Quaternary Science Reviews 73 (2013) 1e14
Contents lists available at SciVerse ScienceDirect
Quaternary Science Reviews journal homepage: www.elsevier.com/locate/quascirev
Numerical analyses of a multi-proxy data set from a distal glacier-fed lake, Sørsendalsvatn, western Norway Jostein Bakke a, b, *, 2, Mathias Trachsel a, b,1, 2, Bjørn Christian Kvisvik b, c, Atle Nesje a, b, Astrid Lyså d a
Department of Earth Science, University of Bergen, Allégaten 41, 5007 Bergen, Norway Bjerknes Centre for Climate Research, Uni Research, Allégaten 55, 5007 Bergen, Norway Department of Geography, University of Bergen, Fosswinckelsgate 6, 5020 Bergen, Norway d Geological Survey of Norway, Leiv Eirikssons vei 39, 7040 Trondheim, Norway b c
a r t i c l e i n f o
a b s t r a c t
Article history: Received 11 December 2012 Received in revised form 30 April 2013 Accepted 6 May 2013 Available online xxx
Here we present a Holocene record of glacier variability as documented through physical sediment properties analysed on sediments from the distal glacier-fed Lake Nedre (Nedre ¼ Lower) Sørsendalsvatn (918 m a.s.l.), located 35 km inland from the coast in western Norway. We emphasise comparing different sediment parameters by means of statistical methods as well as integrating chronological uncertainties along with uncertainties of reconstructed glacier variability. A multi-proxy data set consisting of sedimentological, physical, and geochemical data shows one main process, as extracted by means of principal component analysis (88% of the variance explained by the first PC), driving sediment variability in Nedre Sørsendalsvatn. The common signal extracted from the sediment data is indicative of glacial activity in the catchment and is interpreted to vary in concert with the changing glacier equilibrium-line altitude. The reconstruction of former glacier activity is in accordance with glacier variability reconstructed from other sites in western Norway, including the termination of the deglaciation at approximately 10,000 cal yr BP, the 8.2 ka BP (Finse) event, the Holocene thermal optimum between w8000 and 5500 cal yr BP, and the onset of the Neoglacial at 5500 cal yr BP. The largest glacial extent during the Neoglacial time period took place during the ‘Little Ice Age’. The combined radiocarbon chronologies from three different sediment cores provide insight into the duration of the “8.2 ka event” in the terrestrial system. The maximum glacier activity at approximately 8.2 cal BP is the culmination of a glacier advance that began around 9 cal BP and accelerated at 8.4 cal BP. The glacier advance ended abruptly at 8.0 cal BP. Ó 2013 Elsevier Ltd. All rights reserved.
Keywords: Holocene Glacier variability Lake sediments XRF Norway Numerical analyses
1. Introduction Alpine glaciers are commonly located in remote and highaltitude regions which are rarely covered by instrumental or historical records. Past and present sizes of alpine glaciers hold an integrated signature of atmospheric processes as their size is related to changes in both the ablation season temperature and the amount of solid winter precipitation. Robust glacier reconstruction can be an important source of knowledge regarding past climate and can provide a better understanding of natural climate variability (Jansen et al., 2007).
* Corresponding author. Department of Earth Science, University of Bergen, Allégaten 41, 5007 Bergen, Norway. E-mail address:
[email protected] (J. Bakke). 1 Currently at: Department of Biology, University of Bergen, Thormøhlensgate 53a, 5006 Bergen, Norway. 2 These authors contributed equally to this manuscript. 0277-3791/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.quascirev.2013.05.003
The Swedish scientist Wibjörn Karlén initially suggested that glacial erosion, and the associated production of rock-flour deposited in downstream lakes, could provide a continuous record of glacial fluctuations, overcoming the problem of incomplete reconstruction obtained by e.g. dating of marginal moraines or mega-fossils in glacier forelands (Karlén, 1976). The method of reconstructing glaciers based on analyses of sediments from distal glacier-fed lakes was later applied in other glaciated areas and has been further developed since, through several different approaches (e.g. Leemann and Niessen, 1994; Nesje et al., 2001; Bakke et al., 2010). The readings regarding glacial signals preserved in lake sediments now include applications of various methods. These methods measure the amount of minerogenic versus biologic matter (typically inferred from loss-on-ignition (LOI)), grain-size analysis, magnetic properties, Rare-Earth Elements, dry bulk density (DBD), and analysis of elemental composition by means of ITRAX-XRF, resulting in multivariate data sets suited to analyses by numerical methods.
2
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
While numerical analytical methods have been widely and successfully applied to biological proxy data sets (e.g. Birks et al., 2010; Borcard et al., 2010), these methods are more rarely applied to sedimentological data sets (e.g. Olsen et al., 2010; Vasskog et al., 2012). Numerical methods introduce objectivity, allow for a better understanding of interrelations among proxies, provide a sound base for subsequent data interpretation, and enable us to interpret more than one proxy at a time. Applying these numerical methods to sedimentological data will therefore improve the understanding and interpretation of large data sets and thereby objectify and strengthen interpretations of past climate. Important advances have been made in ageedepth modelling and uncertainty estimations of ageedepth models (e.g. Haslett and Parnell, 2008; Blaauw, 2010). Chronological uncertainty is one of the most critical uncertainties in climate reconstruction, based on sedimentary archives. This uncertainty is commonly acknowledged when presenting ageedepth models (e.g., Blaauw et al., 2007). However, only a few studies explore the transformation of age uncertainty into variable uncertainty (e.g. Blaauw et al., 2007; Parnell et al., 2008). In this study, we present a lacustrine sediment record based on three c. 3.5 m-long sediment cores from the distal glacier-fed lake Nedre Sørsendalsvatn located downstream of Blåbreen in Nordfjord, western Norway (Fig. 1). A suite of sediment analyses and numerical methods has been invoked in order to identify the glacial signal and quantify the precision of the glacier reconstruction. The first objective of this study is an analysis of the sedimentary proxy data sets obtained from the three sediment cores from Nedre Sørsendalsvatn by means of numerical methods in order to better understand the interrelations of sediment variables and improve subsequent inferences on glacier size variability. To achieve this, the cores have been objectively divided into stratigraphic zones of significantly different sediment variables. We then compared three pre-defined data sets (sedimentology, geochemistry and grain-size analyses) both internally and among one another. We thereby identified phases of agreement and disagreement between different variables. These differences were attributed either to differing secondary processes in the catchment or to non-linearity between the variables. Finally, a common signal among the different variables was extracted by means of principal component analysis (PCA). A second objective of this study was to transform uncertainty in the radiocarbon-based chronology into uncertainties in our knowledge of past glacier size variability in the catchment of Nedre Sørsendalsvatn. This was achieved by using existing ageedepth modelling algorithms based on bootstrapping techniques (Blaauw, 2010) in order to obtain a multitude of possible ageedepth models and thereafter transform the different agee depth models into variable uncertainties. Finally, we discuss and compare the findings from Nedre Sørsendalsvatn with Holocene glacier variability elsewhere in Scandinavia and in relation to selected palaeoclimate proxy archives within the Atlantic Ocean. 2. Study area 2.1. Glacier, climate and bedrock The lakes Øvre (¼Upper) and Nedre Sørsendalsvatn (61670 3500 N, 6 280 8500 E) are located in the upland mountain area to the south of Gloppenfjorden, a south-easterly trending branch of Nordfjord, at an elevation of 928 m and 918 m (Figs. 1 and 2). The topography of the area is dominated by several individual summits, with the highest, Botnafjellet (1572 m), to the southeast of the Sørsendalsvatn lakes (Figs. 1 and 2). Deep glacial troughs and cirque basins dissect the landscape. Some of the north-facing cirques and headwalls host small glaciers and large perennial snow patches. The catchment of Nedre Sørsendalsvatn covers an area of 8.5 km2,
which includes the glacier Blåbreen occupying an area of approximately 2 km2. The bedrock in the area is of Precambrian age and is dominated by gneissic bedrock (Bryhni and Grimstad, 1970). The equilibrium-line altitude (ELA) at Blåbreen is c. 1050 m a.s.l. in years when the net balance is close to zero (Østrem et al., 1988). The present climate is semi-continental to maritime with a mean (1961e1990) summer ablation season temperature (1 Maye30 September) of 12.12 C at the meteorological station Sandane (station no. 58,070, 51 m a.s.l.) (eKlima.no). Using an environmental lapse rate of 0.65 C/100 m (Sutherland, 1984) provides a mean summer temperature (Ts) at the present ELA (1050 m) of Blåbreen of c. 5.5 C. Winter precipitation (Pw) (1 Octobere30 April) based on the Myklebust station (station no. 58,320) shows a mean (1961e 1990) of 1023 mm at 315 m a.s.l. (eKlima.no). Using a suggested mean exponential increase in winter precipitation with altitude of 8%/100 m in southern Norway (Haakensen, 1989), the precipitation is calculated as close to 1800 mm at the ELA of Blåbreen. According to the “Liestøl-equation” (Liestøl in Sissons, 1979) there should be no glacier at Blåbreen based on the summer temperature and winter precipitation values. It is therefore evident that the supply of solid snow from wind drift and avalanching from Botnafjellet is important for sustaining a positive net mass balance at Blåbreen (Fig. 3). The glacier foreland and the glacial geomorphology of Blåbreen and its surroundings were intensively studied during the 1990’s as there was discussion regarding the age of the Neoglacial moraines in the area, with implications for the moraine chronologies of entire western Scandinavia (Evans et al., 1994; Matthews et al., 1996; Evans, 1997). A well-preserved recessional moraine sequence is mapped in front of Blåbreen (Evans et al., 1994). Marginal moraines on the northern side of Lake Øvre Sørsendalsvatn are well vegetated and not dated, whereas terminal moraines closer to Blåbreen have been subject to lichen measurements by Evans et al. (1994) and Matthews et al. (1996). Evans et al. (1994) distinguished three moraine stages. Both author groups agree that the stage 3 moraines are of late-glacial age, and that stage 1 moraines formed in the early 19th century, but disagree on the age of stage 2 moraines (Fig. 1). While Evans et al. (1994) dated stage 2 moraines between 400 and 800 cal yr BP, Matthews et al. (1996) questioned the lichen measurements of Evans et al. (1994) and concluded that these moraines are at least 5000 years old, based on Schmidt hammer measurements. Based on the lake sediment study presented below, we support the view of Matthews et al. (1996) that the stage 2 moraines are of early Holocene age (see Section 5.2). 2.2. Catchment lakes In this study, we present analyses of sediment cores retrieved from Nedre Sørsendalsvatn (Figs. 1 and 2). The present meltwater draining from the glacier Blåbreen is routed through the Øvre Sørsendalsvatn. This is a small lake of 0.34 km2 with one main inlet, three smaller tributary streams, and one outlet in the eastern part. The river from Blåbreen enters the lake through a well-developed glaciofluvial fan delta, where the direction of the river entering the lake changes between a northerly and westerly position from year to year due to changing sediment supply and runoff over the outwash plain. Both the glaciofluvial fan delta and the lake basin of Øvre Sørsendalsvatn are traps for the sediments transported with the glaciofluvial meltwater stream, implying that only sediments in suspension are transported further downstream and into Nedre Sørsendalsvatn. Nedre Sørsendalsvatn is 460 m along its longest axis and covers an area of 0.065 km2. The lake has two inlets, the main river from Øvre Sørsendalsvatn enters the lake in the NW corner and three smaller tributary streams enter the lake in the SW
Fig. 1. Study area showing coring locations (red dots), and late-glacial glacier extension margins (stages 2 and 3) and extent during the Little Ice Age (LIA), between AD 1750 and 1890. Inset shows an overview map over southern Norway. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
4
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
Fig. 2. Overview of the Lakes Øvre and Nedre Sørsendalsvatn and the glacier Blåbreen looking southward. Photographer: Atle Nesje.
part. The outlet from the lake is in the NE part, draining into the valley Sørsendalen before it enters Lake Breimsvatnet, which drains into the inner part of the fjord Gloppenfjorden (Fig. 1). Using lacustrine sediments retrieved from downstream glacier-fed lakes to reconstruct past glacier activity requires careful validation of records in question, especially as gravity driven processes in the alpine catchment can introduce noise when such archives are used to reconstruct glacier size variability. The effect of catchment processes must therefore be included in the interpretation of the sedimentary records. The slopes surrounding Nedre Sørsendalsvatn are gentle and covered with surficial material, mainly consisting of till and lateral moraines from the glacier reconstructed in stage 3 (Section 2.1). No active or relict avalanche tracks are mapped in the slopes around the lake. However, the relief around Nedre Sørsendalsvatn is substantial, especially with regard to the two summits Botnafjellet (650 m southeast of the lake) and Storevarden (500 m north of the lake), and possible snow avalanching from these areas cannot be ruled out. Around Øvre Sørsendalsvatn, and towards the glacier Blåbreen, several talus fans are mapped, two of which are situated at the southern shore of the lake and two are at the south side of the valley down valley from Blåbreen. The talus fans can influence the sedimentation in Øvre Sørsendalsvatn, and may also influence the sediments in Nedre Sørsendalsvatn, as sediments in suspension can be transported down valley with the river. 3. Methods The reconstruction of Holocene glacier fluctuations and the utilisation of the sediment parameters describing the sediments from Nedre Sørsendalsvatn are based on a combination of geomorphological mapping, lake coring, and various laboratory analyses. A firm chronology has been obtained by AMS radiocarbon dating (see below) of macrofossils. In addition, we have used earlier studies of lichen growth on marginal moraines in front of Blåbreen
as an independent dating of when the glacier was larger than it is at present (Evans et al., 1994; Matthews et al., 1996; Evans, 1997). 3.1. Coring and laboratory analyses In September 2005, three cores were retrieved from Nedre Sørsendalsvatn. The cores were retrieved from a raft using a piston corer with a 110-mm-diameter core tube constructed to obtain up to 5.75 m of sediments (Nesje, 1992). Many attempts were done for retrieving a short core covering the water sediment interface using a small gravity corer. The upper soft sediments are very hard and we only managed to take one core in the southern part of the lake. After splitting the cores horizontal in two parts, one was stored as reference and the other was carefully cleaned prior to photographing. Lithofacies and sedimentological structures and textures were described before the cores were scanned and sub-sampled for further analysis. The samples for loss-on-ignition (LOI), dry bulk density (DBD), and water content (WC), sampled contiguously at 0.5-cm intervals (SOP105; n ¼ 769, SOP205; n ¼ 322, SOP305; n ¼ 754), were dried overnight at 105 C in ceramic crucibles before the dry weight was measured (normally 0.5e2 g). Water content was calculated in % of total weight, and dry and wet bulk density (g/cm3) was measured using a syringe for fixed volume sample extraction. In the furnace, the samples were subjected to gradually rising temperatures for 0.5 h and then ignited at 550 C for 1 h. The crucibles were then cooled in a desiccator for approximately 0.5 h and then weighed at room temperature (w18e20 C). The weight LOI was calculated in percent of dry weight. Grain-size distributions were analysed using a Micromeretics Sedigraph 5100 (X-ray determination) in conjunction with a Micromeretics auto sampler. Samples were disaggregated in an ultrasonic bath for 30 s prior to analysis. Grainsize data were processed using the Gradistat 4.0 software for sediment parameters (Blott and Pye, 2001). The sediment cores were analysed at 1 cm sampling interval (SOP105; n ¼ 288;
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14 3
0
SOP105 DBD (g/cm ) 0.4 0.8 1.2
5 3
1.6
2
0
SOP205 DBD (g/cm ) 0.4 0.8 1.2
1.6 0
0
A 40
C 100
D 80
E F 200
120
G
Depth (cm) SOP205
Depth (cm) SOP105 and SOP305
B
160 300
H 200
400 0
0.4 0.8 1.2 1.6 SOP305 DBD (g/cm3)
2
Fig. 3. Correlation of the three cores based on DBD cores is correlated into eight units (AeH). Data are listed in Supplementary (web link).
SOP205; n ¼ 120; SOP305; n ¼ 66) covering the upper parts of the sediment cores. Magnetic susceptibility (MS) was measured using a Bartington MS2E sensor at 0.5 mm intervals on the cleaned face of the three sediment cores covered with polyethylene. The X-ray florescence (XRF) analyses were completed with an ITRAX, X-ray Fluorescence Core Scanner (ITRAX-XRF) (Croudance et al., 2006). The analysis was performed with a chromium tube, and the power was set to 30 kV and 50 mA using 10 s counting time at 500 mm increment. In situ reflectance spectroscopy was measured on cores SOP205 and SOP305 with a Gretag-Spectrolino (Gretag Macbeth) at intervals of 2 mm on the polyethylene-covered surface of split cores. Reflectance spectra were recorded at intervals of 10 nm between 380 and 730 nm. Each spectrum was corrected, accounting for illumination and transparency effects by dividing it by the spectrum of a transparency-covered white standard (BaSO4). VIS-RS has been applied successfully in many studies to infer the amount of organic material (Rein and Sirocko, 2002; Wolfe et al., 2006; Michelutti et al., 2010), the amount of clastic material (Rein et al., 2005) and for general characterisations of sediments (Debret et al., 2011). Organic matter, specific chemical compounds, and specific minerals (e.g. mica and chlorite) have distinct absorption/reflectance properties (e.g. Rein and Sirocko, 2002; Wolfe et al., 2006; Debret et al., 2011). Clastic sediments show a decrease in reflectance between 590 and 690 nm. This decrease is attributed to the amount of illite, chlorite, and biotite; thus, we used the ratio between reflectance at 590 and 690 nm as an indicator of clastic input (Rein and Sirocko, 2002; Trachsel et al., 2010). We expect an inverse relation between this indicator of clastic input and LOI. Subsequently, this data series was resampled to 5 mm resolution. Hereafter this data series is referred to as VIS-RS data.
3.2. Radiocarbon dating and ageedepth modelling For radiocarbon dating, plant macrofossil fragments were wet sieved (125 mm mesh) from the lake sediments. Thirteen samples of macrofossils and four bulk sediment samples were analysed using accelerator mass spectrometry (AMS). Radiocarbon dating was carried out by the Poznan Radiocarbon Laboratory in Poland and by Beta Analytic in Florida. Ageedepth models were constructed using R-source code provided by Blaauw (2010). For each radiocarbon date, the actual year used for ageedepth modelling was drawn based on the calibrated radiocarbon age, with sampling probability proportional to the probability density of the calibrated radiocarbon age. As suggested by Heegaard et al. (2005) and further developed by Blaauw (2010), we included a term taking into account the uncertainty of the uncalibrated radiocarbon date. The year that was later on calibrated was drawn from the probability density function of the uncalibrated radiocarbon age, with sampling probability proportional to the probability density of the uncalibrated radiocarbon age. A smooth spline (smoothing parameter ¼ 0.3) was then drawn through the age estimates, with the spline weighted by the calibration probabilities of the individual age estimates. This procedure was repeated 1000 times for each core. Age-models showing age reversals were removed prior to further analysis.
3.3. Statistical methods Statistical analyses were carried out in order to: i) Sub-divide the record into different, statistically significant stratigraphical sequences
6
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
ii) Compare the different data sets on physical sediment variability (DBD, VIS-RS, grain-size and MS), geochemistry (XRF measurements) and organic variability (LOI) internally and among each other iii) Extract a signal common to all variables iv) Incorporate the chronological uncertainty into the signal uncertainty Down-core sedimentological variability is usually divided into different stratigraphic units by visual inspection and variability in MS (e.g. Bakke et al., 2010). In this study, the different units were identified by means of objective clustering methods that are constrained by the stratigraphy (i.e. depth). For this purpose, the constrained incremental sum of squares clustering (CONISS) (Grimm, 1987) was used and the number of significant zones was determined by the broken stick model (Bennett, 1996). The data sets were internally compared by calculating correlation coefficients (Pearson’s product-moment and Spearman’s rank correlation coefficients) among the different variables. PCA, (Hotelling, 1933) was applied in order to detect the amount of common variance in the data sets, and eigenvalue decompositions were calculated on the correlation matrix (i.e. standardized variables were used for PCA). The number of significant principal component axes was based on the broken stick model (Frontier, 1976). As PCA assumes linearity between variables (e.g., Legendre and Legendre, 1998), non-linearity between variables might reduce the common variance detected by PCA. Non-linearity was detected by comparing Pearson’s and Spearman’s correlation coefficients, which test for linearity and monotony, respectively, and by visual inspection of scatter plots. Non-linearity was accounted for by looking for an appropriate function in order to remove it (e.g. a function LOI ¼ b1*DBD^b2). Parameters b1 and b2 were estimated by means of simulated annealing (Kirkpatrick et al., 1983) and aimed to reduce root mean square error of prediction between transformed variable and target variable. Cluster analysis is an objective method of dividing variables or samples into groups based on a distance matrix between variables and samples, respectively (e.g., Legendre and Legendre, 1998). We applied cluster analysis (average linkage, euclidean distances) to test whether non-linearity between variables resulted in clustering variables linear to each other in the same clusters. We then compared the data sets, sedimentology (LOI, DBD, MS and VIS-RS), geochemistry (ITRAX-XRF data), and grain-size among one another by calculating RV coefficients (e.g. Abdi, 2007), a multivariate extension of Pearson’s correlation coefficient and by applying co-inertia analysis (Dolédec and Chessel, 1994) and redundancy analysis (RDA, van der Wollenberg, 1977). Both RDA and co-inertia analysis are constrained ordination techniques and aim at comparing two data sets. In contrast to PCA were the ordination is exclusively based on the data set on which an ordination is sought, external variables influence the ordination in RDA and coinertia analysis. This influence is fundamentally different in RDA and co-inertia analysis. In RDA ordination is calculated in one of the data sets compared. The data set in which ordination is calculated is slave to the other (constraining) data set. The ordination axes are linear combinations of the constraining data set (Legendre and Legendre, 1998). In co-inertia analysis the two data sets compared are equal. In both data sets ordination axes are sought, so that the covariance between the scores of these axes is maximised (Dolédec and Chessel, 1994). Hence RDA is appropriate when comparing two data sets with clear causal relationships and dependencies, whereas co-inertia analysis is the method of choice when comparing two independent data sets. As ITRAX-XRF data are affected by water content, organic carbon and variations in grainsize [specific minerals (and thereby specific elements) more
abundant in specific grain-size categories (and thereby as well specific elements)] it seems appropriate to compare these data sets by RDA, whereas LOI and grain-size that are independent should be compared by co-inertia analysis. As grain-size data are affected by the closure effect, conventional data analysis techniques are not applicable to this data set. Therefore, we applied log transformations prior to statistical analysis (Aitchison, 1983). When comparing grain size and ITRAX-XRF data, we only used the internal distribution of geochemical data (i.e. we divide the count rate of one element by the sum of the counts of all elements). After comparing data sets among each other, we aimed to extract a signal common to those compared using PCA. In order to combine chronological uncertainty into signal uncertainty, we used the chronology replicates obtained in the Monte Carlo ageedepth modelling (Blaauw, 2010), which were resampled to a common 40-year resolution. The rates of change (ROC, derivatives) of each time series were calculated to compare their relative timing and magnitude change. To calculate ROC, data were re-sampled to a common temporal resolution equal to the lowest temporal resolution in the entire core, in core SOP105 it was set to 40 years (e.g., Birks and Amman, 2000). Evenly sampled time-series are a fundamental prerequisite for calculating interpretable ROCs. ROCs were calculated for all models obtained in the ageedepth modelling using Clam. For the time span between 8500 and 7600 cal yr BP, we used a local linear mixed model to examine the composite sedimentological and geochemical data set (Heegaard and Nilsen, 2007). We then used the shared local predictions to determine the onset of the rapid increase in glacial size towards the maximum glacial extent associated with the 8.2 ka event and the end of the rapid decrease in glacial size following the 8.2 ka event, respectively (e.g., Rohling and Pälike, 2005) based on the first derivatives of the curve of shared local predictions. Statistical analysis was performed using the open-source software R (r-project.org) and its add-on packages Clam (Blaauw, 2010), vegan (Oksanen et al., 2011), rioja (Juggins, 2009), ade4 (Chessel and Duffour, 2011), compositions (van der Boogaart et al., 2011) and llmm (Heegaard and Nilsen, 2007). 4. Results 4.1. Radiocarbon dating and ageedepth modelling AMS radiocarbon ages on macroscopic plant remains from 13 levels (Table 1) form the basis for ageedepth relationship for the Holocene part of the cores. Sufficient organic material could not be extracted for AMS radiocarbon dating from some selected intervals of the upper section of the cores, and dating on bulk sediments was therefore attempted. The results include several stratigraphically inverted ages, and we therefore rejected all the radiocarbon dates on bulk sediment. We used the radiocarbon ages from SOP305 (6 dates) to obtain a master chronology. As the three cores are easily correlated (Fig. 4), radiocarbon dates obtained from the two other cores were projected on to the stratigraphy of SOP305 (SOP105 ¼ 3, SOP205 ¼ 2 and SOP305 ¼ 6). The ageedepth model show high accumulation rates prior to 8200 cal yr BP, reduced sediment accumulation between 8000 and 2000 cal yr BP, and high accumulation rates during the last 2000 years (Fig. 4). Uncertainties in the ageedepth model were moderate before 8000 cal yr BP (in core 305 at 170 cm depth for example, the 2.5 and 97.5 percentiles of the estimated ages differed by 200 years) but increased substantially in the earliest parts of the record (in core 305 at 210 cm depth, the 2.5 and 97.5 percentiles of the estimated ages differed by 600 years).
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
7
Table 1 Radiocarbon dates obtained from the cores studied. Core
Depth (cm)b
Lab-no
14
SOP-105 SOP-105 SOP-105 SOP-105 SOP-205 SOP-205 SOP-205 SOP-205 SOP-305 SOP-305 SOP-305 SOP-305 SOP-305 SOP-305 SOP-305 SOP-305 SOP-305
63 131 175 229 16 41 70 98 11 20 51 51 99 120 153 161 205
Poz-28659 Poz-28700 Poz-28701 Poz-28702 Beta-304584 Beta-304585 Beta-302970 Beta-304586 Beta-304587 Poz-28660 Beta-304588 Beta-304951 Beta-302971 Poz-28704 Poz-28705 Poz-28706 Beta-302972
1890 4654 7050 7430 3180 2170 4660 6960 2590 330 2730 2180 4090 5250 7040 6930 8910
a b c
C age
30 35 50 70 40 30 40 40 40 140 40 30 30 40 60 50 50
Calibrated agec (cal yr BP) (min)
Calibrated age (cal yr BP) (max)
Leaf fragments, twig and bark Bark and leave fragments Seed, leave fragments, trig, bark Leaf fragments, trig, barka Organic sedimenta Organic sedimenta Salix twig Wood Organic sedimenta Bark and small twig Organic sedimenta Chitin Salix bark Moss Leaf fragments, trig, bark, seed Leaf fragments, barka Salix twig
1735 5311 7762 8052 3339 2066 5311 7691 2504 3 2756 2118 4448 5923 7732 7672 9799
1893 7772 7970 8387 3475 2310 5572 7922 2778 617 2922 2312 4808 6178 7971 7920 10,203
Samples were rejected for ageedepth modelling. All samples extracted from 1 cm slices. Calibrated using Calib 6.0.
4.2. Lithostratigraphy
40 60
100 150 200
100
80
SOP 305 SOP 205 SOP 105
SOP-205 depth (cm)
50
20
0
0
Cores SOP105 (N61.67406 E6.28868) and SOP305 (N61.67401 E6.8859) were retrieved 10 m apart in the north-eastern part of the lake, whereas the core SOP205 (N61.67367 E6.28937) was taken 100 m further to the southeast in the lake (Fig. 1). The lithostratigraphy in the three cores was correlated into eight units defined by CONISS applied to DBD and LOI (Fig. 4, Supplementary Fig. 1); these were mimicked in other variables such as MS, geochemical, and grain-size data. We used SOP105 to describe the lithostratigraphy and define the units as well as for a detailed description of the sediment parameters, as this core showed the highest sedimentation rate (Fig. 5).
SOP-105/305 depth (cm)
Material dated
10000
8000
6000
4000
2000
0
Age (cal yr BP) Fig. 4. Ageedepth relationship for the cores from Nedre Sørsendalsvatn. Thick center lines are best estimate models and grey shadings are 95% confidence intervals. Blue shaded areas are probability density functions of calibrated radiocarbon dates used in ageedepth modelling. Black histograms are probability density functions of calibrated radiocarbon dates rejected in the ageedepth modelling. Radiocarbon ages are listed in Table 1. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Unit H consists of massive minerogenic sediments of clay and silt. DBD is above 1.2 g/cm3 throughout the unit. The organic content is very low in the sediments and they are not dated (Unit H: SOP105: 229.5e380 cm; SOP205: 110.5e161 cm and SOP305: 201.5e374 cm). Unit G consists of light brown to grey massive gyttja, with DBD values between 0.4 and 0.6 g/cm3 and LOI increases from 6 to 12% with high frequent fluctuations (Unit G: SOP105: 187e229.5 cm; SOP205: 105.5e110.5 cm and SOP305: 167e201.5 cm). Unit F consists of grey silty clay grading to more organic-rich sediment at both upper and lower boundaries. The various sediment parameters show an asymmetric peak, with a moderate increase and a steep decline upward in the core. DBD shows a sharp peak with values of 0.7 g/cm3, and LOI decreases to 4% (Unit F: SOP105: 175.5e187 cm; SOP205: 95.5e105.5 cm and SOP305: 154.5e167 cm). Unit E consists of greyish to dark brown massive gyttja and has the lowest DBD values throughout the core, with values stable at 0.3 g/cm3 and LOI fluctuating between 15 and 18% (Unit E: SOP105: 130.5e175.5 cm; SOP205: 72.5e95.5 cm and SOP305: 108.5e 154.5 cm). Unit D consists of a weakly laminated grey to brown silty gyttja. The laminations are separated by gradational contacts and are irregular in thickness (0.1e0.6 cm); the transition towards Unit C is gradual towards browner sediment. The DBD values fluctuate around 0.4 g/cm3 and LOI is between 5 and 11%, with highfrequency fluctuations (Unit D: SOP105: 87.5e130.5 cm; SOP205: 59.5e72.5 cm and SOP305: 63.5e108.5 cm). Unit C consists of a grey to brown silty gyttja with DBD values fluctuating around 0.4 g/cm3 and LOI decreasing from 8 to 5% with high-frequency fluctuations (SOP105: 63.5e87.5 cm; SOP205: 48e 59.5 cm and SOP305: 41.5e63.5 cm). Unit B consists of grey massive minerogenic silty clay without visible structures and with a graded transition from unit C; DBD has sharply increasing values from c. 0.4 to 1.2 g/cm3 and LOI decreases from 12 to 4% (Unit B: SOP105: 47e63.5 cm; SOP205: 35.5e48 cm and SOP305: 21.5e41.5 cm). In unit A the sediments are grey, massive, and have graded transition from browner sediments in unit B, intersected by layers of light grey silt visible in the x-ray photographs (Supplementary Fig. 2). DBD values increase from 0.9 to 1.2 g/cm3, and LOI
8
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14 Bulk density (g/cm ) 0,2
1,2
Water content (%) 350
200
18
10
Coarse silt (%)
Clay (%)
Loss-on-ignition(%) 0
0
60
30
0
0
10
Mag.Suceptibility (10 ) Ti-count rate 20 0
10
1000 5
100
10
15
20
25
30
0
0
2000
50
1890±30
4000
100 4655±35
6000
7050±50
Depth (cm)
Age (cal yr BP)
0,6
Measurement error on lower section
150
8000
200 7430±50 10000
250
300
350
400
Fig. 5. Sediment variables measured on core SOP10 along with line-scan image of core face and radiocarbon ages (14C yr BP). Data are listed in Supplementary Table 1.
decreases from 4 to 2% (Unit A: SOP105:0e47 cm; SOP205: 0e 35.5 cm; SOP305: 0e21.5 cm).
4.3. XRF analyses The six chemical elements, silicon (Si), potassium (K), calcium (Ca), titanium (Ti), iron (Fe) and strontium (Sr), have count rates above background. All elements show similar progressions: i.e. high count rates in unit A, generally decreasing count rates in units B, C, and D, with lowest values in Section E. Section F exhibits a sharp increase in count rates, whereas counts are low again in unit G. At the transition between units G and H, count rates increase sharply and are highest in unit H. The count rates are in general agreement with the DBD values as opposed to LOI values (see Section 4.5).
4.4. Reflectance spectroscopy VIS-RS data measured on core SOP205 are in agreement with LOI (r ¼ 0.93) (Fig. 6). The VIS-RS values were particularly high in units D to G. In units A to C, VIS-RS fluctuates whereas LOI show no variability. In units A to D, VIS-RS is highly covariant with Fe count rates (r ¼ 0.86, entire core r ¼ 0.92). 4.5. Data inter-comparison In this section, we analyse and compare the sedimentological data set (DBD, LOI, VIS-RS and MS), geochemical data (ITRAX-XRF) and grain-size data internally and among one another. We restrict this analysis to core SOP205 as this is the only core analysed with the full suite of methods, and results obtained from the other cores are generally similar to the results presented below.
B
depth [cm]
C
Coarse Silt Medium Silt Fine Silt Very Fine Silt Clay
percent
z−scores
z−scores
A
depth [cm]
depth [cm]
Fig. 6. (A) Sediment variables measured on core SOP205 standardized to have a mean of zero and unit variance (z-score), DBD and log MS are plotted upside down. (B) ITRAX-XRF count rate, all count rates plotted upside down. (C) Grain-size measurements. Dashed vertical lines: zone boundaries as determined by CONISS. Data are listed in Supplementary Table 1.
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
Agreement is especially high between LOI and VIS-RS, between DBD and log MS (Fig. 6a). These two groups differ considerably between 40 and 70 cm depth in the core, which is confirmed by cluster analysis, where LOI and VIS-RS, and DBD and log MS are merged first. Focussing on LOI and DBD, we find that both variables are monotonically, but non-linearly, related (Supplementary Fig. 3a), which is also expressed in differing Pearson’s and Spearman’s correlation coefficients of 0.89 and 0.98, respectively. LOI and DBD are related by a function (Supplementary Fig. 3a):
LOI ¼ b1*DBDb2 ; where b2 < 0 Estimating parameters b1 and b2 by means of simulated annealing yields values of 1.95 and 1.29 respectively, RMSEP is 0.72%, the correlation coefficient increases to r ¼ 0.98, and the transformed DBD is perfectly linear to LOI (Supplementary Fig. 3b). PCA was applied to both raw and linearised data (LOI, DBD, VISRS and log MS). Both PCAs resulted in one significant axis, explaining 84% and 91% of the variance for the raw and linearised data. The ITRAX-XRF data set of the chemical elements Si, K, Ca, Ti, Fe, and Sr are highly covariant (Fig. 6b). CONISS divides the data into six zones, with borders at 41, 58, 70, 97.5 and 111 cm. Ca, Si, and Sr show a perfect monotonic but slightly non-linear relationship. Fe, K and Ti are perfectly linear between 41 and 120 cm, whereas between 1 and 41 cm the values do not vary in concert (Supplementary Fig. 5). The cluster analysis also reveals two groups consisting of Si, Ca and Sr, and Fe, K and Ti, respectively. The only significant PC-axis explains 86% of the variance of the six chemical elements. CONISS and the broken stick model divide the grain-size data into six zones (boundaries at 22, 37, 69, 84, and 99 cm) (Fig. 6c). The first two are characterised by gradually decreasing percentages of coarse (12%e5%) and medium silt (30%e16%) and a concomitant increase in clay (17%e22%). The third zone is characterised by significantly increasing amounts of clay (20%e55%), which decrease again towards the end of the third zone, while medium and fine silt show exactly opposite progression. Zones four to six show very rapid, high-amplitude fluctuations in the content of clay, very fine silt, and fine silt. A marked increase in coarse and medium silt is detected between 99 and 105 cm. Comparing CONISS applied to the three data sets, we find three common zone boundaries at ca. 37 cm (ca. 2000 cal yr BP), ca. 69 cm (ca. 5700 cal yr BP) and at ca. 99 cm (ca. 8300 cal yr BP) in depth. There is high accordance between the sedimentological and geochemical data sets (Fig. 6a and b). This is expressed in an RV coefficient of RV ¼ 0.95. Comparing individual variables, it seems DBD, log-transformed MS, and the elements Si, Ca, and Sr are linearly related to one another, whereas LOI, VIS-RS, Fe, K and Ti are linearly related (Supplementary Fig. 5). This division is supported by cluster analysis which merges LOI, VIS-RS, Fe, Ti and K on one side and DBD, log-transformed MS, Si, Ca and Sr on the other. As ITRAX-XRF measurements are affected by organic matter and water content (see discussion), we tested how strongly these variables influence count rates of detected measurements. The count rate of all elements detectable above background levels are negatively correlated with LOI (r < 0.81, i.e. absolute values >0.81), the total count rate correlates to LOI with r ¼ 0.93. The first RDA-axis explains 89% of the variance of the geochemistry data when constrained by water content (in percent of total weight) and LOI together and 82% of the variance when the XRF data are constrained by LOI. When comparing the geochemistry data with grain size, we find no clear relationship (RV ¼ 0.23). Calculating RDA, where the geochemical data set is constrained by grain size, the only
9
significant RDA axis explains 66% of the variations in the geochemistry. This relatively low accordance between grain-size and geochemistry, however, is only valid when considering the entire record. When only considering data between 1 and 69 cm in depth (depth is constrained by CONISS), where LOI is lower than 10%, the RV coefficient is considerably higher (RV ¼ 0.81). In RDA, where geochemistry is constrained by grain-size variations, the first RDA axis explains 90% of the variation in geochemistry. This high agreement is mainly due to positive (negative) correlations between coarse, medium and fine silt and Si, Ca and Sr (Fe and Ti) and negative (positive) correlations between clay and very fine silt and Si, Ca, Sr (Fe, Ti), as shown in Table 2. When concentrations of Si, Ca, and Sr (Fe, Ti) are high (low), the relative amount of silt is high whereas the relative amount of clay is low (high) provided Si, Ca, Sr (Fe, Ti) concentrations are low (high). For depths between 1 and 69 cm, grain-size composition strongly influences DBD (the first RDA axis explains 93% of variance of DBD). LOI explains exactly the same amount of variance in DBD. Comparing LOI and MS and grain-size between 1 and 69 cm depth (RV ¼ 0.72), the correlation between the two first co-inertia axis scores is r ¼ 0.88. The scores from the first co-inertia axes explain 71% of the variance in grain-size data and 92% of the variance in LOI and MS. As the relationship between the grain-size distribution and the other variables changes, we only used sedimentological and geochemical data for extracting a common signal by means of PCA. Applying PCA to the sedimentological and geochemical data, we find one significant PC-axis explaining 88% of the variance (Fig. 7). Focussing on depths between 1 and 69 cm, including grain-size data, we find two significant PC-axes explaining 70% and 17.5% of the total variance. Hence sediment data measured between 1 and 69 cm (5700 cal BP and the present) are in high accordance. 4.6. DBD variability We used the variable DBD to demonstrate the transformation of chronology uncertainty into variable uncertainty. For our experiment, we use DBD1.5 (i.e. 1/DBD1.5) as this is a likely way in which DBD is translated into ELA altitude variations (Bakke et al., 2010). The DBD data set represents the sum of the inorganic input into the lake, including sediments produced by glacier abrasion and potential gravity driven catchment processes (Bakke et al., 2005) (see Section 2.1). We combined the records of all three cores to assess the influence of different coring sites (Fig. 8). Among the different cores the timing of the transition from Unit H to Unit G has an uncertainty of 400 years and occurred between 10,300 and 9900 cal yr BP. Thereafter, DBD stays low up to approximately 9000 cal yr BP. After 9000 cal yr BP, DBD values increased gradually, accelerated at 8400 cal yr BP, and culminated at approximately 8200 cal yr BP. This event recorded in the DBD parameter ends abruptly at 8000 cal yr BP seen in all cores. During the period between 7900 and 5500 cal yr BP, the lowest DBD values is recorded. DBD values increased at approximately 5500 cal yr BP.
Table 2 Pearson’s product moment correlation coefficients (r) between element composition and grain-size classes (depth 1e69 cm). Very coarse silt Coarse silt Medium silt Fine silt Very fine silt Clay Si 0.20 K 0.40 Ca 0.13 Ti 0.50 Fe 0.21 Sr 0.02
0.75 0.64 0.80 0.13 0.77 0.77
0.94 0.88 0.96 0.39 0.95 0.85
0.86 0.88 0.85 0.64 0.86 0.71
0.56 0.44 0.59 0.07 0.56 0.54
0.95 0.91 0.97 0.51 0.96 0.86
10
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
Fig. 7. Sediment variables and PCA axis 1 scores for raw and linearised data. a) raw data b) linearised data. Thick orange lines are PCA axis 1 scores. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
The increase in DBD is difficult to quantify because of the large spread in the preceding period. Between 5500 and 2000 cal yr BP, DBD values remained relatively stable. Due to dating uncertainties, no distinct phases of increased or decreased DBD values are detectable. We then find a 1000-year long increase in DBD values, culminating in high values during the last 1000 years (Fig. 8). ROCs were calculated for all variables considered in PCA. Two periods with increased ROCs in all variables were found around 10,000 and 8100 cal yr BP. 5. Discussion 5.1. Ageedepth modelling Ageedepth models calculated using Blaauw’s (2010) algorithm are based on eleven radiocarbon dates. These radiocarbon dates are not evenly spread over time and there are fewer dates between 5000 and 2000 cal yr BP. The width of the confidence intervals summarizes the accuracy as a function of the dates sample density and sedimentation rate. Age control is particularly solid in phases where sediment parameters/variables indicate strong changes. More radiocarbon ages were obtained in these phases (4 dates
between 6000 and 4500 cal BP, 3 dates around 8200 cal BP, and 2 dates around 2000 cal BP) and correlations between the different cores are stronger in during these timespans. The increase in LOI values at approximately 8000 cal yr BP was for example dated at exactly the same age in all three cores (base of unit F). The strong change in LOI at approximately 5500 cal yr BP is well constrained by four 14C dates between 6000 and 5500 cal yr BP. Finally, the decrease in LOI at approximately 2000 cal yr BP is constrained by two dates. The increased sedimentation rates between 10,000 and 8000 cal yr BP and 2000 cal yr BP to the present are only constrained by one radiocarbon date or one radiocarbon date and the sediment surface considered to represent present day, respectively. These phases are therefore less reliable based on the spacing of dates and the confidence interval in the ageedepth models. 5.2. Holocene glacier variability In this section, we discuss the interpretation of glacier variability based on the DBD variability. As the nature of glacial erosion is reflected by the supply of insoluble particles to a river system, analyses of physical properties of the glacial sediments may be a diagnostic parameter for variations in glacier size. Warm-based
Fig. 8. Holocene DBD variability in the three cores from Nedre Sørsendalsvatn. DBD is used to quantify glacier size variability in the catchment of Nedre Sørsendalsvatn. Results are based on all ageedepth models calculated for the three cores. Red indicate high probability for glacial-activity, yellow and green indicate low probability and white indicate less probability in the reconstruction of glacial variability. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
glaciers produce abundant clay-silt size fractions that are transported downstream to produce characteristic signatures in glaciolacustrine sediments. Bakke et al. (2005, 2010) showed how variability in DBD could be linked to glacier size (and hence ELA) fluctuations through time. By definition, bulk density expresses the ratio of the mass of dry solids to the bulk volume of the sediment (Blake and Hartge, 1986). Commonly, this parameter defines how granular, fibrous and powdery materials pack or consolidate under a variety of conditions and can be used to calculate the porosity of the sediment. After changes in organic content (reflected in LOI), changes in flux and packing (reflected in grain-size composition) are probably the most important parameter affecting DBD values in a glacial-fed lake (Webb and Orr, 1997). Based on our interpretation of DBD variability as a measure on glacier size variability we conclude that the glacier advance during the 8.2 ka BP event was smaller than during the “Little Ice Age”. Based on this the, age of moraine stage 2, which was subject for a discussion between Matthews et al. (1996) and Evans 1997, must pre-date the 8.2 ka BP event. The most likely age of this moraine stage is therefore the glacier advance called the “Erdalen Event” dated to 10,200e9700 cal yr BP at Jostedalsbreen (Nesje et al., 1991). This time interval is not possible to resolve based on the sediments deposited in Nedre Sørsendalsvatn. 5.2.1. The 8.2 ka BP event Alley and Augustdottir (2005) described the 8.2 ka BP event as ‘a prominent, abrupt climate event about 8200 years ago’ that ‘brought generally cold and dry conditions to broad northernhemisphere regions especially in wintertime, in response to a very large outburst flood that freshened the North Atlantic.’ In the following, we describe the 8.2 ka BP event as detected in the sediments of Nedre Sørsendalsvatn. For comprehensive discussions of the proxy data available for the 8.2 ka BP event, refer to Rohling and Pälike (2005), Alley and Augustdottir (2005), and Daley et al. (2011). Additionally, numerous modelling studies discussing the 8.2 ka BP event have been presented [see Born and Levermann (2010) and references therein]. The 8.2 ka BP event, in Norway termed the “Finse-event” (Dahl and Nesje, 1994, 1996; Nesje and Dahl, 2001), is clearly visible and chronologically well constrained in all three cores. It is characterised by two differentiable sharp increases in glacial size followed by a sharp decrease in glacier activity. Maximum glacial size was recorded between 8280 and 8060 cal yr BP, depending on the age model used. The strong increase in glacial activity, as determined by a local linear mixed model, started at 8390 cal yr BP (8300e8470), and the rapid decrease in glacial activity after the 8.2 ka BP event ended at 7890 cal yr BP (7810e7950). When considering the individual age models, this interval lasted 480 years (450e510). The increase in glacial activity lasted 250 years (230e270 years), and the decrease in glacial activity lasted 230 years (220e240 years). The timing of the different features related to the 8.2 ka BP (Finse) event in Nedre Sørsendalsvatn is largely in agreement with the age of similar features in five lake records in southern Norway (Nesje and Dahl, 2001). The lowest v180 values were recorded 8186 calendar years BP with uncertainties 8139e8233 cal yr BP (8236 47 B2K) in the Greenland ice-cores (Vinther et al., 2006). This allows for simultaneous events but would also allow for a glacier response lagging behind the event recorded in Greenland ice-cores. Like Rohling and Pälike (2005), we find evidence for a glacial advance prior to 8200 cal yr BP (as indicated by ROCs). Rohling and Pälike (2005) as well as Daley et al. (2011), date the onset of the gradual cooling to 8600 cal yr BP. In Nedre Sørsendalsvatn, the increase in glacier size seems to have lasted even longer. Minimum glacier size in the early Holocene was reached around
11
9000 cal yr BP, followed by a period with advancing glacier size that accelerated at approximately 8400 (8300e8470) cal yr BP and culminates at 8200 cal yr BP. The most prominent feature in Nedre Sørsendalsvatn is, however, the abrupt decrease of glacier size after the 8.2 ka BP event. ROCs are considerably higher than during increasing glacier size. Hence, this abrupt end of the 8.2 ka BP event might be exclusively found in glacier records. For instance, it is at least partly reproduced for Myklebustbreen 30 km to the east of Blåbreen (Nesje et al., 2001). 5.2.2. Neoglaciation Glacial size seems to increase between 6000 and 5500 cal yr BP (see Section 5.3). This onset of glaciation is not marked by increased ROCs. An increase in glacier size in western Norway at approximately 5500 cal yr BP was recorded for the nearby glacier Jostedalsbreen (e.g., Nesje, 2009) and for glaciers in the catchment of Lake Nerfloen, Stryn (Vasskog et al., 2012). Similar ages for the onset of the Neoglaciation are found for the Folgefonna icecap (Bakke et al., 2005) located approximately 180 km south of our study area (Fig. 9). According to Geirsdottir et al. (2009), a period of ‘Neoglacial cooling’ occurred in Iceland between 6000 and 5500 cal yr BP, which points to a synchronous, large-scale climate change in the North Atlantic region. The onset of the Neoglacial at different sites in Southern Norway does not depend on altitude of the glaciated areas and therefore points towards an abrupt change in boundary conditions in the eastern sector of the North Atlantic. Dating uncertainties and differences between cores do not allow for detailed detection of changes in glacier size between 5700 and 2000 cal yr BP. Data obtained from SOP205 indicate rapid increase in glacier size at approximately 5700 cal yr BP, whereas cores SOP105 and SOP305 suggest rapid increase in glacial size between 2000 and 1000 cal yr BP. Between 2000 and 800 cal yr BP, the glacial activity in the catchment of Nedre Sørsendalsvatn increased. Available summer temperature reconstructions are ambiguous for this time period. Though no general cooling trend is found in treering data from Scandinavia (e.g. Grudd et al., 2002; Grudd, 2008; Linderholm et al., 2010), a combined pollen record based on 36 individual records (Seppä et al., 2009) shows decreasing temperatures. It is therefore impossible to assess how much of this increase in glacial size that can be attributed to changes in summer temperature versus winter precipitation. The largest reconstructed glacier extent at Blåbreen was during the last 1000 years, especially during the Little Ice Age (LIA). This is in accordance with all glacier reconstructions available for western Norway (e.g., Nesje et al., 2001; Nesje, 2009; Vasskog et al., 2012), northern Norway (e.g., Bakke et al., 2010, see Fig. 9) and is also in accordance with glacier fluctuations in Iceland (Geirsdottir et al., 2009) and in the European Alps (e.g. Holzhauser, 2007).
5.3. Data inter-comparison The data from Nedre Sørsendalsvatn show high internal accordance, which is especially pronounced between 5500 cal yr BP and the present. Sedimentological and geochemical data are in high accordance, as expressed by the first PC-axis of the combined data set explaining 85% of the variance and an RV-coefficient RV ¼ 0.95. Regarding the study site, based on the high ratio of glacier to nonglacier covered area of the drainage area to Nedre Sørsendalsvatn (2/7 km2), it is assumed that glacial activity in the catchment is the most important process governing clastic sedimentation in Nedre Sørsendalsvatn. Blåbreen currently cover ca. 25% of the catchment, and the sediment-laden meltwater stream from Blåbreen and Øvre Sørsendalsvatn is the main conveyor of sediment into Nedre Sørsendalsvatn.
12
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
Fig. 9. Comparison of glacial activity reconstructed for Blåbreen and other records of glacial activity in Scandinavia and one record showing the 8.2 ka event in the North Atlantic as expressed by sortable silt. A; this study, B; ELA at Folgefonna (Bakke et al., 2005); C; sortable silt (Ellison et al., 2006), D; glacier activity in Nerfloen (Vasskog et al., 2012), E; ELA at Okstindan (Bakke et al., 2010).
Accordance in sediments variables between ca. 5500 cal yr BP and the present (Neoglacial, 2 PCs explaining 89% of the variance of these 15 variables), and the detection of a major change in all three data sets around 5500 cal yr BP is indicative of a major change in sedimentation regime at that time interpreted as the onset of the
Neoglacial. Larger glacier will affect the sediment regime in Nedre Sørsendalsvatn in two ways: 1) Larger glacier will be more erosive and hence deliver more inorganic sediments to the lake system quantified as higher DBD values. 2) Larger glacier will release more melt-water to the glaciofluvial system, hence changing the
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
transport competence of the river system, quantified in the grainsize distribution as an increase in coarser grain sizes deposited distal in the lake system. The high frequency noise apparent in grain-size data can be attributed to the influence of other catchment processes such as slope wash and flooding; however, the main trend in grain size distribution is superimposed on this noise and is interpreted to reflect changes in glacier size. This interpretation is supported over the interval between 5500 cal yr BP and the present (for core 205 from 1 to 69 cm depth and between 1 and 130 cm for core 105). The amount of silt is high when sedimentological and geochemical data indicate large glacier extent (i.e. in the uppermost part of the core) and decreases when the glacier front retreats. The relative amount of coarse and medium silt decreases throughout the entire core section, whereas the relative amount of fine silt remains constant between 0 and 38 cm and only begins to decrease at the point when LOI values are higher than 3%. This gradual reaction of the different silt classes, as well as the continuous increase in relative clay amount, reflects the transport competence of the glaciofluvial river running through the lake Nedre Sørsendalsvatn. Between 99 and 120 cm depth in core 205 (from deglaciation to 8000 cal yr BP) grain-size and geochemical data are related. The lack of accordance between grain-size and geochemistry between 5500 and 8000 cal yr BP (Holocene climate optimum, 69e98 cm depth) indicates low or no glacial activity in the catchment, which is supported by the highest LOI values, low DBD, and lowest counts-rates of chemical elements. In this interval, LOI values show high variability, whereas DBD and geochemical data remain constant. This indicates factors other than glacial activity, most likely summer temperatures and/or rain-/snowfall induced river runoff, influencing LOI. The common signal shared between geochemical and sedimentological data is remarkably high (PCA 1 explaining 88% of the variance). The common signal extracted in PCA 1 is affected by the variance in LOI as organic content strongly influences all element intensities (LOI explaining 82% of the variance of geochemistry). Löwemark et al. (2011) explain this effect: ‘As XRF scanners are largely insensitive to organic material in the sediment, increasing levels of organic material effectively dilute those components that can be measured, such as the lithogenic material (the closed-sum effect). Consequently, in sediments with large variations in organic material, the measured variations in an element will to a large extent mirror the changes in organic material.’ PCA assumes linearity between variables (eigenvalue decompositions are calculated on correlation matrices of Pearson’s product-moment correlation coefficients). We demonstrated that LOI and DBD, for example, are perfectly, but non-linearly related by calculating Spearman’s rank correlation coefficient and linearising DBD to LOI. This also means that a large part of the mismatch between LOI, DBD and MS between ca. 5500 and 1500 cal yr BP could be due to non-linearity between variables (i.e. different reactions to the same forcing) and is not indicative of an ambiguous signal. Linearising DBD and MS to LOI results in a first PC-axis explaining 91% of the variance. 6. Conclusions In this paper, we have presented a continuous Holocene glacier reconstruction for Blåbreen, Nordfjord, based on a multi-proxy data set derived from sediments of the distal glacier-fed lake Nedre Sørsendalsvatn. We assessed the evolution of sediment-derived variables through time and changes in sedimentation regimes were detected by means of multivariate statistics. With the common signal of sediment-derived proxies, the first PC-axis explained 88% of the variance, which is indicative of glacier variability. Deglaciation occurred at approximately 10,000 cal yr BP, and there
13
is firm evidence of increased glacial size at approximately 8200 cal yr BP; minimum in glacial activity is detected between 7900 and 5500 cal yr BP. A significant shift in the sedimentation regime, as indicated in grain-size data, is detected at approximately 5500 cal yr BP and is indicative of the onset of Neoglaciation in the catchment of Nedre Sørsendalsvatn. We emphasised assessing the uncertainties of the reconstruction of glacial variability. In particular, we assessed the effects of chronological uncertainty, resulting in high uncertainties around 10,000 cal yr BP and between 7900 and 5500 cal yr BP. We further accounted for homogeneities within the lake by including data from all three cores, which resulted in differing inferences of glacial variability between 5000 and 2000 cal yr BP. Acknowledgements We thank Kristian Vasskog and Olav Haavik for help during fieldwork, Ingelinn Aarnes for identifying macrofossils for AMS radiocarbon dating and Hella Wittmeier for help with sampling for radiocarbon dating. MT thanks Christian Kamenik for introducing him to R. Further thanks are due to all persons involved in programming R-codes and R-packages used in this study. We would like to thank Christian Kamenik, Richard Telford and John Birks for discussion on statistics. We thank Darrell Kaufman and Jesper Olsen for comments that greatly improved the clarity of this manuscript. Funding was provided by the Norwegian Research Council (Norges Forskningsråd, Project SHIFTS and NORPAST-2), the Swiss National Science Foundation through a personal grant to MT and by the Bjerknes Centre for Climate Research. This is publication no. A424 from the Bjerknes Centre for Climate Research. Appendix A. Supplementary data Supplementary data related to this article can be found online at http://dx.doi.org/10.1016/j.quascirev.2013.05.003. References Abdi, H., 2007. RV coefficient and congruence coefficient. In: Salkind, N. (Ed.), Encyclopaedia of Measurements and Statistics. Aitchison, J., 1983. Principal component analysis of compositional data. Biometrika 70, 57e65. Alley, R.B., Augustdottir, A.M., 2005. The 8 k event: cause and consequences of a major Holocene abrupt climate change. Quaternary Science Reviews 24, 1123e 1149. Bakke, J., Dahl, S.O., Paasche, O., Simonsen, J., Kvisvik, B., Bakke, K., Nesje, A., 2010. A complete record of Holocene glacier variability at Austre Okstindbreen, northern Norway: an integrated approach. Quaternary Science Reviews 29, 1246e1262. Bakke, J., Lie, Ø., Nesje, A., Dahl, S.O., Paasche, Ø., 2005. Utilizing physical sediment variability in glacier-fed lakes for continuous glacier reconstructions during the Holocene, northern Folgefonna, western Norway. The Holocene 15, 161e176. Bennett, K.D., 1996. Determination of the number of zones in a biostratigraphical sequence. New Phytologist 132, 155e170. Birks, H.H., Amman, B., 2000. Two terrestrial records of rapid climatic change during the glacial-Holocene transition (14,000e9000 calendar years B.P.) from Europe. PNAS 97, 1390e1394. Birks, H.J.B., Heiri, O., Sepp, H., Bjune, A.E., 2010. Strengths and Weaknesses of quantitative climate reconstructions based on Late-Quaternary biological proxies. Open Ecology Journal 3, 68e110. Blaauw, M., Christen, J.A., Mauquoy, D., van der Plicht, J., Bennett, K.D., 2007. Testing the timing of radiocarbon-dated events between proxy archives. The Holocene 17, 283e288. Blaauw, M., 2010. Methods and code for ‘classical’ age-modelling of radiocarbon sequences. Quaternary Geochronology 5, 512e518. Blake, G.R., Hartge, K.H., 1986. Particle density. In: Klute, A. (Ed.), Methods of Soil Analysis. Part 1 e Physical and Mineralogical Methods, second ed. American Society of Agronomy, Madison WI. Blott, S.J., Pye, K., 2001. Gradistat: a grain size distribution and statistics package for the analysis of unconsolidated sediments. Earth Surface Processes and Landforms 26, 1237e1248. Borcard, D., Gillet, F., Legendre, P., 2010. Numerical Ecology with R. Springer, Dordrecht.
14
J. Bakke et al. / Quaternary Science Reviews 73 (2013) 1e14
Born, A., Levermann, A., 2010. The 8.2 ka event: abrupt transition of the subpolar gyre toward a modern North Atlantic circulation. Geochemistry, Geophysics, Geosystems 11, Q06011. Bryhni, I., Grimstad, E., 1970. Supracrustal infracrustal rocks in the Gneis region of the Caledonides west of Breimsvatn. Norges Geologiske Undersökelse 266, 105e140. Chessel, D., Duffour, A., 2011. Ade4: Analysis of Ecological Data: Exploratory and Euclidean Methods in Environmental Sciences. Version1.4-17. http://cran.rproject.org/web/packages/ade4/index.html. Croudance, I.W., Rindby, A., Rothwell, R.G., 2006. ITRAX: description and evaluation of a new multi-function X-ray core scanner. In: Rothwell, R.G. (Ed.), New Techniques in Sediment Core Analysis. Geological Socety, London, pp. 51e63. Dahl, S.O., Nesje, A., 1994. Holocene glacier fluctuations at Hardangerjøkulen, central-southern Norway: a high resolution composite chronology from lacustrine and terrestrial deposits. The Holocene 4, 269e277. Dahl, S.O., Nesje, A., 1996. A new approach to calculating Holocene winter precipitation by combining glacier equilibrium-line altitudes and pine-tree limits: a case study from Hardangerjokulen, central southern Norway. Holocene 6, 381e398. Daley, T.J., Thomas, E.R., Holmes, J.A., Street-Perrott, F.A., Chapman, M.R., Tindall, J.C., Valdes, P.J., Loader, N.J., Marshall, J.D., Wolff, E.W., Hopley, P.J., Atkinson, T., Barber, K.E., Fisher, E.H., Robertson, I., Hughes, P.D.M., Roberts, C.N., 2011. The 8200 yr BP cold event in stable isotope records from the North Atlantic region. Global and Planetary Change 79, 288e302. Debret, M., Sebag, D., Desmet, M., Balsam, W., Copard, Y., Mourier, B., Susperrigui, A.S., Arnaud, F., Bentaleb, I., Chapron, E., Lallier-Verges, E., Winiarski, T., 2011. Spectrocolorimetric interpretation of sedimentary dynamics: the new “7/4 diagram”. Earth-Science Reviews 109, 1e19. Dolédec, S., Chessel, D., 1994. Co-inertia analysis e an alternative method for studying species environment relationships. Freshwater Biology 31, 277e294. eklima.met.no eklima monthly normal values (05.12.2012). Ellison, C.R.W., Chapman, M.R., Hall, I.R., 2006. Surface and deep ocean interactions during the cold climate event 8200 years ago. Science 312, 1929e1932. Evans, D., Butcher, C., Kirthisinga, A., 1994. Neoglaciation and early “Little Ice Age” in western Norway: lichenometric evidence from the Sandane area. The Holocene 4, 278e289. Evans, D., 1997. Reassessment of supposed early ‘Little Ice Age’ and older neoglacial moraines in the Sandane area of western Norway: a reply. The Holocene 7, 121e 124. Frontier, S., 1976. Etude de la croissance des valeurs propres dans une analyse en composantes principales: comparaison avec le modèle du bâton brisé. Journal of Experimental Marine Biology and Ecology 25, 67e75. Geirsdottir, A., Miller, G.H., Axford, Y., Olafsdottir, S., 2009. Holocene and latest Pleistocene climate and glacier fluctuations in Iceland. Quaternary Science Reviews 28, 2107e2118. Grimm, E.C., 1987. CONISS: a FORTRAN 77 program for stratigraphically constrained cluster analysis by the method of incremental sum of squares. Computers & Geosciences 13, 13e35. Grudd, H., Briffa, K.R., Karlen, W., Bartholin, T.S., Jones, P.D., Kromer, B., 2002. A 7400-year tree-ring chronology in northern Swedish Lapland: natural climatic variability expressed on annual to millennial timescales. The Holocene, 12657e12665. Grudd, H., 2008. Torneträsk tree-ring width and density AD 500e2004: a test of climatic sensitivity and a new 1500-year reconstruction of north Fennoscandian summers. Climate Dynamics 31, 843e857. Haakensen, N., 1989. Akkumulasjon på breene i Norge vinteren 1988e89. Været 13, 91e94. Haslett, J., Parnell, A.C., 2008. A simple monotone process with application to radiocarbon-dated depth chronologies. Applied Statistics Series C 57, 399e418. Heegaard, E., Birks, H.J.B., Telford, R.J., 2005. Relationships between calibrated ages and depth in stratigraphical sequences: an estimation procedure by mixedeffect regression. The Holocene 15, 612e618. Heegaard, E., Nilsen, T., 2007. Local linear mixed effect modelsdmodel specification and interpretation in a biological context. Journal of Agricultural, Biological, and Environmental Statistics 12, 414e430. Holzhauser, H., 2007. Holocene glacier fluctuations in the Swiss Alps. In: Mordant, C., Richard, H., Magny, M. (Eds.), Environnements et cultures a` l’A^ge du Bronze en Europe occidentale. Comité des travaux historiques et scientifiques (CTHS), Paris, pp. 29e43. Hotelling, H., 1933. Analysis of a complex of statistical variables into principal components. Journal of Educational Psychology 24 (417e441), 498e520. Jansen, E., Overpeck, J., Briffa, K.R., Duplessy, J.-C., Joos, F., Masson-Delmotte, V., Olago, D., Otto-Bliesner, B., Peltier, W.R., Rahmstorf, S., Ramesh, R., Raynaud, D., Rind, D., Solomina, O., Villalba, R., Zhang, D., 2007. Palaeoclimate. In: Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (Eds.), Climate Change 2007: the Physical Science Basis. Juggins, S., 2009. RIOJA: Analysis of Quaternary Science Data, Version 0.5e6. http:// cran.r-project.org/web/packages/rioja/index.html. Karlén, W., 1976. Lacustrine sediments and tree-line variations as indicators of climatic fluctuations in Lappland, northern Sweden. Geografiska Annaler 58 A, 1e34. Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P., 1983. Optimization by simulated annealing. Science 220, 671e680.
Leemann, A., Niessen, F., 1994. Holocene glacial activity and climatic variations in the Swiss Alps: reconstructing a continuous record from proglacial lake sediments. The Holocene 4, 259e268. Legendre, P., Legendre, L., 1998. Numerical Ecology. Elsevier, Amsterdam. Linderholm, H.W., Bjorklund, J.A., Seftigen, K., Gunnarson, B.E., Grudd, H., Jeong, J.H., Drobyshev, I., Liu, Y., 2010. Dendroclimatology in Fennoscandia e from past accomplishments to future potential. Climate of the Past 6, 93e114. Löwemark, L., Chen, H.F., Yang, T.N., Kylander, M., Yu, E.F., Hsu, Y.W., Lee, T.Q., Song, S.R., Jarvis, S., 2011. Normalizing XRF-scanner data: a cautionary note on the interpretation of high-resolution records from organic-rich lakes. Journal of Asian Earth Science 40, 1250e1256. Matthews, J.A., Nesje, A., Dahl, S.O., 1996. Reassessment of supposed early ‘Little Ice Age’ and older Neoglacial moraines in the Sandane area of western Norway. The Holocene 6, 106e110. Michelutti, N., Blais, J.M., Cumming, B.F., Paterson, A.M., Ruhland, K., Wolfe, A.P., Smol, J.P., 2010. Do spectrally inferred determinations of chlorophyll a reflect trends in lake trophic status? Journal of Paleolimnology 43, 205e217. Nesje, A., 1992. A piston corer for lacustrine and marine sediments. Arctic and Alpine Research 24, 257e259. Nesje, A., Kvamme, M., Rye, N., Løvlie, R., 1991. Holocene glacial and climate history of the Jostedalsbreen region, Western Norway – evidence from lake-sediments and terrestrial deposits. Quaternary Sciene Reviews 10, 87e114. Nesje, A., Dahl, S.O., 2001. The Greenland 8200 cal. yr BP event detected in loss-onignition profiles in Norwegian lacustrine sediment sequences. Journal of Quaternary Science 16, 155e166. Nesje, A., Matthews, J.A., Dahl, S.O., Berrisford, M.S., Andersson, C., 2001. Holocene glacier fluctuations of Flatebreen and winter-precipitation changes in the Jostedalsbreen region, western Norway: evidence from pro-glacial lacustrine sediment records. Holocene 11, 267e280. Nesje, A., 2009. Latest Pleistocene and Holocene alpine glacier fluctuations in Scandinavia. Quaternary Science Reviews 28, 2119e2136. Olsen, J., Bjärck, S., Leng, M.J., Gudmundsdottir, E.R., Odgaard, B.V., Lutz, C.M., Kendrick, C.P., Andersen, T.J., Seidenkrantz, M.S., 2010. Lacustrine evidence of Holocene environmental change from three Faroese lakes: a multiproxy XRF and stable isotope study. Quaternary Science Reviews 29, 2764e2780. Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., O’Hara, R.B., Simpson, G.L., Solymos, P., Stevens, M.H.H., Wagner, H., 2011. vegan: Community Ecology Package, Version1.17-11. http://cran.r-project.org/web/packages/vegan/index.html. Østrem, G., Dale Selvig, K., Tandberg, K., 1988. Atlas over breer i Sør-Norge (Atlas of glaciers in south Norway). Meddelelse nr. 61 fra Hydrologisk avdeling 1988, Norges vassdrags- og energiverk. Vassdragsdirektoratet, p. 248. Parnell, A.C., Haslett, J., Allen, J.M.R., Buck, C.E., Huntley, B., 2008. A flexible approach to assessing synchroneity of past events using Bayesian reconstructions of sedimentation history. Quaternary Science Reviews 27, 1872e1885. Rein, B., Sirocko, F., 2002. In situ reflectance spectroscopy e analysing techniques for high-resolution pigment logging in sediment cores. International Journal of Earth Science 91, 950e954. Rein, B., Lueckge, A., Reinhardt, L., Sirocko, F., Wolf, A., Dullo, W., 2005. El Nino variability off Peru during the last 20,000 years. Paleoceanography 20, PA 4003. Rohling, E.J., Pälike, H., 2005. Centennial-scale climate cooling with a sudden cold event around 8,200 years ago. Nature 434 (7036), 975e979. Seppä, H., Bjune, A.E., Telford, R.J., Birks, H.J.B., Veski, S., 2009. Last nine-thousand years of temperature variability in Northern Europe. Climate of the Past 5, 523e535. Sissons, J.B., 1979. Palaeoclimatic inferences from former glaciers in Scotland and the Lake District. Nature 278, 518e521. Sutherland, D.G., 1984. Modern glacier characteristics as a basis for inferring former climates with particular reference to the Lomond Stadial. Quaternary Science Reviews 3, 291e309. Trachsel, M., Grosjean, M., Schnyder, D., Kamenik, C., Rein, B., 2010. Scanning reflectance spectroscopy (380e730 nm): a novel method for quantitative highresolution climate reconstructions from minerogenic lake sediments. Journal of Paleolimnology 44, 979e994. van der Boogaart, G., Tolosana, R., Bren, M., 2011. Compositional Data Analysis, pp. 1.10e1.12. van der Wollenberg, A.L., 1977. Redundancy analysis: an alternative for canonical correlation analysis. Psychometrika 42, 207e219. Vasskog, K., Paasche, Ø., Nesje, A., Boyle, J.F., Birks, H.J.B., 2012. A new approach for reconstructing glacier variability based on lake sediments recording input from more than one glacier. Quaternary Research 77, 192e204. Vinther, B.M., Clausen, H.B., Johnsen, S.J., Rasmussen, S.O., Andersen, K.K., Buchardt, S.L., Dahl-Jensen, D., Seierstad, I.K., Siggaard-Andersen, M.L., Steffensen, J.P., Svensson, A., Olsen, J., Heinemeier, J., 2006. A synchronized dating of three Greenland ice cores throughout the Holocene. Journal of Geophysical Research e Atmospheres 111, D13102. http://dx.doi.org/10.1029/ 2005JD006921. Webb, P.W., Orr, C., 1997. Analytical Methods in Fine Particle Technology. Micromeritics Instrument Corporation, Norcross, GS, USA. Wolfe, A.P., Vinebrooke, R., Michelutti, N., Rivard, B., Das, B., 2006. Experimental calibration of lake-sediment spectral reflectance to chlorophyll a concentrations: methodology and paleolimnological validation. Journal of Paleolimnology 36, 91e100.