponding end moraines were built behind the LMIE limits or merged with it, forming close-nested moraines.

Asymmetries arise when comparing both sides of the mountain range. On the one hand, we have the example of the SW Pyrenees, in which the MIE is from the penultimate glaciation (Lewis et al., 2009;García-Ruiz et al., 2013), but this is not the case in the SE Pyrenees, where glacial extent during the penultimate glacial cycle (PGC) was less than during the LMIE (Turu et al., 2007;Ventura and Turu, 2022). On the other hand, evidence from the LGM is somewhat elusive in the SW Pyrenees (García-Ruíz et al., 2003Lewis et al., 2009;Bartolomé et al., 2021). Aridity (Höbig et al., 2012;Allard et al., 2021) and moisture variations (Jalut et al., 2010;Delmas et al., 2015;Turu, 2018) seem to have provided strong gradients between the Atlantic and the Mediterranean and exerted a significant influence on glacial extent across the Pyrenees. A W-E moisture gradient during the Holocene has been demonstrated in Iberia (Liu et al., 2021) and during the last termination period in the Iberian Central System mountain range (Turu et al., , 2021. A similar NW-SE gradient is expected to have existed across the Pyrenean range during glaciations. Before attempting any discussion comparing the existing data identifying glacial cycles or phases in the Pyrenees, some additional chronological data from relevant sites is needed. To do this task, we chose end moraines and terminal complexes within glacial valleys from the south-central and southeastern Pyrenees, especially far-flung moraines (Anderson et al., 2012) and nested moraines (Kirkbride and Winkler, 2012).

GEOMORPHOLOGICAL SETTINGS AND GLACIAL ANTECEDENTS
The Segre River tributary catchments are located on the southern slope of the French-Spanish-Andorran Pyrenees, draining to the Ebro basin and thus to the Mediterranean Sea (Fig. 1). From east to west, the Segre tributaries are the Valira, Noguera Pallaresa, Noguera Ribagorçana, and Cinca Rivers. Most of the southern Pyrenean glacial valleys are U-shaped, flanked by lateral moraines, kame-type deposits, and former ice-dammed valleys. From west to east, the highest peaks in the studied area are the Pic de Margalida (3251 m) in the Noguera Ribagorçana, followed by Pica d'Estats (3143 m) in the Noguera Pallaresa on the France-Spain border, the Comapedrosa (2942 m) in the northwest of Andorra, and the Carlit (2921 m) in the French Cerdagne (Fig. 1). Most of the end-moraine complexes from the southeastern and south-central Pyrenees did not reach the foreland foothills, with only the Valira and Querol palaeoglaciers having reached the intramountain grabens of the Urgellet and Cerdagne, respectively (Calvet, 1998;Turu et al., 2007).

The Upper Noguera Ribagorçana valley
This valley is located on the westernmost side of the studied area ( Fig. 1, sector 1). Geologically speaking, three rock types are present in this valley, distributed following a W-E trend (Mey, 1965): late Variscan crystalline rocks (granites) on the northern part of the valley (north from the Baserca reservoir; Fig. 2), Mesozoic sedimentary rocks southward from the confluence of the Baliera River and the Noguera Ribagorçana River (south Vilaller; Fig. 2), and finally, Palaeozoic low-grade metamorphic rocks in between the two previous lithological areas. Geomorphologically speaking, the valley harbours thick lateral moraines and kame deposits well suited to study glacial evolution.
The upper Noguera Ribagorçana was the first glacial valley from the southern slope of the Pyrenees to be dated, specifically in the Llauset (Fig. 2) glaciolacustrine deposits (Vilaplana, 1983a(Vilaplana, , 1983bMontserrat-Martí, 1985) when the lake water level was artificially raised to become the reservoir for a pumpedstorage hydroelectric power plant. In the same area, the former Llestui ice-dammed glaciolacustrine complex was the next to be dated Vilaplana and Bordonau 1989;Bordonau et al., 1993) along with the Seminari de Vilaller (Fig. 2) overdeepened trough (Bordonau, 1992). Half a kilometre northward from Sant Mamés (Fig. 2), a terminal-moraine complex was identified at Seminari de Vilaller (Vilaplana, 1983a(Vilaplana, , 1983bBordonau et al., 1989;Bordonau, 1992;Pallàs et al., 2006). Nearby, erratic boulders at Tinabre (Fig. 2) have produced the oldest 10 Be cosmic ray exposure (CRE) ages in the valley (Pallas et al., 2006). No end-moraine exposures south of Vilaller are currently known. However, the remains of glaciofluvial and kame deposits help us to reconstruct the LMIE glacial front in Pont de Suert (Fig. 2).

The upper Noguera Pallaresa valley
Surrounded mainly by low-grade metamorphic rocks (Zandvliet, 1960), the former Noguera Pallaresa glacier benefitted from the contribution of the former Garonne glacier coming from the NW side of the valley (the Beret and Bonaigua pass; Fig. 1) and produced the most extensive glacial tongue (63 km) in the southern Pyrenees (Ventura and Turu, 2022). The terminalmoraine complex from the LGC was identified ∼2 km upstream from Sort (Fig. 1, sector 2) at 748 m above sea level (m asl; Bastida de Sort; Furdada, 1988;Ventura and Turu, 2022). Unfortunately, no other end moraine has been identified from the Noguera Pallaresa glacier. However, a thick glaciofluvial terrace is located in Sort (Fig. 1).

The Valira valleys
The main Valira glacier ( Fig. 1, sector 3) was formed by the confluence of two tributaries, the former Valira del Nord glacier and the Valira d'Orient glacier (Fig. 3). Both tributaries flowed over metamorphic rocks grading from slates to gneiss. However, the floors of these valleys are in granite bedrock (Zwart, 1979). The resulting ice tongue (Fig. 3) extended from Escaldes-Engordany (1050 m) to beyond Sant Julià de Lòria (900 m) (Turu et al., 2007) and formed several terminal-moraine complexes (Cal Tolse, Sant Julià de Lòria, La Margineda, and Santa Coloma; Fig. 3), mainly related to readvances within general deglaciation (Turu et al., 2007(Turu et al., , 2017. The southernmost deposits are the Cal Tolse terminal-moraine complex (Turu, 1994), exposed in an ancient quarry located +85 m above the riverbed (m arb), which exploited aggregates from glaciofluvial deposits southward from Sant Julià de Lòria (Fig. 3). This area was exposed by the triggering of a landslide in 2018 (Luzi et al., 2021), allowing identification of a supraglacial till that is partially eroded and overlain by glaciofluvial deposits.
The subsequent deposits are in Sant Julià de Lòria (Fig. 3), where Prat (1980) identified erratic boulders on the eastern Valira riverbank. Later, building works allowed a better view of the terminal complex (Jalut and Turu, 2008).
Upstream from Sant Julià de Lòria is the La Margineda terminal complex, long established as an end moraine from the Valira glacier (Panzer, 1926;Nussbaum, 1934;Llobet, 1947). Here colluvium overlies glaciofluvial deposits located at 905 m (45 m arb; Turu, 1994); however, only occasional exposures of a supraglacial till have been seen during building excavations. Weathered granite boulders were identified at 980 m and attributed to an old moraine (the Llumaneres moraine from Turu and Peña-Monné, 2006). Nevertheless, a large outcrop to the south of La Margineda allows the identification of a fresher end moraine.
In Santa Coloma (Fig. 3), an open pit cut into a fluvial terrace +15 m arb by an ancient quarry provided an opportunity for Nussbaum (1956) to recognize a moraine from the Valira glacier when it had a length of 29 km (Chevalier, 1906(Chevalier, , 1907. Upstream from Santa Coloma many other retreat moraines have been recognized (Bladé, 1875;Penck, 1884;Turu et al., 2017 and references therein), but these were not revisited in this study.

The upper Segre and Cerdagne
In the upper Segre and Cerdagne basin (Fig. 1, sectors 3 and 4), exposures are found at the confluence of the Segre River with its tributaries (Adrall and Ur sites) or within the Segre River terrace system (Sanavastre).
The Cerdanya is located in the uppermost part of the Segre basin and drains low-grade metamorphic rocks and granites on its northern border (ICGC, 2006). This area was divided by the treaty of the Pyrenees in 1659, so we use "Cerdagne" for the French part (in the north) and "Cerdanya" for the Spanish zone. Within the latter area, we found the abandoned quarry of Albellorols (1060 m) close to the village of Sanavastre, in which a weathered coarse deposit, some 7-8 m thick, is located at ∼20-35 m arb and belongs to the Segre T4 (the Mala Mort level; Calvet, 1998). Terrace T4 overlapped by the Escadarchs fan ( Fig. 4) and was fed by limestone-sourced sediments. In Cerdagne, T4 is decalcified by intense weathering (Calvet, 1998), but this is not the case for the same terrace on the southern flank of the Segre River (Poch et al., 2013). The fluvial terrace system continues downstream in the form of two weathered outliers (terrace T3; Fig. 4). Pedogenic geopetal structures, including carbonate nodules and minor clay accumulation, are common in terrace T3 (Poch et al., 2013), with gruss from weathered cobbles of granite; however, rich-iron minerals remain unweathered. The well-known nested end-moraine complex of the Carol glacier surrounding Puigcerda is found in Cerdagne (Calvet, 1994(Calvet, , 1998Calvet et al., 2011b;Pallàs et al., 2010 and references therein;Fig. 4). Upstream of the Carol-Segre confluence is the Ur-Llaurar rest area (Fig. 4) close to the N20 highway, in which we sampled a sandy layer from a thick deposit of glaciofluvial gravels.

Fieldwork
Fieldwork included detailed geological characterisation of each exposure, consisting of stratigraphic logging and interpretation of sedimentological features. Sample collection (Table 1) from previously cleaned profiles was undertaken at the sites mentioned above ( Fig. 1) from coarse glaciofluvial, colluvial, or fine-grained glaciolacustrine sediments. For luminescence dating, we hammered steel tubes into the sections and then extracted them, avoiding any exposure of the contents to daylight. Adjacent to each tube, a subsample of sediment was collected to determine the dose rate at the laboratory.

Grain-size analysis
The texture of the sediment samples was determined at the Sedimentology Laboratory of the Department of Earth Sciences, University of Coimbra, Portugal (Supplementary Material 1). The grain-size distribution and statistical parameters of each sample were obtained through the integration of two methods: by sieving the >63 μm fraction and by laser diffraction analysis of the <63 μm fraction with a Coulter laser granulometer (2 mm-0.04 μm). The grain-size distribution provides information about the sediments' maturity, transportation history, and deposition. Because particles transported under subaerial conditions are exposed to daylight, near-complete bleaching (zeroing of the optically stimulated luminescence [OSL] signal during transport) is expected, avoiding luminescence-dose inheritance (see Hoey, 2004). Sediments suitable for OSL are those of fluvial or aeolian origin, showing skewed, unimodal, well-sorted grain-size distributions and associated with transport histories in which subaerial conditions are met. Because such ideal conditions would not always be represented in a glacial-type or related deposit (Walker, 2005), two modern samples from the Segre riverbed were measured to estimate the offset of possible inheritances (Table 1, SANAV-0 and STPCodinet-1).

Water-content estimations
For each sample, we calculated the field water content (w). First, we estimated the void ratio (e), a parameter related to packing the particles within the sediment. This parameter is readily estimated concerning published tables (Jiménez-Salas and de Justo-Alpañes, 1975). For example, e = 0.35 equates with the densest packing of spherical particles of equal size bearing on top of each other, whereas e = 0.91 corresponds to the weakest packing of spheres. The void ratio and the porosity (n) are related by n = e (1 + e) −1 . Because the density of igneous and metamorphic particles forming the framework of unconsolidated sands are known (ρ ≈ 26.5 kN/m 3 ), the dry density (δ d ) is associated to the porosity (n) related by δ d = ρ ⋅ (1 − n), while the density of a fully saturated soil (δ s ) is: δ s = δ d + n ⋅ δ w , where δ w is the density of liquid water. Estimating the water content (w%) at full saturation may also be an estimate of the maximum water content during the burial period, probably below the water table; thus w% = 100 ⋅ (δ s − δ d ) ⋅ δ d −1 . However, once the deposits have been uplifted or exposed by erosion, the water content will be lower than in fully saturated conditions. The minimum water content corresponds with the highest void ratio because these are inversely related. When the deposits are unsaturated, capillarity action retains moisture at a rate inversely proportional to their void ratio. Void ratio values for the most common deposits are shown in Supplementary Material 2. Knowledge of water content is important, as water absorbs radiation differently from mineral sediment and thus must be accounted for in dose-rate calculations (Walker, 2005).

Luminescence dating
In this section, we include both OSL and thermally stimulated luminescence methods. We dated several deposits twice to identify any biases between these methods. Luminescence dating covers those analytical methods that measure the time elapsed since grains of particular minerals (e.g., quartz or feldspars) were last exposed to significant heating or daylight (Duller, 2004). The method was developed mainly by using quartz as the dose meter (Qtz-OSL), but K-feldspar can be used when the Qtz-OSL signal is saturated.

OSL dating
Exposure to daylight during sediment transport removes the latent luminescence signal from the quartz or feldspar crystals. After the burial, the luminescence signal (trapped charge) starts to accumulate in the mineral grains due to ionising radiation arising from the decay of 238 U, 232 Th, and 40 K present in the sediment and from cosmic-ray bombardment (Murray et al., 1987). In the laboratory, the equivalent dose (D e , assumed to be the dose absorbed since the last light exposure; i.e., the burial dose, expressed in Gy) is determined by comparing the natural luminescence signal resulting from the charge trapped during a burial with that trapped following laboratory irradiation (Olley et al., 1996). For the calculation of the dose rate of sand-sized K-feldspar grains, an internal K content of 12.5 ± 0.5% was assumed (Huntley and Baril, 1997).
Sample preparation for luminescence analyses was undertaken in darkroom conditions. Samples were wet-sieved to separate the 180-250 μm grain-size fraction, followed by HCl (10%) and H 2 O 2 (10%) treatments to remove carbonates and organic matter, respectively. The K-feldspar-rich fraction was floated off using a heavy-liquid solution of sodium polytungstate (density = 2.58 g/cm 3 ). The quartz fraction was obtained by etching another portion with concentrated hydrofluoric acid (HF) (40%). The K-feldspar fraction was treated with 10% HF for 40 minutes to remove the outer alpha-irradiated layer and clean the grains. After etching, the quartz and K-feldspar fractions were treated with hydrochloric acid (HCl) (10%) to dissolve any remaining fluorides. Quartz purity was confirmed by the absence of a significant infrared-stimulated luminescence (IRSL) signal.
Equivalent doses were measured on automated Risø TL/OSL DA-20 readers, each containing a beta source calibrated for irradiation on stainless steel discs and cups. Quartz measurements were made on large (8 mm) aliquots containing several thousands of grains mounted on stainless steel discs. Small (2 mm) aliquots of K-feldspar were mounted on stainless steel cups. Quartz dose estimates were made using a standard single-aliquot regenerative- dose (SAR) protocol using blue-light stimulation at 125°C for 40 s with a 240°C preheat for 10 s, a 200°C cut the heat, and an elevated temperature (280°C) blue light-stimulated clean-out step Wintle, 2000, 2003). The OSL signal was detected through a U-340 filter. All samples had a strong, fast component. The net OSL signal was calculated from the stimulation's initial 0.0-0.8 s and an early background between 0.8 and 1.6 s. The K-feldspar D e estimates were measured with a post-IR IRSL SAR protocol using a blue-filter combination (Thomsen et al., 2008;Buylaert et al., 2012). The preheat was 320°C for 60 s, and the cut heat was 310°C for 60 s. After preheating, the aliquots were IR bleached at 50°C for 200 s (IR50 signal) and subsequently stimulated with IR at 290°C for 200 s (pIRIR290 signal). It has been shown by Buylaert et al. (2012) that the post-IR IRSL signal measured at 290°C can give accurate results without the need for correction for signal instability. For all IR50 and pIRIR290 calculations, the initial 2σ of the luminescence decay curve, less a background derived from the last 50 s, was used. Dividing the D e by the environmental dose rate (in Gy/ka) gives the luminescence age of the sediment.

Thermoluminescence dating
Samples were treated in the laboratory following conventional procedures for fine-silt (4-20 μm) sample preparation (Mauz et al., 2002) and etched using 20% and 10% HF for various rounds to remove the feldspar-derived luminescence component (Mauz and Lang, 2004a). After the grains were etched, the silt of 4-15 μm grain size was separated by settling in acetone. The aliquots of each sample consisted of 2 mg of material pipetted onto 1 cm aluminium discs. All samples were tested using infrared stimulation (Duller, 2003), which examines the effect of the feldspar on normalised thermoluminescence (TL) measurements and gives an estimate of the amount of feldspar remaining in the sample after chemical treatment (Mauz and Lang, 2004a). According to these tests, all samples used for further experiments were pure quartz. The self-attenuation of the samples was calculated in a simplified approach using the following equation: Where D is the attenuated dose of gamma radiation, DO is the dose delivered by the source, μ is the linear attenuation coefficient of the sample, and r is the inner radius of the container. Equation 1 was derived for homogenous and isotropic radiation around a small container, compared with the mean free path of gamma photons (μ ⋅ r << 1). Any attenuation due to the thickness of the "sample" is assumed to result from a homogenous and isotropic radiation field. The 90-160 μm fraction was obtained by drysieving, and the 4-11 μm fraction was extracted by settling in acetone. TL measurements were performed using a blue detection window. Some measures were made with and without silicon oil as a fixing agent. An insignificant tendency toward slightly lower results with silicon was observed. All analyses were performed at the ITN (Instituto Tecnologico Nuclear, C2TN, Campus Tecnológico Nuclear, in Portugal, following laboratory procedures from Mauz and Lang (2004b) and Richter et al. (2003).

Radiocarbon accelerator mass spectrometry (AMS) dating
Radiocarbon AMS dating was restricted to bulk sediment samples from the Valira valleys. Measurement of 14 C activity and the isotope ratio 13 C/ 12 C was carried out at the Beta Analytic Inc.
(Florida, USA) mass spectrometer facility and used the bulk organic fraction from sedimentary samples. Calib software based on IntCal20 curves was used for calibration (Reimer et al., 2020). Because radiocarbon dating is routinely applied to the final part of the LGC and the Holocene (Walker, 2005), the method is explained in Supplementary Material 3.

RESULTS
Luminescence results are presented from west to east across the study area (Fig. 1), including the sedimentary records and their ages (Tables 1 and 2), while the radiocarbon ( 14 C) ages are exclusively from the Valira valleys (Table 3). Some deposits were sampled twice to assess the coherence between dating methods. These TL samples are LUM-26, LUM-273, and LUM-274 (Table 2); compared with OSL results of samples 122242, 122244, and 122247 (Table 1), respectively, which show good agreement.
Qtz-OSL dating provided finite ages only in a few ancient samples (122238, 122239, 122242, and 122246; Table 1) and two modern samples (122235 and 122248; Table 1) from the Segre River. In all the other samples, the Qtz-OSL signal was saturated or nearly saturated and only provided minimum ages (122236,122237,122240,122241,122243,122244,122245,122247,122256; Table 1). Therefore, signals from K-rich feldspars were used instead because of the higher saturation dose of the dose-response curve for feldspar (Wintle and Murray, 2006). The pIRIR290 signal was chosen because of its stability (Buylaert et al., 2012). We also made several direct tests of the stability of the pIRIR290 signal and checked the suitability of the applied measurement protocol and the completeness of signal bleaching at deposition by using modern samples (122235 and 122248; Table 1). The pIRIR290 signal of several samples is saturated or nearly so; only minimum ages were obtained (122236, 122237, and 122241; Table 1). More complete information about the laboratory measurements of the OSL samples is available in Supplementary Material 4.
Qtz-OSL dating provided finite ages in samples VILALLER-1, STA.COLOMA-2 and FAUCELLES-1 (Table 1), and the two current Segre River bed-load samples, SANAV-0 and STPCodinet-1. The Qtz-OSL signals were too close to saturation for suitable dating for all other samples. For older samples, signals from K-rich feldspars were used instead because of the higher saturation dose of the feldspar dose-response curve (Wintle and Murray, 2006). These samples are UR-1, CMARG-4, STJULIA-2, SEGUDET-1, ADRALL-4, and GINEBROSA-1 (Table 1). However, several samples were too close to saturation, and only minimum ages were obtained (SANAV-1, SANAV-2, and SORT-1; Table 1). Further empirical information about the offset and saturation of the luminescence signal is provided in Supplementary Material 5.

Results from the upper Noguera Ribagorçana valley
The results from the westernmost side of the studied area (Fig. 1, sector 1) are presented in this section. We briefly summarize the existing knowledge of the area and set our results in context. Radiocarbon dates performed on bulk sediment samples retrieved from glaciolacustrine sediments recovered by coring at a site 500 m north of Sant Mamés (Fig. 2) returned inverted ages (Bordonau, 1992). Contamination probably affected one or all of these samples, as was presumed to be the case with glaciolacustrine deposits from the Llestui lateral ice-dammed lacustrine complex (Bordonau et al., 1993), located upstream in the same valley (Fig. 2). The problem was apparently solved at Llestui, where Bordonau et al. (1993) reported two new ages (23,497-20,612 b2k and 27,671-23,993 b2k) from the LGM glaciolacustrine deposits (Pallàs et al., 2006).
The exposure of the abandoned Sant Mamés de Vilaller quarry has two divisions: a non-deformed glaciofluvial unit and a synsedimentary deformed unit. Within the deformed unit, lenses of laminated sand are embedded in a matrix-supported diamicton, which is interpreted as a melt-out till. An overlapping contact divides the Sant Mamés deposit into two units between glaciofluvial sediments that overlap a diamicton (Fig. 5a). The glaciofluvial unit is predominantly sandy but also includes layers of imbricated gravels, indicating that the sedimentation was influenced by episodes of high meltwater discharge (Fig. 5b). Within the diamicton, we identified a massive lodgement till at the base of the outcrop that grades to crudely bedded sands and gravels including boulders ( Fig. 5c) (type C1b facies based on Krzyszkowski and Zieliński [2002]), whereas deformed sandy layers are interbedded at the top of the outcrop (Fig. 5d) (melt-out till facies; Bordonau, 1992).
A sample from the glaciofluvial unit provided an age of 15.7 ± 1.2 ka (VILALLER-1; Table 1), whereas the melt-out till gave an age of 73.9 ± 4.5 ka (VILALLER-2; Table 1). The time interval between these ages sheds light on an unconformity between the melt-out till and the later glaciofluvial unit. Further field descriptions are available in Supplementary Material 6.

Results from the upper Noguera Pallaresa Valley
The neighboring glaciated valley eastward from the upper Noguera Ribagorçana is the upper Noguera Pallaresa valley (Fig. 1, sector 2). In this valley, we analysed a glaciofluvial terrace exposure in Sort. Southward from the village of Sort (692 m), a 120-m-long outcrop was visible in 1992, on the right side of the widened N260 highway, revealing a glaciofluvial terrace Ventura and Turu, 2022) at the toe of a large and currently inactive landslide . A sandy layer from this terrace was sampled for OSL dating (SORT-1; Fig. 6, Table 1) and provided a minimum age >162 ka. The sampled layer overlies coarser deposits and is partly eroded in its upper part (Fig. 6). From bottom to top, we distinguished massive and crudely bedded gravels arranged in sets of low-angle sheets at the bottom of the exposed deposit (Fig. 6); these are partly cut by erosion with a broad channel bar sheet formed by imbricate massive gravels overlying its erosive surface. Crudely bedded sands deposited in low-angle sheets sit concordantly above the coarse deposits. Unfortunately, an unexposed section between this sandy layer and the top of the sequence prevented a complete determination of the stratigraphic record (Fig. 6). However, no significant changes in the sedimentary environment are apparent, there being a profusion of existing imbricated gravels in all visible parts of the sequence. At the top, a poorly consolidated coarse-grained diamicton marks the end of the sedimentary sequence at Sort.
We attribute the Sort deposit to a type C end-moraine fan, as described by Krzyszkowski and Zieliński (2002). The crudely bedded gravels at the base of the deposit corresponding to facies C1a, the channel bars to C1b, and the low-angle sand sheet to C2a. The broad range of facies represents a proximal end-moraine fan (Krzyszkowski and Zieliński, 2002), above which are the remains of the frontal moraine (uppermost facies B1b; Fig. 6).

Results from the Valira valleys
The Valira River joins the Segre at La Seu d'Urgell ( Fig. 1, sector  3). Its basin (592 km 2 ) covers most of the Principality of Andorra (Fig. 3).

The Segudet kame
The Segudet kame complex is located at the confluence of the Cassamanya and Les Aubes streams (Fig. 7a). A subglacial till deposited at Segudet (1350 m; Fig. 7a) may indicate the limits of the ice-dammed valley impounded by the Ordino glacier (Turu et al., 2017). A supraglacial till exists on the hillslope (1520 m; Fig. 7b), deposited by the 200-m-thick trunk glacier. The kame comprises thick glaciofluvial gravels, mainly located in its central part. It is possible to trace this gravely unit to the edge of the sedimentary complex, where it grades to sands and silts. The OSL sample was taken from a laminated fine sand layer ( Fig. 7c and d) and provided an age of 141 ± 16 ka (SEGUDET-1; Table 1). Complementary field descriptions are provided in Supplementary Material 6.

The Ginebrosa kame
The Ginebrosa deposits are found 200 m above the Arinsal riverbed (1449 m; Fig. 3) and are related to a large kame terrace formed in a lee-side position between the Arinsal glacier ( Fig. 3) and adjacent bedrock. Silts and sand lenses are located away from the former slope-glacier ice contact (Fig. 8a). These lenses are interbedded with colluvium from nearby bedrock erosion. On the opposite side of the kame deposits, close to the former ice contact, deformed sediments include intrusions of cohesive clay lenses. This small valley has been accessible since 1997, and several outcrops created by building work have been available for analysis. The sedimentary sequence was emplaced directly over polished bedrock from a previous MIE of unknown age (Fig. 8b,  1). Four sedimentary units were identified (Figure 8b, 2-5). Sandy lenses (Shr) and horizontally bedded imbricated gravels (Gt) form the lowermost unit (Fig. 8b, 2), draped over the bedrock and covered by massive gravels (Gm) beneath a diamicton layer (Dmm, Dcs) deposited by a glacial advance. A final sedimentary unit (Fig. 8b, 3), overlapped by sands and silts (Sh) grading to coarse gravels (Gp facies) (Fig. 8b) filled the irregular surface (Catuneanu, 2006;Fig. 8b, MRS). Instability caused by lateral glacial thrusting produced lenticular injections of cohesive clay. The age of the interbedded laminated sand lenses is 42 ± 8 ka (GINEBROSA-1; Table 1). A Holocene postglacial unit (Fig. 8b, 5) formed by fine-grained colluvium gave dates of 9.17 ± 0.15 ka cal BP using a 14 C AMS determination on bulk organic matter (GINEBROSA-2; Table 3).

La Margineda
The main Andorran valley becomes narrower southward of Santa Coloma (Fig. 10a), and an urbanized hill, known as La Margineda (990 m), stands on the valley's eastern side eastern. We examined an existing outcrop rich in sedimentary facies located at +65-75 m arb (1008 m asl) in a road cutting (Fig. 10b). Here, colluvial facies overlie a glaciofluvial deposit, both densely compacted (Fig. 10c), from which samples were collected (Fig. 10c) for AMS and luminescence determinations. Two sedimentary discontinuities are identified in the road cut (Fig. 10d), differentiating three units (Fig. 10e). The lowermost unit, sampled for OSL dating, comprises imbricated gravels with a significant presence of allochthonous lithologies (e.g., granites), probably transported as bedload by meltwater mixed with colluvium from a melting glacial front and the end-moraine (Llobet, 1947;Turu and Peña-Monné, 2006) reworking. This unit is from MIS 5c (91 ± 9 ka; CMARG-4; Table 1); however, the mentioned end-moraine buildup occurred earlier, probably during MIS 5d. In the middle unit (Fig. 10e), clasts are arranged with their long axes parallel to the slope (colluvium). However, some clasts are imbricated, and others set as though belonging to palaeochannels (lower unit). The occasional presence of allochtonous lithologies (e.g., granites) may come from the erosion of the terminal-moraine complex. The age of the colluvium (Fig. 10e, LM-1) is 26.931-26.201 ka cal BP (22.290 ± 0.090 ka BP; ß-489299; Table 3). The upper unit (Fig. 10e) is formed mainly by angular locally sourced gravels supported by a sandy-silt matrix belonging to the slope's current scree. The general glacial recession from La Margineda predated the deposition of the upper unit (unit 3; Fig. 10e). We were able to date this last unit in an existing road cutting 100 m south of the one described above. Here, unit 3 is very coarse and may come from a rock avalanche. We could date the matrix of this rock avalanche deposit using 14 C AMS. The age of unit 3 is 16.018 ± 0.183 ka cal BP (sample LM2b; Table 3). These stratigraphic units belonging to MIS 2 are also identified 650 m southward from La Margineda, in Aixovall (Fig. 3, sector 3) from an existing open pit excavation (Fig. 11). Complementary field descriptions are provided in Supplementary Material 6.

The Sant Julià de Lòria outcrop
The village of Sant Julià de Lòria (900 m) is located in a V-shaped valley (Fig. 12). Here, the narrow Valira valley is a fluvial gorge carved in calcshists running southward from La Margineda. Cross-stratified sands were dated using TL and OSL (Fig. 13a). When error bars are considered, the dates from different methods converge: the TL age is 32.8 ± 1.2 ka (STJULIA-1; Fig. 13b, Table 2), and the OSL age is 36 ± 3 ka (STJULIA-2; Fig. 13c, Table 1). Geotechnical boreholes for a new bridge connecting a local road to the highway (Fig. 13d, profile) allowed us to identify a glacial diamicton 6 m beneath the Valira riverbed and beneath the studied exposure. The dated melt-out till (Fig. 13d, Till 2) should be younger than the diamicton located beneath the riverbed (Fig. 13d, Till 1). A previous southward terminal glacial-front position is deduced by erratics accumulated on the eastern margin of the valley at Cal Tolse (Fig. 12) and farther down the valley (Fig. 3), close to the Spanish-Andorran border (Bastida de Ponts glacial front; Fig. 3), but in an inner position regarding the LMIE far-flung moraine (Barr and Lovell, 2014) at Pont Trencat (Turu et al., 2007(Turu et al., , 2017 Fig. 3). Further field descriptions can be found in Supplementary Material 6.

Results from the upper Segre
This section refers to three exposures, one close to the River Segre and two in the Cerdanya depression. a Calibration using IntCal20 (Reimer et al., 2020), δ 13 C carbon isotopic fractionation. asl, above sea level; BK, bulk organic matter; C, colluvium; Of, organic fraction.

The upper Segre at Adrall
The exposure here is located 4.5 km below the Valira-Segre confluence (Fig. 1, sector 3). In Adrall (635 m asl), a 75-m-long outcrop was opened in 1986 during the widening of the N260 highway; however, it was not until 30 years later that the fluvial deposits were studied (Turu and Peña-Monné et al., 2006) and were shown to belong to the glaciofluvial Segre-Valira terrace system (Turu et al., 2007;Peña-Monné et al., 2011). The lowermost terrace level at Adrall (+38 m arb) incorporates three fining-upward sequences, each starting above an erosive surface (Fig. 14). These sequences grade upward from granite-rich gravels of Segre origin to schist-rich deposits of local derivation (Fig. 14).

The upper Segre at Sanavastre
In the southern part of the Cerdagne basin (Fig. 1, sector 4), the Abellorols quarry in Sanavastre provided appropriate outcrops for OSL sampling in cross-stratified sandy layers surrounded by lenses of large cobbles. Samples from two levels (SANAV-1 and SANAV-2; Fig. 4b and c, Table 1) proved to be older than 200 ka. Available field descriptions are provided in Supplementary Material 6.

The Ur-Llaurar rest area
The Ur exposure is located 3 km above the Rahur-Segre confluence, on the eastern side of the N20 highway at the French village of Ur (Fig. 1, sector 4). An ancient quarry from the French N20 highway widening work in the mid-1980s was restored in a rest area. Exposures reveal several units (Fig. 15) of glaciofluvial channel deposits (units 1, 2, 4, and 5) and diamictons (units 3 and 6) belonging to an old glaciofluvial terrace from the Segre-Rahur river system (Fig. 4). The lowermost layer, unit 1, is a gravelly layer with trough cross-bedding. In contrast, unit 2 is a pebbly sand body with some imbricated rolled cobbles in an ochrous sandy matrix and with a ribbon geometry. It was partly eroded before or during the emplacement of the overlying units 3 and 4. An intercalated sand layer was dated to 103 ± 13 ka (UR-1; Table 1). Unit 3 is a massive diamicton with angular boulders, and unit 4 is composed of cross-stratified gravels with intercalated sheet sands. Unit 5 is mainly formed by gravels following a trough cross-bedding pattern, with a maximum thickness of several tens of meters. Unit 6 corresponds to a diamicton that includes striated clasts (Calvet et al., 2013).

Correlation between valleys
Based on AMS 14 C dates from kame deposits and in situ 10 Be CRE ages from both erratic boulders and moraines, the LGM phase is now well-defined in the eastern Pyrenees (Delmas, 2005;Delmas et al., 2008;Pallàs et al., 2010;Palacios et al., 2015;Turu et al., 2017). However, the number and the magnitude of glacial recession events during the LGC, their chronology, and those of the PGC still remain obscure.

The western area
The Sant Mamés melt-out till (73.9 ± 4.5 ka; VILALLER-2; Table 1) provides information about the upper age limit for the LMIE phase, belonging to the end of the MIS 5a-MIS 4 interval. The distribution of erratic boulders over the slopes in Vilaller (Mey, 1965;Vilaplana, 1983aVilaplana, , 1983bRodés, 2008), as well as a few remains of kame terraces close to the Noguera de Tor confluence, indicates that the tributary glacier joined the main glacier (Noguera Ribagorçana) during the LMIE and the resulting glacial front finished at the village of Pont the Suert. Unfortunately, no end moraine could be identified there. However, the former glacial front should be close to its outwash, which is the glaciofluvial terrace located south of Pont de Suert (Fig. 2).
Once the Noguera Ribagorçana glacier retreated from the LMIE position, it emplaced the southernmost till identified from outcrops in the study area that was 2 km south of Vilaller (Fig. 2); this is the Sant Antoni till (∼940 m) (Vilaplana, 1983a(Vilaplana, , 1983bBordonau, 1992;Pallàs et al., 2006). Located ∼15 m above the Noguera Ribagorçana riverbed, it includes abundant rounded to subangular granodiorite boulders ≈1 m in diameter (Pallàs et al., 2006 and references therein). However, the absolute age of the Sant Antoni till is still unknown.

The eastern area
In the Valira valley, the CMARG-4 (91 ± 9 ka; Table 1) sample gave an age within MIS 5c for the Valira glacier terminal complex at La Margineda, which is possibly equivalent to the Ur glaciofluvial terrace (103 ± 13 ka, MIS 5c; UR-1; Table 1). The early Würm (MIS 5c) glacial front of La Margineda (Fig. 10) is 4.6 km upstream from the Cal Tolse Rissian terminal complex (Fig. 3); thus, in Cerdagne, the corresponding T3 end moraine would have been located upstream of the Ur locality and would  have been surpassed by later glacial advances forming the M1 and M2 close-nested moraine complex (Barr and Lovell, 2014) of the River Querol (Pallàs et al., 2010;Fig. 4).

DISCUSSION
We found some key deposits from the Segre River tributaries that are beyond the range of the K-feldspar OSL dating method (Fig. 16), always weathered, and mainly located beyond the limits of the maximum Würmian glaciation (LGC) (Turu and Peña-Monné, 2006;Lewis et al., 2009;Jalut et al., 2010;García-Ruiz et al., 2013;Oliva et al., 2019). An example of this is in Cerdagne, where moraine M3 was assigned to the middle Pleistocene (Calvet et al., 2013), and we now correlate it with the older than 200 ka Sanavastre terrace (Fig. 17a, T4).
Deposits that are beyond the range of the Qtz-OSL dating method (>80 ka) are found beneath the LMIE glacial front limits and show weathering according to their ages (Rixhon, 2022). The best example in our area corresponds to the multi-glacial cycle nested moraine of La Margineda, where the early Würm end moraine (+65-75 m arb; 91 ± 9 ka, MIS 5b-5c; CMARG-4; Fig. 10e; Table 1) was surpassed by later glacial advances (Fig. 17c). In the NE Pyrenees (Niaux Cave, NE Pyrenees, 678 m; Fig. 1) an early Würm glacial progression has also been dated using U-Th methods (Bakalowicz et al., 1984), leading to the conclusion that the Ariège glacier was in the vicinity of the cave at ∼95 ka, within the MIS 5d-5c interval (Sorriaux et al., 2016), 13 km upstream from the MIS 6 terminal complex (Delmas et al., 2011). The onset of the LGC was probably synchronous on both sides of the Pyrenees because a similar age comes from the glaciofluvial terrace at Ur (Fig. 1, sector 4).
Unexpectedly, some of the deposits within the MIS 3 range of the AMS 14 C-dating method (≈45 ka) are found in an inner position within the glacial valley far from the LMIE and LGM glacial front limits. We refer to these deposits as resulting from an ice or glacial thinning recession within the LGC and corresponding to the relevant palaeoenvironmental evidence from the northern foothills of the Pyrenees, which were occupied by the Neanderthals during the Châtelperronian, spanning from ca. 44.5 ka cal yr BP to 38.5 ka cal yr BP (Bricker, 2014). The archaeological site of Tambourets is only a few kilometres beyond Barbazan (Fig. 1), where a close-nested end moraine from the Garonne glacier is located (Andrieu et al., 1988;Fernandes et al., 2021). Bricker (2014) described archaeological remains within loess-type deposits from the beginning of Heinrich Event 4 (GS-10; Rasmussen et al., 2014). However, a brief episode of milder climate (Bricker, 2014) during GI-9 (Rasmussen et al., 2014) preceded the resumption of severe cold and dry conditions during stadial GS-9 (Rasmussen et al., 2014). We attempt to correlate this to the available sedimentary record from the southern slope of the Pyrenees as follows: 1. The highest flooding level in Ginebrosa (Fig. 8, unit 3), recorded after 42 ± 8 ka, may be related to an increase in icedamming that we assign to stadial GS-11 (42,240-41,460 b2k;Rasmussen et al., 2014). Unit 4 (Fig. 8) indicates that the ice tongue thinned rapidly, probably during GI-10 (41,460-40,800 b2k; Rasmussen et al., 2014). 2. The Sia (SW Pyrenees; Serrano et al., 2011) and Gavin lateral moraine complex (Fig. 1) was built (Fig. 17e) Fig. 1). The erstwhile Ara glacier thinned shortly after unblocking the Linas de Broto ice-dammed palaeolake (Sancho et al., 2018;Bartolomé et al., 2021). Daura et al. (2013) indicated a significantly arid climate during GS-9 in NE Iberia, matching the cold and dry climate reported at Tambourets by Bricker (2014) and the recession mentioned above of the Gállego, Ara, and Valira glaciers. At the end of MIS 3 (ca. 39.7-34.2 ka cal BP; Turu et al., 2017), the retreat of the Valira d'Orient glacier at Canillo (NE Andorra; 1530 m asl; Fig. 1) may have been responsible for an early triggering of the El Forn landslide (Corominas et al., 2015). Between GI-10 and GI-7 (Rasmussen et al., 2014), glaciers receded synchronously at both extremities of the southern slopes of the Pyrenees.
In La Margineda (Fig. 10c), a set of colluvia (26.65 ± 0.3 ka cal BP; LM1; Table 3) were consolidated by the regrowth of the Andorra glacier when it reached its LGM positions 1 km downstream of La Margineda (Fig. 11). An early LGM (Clark et al., 2009) was registered in the southern slope of the Pyrenees by the buildup of the Tinabre lateral moraine (Fig. 2), and it is also evident from the buildup of the innermost lateral moraine ridge from Gavin (Fig. 17e) at 28 ± 3 ka (sample X1595; Supplementary Material 8). However, the LGM period (Clark et al., 2009) was quite unstable in the southern slope of the Pyrenees because the glacial recession and regrowth produced reworking and deformation of the sediments in the La Massana palaeolake (Turu et al., 2017); slope instability in Andorra (VB05-30.2, 26.010-25.828 ka b2k, 21.52 ± 0.070 ka BP; Planas and Torrebadella, 2022); moraine ridge duplication (Tomkins et al., 2021), as in the case of the Artigallonga moraines (Fig. 2); and the recurrence of ice-damming at Llestui (Fig. 2). Evidence from the late LGM period was also reported upstream Figure 17. Classification of the glacial terminal complexes from the south-central and southeastern slope of the Pyrenees, including, for comparison, a case from the SW Pyrenees and the north-central slope of the mountain range. (a) Querol terminal complex formed by nested end moraines (Pallàs et al., 2010) and the related glaciofluvial system of terraces (Poch et al., 2013). Terrace T3 onlap, the oldest end-moraine M3 correlated with the oldest glaciofluvial level (T4). (b) The Garonne terminal complex, formed by close-nested and nested end moraines (Stange et al. (2014) and Fernandes et al. (2017)). The oldest ages from the outer end moraine are from MIS 6; however, younger ages were also reported by Fernandes et al. (2021) in this terminal complex that can be correlated with the glaciofluvial terraces (Bricker et al. 2014) and the chronology of lake Barbazan (Andrieu et al., 1988) and their surrounding end moraines. (c) The Valira terminal complex is the most diverse case (Turu and Peña-Monné, 2006). The Valira glacier partially eroded MIS 6 end moraines during the LGC, building multifold-nested and close-nested moraines (Turu et al., 2017). Beyond, multiple far-flung end moraines correlated with the Segre-Valira staircase system of glaciofluvial terraces (Turu et al., 2007). (d) The Gállego terminal complex centred around Senegüé (Turu et al., 2007), the bestpreserved end moraine (Lewis et al., 2009). Unpublished dates (Supplementary Material) from the Gavin lateral moraine ridges (Serrano et al., 2011) allow the reconstruction of glacial evolution behind and beyond Senegüé . (e) The Ribagorçana terminal complex around its best-preserved end moraine (Mey, 1965;Vilaplana 1983a;Bordonau 1992), the Seminari de Vilaller (Bordonau et al. 1993;Pallàs et al. 2006). Calibrated and recalculated dates (this work) from the lateral moraine ridges and kames behind the end moraine (Delmas et al. 2021b). Glacial front in Pont de Suert (reconstructed). The ages of the glaciofluvial deposits of Sextas (Fig. 1) by Gonzalez-Sampériz et al. (2006), and for the Santa Coloma end moraine (Turu et al., 2007(Turu et al., , 2017Fig. 9).
Glacial tongue asymmetries within the LGM period (22.25 ± 4.25 ka; Clark et al., 2009) and between the LMIE and LGM phases occur between Pyrenean valleys; only in Cerdagne does the close-nested end moraine (Fig. 4, M1 and M2) from the Querol glacier seem to span from MIS 4 (Delmas et al., 2008;Calvet et al., 2011a) to MIS 2 (Pallàs et al., 2010), highlighting that the LMIE and the LGM had similar glacial extents in the southeasternmost slope of the Pyrenees (Fig. 17a). However, this is not seen in the south-central Pyrenees, where an LGM end moraine (Fig. 17d) in the Seminari de Vilaller is far from the LMIE limits at Pont de Suert (Fig. 2). Subsequently, the final glacial retreat occurred toward the end of the last termination. In the Noguera Ribagorçana (Fig. 17d), it is represented by the end of the infilling of the overdeepened trough at around 15.7 ± 1.2 ka (VILALLER-1; Table 1), when the local glacial fronts were located at Santet and Bissiberri (Pallàs et al., 2006;Fig. 2;10 Be exposure ages recalculated, Supplementary Material 7). However, in the neighboring eastern valley, the Noguera de Tor (Fig. 2), the disappearance of cirque glaciers took place at 16.4-16.0 cal ka BP (Copons and Bordonau, 1996), and only at ≈2200 m did small glaciers survive in this valley until 12.05 ± 0.4 ka (Tomkins et al., 2021). In the Valira d'Orient valley (Fig. 3), the glacial thickness was less than 250 m at 15.3 ± 1.1 ka, the timing of the triggering of the Encampadana sackung (1625 m asl; McCalpin and Corominas, 2019; Fig. 1). The retreat of the Valira d'Orient glacier from Canillo preceded the second triggering of the Forn landslide at 13.144 ± 0.175 ka cal BP (Turu and Planas, 2005; Fig. 1, sector 3) and the blockage of the Valira d'Orient valley at Canillo (1530 m asl).

CONCLUSIONS
We distinguish between two types of terminal complexes, those in which there is at least one far-flung end moraine ( Fig. 17d and e) and those in which there is not (Fig. 17b); however, both include nested moraines and/or a close-nested end-moraine complex ( Fig. 17a and b). Close-nested end-moraine complexes encompass at least two glacial cycles: the LGC and the penultimate glacial cycle (Fig. 17b), and these moraines could be termed "multiple nested moraines" (Fig. 17c). Of note, the end moraines from the penultimate glacial cycle (MIS 6) are common in the whole range of the Pyrenees and were formed during the Western Mediterranean Humid Period 6 and 5.
The starting point for cooling in the LGC was generally at ∼97 −15/+19 ka (Niaux, UR-1; Fig. 16), in MIS 5d, synchronous on both sides of the Pyrenees, and we refer to the early Würm (Fig. 16, EW). However, the MIS 5d Valira glacier front (Fig. 17c) was far from the LMIE far-flung moraine, and this could also be the case for other glaciated valleys in the Pyrenees (Fig. 17a).
The LMIE advance occurred at the beginning of MIS 4 (Fig. 16, VILALLER-2), after an early glacial recession (Fig. 16, EGR) after/following MIS 5d. Literature consultation enabled us to compare the glaciated Noguera Ribagorça valley with its neighbouring valleys, allowing the identification of a significant glacial thinning (Fig. 16, GTP) and glacial front retreat during the second half of MIS 3 (Fig. 16, GINEBROSA-1), during which glaciers thinned and receded or almost disappeared (no large valley glacial period [NLVGP]; Fig. 16). Evidence of this glacial ice thinning pertains to the Gállego glacier (SW Pyrenees) and almost disappeared in the Valira glacier (SE Pyrenees). It probably affected both slopes of the Pyrenees. However, the duration of the NLVGP was short on the eastern side of the mountain belt. Most of the nested moraines within the terminal complexes were formed during the latest part of MIS 3 (Fig. 16, STJULIA-2) and progressed during the LGM period of MIS 2 (Fig. 16, LM-1).