The deglaciation of Upernavik trough, West Greenland, and its Holocene sediment infill: processes and provenance

Under glacial climates, continental ice sheets such as, e.g., the Greenland Ice Sheet, extended onto the continental shelves and often carved out deep cross‐shelf troughs. The sedimentary infill of such troughs commonly is a product of the complex interactions between the ice sheets, largely driving sediment input into the ocean, and the surrounding water masses. Off West Greenland, research has focused on the Disko and Uummannaq troughs, leaving the northerly adjacent Upernavik trough relatively understudied. Hence, neither the chronology of deglaciation nor the details of its postglacial infill are sufficiently well understood. Here, we combine computed tomography image‐derived information with geochemical and granulometric data from four sediment cores recovered from the Upernavik trough that point to (i) deglaciation of the mid‐shelf probably around 13.4 cal. ka BP that was most likely driven by a northward advection of warmer Atlantic waters during the Bølling–Allerød, (ii) the presence of widespread mass wasting around 8 cal. ka BP on the inner shelf and (iii) the complex interplay between various modes of sediment input, transport and deposition under hemipelagic sedimentation afterwards. While this interplay complicates provenance studies, we identify two major sediment delivery mechanisms that control transport and deposition from four sediment source areas. Through the Early Holocene the relative contributions of sediments from the various sources changed from a predominantly local origin to more southerly sources, mainly driven by decreasing input from the local sources. The integration of relative sediment source contributions with varying sedimentation rates challenges previous studies postulating intensified sediment delivery from the south through a greater influence of the West Greenland Current and highlights the need for the integration of sediment input and transport mechanisms into provenance studies in the area.

The glaciated margin surrounding Greenland was shaped through the large fluctuations of the Greenland Ice Sheet (GrIS) during past glacial-interglacial cycles (Aksu & Piper 1987;O Cofaigh et al. 2013b;. It is characterized by deep cross-shelf troughs, extending to the shelf break and bathymetric highs, the shelf platforms, between them (Batchelor & Dowdeswell 2014). The cross-shelf troughs off west Greenland are efficient sediment traps and proximal depocentres, making them well suited for studying both oceanographic and cryospheric environmental change in Baffin Bay at high temporal resolution Jennings et al. 2014). Having been carved out by repeated ice advances during past glacials, when the ice margin of the GrIS extended to the shelf break, they hosted highly dynamic ice streams, whose decay was closely linked to the inflow of relatively warm waters via the West Greenland Current (WGC; Sheldon et al. 2016). Especially given (i) the drastic changes that the GrIS has experienced in the past decades and will experience in the future (Rignot et al. 2010;Straneo & Heimbach 2013;Briner et al. 2020) and (ii) the ongoing and potential perturbations of the Atlantic Meridional Overturning Circulation (AMOC; Rahmstorf et al. 2015;Bakker et al. 2016;B€ oning et al. 2016), which largely drives the WGC, a refined comprehension of the palaeoenvironmental development in Baffin Bay is needed. As this is manifested in the sedimentary archives that have accumulated in the shelf troughs, unravelling and understanding their depositional history is a fundamental prerequisite for a reconstruction of the varying impacts of ocean currents and continent/ice-derived sediment inputs, which are mostly driven by climatic change from the deglaciation through the Holocene Thermal Maximum into the Neoglacial (e.g. Weidick et al. 2012;Briner et al. 2013;McKay et al. 2018).
Previous studies of postglacial sedimentation and environmental history focused predominantly on the Uummannaq ( O Cofaigh et al. 2013a, b;Roberts et al. 2013;Lane et al. 2014;Sheldon et al. 2016;Jennings et al. 2017;Philipps et al. 2017), Disko (Hogan et al. 2011(Hogan et al. , 2016O Cofaigh et al. 2013b) or Melville Bay troughs (Freire et al. 2015;Slabon et al. 2016Slabon et al. , 2018a, establishing chronologies for the deglaciation as well as outlining the general postglacial sediment infill. In between lies the still poorly investigated Upernavik trough that is fed by one of the major ice streams in West Greenland, the Upernavik Upernavik trough is limited to radiocarbon-dated sediment cores from proglacial threshold lakes and 10 Be-datings of erratic blocks from the northern margin of the Upernavik fjord that indicate the retreat of the icemargin beyond modern limits by 9.9 ka BP ) and combined 10 Be and 26 Al-measurements on adjacent bedrock-boulder pairs constraining the deglaciation in the southern Upernavik area to around 11.3 ka . Investigations of postdeglacial sediment transport and deposition within Upernavik trough are based on one single sediment core (Caron et al. 2018;Giraudeau et al. 2020) and revealed that sedimentation was dominated by deglacial melt in the Early Holocene and increasingly controlled by oceanic sediment delivery from southern sources over the course of the Holocene. However, neither the timing of the troughs' deglaciation nor the details of the processes governing the sediment infill, especially in the Middle to Late Holocene, along the length of the trough are currently understood or documented.
With this study, we present high-resolution records of sedimentological and geochemical variability in four sediment cores in a proximal to distal transect along the Upernavik trough. The data allow us to (i) elucidate the timing of deglaciation of the Upernavik trough, (ii) identify numerous depositional sedimentary processes and (iii) reconstruct variations in sediment provenance, both temporally and spatially. We highlight the complex hemipelagic sedimentation patterns within polar crossshelf trough systems and discuss potential consequences for their interpretation.

Study area
The Upernavik trough is one of the main cross-shelf troughs on the West Greenland shelf running approximately perpendicular to the coast (Nielsen et al. 2012;Briner et al. 2013). It is overdeepend, i.e. deeper at the inner (>900 m water depth) than at the outer shelf (<800 m water depth; Slabon et al. 2016Slabon et al. , 2018aFig. 1). Its sedimentary infill is primarily derived from glacial erosion of the surrounding landmasses. The composition of these sediments depends on the bedrock geology of the surrounding landmasses, which generally comprises rocks from the Precambrian Shield, mostly Archaean to Proterozoic in age. The oldest geological unit cropping out in our study area is the early Proterozoic schists of the Karrat Group, which are mainly present around the Upernavik trough (e.g. Kalsbeek et al. 1998). The Karrat Group metasediments were intruded by Charnocktites (orthopyroxene-bearing granites; Escher & Pulvertaft 1995;Henriksen et al. 2009) during the Precambrian (Henderson & Pulvertaft 1967;Kalsbeek et al. 1998), which now make up large parts of the Prøven Igneous Complex (Thrane et al. 2005) just south of Upernavik. Rifting during the late Cretaceous and early Tertiary produced the Palaeogene Nuussuaq Basin volcanics (Escher & Pulvertaft 1995;Henriksen et al. 2009;Larsen & Pedersen 2009) in central and northern Disko Bay.
The oceanic regime in the Upernavik area is controlled by the WGC that flows north along the shelf in intermediate depths carrying a mixture of cool, relatively fresh water originating from the East Greenland Current and warmer, more saline water originating from the Irminger Current (the West Greenland Intermediate Water,~200-800 m). The WGC is typically overlain by a thin lens of less saline, cold water, originating mainly form melting (the Arctic Water, AW). Below 800 m water depth these water masses are underlain by cold, but more saline Baffin Bay Deep Water (Tang et al. 2004).

Material and methods
Sediment cores GeoB19920-1, GeoB19927-3, GeoB 19969-1 and GeoB19973-2 were raised from the Upernavik trough off West Greenland during RV 'Maria S. Merian' expedition MSM44 in July 2015. The core sites are approximately oriented along a transect perpendicular to the coast with GeoB19973-2 representing the central shelf, GeoB19920-1 and GeoB19927-3 representing the over-deepened inner shelf and GeoB 19969-1 representing the setting close to the present-day coast (Table 1, Fig. 1).

Stratigraphy
The chronostratigraphies of the sediment cores used in this study are based on Accelerator Mass Spectrometry (AMS) 14 C-datings of mixed benthic foraminifera from the >100 lm fraction or mollusc shells. The AMS measurements of the foraminifera were performed directly on the CO 2 gas with the MICADAS (Mini Carbon Dating System) at the Alfred Wegener Institute in Bremerhaven, Germany (for methodological details see Wacker et al. 2013) and at the Laboratoryof Ion Beam Physics at the ETH Zurich. The mollusc shells were measured at the Poznan Radiocarbon Laboratory, Poland. We present new age models for all of the cores used in this study. The basal age from core GeoB19920-1 was previously published by Slabon et al. (2016), while Saini et al. (2020) published an age model for core GeoB19927-3 using 12 of the 14 radiocarbon ages presented here and radionuclide analyses of the core top as well as a different modelling approach. We are updating the existing age model using two additional ages in the lower part of the core (1053 and 1117 cm core depth).
Choosing the right approach for calibration and reservoir correction is an ongoing matter of debate (see below), but here we opted to employ the Marine13 calibration curve (Reimer et al. 2013) with an additional reservoir correction of DR = 140AE30 a using Calib v7.0.4 for the calculations for the sake of comparability with previously published ages from the region. Subsequently, each of the age models was constructed in accordance with the lithological structure of the respective sediment cores, combining Bayesian age-depth modelling performedwith the open-source software package Undatable (Lougheed & Obrochta 2019) and linear extrapolation beyond the intervals constrained by radiocarbon dates. The units that were characterized as mass-wasting deposits, transitional unit or glacial till (see Results and Discussion) are considered as near-instantaneous Map of the study area and core locations. Left panel shows overview of Labrador Sea and Baffin Bay with locations mentioned in the text as well as the core locations and the relevant hydrography: relatively warm West Greenland Current (WGC) in red, cooler Baffin Current (BC) in blue. DS, Davis Strait; DT, Disko trough; UmT, Uummannaq trough; UpT, Upernavik trough; MB, Melville Bay; NS, Nares Strait. Panel on the right shows a closeup of the study area with colour-coded extents of the prevalent geological provinces (after Henriksen et al. 2009), glacier velocities in red (Joughin 2017) and core locations. Core site labels abbreviated (see Table 1); '204' refers to core AMD14-204 (Caron et al. 2020). Map data: GEBCO (GEBCO Compilation Group 2020), IBCAOv4 (Jakobsson et al. 2020), TanDEM-X (German Aerospace Center (DLR) 2018) and BedMachine v3 (Morlighem et al. 2017). deposits and are therefore given one 'event age' equivalent to the basal age of the overlying hemipelagic units. Following the most conservative approach, sedimentation rates are therefore only calculated between the age constraints within the hemipelagic units (Table 2). While this approach is in overall agreement with the majority of studies from the eastern Baffin Bay O Cofaigh et al. 2013a;Perner et al. 2013;Jennings et al. 2014;Ouellet-Bernier et al. 2014;Hogan et al. 2016;Sheldon et al. 2016;Jackson et al. 2017;Saini et al. 2020), we stress that larger reservoir ages are to be expected off Greenland owing to local effects of sea ice, wind stress and freshwater fluxes impacting on the oceanic 14 CO 2 uptake, which are not considered in the global marine calibration curves (Heaton et al. 2020), questioning their applicability here. An alternative approach might be found in the use of the modelled reservoir ages of Butzin et al. (2017). The resulting very high reservoir ages, however, lead to very young calibrated ages, especially in the deglacial and Early Holocene sections of the cores, which we found to be incompatible with surrounding bathymetric and seismic evidence (see Discussion).

Granulometry
The disaggregated siliciclastic grain-size distributions of cores GeoB19927-3 and GeoB19973-2 were determined in the Particle-Size Laboratory at MARUM, University of Bremen with a Beckman Coulter Laser Diffraction Particle Size Analyser LS 13320, according to the protocol outlined by, e.g., McGregor et al. (2009), Bartels et al. (2017) and Weiser et al. (2021). This includes the chemical removal of organic carbon, calcium carbonate and biogenic silica as well as chemically disaggregating any remaining aggregates prior to the measurement. The use of deionized, degassed and filtered water reduces influence of gas bubbles or particles within the water.
Reproducibility of the results is checked regularly by replicate analyses of three internal glass-bead standards and is found to be better than AE0.7 lm for the mean and AE0.6 lm for the median particle size (1r). The average standard deviation integrated over all size classes is better than AE4 vol% (note that the standard deviation of the individual size classes is not distributed uniformly).
Sortable silt mean grain size (SS) was calculated following the protocol proposed by McCave & Andrews (2019b) outlining the applicability of laser scanner-derived grain-size data for sortable silt calculations.

Computed tomography
Archive halves of the sediment cores GeoB19920-1, GeoB19927-3, GeoB19969-1 and GeoB19973-2 were scanned using a Philips Brilliance iCT Elite 256 computer tomograph (CT) at the hospital Klinikum Bremen-Mitte, with an X-ray source voltage of 120 kV and a current of 300 mA. The CT image stacks have a physical resolution of 0.293 mm in the x and y directions and 0.625 mm resolution in the z direction (collimation; 0.3 mm reconstruction interval). Images were reconstructed using the filtered Back Projection (fBP) mode and a bone kernel (YB (Enhanced)).
The obtained CT data were processed using the ZIB edition of the Amira software (version 2021.08; Stalling et al. 2005) and followed in general the approach of Bartels et al. (2017). Within Amira, the CT scans of the core sections were merged and core liners, including about 2 mm of the core rims, were removed from the data set. High-density constituents >~1 mm, open (air-or water-filled) bioturbation traces and the matrix sediment were segmented with the Segmentation Editor and the marker-based watershed algorithm. The high-density constituents comprised lithic clasts (predominantly icerafted detritus) and lithified/pyritized bioturbation traces. Separation of these two constituents is based on object shape parameters (minimum length, >2 mm; ratio of maximum inner radius/maximum diameter <0.5), effectively separating the spherical clasts from the more tubular bioturbation traces.
Additionally, the X-ray attenuation of the matrix sediment density and its standard deviation (measured in Hounsfield unit (HU)) as proxy for wet bulk density was determinedwith the Material Statistics module (statistics per slice and label) after reducing the matrix sediment label by two voxels to exclude potential marginal effects. The data were subsequently grouped into 10 mm bins with the NumPy (Harris et al. 2020) and pandas (McKinney 2010; The Pandas Development Team 2020) cut and GroupBy-methods.
X-ray fluorescence X-ray fluorescence (XRF) core scanner data were collected every 1-2 cm downcore over an active area of 15 mm 2 with a downcore slit size of 10 mm using generator settings of 10 and 30 kV directly at the split core surface of the archive half with a XRF Core Scanner II (AVAATECH Serial No. 2) at MARUM, Bremen. The split core surface was covered with a 4-lm thin SPEX Certi Prep Ultralene foil to avoid contamination of the measuring unit and desiccation of the sediment. Raw results were processed by the analysis of X-ray spectra with the Iterative Least Squares software package (WIN AXIL) by Canberra Eurisys.
A total of 25 elements were measured during the XRF scanning. Nine elements (Al, Si, K, Ti, Fe, Zr, Zn, Ca and Sr) were considered reliable while 13 were deemed unreliable owing to low counts (Ni, Cu, Zn, Ga, Y, Nb, Mo, Bi, P, Cr and S) or interferences (Cl, Rh). Ca and Sr were excluded from the further analysis to avoid the influence of biogenically produced carbonates. However, BOREAS they were checked for the potential presence of Baffin Bay detrital carbonate layers Simon et al. 2014;Jackson et al. 2017), for which no evidence could be found (Fig. S2). Zn, even though generally considered unreliable, is included because (i) it is the most prominent marker for one of the four provinces used in the provenance analysis (see Results) and used accordingly by Giraudeau et al. (2020) and (ii) because Boxberg et al. (2020) showed that Zn counts from the instrument used in this study compare well with Zn concentrations measured via wavelength dispersive XRF. For the purpose of this study, we are focusing on the ratios of Al/Ti, Fe/K as well as the Zn and Zr counts. Given that XRF scanning is a semiquantitative measurement, the individual element counts were normalized by division with the sum of all counts per sample to improve comparability and minimize the effects from varying porosity, grain size or water content (Tjallingii et al. 2007;Weltje & Tjallingii 2008;Lyle et al. 2012;Bahr et al. 2014;Boxberg et al. 2020). Table 2. Overview of the samples used as age constraints as well as the reservoir correction and calibration procedures. Samples marked with an asterisk were previously published by Saini et al. (2020;GeoB19927-3) and Slabon et al. (2016;GeoB19920-1 Statistical analyses Provenance analysis. -In order to link the sediment core data to the onshore geology, we followed and extended the approach of Giraudeau et al. (2020) in that we used the stream-sediment data compiled by Steenfelt et al. (1998) and Steenfelt (1999Steenfelt ( , 2001 but also added data from whole-rock analyses available via the PetDB/EarthChem portal (PetDB: https://search. earthchem.org/; EarthChem: http://portal.earthchem. org/) containing data from Carter et al. (1979), Nelson (1989), Holm et al. (1993), Larsen and Pedersen (2009) and Larsen et al. (2016). This is done primarily in an effort to mitigate potential sortinginduced biases present in the stream-sediment data, besides simply increasing the statistical robustness of the database through a larger number of data points. A number of data points of these resource contain results for different oxides of Fe (FeO, Fe 2 O 3 ) or (additionally) report the total amount of iron oxides as FeO_tot. In order to avoid overlap between these results and maintain comparability, only the data for Thus, these four provinces are considered here as potential sediment sources for the deposition in Upernavik trough (see Fig. 1).
In the following, the whole-rock and stream sediment geochemical data were linked to the extent of these four provinces in QGIS (QGIS Development Team 2022) using the Intersection-function, which means that each sample was 'labelled' with the province it was taken from.
Core data. -We employed a combined approach using both principal component analysis (PCA) and end member analysis (EMA) to investigate the sediment core elemental data, with both approaches relying on the semiquantitative XRF data. With the PCAwe are able to identify important covariances within the dataset and identify dominant source regions (see Discussion), while the EMA yields compositional data for the contribution from these source regions.
Linear dimensionality reduction of the XRF-data was achieved through a PCA, using Python's machine learning library scikit-learn (Pedregosa et al. 2011;Buitinck et al. 2013) after data standardization (centre log-ratio transformation). The centre log-ratio transformation removes the constant sum constraint from compositional data, which otherwise prohibits the use of correlation or covariance matrix-based statistical analyses (like PCA), while retaining the proper covariance structure (Aitchison 1981(Aitchison , 1982Kucera & Malmgren 1998). Unit-wise confidence ellipses for the first two principal components were calculated after Schelp & St€ oßer (2021).
An EMAwas performed on the XRF elemental data to statistically unmix the sediment geochemical composition into probable source signals. The input data consisted of the elemental counts/total counts, hence fractions from 0 to 1. The elemental ratios shown in Table 3 were derived after the modelling and are not part of the input data. Calculations were performed with the DRS Unmixer (Heslop et al. 2007) using the default value of 0.8 for the expansion coefficient. In an attempt to unmix the four source areas, we initially ran a fourcomponent end member, but the results from a threecomponent EM model were more genetically meaningful (see Results and Discussion).

Provenance analysis
In order to ensure comparability, only elements that are present in the XRF-scanning data from our sediment cores are considered for the geochemical characterization of the potential onshore source areas. Based on previously published compositional data of onshore sediments/rocks (Carter et al. 1979;Nelson 1989;Holm et al. 1993;Steenfelt et al. 1998;Steenfelt 2001;Larsen & Pedersen 2009;Larsen et al. 2016), the Nuussuaq basin volcanics are characterized by relatively low SiO 2 contents and highest Fe 2 O 3 and TiO 2 contents, as well as lowest K 2 O and Zr concentrations. Accordingly, the Al/Ti ratio is lowest of all provinces, while the Fe/K is the highest of all provinces. Samples from Karrat South Table 3. Compositions and derived elemental ratios from the three end-member model calculated from the X-ray fluorescence data of the four studied sediment cores. Values in italics denote comparably low values, while underlined numbers are comparably high. Combined, these allow the linkage of the EMs to onshore source regions. and Zr contents as well as above average Al and Ti contents. It is nonetheless also characterized by a low Al/Ti ratio. The samples from Karrat North are depleted in Ti and Zn, while being enriched in Al and, to a lesser degree, K 2 O and Zr. Accordingly, these display the highest Al/Ti ratios of all samples (Fig. 2).

GeoB19969-1
Core GeoB19969-1 consists of two main lithological units. In the lower unit (>874 cm core depth) the CT image reveals a mottled sediment structure, inversely graded mudclasts, synsedimentary faulting and erosive contacts between two layers of strongly elevated matrix sediment density at 888 and 898 cm core depth, classifying it as a mass-wasting deposit (mwd), specifically as a debris flow deposit in association with turbidites (cf. Shanmugam & Benedict III 1978;Fig. 3).
No indication for bioturbation is observed in this unit, the lithic clast content is around 1 vol% and lithic clast  Steenfelt et al. (1998) and Steenfelt (1999Steenfelt ( , 2001 and the PetDB/EarthChem portals (PetDB, https://search.earthchem.org/; EarthChem, http://portal.earthchem.org/). counts are around 10 clasts cm À3 . The two high-density layers are further marked by elevated Zr counts, while the unit as a whole is characterized by low Fe/K and high Al/ Ti ratios.
Within the upper unit the CT data reveal initially low bioturbation, while shell debris accounts for up to 3 vol % of the sediment (Fig. 3). The disappearance of the shell debris above 450 cm core depth is mirrored by a marked increase in bioturbation, with values partially surpassing 20 vol% classifying these as hemipelagic sediments. Increasing lithic clast counts and volumes further characterize the upper 300 cm, with a pronounced increase in the upper 50 cm of the core. Throughout this unit the Al/Ti (Fe/K) ratio decreases (increases) gradually, while a less pronounced decrease (increase) is visible in the Zn (Zr) counts (Fig. 3).
The age model of core GeoB19969-1 is based on a total of 11 AMS 14 C datings (Fig. 4). Three ages were acquired from the lower unit, constraining the basal age of the core to~8.5 cal. ka BP, but also showing an age reversal at 927 cm core depth, substantiating the classification as mwd. The eight ages from the upper unit get progressively younger towards the core top, indicating continuous hemipelagic sedimentation since~8 cal. ka BP. Fig. 3. Results of the computer tomograph (CT) and X-ray fluorescence (XRF) analyses carried out on core GeoB19969-1. Top panel shows a contrast enhanced, distorted grey-scale image of the core next to records of mean matrix sediment density, lithic clast counts and volumetric sediment composition. The main elemental records used in this study (Al/Ti, Zr, Zn and Fe/K) are shown in grey towards the right, overlain by fivepoint running averages. Note the broken axis for lithic clast counts, sediment composition and Zr record. The column on the far right indicates the depth intervals for the hemipelagic (hemi) section and the mass-wasting deposit (mwd). Black triangles show the depths of radiocarbon measurements. Red boxes in the greyscale image indicate core sections detailed in the bottom panel, which shows undistorted grey-scale images of characteristic core sections next to 3D reconstructions of sediment constituents (orange, lithic clasts; green, bioturbation; red, shell debris; multicoloured, segmented, individual mudclasts).

Fig. 4.
Age-depth models for the studied cores. Each dated sample is shownvia its probability density function (pdf), which is subsequently used for the modelling. Grey intervals show 68.2% confidence intervals (CIs); these are overlain by colour-coded line plots showing median ages for each core. Stippled lines indicate extrapolated sections. Error bars for GeoB19973-2 indicate modelled 1-and 2r ranges for extrapolated ages at 597 cm (top of transitional unit) and 657 cm (top of till). Bottom panel shows sedimentation rates calculated for hemipelagic, non-extrapolated sections, overlain by 11-point running averages. Sedimentation rates are highest between~5 and 7 cal. ka BP (>150 cm ka À1 ) and slightly lower throughout the rest of the record (~100 cm ka À1 ; Fig. 4).

GeoB19927-3
Core GeoB19927-3 consists of two main lithological units. In the lower unit below 804 cm core depth a chaotic sediment structure, significant sediment deformation (slump folds, fluid escape structures) and several high-density layers are evident in the CT data (Fig. 5), classifying this unit as a mwd. This unit is barren of traces of bioturbation, but enriched in lithic clast, with highest values exceeding 4 vol% or 20 clasts cm À3 below 1050 cm and exhibits a lithic clast-rich layer between 930 and 940 cm (~20 vol%). Geochemically, this unit is characterized the overall highest Al/Ti ratio and several pronounced Zr excursions (820, 1080 and 1140 cm core depth, marking the high-density layers), as well as low, yet increasing Zn counts and Fe/K ratio. The overall trend is interrupted by positive (negative) excursions in the Al/Ti (Fe/K) ratios at 920 cm core depth (Fig. 5).
The upper unit (<804 cm) shows a generally homogeneous, yet bioturbated sediment structure typical for hemipelagic sediments with lithic clast counts remaining below 4 clasts cm À3 (Fig. 5). Bioturbation increases to >2 vol% between 200 and 50 cm core depth. The Al/Ti ratio shows a continuous, albeit slight decrease that levels out above 300 cm. This is paralleled by a slight increase in Zn counts. A more pronounced change becomes apparent in the Fe/K ratio, which shows a notable increase above 750 cm that levels out at maximum values above 300 cm. Zr shows a very slight decrease throughout the unit. Within the unit, a turbidite layer at~775 cm was identified in the CT data.
The age model of core GeoB19927-3 is based on a total of 14 AMS 14 C datings (Fig. 4). Three ages were acquired from the lower, mass-wasting unit, constraining the basal age of the core to >8 cal. ka BP, but also exhibiting an inconclusive age pattern and an age reversal at 1053 cm core depth, substantiating the classification as MWD based on the CT-data described above. The 11 ages from the upper unit get progressively younger towards the core top, indicating continuous hemipelagic sedimentation since~8 cal. ka BP. Sedimentation rates are variable, but show an overall decreasing pattern from elevated values in the order of 150 cm ka À1 prior to 3 cal. ka BP to <50 cm ka À1 in the late Holocene (Fig. 4). The turbidite at~775 cm is excluded from the age model, leading to a hiatus of~200 years, which is supported by the CT data showing its erosive base, i.e. missing sediment.
The grain-size data classify the majorityof the material as fine to medium silt. Below~1000 cm and at~775 cm core depth coarser sediments (coarse silt or fine to medium sand) are recovered. These are polymodal, showing a strong fine (negative) skew and a modal grain size >100 lm. The sediments above~1000 cm core depth are finer than 200 lm, mesokurtic and polymodal with a main modal grain size around 6 lm (Fig. 6).

GeoB19920-1
Core GeoB19920-1 consists of three lithological units. The lowermost unit (>1087 cm core depth) shows a homogeneous, yet slightly bioturbated (0.2 vol%) sediment structure with lithic clast counts remaining below 1 clast cm À3 (Fig. 7). Given that this hemipelagic unit is only~20 cm thick, we refrain from describing trends within the elemental records.
The middle unit (1087-690 cm) comprises a fairly lithic clast-rich section that is free of bioturbation, but shows increasingly abundant mudclasts (50-100 vol%) with inverse grading (1087-950 cm) and a very homogenous layer on top (950-690 cm) that neither features signs of bioturbation nor contains lithic clasts; only rarely small were synsedimentary faults observed (Fig. 7). The bottom part of this unit is characterized by several excursions (both positive and negative) in all geochemical markers that rarely appear synchronously, while all markers remain stable in the upper, homogeneous part. Accordingly, this unit represents a masswasting deposit, specifically a stacked debris flow deposit that is overlain by the debris flow 'tail' (i.e. the suspension cloud), possibly in conjunction with turbidites.
The upper unit consists of homogeneous, yet bioturbated hemipelagic sediments with elevated lithic clast counts (>1 clast cm À3 ) between 690 and 520 cm (Fig. 7). The section between 220 and 40 cm core depth is strongly bioturbated with values partially surpassing 10 vol%. Geochemically, the onset of the upper unit is marked by a sharp increase in the Fe/K ratio between~670 and 620 cm. This increase continues towards the core top, albeit less pronounced. Al/Ti (Zn) shows a gradual decrease (increase) throughout the unit, while the Zr counts remain stable. The topmost~30 cm of the core Fig. 5. Results of the CTand XRFanalyses carried out on core GeoB19927-3. The top panel shows a contrast enhanced, distorted grey-scale image of the core next to records of mean matrix sediment density, lithic clast counts as well as volumetric sediment composition. The main elemental records used in this study (Al/Ti, Zr, Zn and Fe/K) are shown in grey towards the right, overlain by five-point running averages. Note the broken axis for the sediment composition and Zr record. The column on the far right indicates the depth intervals for the hemipelagic (hemi) section and the mass-wasting deposit (mwd); a turbidite layer (T) at~775 cm within the hemipelagic unit is indicated too. Black triangles show depths of radiocarbon measurements. Red boxes in the greyscale image indicate core sections detailed in the bottom panel, which shows undistorted greyscale images of characteristic core sections next to 3D reconstructions of sediment constituents (orange, lithic clasts; green, bioturbation) as well as slump folds (dark blue) and turbidites (light blue).
show an increase in all elemental markers except Al/Ti, paralleled by increasing lithic clast counts.
The age model of core GeoB19920-1 is based on a total of seven AMS 14 C datings (Fig. 4). Two ageswere acquired from the lowermost hemipelagic unit, constraining the basal age of the core to 8.2 cal. ka BP. The two ages from the middle, mass-wasting unit reveal an inconclusive age pattern with an age reversal at 959 cm core depth. The three ages from the upper unit get progressively younger towards the core top, indicating continuous hemipelagic sedimentation since~8 cal. ka BP. Here, sedimentation rates range between~90 prior to 4.5 ka BP and 60 cm ka À1 in the Late Holocene (Fig. 4).

GeoB19973-2
Core GeoB19973-2 consists of three lithological units. The lowermost unit (>657 cm) shows a high-density matrix (average~1400 HU) with abundant (~25 clasts cm À3 ) and large lithic clasts accounting for up to 30 vol% of the sediment and no traces of bioturbation, classifying it as a till (Fig. 8). Geochemically, it is characterized by low Al/Ti, Zn and Fe/K values and highest Zr counts. The middle unit between 597 and 657 cm core depth exhibits alternating layers of high-and low-density sediments evident from the CT data (Fig. 8). The unit shows decreasing lithic clast abundance and is free from traces of bioturbation. Hence this is a transitional unit, comprising 'glacimarine laminae' as found in other proglacial environments (Jennings & Weiner 1996;Cowan et al. 1997;Dowdeswell et al. 2000;Sheldon et al. 2016). The geochemical data show a pronounced maximum in the Fe/K ratio. Zn counts and the Al/Ti ratio begin to increase, while Zr counts start to decrease.
The upper unit (<597 cm) comprises hemipelagic sediments with a low-density matrix that are slightly (<1 vol%; 597 to~320 cm) to moderately (up to 6 vol%; 320-50 cm core depth) bioturbated (Fig. 8). The onset of this unit is marked by a sharp increase in the Al/Ti ratio, a similarly pronounced decrease in the Zr counts and a notable reduction in the Fe/K ratio. Afterwards, Al/Ti slightly decreases towards the core top, while Zn continuously increases. The Fe/K remains largely stable until~400 cm core depth and continuously increased afterwards. Zr counts increase slightly, but continuously, towards the core top.
The age model of core GeoB19973-2 is based on 10 AMS 14 C datings (Fig. 4). The two ages acquired from the lower, over-compacted till unit show unusually high 14 C   7. Results of the CTand XRFanalyses carried out on core GeoB19920-1. The top panel shows a contrast-enhanced, distorted grey-scale image of the core next to records of mean matrix sediment density, lithic clast counts and volumetric sediment composition. The main elemental records used in this study (Al/Ti, Zr, Zn and Fe/K) are shown in grey towards the right, overlain by five-point running averages. Note the broken axis for the lithic clast counts, sediment composition and Zr record. The column on the far right indicates the depth intervals for the hemipelagic (hemi & h) sections and the mass-wasting deposit (mwd). Black triangles show the depths of radiocarbon measurements. Red boxes in the greyscale image indicate the core sections detailed in the bottom panel, which shows undistorted grey-scale images of characteristic core sections next to 3D reconstructions of sediment constituents (orange, lithic clasts; green, bioturbation; multi-coloured, segmented, individual mudclasts). BOREAS ages (720 cm, 47.1 cal. ka; 817 cm, carbon dead), suggesting the presence of older, reworked material. The remaining eight dates obtained from the upper, bioturbated unit get progressively younger towards the core top and suggest continuous hemipelagic sedimentation since at least 11.3 cal. ka BP. Sedimentation rates range between~30 cm ka À1 after 6 cal. ka BP and 50 cm ka À1 before (Fig. 4). The fairly constant Early Holocene sedimentation rate was extrapolated to the base of the hemipelagic unit at 657 cm core depth, yielding an estimated age of 13.4 cal. ka BP. Further linear extrapolation to the base of the transitional unit would date the top of the till to around 14.4 cal. ka BP. This should, however, be considered as a maximum age estimate for the deposition of the till only (see Discussion).
The grain-size data show two fundamentally different grain-size compositions. Below~600 cm (i.e. in the lowermost and middle unit) the sediment is coarse (mostly medium sand) and poorly sorted with a strong fine (negative) skew and has modal grain sizes >200 lm. A tertiary modal grain size around 6 lm shows substantial additional abundances of finer material. Above~600 cm the material is significantly finer (fine to medium silt), meso-to leptokurtic and polymodal. The primary mode is centred around 4-6 lm, very similar to the tertiary mode of the lower units (Fig. 9).

End member analysis
Both the three-and four-component end member models we tested explain >99% of the datasets' variance. In combination with the results of the grain-size analysis and provenance reconstructions, we concluded that the three-EM model is more genetically meaningful, despite the presence of four source areas (see explanation in the Discussion). EM 1 is primarily characterized by high contributions of Fe and Zn and low contributions of Al and K (Table 3). Accordingly, it exhibits the highest Fe/K ratio. EM 2 shows overall lowest contribution of Fe and Ti combined with the highest contribution of K. It exhibits the highest Al/Ti ratio. EM 3 shows elevated contributions for Al and Ti, a comparably low Al/Ti ratio and the highest Zr contribution.

Principal component analysis
The PCA of the geochemical data reveals three principal components (PCs), which explain~95% of Fig. 8. Results of the CTand XRFanalyses carried out on core GeoB19973-2. The top panel shows a contrast-enhanced, distorted grey-scale image of the core next to records of mean matrix sediment density, lithic clast counts aswell as volumetric sediment composition (note the broken axis). The main elemental records used in this study (Al/Ti, Zr, Zn and Fe/K) are shown in grey towards the right, overlain by five-point running averages. The column on the far right indicates the depth intervals for the hemipelagic (hemi) section as well as the transitional (trans) unit and the basal till. Black triangles show depths of radiocarbon measurements. Red boxes in the greyscale image indicate the core sections detailed in the bottom panel, which shows undistorted grey-scale images of characteristic core sections next to 3D-reconstructions of sediment constituents (orange, lithic clasts; green, bioturbation). the variance: PC 1 explains 41.34% of the datasets' variance and primarily differentiates between Al, Si and K vs. Fe and Zn. PC 2 explains 27.38% of the datasets' variance and primarily differentiates between Zr and Ti vs. Zn, Si, Al and K (Fig. 10). The third principal component primarily differentiates between Ti and Fe vs. K, Zn and Zr and explain almost as much of the variance as PC 2 (25.74%). However, since the four source areas can be identified in the biplot of PC 1 and PC 2 fairly well, we are focussing only on the first two PCs.

Source area characterization and sediment provenance
The analysis of the onshore elemental data enables a geochemical differentiation between the four provinces using specific marker elements (or ratios thereof) that are descriptive for these sediment source areas. These markers need to meet two basic criteria: (i) the according elements must be part of the reliable XRF-scanner dataset (including Zn; see above) to allow for a comparison between onshore geochemical signatures and the geochemical signatures in the sediment cores and (ii) none of the elements should be used more than once, since the XRF scanning data are compositional and hence autocorrelated to a certain degree. This is essential to maximizing the degree of independence of the markers and avoiding statistical artefacts. Given these constraints, we determined that Karrat North is best captured by a high Al 2 O 3 /TiO 2 ratio as well as relatively elevated K 2 O and Al 2 O 3 contents, Karrat South is primarily characterized by high Zn concentration, the Nuussuaq Basin volcanics are best captured by the high Fe 2 O 3 /K 2 O ratio and the Prøven Igneous Complex is characterized by highest Zr and K 2 O concentrations (Fig. 2).
With respect to our sediment core data, we combined two statistical methods (EMA and PCA) in order to classify the sediments' elemental composition and link them to onshore source areas. While the two methods share a number of similarities, it is important to take note of their differences to avoid erroneous interpretations. The end-member modelling makes use of compositional data to linearly unmix these into (theoretical) end members (Weltje 1997;Dietze et al. 2012). The outcome remains compositional, hence the fractional abundances of the end members sum to unity. In contrast, PCA requires data to be non-compositional (necessitating the centre log ratio transformation; see Material and methods) and results in new, theoretical variables, the principal components, which maximize variance (Jollife & Cadima 2016). The orthogonality of the principal components' axes makes them by definition uncorrelated. Accordingly, we do not interpret them in a quantitative sense. For this study, this means that the overall sediment geochemical composition is described by the relative contributions of the end members, while the PCA highlights characteristics that are useful in differentiating individual units.
Bearing the elemental composition of the most relevant onshore source regions (see Fig. 2) in mind, the end members calculated from the XRF data can be linked to the onshore source areas via the identified marker elements/ratios. Accordingly, EM 1 mirrors input from the Nuussuaq Basinvolcanics (high Fe/K, low K) or Karrat South (high Zn). EM 2 can be linked to input from Karrat North via the high Al/Ti ratio, while EM 3 mirrors Zr-rich input from the Prøven Igneous Complex (Table 3). A four-component end member model also tested was not able to further unmix the sediments from the Nuussuaq Basin volcanics or Karrat South. Therefore, these two source areas will be treated in unison. Since the sediment from these southern provinces has to be current-transported to the core site (except for occasional ice rafting), a large degree of mixing during transport is to be expected, explaining why these two source signals appear as one mixed signal in the EM analysis (see details on sediment dispersal below).
Qualitatively, this classification is supported by the factor loadings from the PCA and accordingly the orientation of the sample scores in the PCA biplots. Positive scores of PC 1 highlight loading with Al, Si and, to a lesser degree, Zr, while negative PC 1 scores accentuate additional Zn, Fe or Ti. Positive PC 2 scores primarily highlight Zr-rich samples, while negative scores accentuate Zn, Si and/or Al. Accordingly, the quadrants of the PCA biplots highlight contributions from the four identified source areas (Fig. 10). Here, (i) samples in the top right quadrant are Zr rich, suggesting input from the Prøven Igneous Complex, (ii) samples in the bottom right quadrant have a high Al/Ti ratio, i.e. carrying a prevalent input signature of Karrat North, (iii) samples in the bottom left quadrant are characterized by a high Zn content, pointing to input from Karrat South and (iv) data points in the top left quadrant exhibit high Fe/K ratio suggesting input of the Nuussuaq Basin volcanics. Thus, combining the EMA and PCA therefore holds the potential of maintaining the semiquantitative compositional information retrieved from XRF scanning supported by qualitative accentuation of the most prominent contributions from the four source regions.

Palaeoenvironmental reconstructions
The deglaciation of the Upernavik trough. -Core GeoB19973-2 is the only record from Upernavik trough that contains a glacial diamicton, i.e. a till, at its base. It therefore contains sedimentological evidence of the deglaciation of the mid-shelf of this trough. However, establishing the age of the deglaciation at this site is difficult as there is no further dating between 470 cm core depth (11.3 cal. ka BP) and the top of the till in 657 cm. Seeing the homogeneity of the hemipelagic sediments down to the base of the hemipelagic unit at 597 cm core depth, visible in the CT, grain-size and XRF data (Figs 8,9,11), there is little reason to assume significant changes in sedimentation rate. Consequently, we extrapolated these fairly constant sedimentation rates of~50 cm ka À1 (calculated from the hemipelagic sediments older than 6 cal. ka BP) to the base of the hemipelagic unit yielding an age of 13.4 cal. ka BP. Further extrapolation to the top of the till would yield an age of 14.4 cal. ka BP. However, the CT data imply the presence of glacial laminae in the transitional unit (597-659 cm core depth) and thus proglacial, most likely very rapid sedimentation (sensu Hogan et al. 2016). Hence, a top age of 14.4 cal. ka BP for the till must be considered as maximum estimate for the retreat of the glacier past the core site only.
According to this age model, a grounded ice stream retreated from the GeoB19973-2 core site between 14.4 and 13.4 cal. ka BP, leaving behind the overcompacted diamicton of the lowermost core unit. The subsequent deposition of proglacial marine sediments probably took place until~13.4 cal. ka BP documenting the early deglaciation of the mid-shelf of the Upernavik trough. Consequently, ice retreat over the core site would have happened during peak warmth of the Bølling-Allerød (B/A) interstadial (Petit et al. 1999;McManus et al. 2004;Denton et al. 2005)a chronology that would be supported by the presence of a grounding zone wedge (GZW) just landward of site GeoB19973-2 identified by Slabon et al. (2016), which has tentatively been assigned a Younger Dryas (YD) age. Grounding zone wedges are morphological features of the sea floor defined as icemarginal depocentres with steeper ice-distal and flatter ice-proximal sides and commonly interpreted to represent times of glacial stillstands or short-lived readvances during retreat (Evans et al. 2002;Young et al. 2013;Batchelor & Dowdeswell 2015;Sheldon et al. 2016;Slabon et al. 2016;Crump et al. 2017;O'Hara et al. 2017;Philipps et al. 2017;Batchelor et al. 2018). Hence, we propose that the ice stream retreated past the core site during the relatively warm B/A and came to a halt at (or re-advanced to) the position of the GZW just landward of the core site during the YD, while noting that this hypothesis is based on extrapolated age estimates only. Support for the prevalence of this mechanism was presented by, e.g., Jennings et al. (2017), who found episodes of ocean warming around (and prior to) 14 cal. ka BP on the outer Disko shelf in cores JR175-VC29 and HU2008029-12PC and linked these to initial ice margin retreat.
Such a chronology would further place the initial ice retreat from the mid-shelf within the timeframe of the Baffin Bay Detrital Carbonate (BBDC) event 1 (15-13.7 cal. ka BP; Simon et al. 2014), for which an increased advection of Atlantic waters has also been proposed as a triggering mechanism (Andrews et al. 1998;Simon et al. 2014). BBDC layers have been identified in numerous sediment cores from Baffin Bay continental slope and central basin sites by elevated Ca/ Sr ratios and/or high dolomite contents Jackson et al. 2017;Kelleher et al. 2022) or elevated Ca/Ti ratios (Jenner et al. 2018). Neither of these characteristics is evident from our data (see Fig. S2 for a comparison with nearby core GeoT€ u SL 170 of Jackson et al. 2017) and we hence conclude that the detrital carbonates sourced from northern Baffin Bay during these events did not reach our site at the mid-shelf off West Greenland. However, the triggering mechanism that initiated the BBDC events might also have influenced ice retreat and sedimentation on the mid-shelf of the Upernavik trough.
In the southerly adjacent Uummannaq trough Sheldon et al. (2016) proposed a very similar glacier retreat pattern with retreat from the outer shelf during the B/A and stillstand on the mid-shelf during the late B/A and YD. In combination with findings from , the authors dated the mid-shelf stillstand to 13.9-11.5 cal. ka BP, which led Slabon et al. (2016) to postulate a similar GZW formation during a YD ice margin stillstand in the Upernavik and the northerly adjacent Melville Bay troughs. Thus, our new age estimate (in combination with the existing data) supports the hypothesis that deglaciation and ice stream dynamics in NE Baffin Bay shelf troughs are predominantly climatically controlled (Hogan et al. 2016;Sheldon et al. 2016;Slabon et al. 2016). Previous reconstructions of glacier retreat ( 10 Be-dating) constrained the deglaciation of the Upernavik trough to the modern-day coastline at 9.9 ka . As our new age estimate supports a YD stillstand of the ice margin just landward of our site in analogy to the Uummannaq trough (13.9-11.5 ka BP; Sheldon et al. 2016), it is now possible to approximate average retreat velocities for Upernavik trough. These imply a rather swift average of the ice stream (~180 km within 1600 years) from the mid-shelf stillstand site to the modern-day coastline, equalling an average retreat velocity of~0.11 km a À1 . In contrast, Corbett et al. (2013) suggested a deglaciation at 11.3AE0.5 ka based on their 10 Be and 26 Al-data from the outer coast south of Upernavik, which would result in a glacier retreat velocity almost one order of magnitude faster. However, as Larsen et al. (2017) pointed out that glacier retreat is always associated with glacier thinning and considering that the sites of Corbett et al. (2013) cover almost 1000 m in elevation, we consider our retreat velocity estimate based on the Briner et al. (2013) deglaciation age as more reliable. Retreat velocities from modern outlet glaciers in the region, such as Jakobshavn Isbrae or Upernavik North, range between >0.5 km a À1 (Upernavik; Howat & Eddy 2012) and 0.27 km a À1 (Jakobshavn; Khan et al. 2015;Steiger et al. 2018). This, in turn, means that the modern-day, anthropogenically accelerated rate of ice-sheet decay in the area (Rignot & Kanagaratnam 2006;Velicogna & Wahr 2006;Vizca ıno et al. 2008;van den Broeke et al. 2009;Gregory et al. 2020) is happening faster than the average retreat during the last deglaciation, which otherwise presents the intervals with the most profound environmental perturbations. However, it has to be kept in mind that this is based on a comparison of decadal-and millennial-scale data, without having any information on decadal-tocentennial scale variability in the deglacial record.
While the till in core GeoB19973-2 is a clear proof for the presence of grounded ice, the overlying transitional unit marked by alternating high-and low-density layers barren of any signs of bioturbation (Fig. 8) probably reflects the first sedimentation directly in front of the receding ice. A similar facies has also been found by Hogan et al. (2016) and Sheldon et al. (2016) in sediment cores from Disko trough. They similarly interpret these alterations as an expression of two major forms of sedimentation, iceberg rafting and settling from a meltwater plume. Based on the assumption that these laminae are glacimarine varves (as identified in Alaskaby e.g. Cai et al. 1997;Cowan et al. 1997;Ullrich et al. 2009), Hogan et al. (2016 speculate that the deposition of these sediments occurred very quickly, thereby indicating intense melting and high sediment input during the period following initial glacier retreat. This is in support of our age model of core GeoB 19973-2, that considers rapid deposition of these sediments (see above).
In addition to estimating the timing of the deglaciation, the provenance analyses (XRF records, PCA and EMA) allow the main sediment sources to be constrained. The till at site GeoB19973 exhibits low Al/Ti ratios and elevated Zr counts, a large proportion of EM 3 and positive PC 1 and PC 2 scores pointing to the Prøven Igneous Complex as the main source area (Figs 8, 10, 11). We note that Zr elemental records often show a grain-size (sorting) effect (e.g. Wu et al. 2020), with coarser sediments exhibting an enrichment in Zr. Indeed, we see that the coarser material from the till is enriched in Zr in our case. However, the onshore elemental data were mostly measured on stream sediments, which have already seen a certain degree of sorting and hence Zr enrichment, which means that part of grain-size effected is inherently accounted for already. This is supported by the fact that we found a greater correlation between Zr and the vol% of sediments in the 63-125 lm fraction (r 2 = 0.88) than between Zr and the amount of lithic clasts (r 2 = 0.24), indicating that Zr input is governed by the finer, pre-sorted stream sediments rather than the lithic clasts. We are therefore confident that the Zr enrichment in the basal section of the core reflects predominantly a source signal, rather than a representation of grain size.
Interestingly, the EMA of the basal till also revealed up to~40% admixture from the Nuussuaq Basin volcanics and Karrat South. Combined with the presence of mollusc shells, two of them AMS 14 C dated (720 and 817 cm) and yielding a calibrated age of 47.1 cal. ka BP or being 'carbon dead', pointing to the presence of hemipelagic sediments, we suggest that the till comprises both the material directly delivered via the ice stream from the Prøven Igneous Complex and older (i.e. at least marine isotope stage 3) hemipelagic deposits. Parts of the hemipelagicswere most likely delivered from the south by a 'palaeo-WGC', as indicated by their geochemical composition, and after initial deposition in the Upernavik trough, reworked and incorporated into the till during ice stream advance.
The overlying transitional unit still contains high Zr contents, pointing to continuing input from the Prøven Igneous Complex, which might reflect the initial meltout of the glacier after its retreat. Additionally, we record a strong Fe-and Ti-rich input (indicative of the southerly source regions) and/or the disappearance of material from Karrat North (EM 2). This indicates an invigoration of the WGC, which delivered additional material from the south, while at the same time dispersing the local Karrat North sediments (see below). Such an invigoration of the WGC bringing warmer waters from the North Atlantic probably contributed to the melt and retreat of the Upernavik ice stream. An analogous scenario has been proposed for the southerly adjacent Uummannaq and Disko troughs by Jennings et al. (2017), who also invoked subsurface warming (Marcott et al. 2011) and advection of Atlantic waters (Knutz et al. 2011) as a driving force for initial melt of grounded ice on the outer shelf. A large-scale invigoration of the AMOC during the B/A was recorded A large-scale mass-wasting event in the inner Upernavik trough at~8 cal. ka BP. -All three cores collected from the inner Upernavik trough (GeoB19920-1, GeoB19927-3 and GeoB19969-1) contain mass-wasting deposits in their bottom parts. The base of core GeoB19969-1 contains a mass-wasting unit a few decimetres-thick with two debris flows, rich in mudclasts overlain by turbidites (Fig. 3), while in core GeoB19920-1, the mass-wasting unit is several metres thick, consisting of a debris flow deposit rich in mudclasts, overlain by the material settling from its 'tail' (i.e. the suspension cloud; Fig. 7). While there is some evidence of flow structures resembling a turbidite, these are much less pronounced than in the other cores. In core GeoB19927-3, the mass-wasting unit contains a slump deposit overlain by several turbidites (Fig. 5).
The PCA illustrates that these mass-wasting deposits are geochemically distinctive, clearly differing from the hemipelagic units and point to the Prøven Igneous Complex and Karrat North as the principal source areas of the redeposited sediments (Figs 2, 10). This is most obvious for the most inshore core GeoB19969-1, which is located just off the Karrat North region and slightly north of the Prøven Igneous Complex. For the two cores GeoB19920-1 and GeoB19927-3 located slightly further offshore, the PCA also reveals a small, but obvious contribution from the southern provenances as also reflected in the EMA (Fig. 1). This pattern corresponds largely to the provenance analysis of near by core AMD14-204 (Caron et al. 2020;Giraudeau et al. 2020), with the XRF data from Giraudeau et al. (2020) and our cores pointing more towards Karrat North and the mineralogical data of Caron et al. (2020) towards the Prøven Igneous Complex as primary sources. Keeping in mind that these are all reworked sediments and that the original hemipelagic sediments already had mixed signatures, the slight differentiation in the composition of the mass-wasting deposits probably points to slightly different source regions, at least for core GeoB19969-1 on the one hand and cores GeoB19920-1 and GeoB19927-3 on the other. Considering the confined morphology of the inner Upernavik trough and adjacent fjord system (Fig. 1), these have to be rather local sources. The stronger sediment contribution from southern source areas to the more offshore sites, thus, may reflect (i) an enhanced sediment input via the WGC on the original, pre-mass-wasting, sediment composition or (ii) a less efficient sediment transport of the locally sourced material to the outer sites as this would result in similar relative elemental signatures (see details on sediment transport mechanisms below).
The core positions in three different sub-basins of the inner trough, as well as the different transport mechanisms (debris flow vs. slump) and the slightly different geochemical sediment composition, implies at least three spatially discriminative events. Whether the sub-events within the mass-wasting units, preserved in cores GeoB19969-1 and GeoB19927-3, reflect several synchronously initiated mass-wasting events that were subsequently deposited at the core site due to different transport distances and transport velocities (multi-stage event) or whether they reflect temporarily different events cannot be resolvedwith the available data. However, the synchronicity of all three mass-wasting units with a top age of~8 ka BP is remarkable and favours the first scenario. Only in core GeoB19920-1 has the mass-wasting deposit been fully penetrated and the hemipelagic sediments beneath have been dated to 8.2 ka BP, thus constraining the timing of these mass-wasting events to 8.0-8.2 cal. ka BP.
Mass wasting is often caused by instabilities in the sediment column owing to weak layers and/or rapid sediment accumulation (e.g. Huvenne et al. 2002;Bøe et al. 2004;Haflidason et al. 2004;L'Heureux et al. 2012;Locat et al. 2014;Laberg et al. 2016) and represents wellknown phenomena on glaciated margins that are in this region closely linked with ice-margin dynamics (e.g. Syvitski 1991;Dowdeswell et al. 1998;O Cofaigh et al. 2013a, b). However, the postulated synchronicity of the mass-wasting events points to one common trigger, which suggests two potential mechanisms: (i) a seismic trigger in analogy to case studies, where ice-margin instability-induced seismicity was previously reported on from southern Greenland (Steffen et al. 2020) as well as Disko Bay, close to our study site (Olsen & Nettles 2019); and/or (ii) a large mass-wasting event and its associated tsunami initiating multiple quasisynchronous mass-wasting events in the region.
Hemipelagic sedimentation since~8 cal. ka BP: Provenance, transport and deposition. -Throughout the Holocene all four studied cores reveal a consistent pattern BOREAS regarding the changes in sediment source and input, which can be summarized by (i) distally decreasing abundance of EM 2 (Karrat North), (ii) distally increasing contents of EM 3 (Prøven Igneous Complex) that remain stable over time and (iii) temporally increasing contributions of EM 1 (Nuussuaq Basin volcanics and Karrat South) and decreasing EM 2contributions in all cores (Fig. 11). The PCA (Fig. 10) substantiates these findings and highlights Karrat North and the Nuussuaq Basin volcanics as primary sediment sources whose interplay largely governed the sediments' overall composition.
For a meaningful reconstruction of the sediment provenance, however, both the sediment delivery and its depositional processes need to be considered. Recently, Weiser et al. (2021) investigated the sedimentation in a cross-shelf trough in the Labrador Sea and found two dominant sediment delivery processes in general agreement with observational and modelling studies of shelf trough sedimentation (March es et al. 2007;Voigt et al. 2013). The first one is a primarily gravitational, near-bed transport of glacial meltwater plume-derived sediments that effectively transports suspension loads onto the outer shelf within the troughs and therefore independent of the WGC. The second process is entirely current controlled and relies on sediment dispersion by the ocean current (potentially after re-suspension) along the coast until the hydrodynamic energyand hence sediment holding capacitydrops as the water traverses a bathymetric low, i.e. a cross-shelf trough.
Given the fact that the Sukkertop trough studied by Weiser et al. (2021) and the Upernavik trough are two very similar depositional regimes, it appears reasonable that these two processes are dominant here as well. We note, however, that current sorting might play a minor role here, as the overall current speeds of the WGC are significantly lower north of Davis Strait than they are in Labrador Sea. Nonetheless, both the longshore sediment dispersal by the WGC as well as the gravitational sediment transport perpendicular to the coast along the thalweg of the trough have to be considered as important sediment delivery processes.
Bearing this and the geographic location of the identified geological provinces in mind, it becomes obvious that sediment input from the individual provinces inevitably underlies different sediment dispersal processes. Karrat North sediments, for example, cannot reach the core sites of anyofour four studied cores via oceanic longshore dispersal, as longshore transport along the path of the WGC would transport the sediments northward out of our study area. On the other hand, the shallowing water depth between the inner shelf and the mid-shelf also questions transport as bedload, as an 'uphill' transport is unlikely, although not entirely impossible (e.g. V€ olker et al. 2008). Instead, the suspended meltwater-plume sediments from Karrat North and the Prøven Igneous Complex were most likely transported within a water mass that floods the shallow trough, but not the platforms on either side of the upper Upernavik trough, so that they are sheltered from northward flowing WGC waters. Such layered sediment transport has been described from Antarctica by Domack et al. (1994) and would arguably present a third sediment transport mechanism in addition to the two mechanisms identified by Weiser et al. (2021), where sediment is transported in suspension but still independent of the WGC (or at least a variation of the currentindependent mechanism). As mentioned initially, the deposition of Karrat North sediments is further complicated by the fact that this source province is located at the same latitude as the Upernavik trough, which means that a significant fraction of the input sediment is likely be drifted north by the WGC and therefore out of our study area before it has plunged to a depth where it is sheltered from along-shelf current influence. Hence, the amount of recovered Karrat North sediments probably only represents a fraction of the amount of delivered sediment.
In contrast, material from the southerly sources (Nuussuaq basin volcanics and Karrat South) has to be primarily delivered to the study area as suspension load carried by the WGC. As soon as the WGC traverses the Upernavik trough, it loses momentum and therefore sediment-holding capacity, and the suspended sediment settles on the sea floor.
The grain size data of the hemipelagic units of cores GeoB19973-2 and GeoB19927-3 ( Fig. 9) reveal (i) dominantly polymodal grain-size distributions and (ii) largely unreliable SS mean (following the criteria of McCave & Andrews 2019a; Fig. S2). As grain size sorting predominantly occurs through re-suspension after benthic storms (McCave & Hall 2006), this indicates a relatively short transport distance with negligible resuspension and calm hydrographic conditions without significant current influence on the sea floor at either site, illustrating the sheltered hydrographic regime mentioned above. The polymodality further implies that the material retained its original polymodal constitution, characteristic of meltwater plumes carrying glacial flour (Boulton 1978;Haldorsen 1981;Lund-Hansen et al. 2010;Meslard et al. 2018;Pesch et al. 2022), hinting at an efficient deposition of the entire sediment load.
Located between the Karrat North province and the southerly source areas, the sediments from the Prøven Igneous Complex can reach the core side via both transport mechanisms. While some of the material will be transported along the Upernavik fjord and trough and reach the core sites as deep suspension load, parts of the province also border Baffin Bay directly where meltwater plumes can enter the ocean away from the deep thalweg of the Upernavik trough. Here, they will be picked up by the WGC, transported north to the study area and be subject to the depositional mechanisms described above.
Accordingly, the abundances of the end members in our cores are not only controlled by the primary sediment input from the provinces they originate from, but must also be understood as a function of the prevalence and efficiency of these different input and transport mechanisms. Bearing this is mind, the patterns we identified at the beginning of this section are interpreted as follows: • Decreasing abundance of EM 2 (Karrat North) towards the distal sites Given the transport in a sheltered hydrodynamic regime (see above), we interpret the decreasing abundance of Karrat North-sourced sediments (EM 2) from the inner shelf towards the mid-shelf as an expression of the decreasing efficiency of this deep suspension sediment transport along the Upernavik trough. While a large fraction of the sediment that enters the ocean close to the coast can be deposited quickly nearshore (GeoB19969-1) or in the deep inner-shelf basins, probably functioning as effective sediment traps (GeoB19920-1 and GeoB19927-3), it becomes increasingly difficult for the sediment to reach the mid-shelf. Nonetheless, the fact that a large fraction of the sediment that makes up the mid-shelf record is derived from Karrat North despite the potential removal by northward drift implies that sediment delivery from those local sources has played and still plays an important role throughout the entire trough.

• Distally increasing contents of EM 3 (Prøven Igneous
Complex) that remain stable over time The fact that the EM 3-abundances increase towards the distal site and remain stable over time in all sites is interpreted to reflect a balanced interplay between the two transport mechanisms outlined above. While the deep suspension transport of Prøven Igneous Complex sediments presumably became less efficient over time (as it did for the Karrat North sediments; see below), the efficient oceanic delivery of this type of sediment from the south might have compensated for this decrease, resulting in a constant contribution of Prøven Igneous Complex sediments (EM 3) over the course of the Holocene. As the oceanic delivery is more likely to reach the offshore site of GeoB19973-2 than the inner-shelf sites, the distally increasing  Fig. 8; see orientation of the confidence ellipses). Hence the geochemical composition of the hemipelagic sediments is essentially a result of the relative contributions from these two source areas.
The temporally increasing proportions of the Nuussuaq Basin volcanics and Karrat South (EM 1) material suggest a more effective sediment transport from the south to the study area via the WGC, mirroring the pattern identified by Caron et al. (2020) for neighbouring core AMD14-204. This is, however, diametrically opposed to the decreasing WGC speeds reconstructed upstream in Labrador Sea (Weiser et al. 2021) and the generally decreasing sedimentation rates in the Upernavik trough. This apparent contradiction can be reconciled by assuming a strongly decreasing delivery of Karrat North sediments instead of increasing delivery from the south. As the EM abundances are autocorrelated, the relative increase of material from the southern sources can just as well be caused by a pronounced reduction in the abundance of material from Karrat North. Support for this hypothesis is found in the temporally decreasing sedimentation rates of all cores, highlighting that the observed changes are more likely to be caused by a reduction of material, instead of any additional input (see Fig. 2). This also holds true for the data from Giraudeau et al. (2020) and Caron et al. (2020), as sedimentation rates of their core also decrease in the Late Holocene (Caron et al. 2018), thereby also favouring a reductive rather than an additive mechanism.
Interestingly, the Neoglacial slowdown of the WGC recorded in Labrador Sea might have similarly affected Baffin Bay, as two records (GeoB19969-1 and GeoB19927-3) reveal stable or no longer increasing Nuussuaq Basin volcanics and Karrat South sediment (EM 1) abundances after 3 ka BP. Here, slower current speeds might have resulted in a less efficient sediment delivery from the south. However, as outlined above, the cause for any observed shift might just as well be found in a slightly increasing Karrat North (EM 2) input, which would suggest a local cause. In such a scenario, the colder temperatures during the Neoglaciation could have increased ice rafting from the local sources as, e.g., evidenced by the increase in lithic clast-contents in GeoB19969-1 (Fig. 3).

Conclusions
We present four new records of sedimentary dynamics of Upernavik trough covering at least the last 11.3 cal. ka BP (extrapolated to >~13.4 cal. ka BP) based on radiocarbon dating, CT imaging and XRF scanning and grain-size analysis. When combined, our findings offer the opportunity for a chronological framework of the regional palaeoenvironmental development. The late glacial GrIS extent onto the mid-shelf is evidenced by a till, which contains a significant contribution from the Prøven Igneous Province, indicating that the ice stream that covered the trough was fed from the Upernavik Istrøm rather than the northern glaciers in the area (Cornell Glacier, Ussing Braeer). The mid-shelf deglaciation is estimated to have an age slightly older thañ 13.4 cal. ka BP within the B/A. For a presumably short period, glacier proximal conditions prevailed before rather constant hemipelagic sedimentation probably began around~13.4 cal. ka BP. If so, peak sediment delivery from southern sediment sources following the deglaciation at our mid-shelf site would coincide with peak B/A-warmth and AMOC invigoration and suggest far northward advection of Atlantic waters that could have facilitated ice-stream retreat prior to the previously hypothesized YD stillstand on the mid-shelf. On the inner shelf open marine conditions prevailed since at least 8.2 cal. ka BP. At around 8 cal. ka BP the area was subject to intense, presumably synchronous masswasting events, leading to the widespread deposition of debris flow and slump deposits as well as turbidites. Since these are reworked deposits of hemipelagics that themselves present a mixture of different sediment sources, their ultimate origin is difficult to establish, but the available geochemical data point to a mixture of Karrat North and the Prøven Igneous Complex contributions. These deposits are constrained to the inner shelf.
The hemipelagic sedimentation after 8 cal. ka BP is governed by a relatively complex interplay between at least two variable sediment dispersal mechanisms (cross shelf vs. along-shelf transport) and varying sediment input from the source areas. Our findings highlight that the deposition of sediments from the different provinces is governed by location of input and prevalent transport mechanism. Only the deep suspension load from Karrat North reaches the core sites via cross-shelf transport, as the surficial plumes aredrifted out ofour studyareabythe WGC. In contrast sediments from the southern sources (Nuussuaq Basin volcanics and Karrat South) can only reach the Upernavik trough by along-shelf transport via the WGC, while Prøven Igneous Complex sediments are transported to the core sites by both processes.
Spatially, the abundance of Karrat North-sediments decreases with increasing distance from the coast, while the abundance of Prøven Igneous Complex sediments increases. Through time, an increasing admixture of sediment from the southern sources is recorded as Karrat North input decreases. However, given the decreasing sedimentation rates in all cores and decreasing WGC speed (Weiser et al. 2021), this change in the sediment composition is most likely caused by a reductive mechanism (decreasing input from Karrat North) rather than an additive one (increasing delivery from the southern sources).
Acknowledgements. -The master and crew of the RV 'Maria S. Merian' are gratefully acknowledged for support of the work during cruise MSM44. This project was supported by the Deutsche Forschungsgemeinschaft through the International Research Training Group 'Processes and impacts of climate change in the North Atlantic Ocean and the Canadian Arctic' (IRTG 1904 ArcTrain). The authors are indebted to editor J. A. Piotrowski and the anonymous reviewers for their invaluable comments. The authors declare that they have no conflicts of interest to disclose.
Author contributions. -JW, JT and DH designed the study. JW performed the XRF and grain size analyses. JW and JT performed the CTanalyses, JW performed the CT processing underguidance of JT. JW picked the calcareous organisms used for radiocarbon dating and developed the age models. JW wrote the initial draft of the manuscript, performed all statistical analyses and produced all of the figures. DH acquired funding, was chief scientist on R/V 'Maria S. Merian' cruise MSM44 and made all sample material available. JT and DH provided advice on the interpretation of all data. All co-authors commented on the manuscript.
Data availability statement. -All data shown here are or will be made available via PANGAEA (https://www.pangaea.de/).  . Downcore records of XRF-derived Ca/Sr and Ca/Ti-ratios for core GeoB19973-2 and GeoT€ u SL 170 (Jackson et al. 2017). Grey boxes approximate the timing of BBDC events. Note that the scale of respective x-axes is the same for ratios for both cores, illustrating the vastly lower magnitude of change in GeoB19973-2. Further, neither the Ca/Sr nor the Ca/ Ti ratio of GeoB19973-2 displays any form of positive excursion as would be indicative of BBDC events.