A decadal-scale Holocene sea surface temperature record from the subpolar North Atlantic constructed using diatoms and statistics and its relation to other climate parameters



[1] A sediment core from Reykjanes Ridge has been studied at 10- to 50-year time resolution to document variability of Holocene surface water conditions in the western North Atlantic and to evaluate effects of Holocene ice-rafting episodes. Diatom assemblages are converted to quantitative sea surface temperatures (SST) using three different transfer functions. Spectral and scale-space methods are also applied on the records to explore variability at different timescales. Diatom assemblage and SST records clearly show that decaying remnants of the Laurentide ice sheet strongly influenced early Holocene climate in the western North Atlantic. This overrode the predominance of Milankovitch forcing, which played a key role in the development of Holocene climate in the eastern North Atlantic and Nordic Seas. Superimposed on general Holocene climate change is high-frequency SST variability on the order of 1°−3°C. The record also documents climatic oscillations with 600- to 1000-, ∼1500-, and 2500-year periodicities, with a time-dependent dominance of different periodicities through the Holocene; a clear change in variability occurred about 5 ka BP. The SST record also provides evidence for Holocene cooling events (HCE) that, in some cases, correlate to documented southward intrusions of ice into the North Atlantic.

1. Introduction

[2] The ocean can switch between different modes of convection owing to trigger mechanisms in the climate system that remain poorly understood [Lenderink and Haarsma, 1994; Rahmstorf, 1994]. An underlying climatic cycle with a period of about 1500 years pervades the North Atlantic last Glacial period [Bond et al., 1997]. An outstanding issue is whether climate variability has similar underlying cycles during the Holocene even though it lacks the strong feedback mechanisms of glacial environments. Bond et al. [1997, 1999] argued that ice-rafting episodes with 1500-year cycles are recorded in Holocene sediments of the North Atlantic. During these ice-rafting episodes, drift ice and cooler surface waters in the Nordic and Labrador seas advected southward and eastward, each time penetrating warmer strands of the subpolar circulation. On the basis of correlations between the variability of ice-rafting episodes and changes in production rates of the cosmogenic nuclides carbon-14 and beryllium-10, it was suggested that variations in solar output influenced the ocean hydrography in the North Atlantic throughout the Holocene [Bond et al., 2001].

[3] Other records from the North Atlantic give evidence for Holocene climate variability with periodicities of about 1500 years [Grootes and Stuiver, 1997; Alley et al., 2001a, 2001b; Schulz, 2002; Rahmstorf, 2003], 2500 and 950 years [O´Brien et al., 1995], and 1000 and 500 years [Chapman and Shackleton, 2000]. However, these records, which are proxies for atmospheric temperatures, ice-rafting episodes or ocean overturning strength, do not necessarily agree on the timing of the cycles [Schulz, 2002]. Establishing whether these high-frequency Holocene fluctuations are a persistent feature of North Atlantic circulation is clearly an important issue because relatively small changes in thermohaline circulation may trigger climate events of widespread significance [Weaver and Hughes, 1994]. This is an especially important issue in light of whether abrupt climate change may influence future climate.

[4] Changes in Holocene surface water conditions in the western North Atlantic have been documented in records of coccolith assemblages [Giraudeau et al., 2000], dinoflagellate cyst assemblages [Solignac et al., 2004], alkenone-derived SST [Moros et al., 2004] and diatom-derived SST [Andersen et al., 2004a]. However, these records have incomplete Holocene sections, relatively low temporal resolution, or both.

[5] In this study, we aim to document the evolution of Holocene surface water variability southwest of Iceland, and to evaluate effects of 1500-year ice-rafting pulses [Bond et al., 1999, 2001] on regional sea surface temperature (SST). We have therefore increased the resolution of a previously published composite sediment core (core LO09-14) located on Reykjanes Ridge in the subpolar North Atlantic [Andersen et al., 2004a]. The newly investigated core now has a time resolution of 10−50 years. We have generated quantitative SST records by utilizing three different diatom transfer functions and applied spectral and scale-space methods to these records.

2. Physical Setting

[6] North Atlantic circulation is an important contributor to the global climate system, transporting heat northward via the North Atlantic Drift (NAD) and ventilating the world ocean through deep-water formation. The NAD forms two major surface currents in the subpolar North Atlantic, the Norwegian Atlantic Current (NwAC) and the Irminger Current (IC) (Figure 1) [Hopkins, 1991]. The NwAC enters the Norwegian Sea east and west of the Faroe Islands, flows along the Norwegian coast and becomes the Atlantic layer in the Arctic Ocean. The IC crosses the Reykjanes Ridge south of Iceland between 53°N and 60°N, flows northward around Iceland [Krauss, 1986], and separates into two branches west of Iceland. Most of the IC turns southwest where it is incorporated into the West Greenland Current (WGC) [Hurdle, 1986], while a small branch continues northward and flows around Iceland through the eastern Denmark Strait [Hopkins, 1991].

Figure 1.

Location map showing the main surface circulation in the Nordic Seas, the core location of LO09-14, and core location of correlative cores: VM28-14 and VM29-191 [Bond et al., 1997]. NwAC, Norwegian Atlantic Current; IC, Irminger Current; EGC, East Greenland Current; EIC, East Icelandic Current; and SAF, Sub-Arctic Front [Levitus and Boyer, 1994].

[7] Core LO09-14 was retrieved at 1685 m water depth from the western flank of Reykjanes Ridge south of Iceland at 58°56,3N latitude and 20°24,5W longitude (Figure 1). At present, warm waters diverging westward from the NAD overlie the study site; however, the area is highly sensitive to the cold, out-flowing East-Greenland Current (EGC) and the position of the Sub-Arctic Front (SAF), where subpolar water separates from warm Atlantic water [Hopkins, 1991]. The SAF meanders, but is most strongly defined at 53°N latitude. Summer SST at the site are normally 11°−12°C [Dietrich, 1969; Johannessen, 1986], but can be lower or higher owing to variable influence of subpolar water originating from the EGC or from the conjunction area between the West-Greenland Current (WGC) and the Labrador Current (LC).

3. Material and Methods

3.1. Cores

[8] Core LO09-14 represents a composite of four individual sediment cores, collected at the same location. Previously, Andersen et al. [2004a] published a record of diatom SST based on material from a large box core (LO09-14 LBC; 0.4 m long), a giant gravity core (LO09-14 GGC; 2.7 m long), and a conventional gravity core (LO09-14 GC; 5.6 m long). This record was investigated at 5- to 10-cm intervals with a time resolution of 50−350 years. For this study, a fourth gravity core (DS37-2P) has been included to cover the gap between the LBC and the GGC. We have also increased the time resolution of the previous record by analyzing the cores every 2 cm throughout the Holocene except for the Holocene Climate Optimum. This period, between ~5 and 7 ka, was sampled every 1 cm.

3.2. Chronology

[9] The 6.5-m composite record of Core LO09-14 covers the time period from ~350 years BP to 11.2 ka BP. The age model is based on 40 accelerator mass spectrometry (AMS) 14C dates of the foraminifera Globigerina bulloides (Table 1). Dates from cores LO09-14 LBC, LO09-14 GGC and LO09-14 GC have been published [Andersen et al., 2004a; Moros et al., 2004]. For this study, we add 11 AMS dates from core DS97-2P (Table 1). These dates were calibrated to calendar years before present by using OxCal Program v3.8 with a marine calibration [Bronk Ramsey, 1995, 2001; Stuiver et al., 1998]. The ages were normalized for a δ13C of 25‰ and corrected by 400 years for the air-sea reservoir difference. We have used a constant reservoir age correction throughout the Holocene since no information exists to the contrary for the North Atlantic. Current work on this topic (S. Bondevik, personal communication, 2007) supports the use of a constant reservoir age correction through the Holocene. An age-depth model was established by linear interpolation between dated levels in Core LO09-14 LBC, and by polynomial fits between dated levels in cores DS97-2P, LO09-14 GGC and LO09-14 GC.

Table 1. AMS Radiocarbon Dates and Calibrated Ages for the Composite Record LO09-14a
Depth, cmComposite Depth, cmLaboratory Number14C age ± Standard DeviationCalibrated Age, Calendar Years B.P.
  • a

    OS, Woods Hole Oceanographic Institute; AAR, Aarhus University; KIA, Kiel, UtC, Utrecht [Prins et al., 2001]; Poz, Potsdam. All measurements are made on the foraminifer species Globigerina bulloides. Polynomial fits for the conversion from core depth (centimeters) to time domain are: (1) LO09-14GGC: cal. ka BP = 2474–2.9 cm + 0.0222 cm2 + 0.00084 cm3-2.31E-006 cm4, R2 = 0.987 and (2) LO09-14GC: cal. ka BP = 4221 + 33.95 cm-0.085 cm2 + 9.61E-005 cm3, R2 = 0.985.

  • b

    Not included in the age model.

Core LO09-14LBC
00AAR-5049705 ± 40310
22OS-32526785 ± 35425
99OS-321401080 ± 25643
2929OS-324751250 ± 65795
3939OS-324741560 ± 301112
Core DS37-2P
145UtC 96541440 ± 1601280b
246Poz-82472765 ± 351293b
647Poz-82481320 ± 301320b
1048UtC 96551380 ± 601380
1862UtC 96561800 ± 801800
2551Poz-82492125 ± 301620
4068Poz-82502260 ± 301932
57149UtC 96572680 ± 902700
76173Poz-82513130 ± 353130
96234UtC 96584020 ± 604200b
126276UtC 96594840 ± 905325b
Core LO09-14GGC
0.586AAR-66711190 ± 352237b
9123OS-324772770 ± 452499
1779KIA75002495 ± 252151
30151OS-324782990 ± 352760
47108AAR-64372690 ± 402384
69152OS-325243030 ± 352793
79142OS-325252860 ± 352627
101167KIA3165 ± 352945
120170OS-326963260 ± 453081
150240KIA75014250 ± 354346
163262OS-326974680 ± 504907
197283OS5220 ± 505577
240314KIA75026440 ± 356913
276354KIA7180 ± 407636
Core LO09-14GC
51287KIA74975330 ± 505689
103326AAR-44546620 ± 807125
135342KIA207936920 ± 407450
159356KIA74986605 ± 407824b
165367KIA207947435 ± 407860
185388OS-326817730 ± 358190
216405KIA74998005 ± 408457
243449OS-326908420 ± 408926
270449KIA207958385 ± 458920
276477AAR-50508145 ± 459187b
300494AAR-44558790 ± 1009324
303531OS-326919050 ± 409653
315528OS-326929040 ± 459638
333558OS-326939220 ± 459911
370569OS-326949350 ± 4010029
394565AAR-50519310 ± 5510003
437632OS-326959920 ± 5510705
447652AAR-445610410 ± 9011366

3.3. Methods

3.3.1. Diatoms

[10] We utilize marine diatom species to reconstruct past SST because they are good indicators of surface water conditions in the region [Koç-Karpuz and Schrader, 1990; Andersen et al., 2004b]. Diatoms were concentrated in samples through a procedure whereby bulk sediment is treated with HCl and H2O2 to remove carbonate and organic matter. The samples are then neutralized by rinsing with deionized water and separated from clay particles by settling. For detailed descriptions of this method see Koç et al. [1993]. Diatom slides of cleaned samples were prepared as described by Koç-Karpuz and Schrader [1990]. A Leica Orthoplan microscope was used for identification and counting of diatoms at 1000× magnification. At least 300 diatom frustules (in addition to Chaetoceros species) from each sample were counted on random traverses following counting procedures described by Schrader and Gersonde [1978].

3.3.2. Statistics Transfer Functions

[11] Three transfer function methods, were used to quantify past SST. These are (1) Q-mode factor analysis and regression analysis of Imbrie and Kipp (I&K) [Imbrie and Kipp, 1971], (2) maximum likelihood (ML) [Upton and Cook, 2002] and (3) weighted averaging partial least squares (WA-PLS) [ter Braak and Juggins, 1993]. For diatom transfer functions August SST gives the best match. A key assumption of transfer functions is that the environmental variable of interest is an ecologically important variable, or that it is linearly correlated with one that is, and this correlation is stable through time [Birks, 1995]. The surface calibration set is based on 52 species from 139 surface samples from the Nordic Seas and the North Atlantic [Andersen et al., 2004b].

[12] The I&K diatom transfer function is based on eight components and has a root mean square error of 1.25°C, a coefficient of determination between observed and inferred SST of 0.89, and a maximum bias of 0.92°C. The eight different factors defined by diatom assemblages and established according to their relation to modern surface hydrography are [Andersen et al., 2004b]: the Arctic Greenland Assemblage (factor 1), the North Atlantic Assemblage (factor 2), the sub-Arctic Assemblage (factor 3), the Norwegian Atlantic Current Assemblage (factor 4), the Sea Ice Assemblage (factor 5), the Arctic Assemblage (factor 6), the Greenland Current Assemblage (factor 7), and the Mixed Water Masses Assemblage (factor 8) (Appendix A). Because of their close affinity to modern hydrographic regimes these factors also enable us to reconstruct movements of the different water masses.

[13] The ML transfer function method has no parameters to set, and has the least autocorrelation problems [Upton and Cook, 2002; Telford and Birks, 2005]. This transfer function method can be chosen if many of the taxa show a unimodal relationship with the environmental variable of interest [Birks, 1995]. If the environmental gradient is shorter, and most taxa have a linear relationship with the environment, methods as PLS may be more appropriate [Birks, 1995]. This unimodal or linear species response based models assume a global relationship [ter Braak, 1995] between the species and the environment.

[14] WA-PLS can be regarded as the unimodal-based equivalent of multiple linear regression [ter Braak and Juggins, 1993]. This means that a species has an optimal abundance along the environmental gradient being investigated. The WA-PLS diatom transfer function has a root mean square error of 0.75°C, a coefficient of determination between observed and inferred SST of 0.96, and a maximum bias of 0.44°C. Like the I&K method, WA-PLS uses several components in the final transfer function. These components are selected, however, to maximize the covariance between the environmental variables to be reconstructed and hence the predictive power of the method in contrast to the I&K method where the components are chosen irrespective of their predictive value to capture the maximum variance within the biological data. In this analysis, we chose to use four components, and unlike the I&K method, the number of components to incorporate in WA-PLS is based on statistical cross-validation procedure [ter Braak and Juggins, 1993]. SiZer

[15] SiZer (Significance of Zero Crossings of the Derivative) [Chardhuri and Marron, 1999] is applied to explore significant features in the I&K and WA-PLS SST time series at different scales. SiZer assumes that the reconstructed SST are independent random variables. A relevant regression model for this is

equation image

where x(t) is the target curve. In equation (1), we assume that the ti are equally spaced on the range of t, that x smooth and that the εi are independent Gaussian variables with mean 0 (which makes x the regression curve yi on ti) and variance Var (εi) = σi2. In this investigation, we focus on estimation only at the data points, t1,…, tn. At each temperature estimate, a local linear kernel estimator is used to produce smooths of the estimated temperatures. More precisely, at point tj, equation imageh (tj) equals the fit equation image0 where (equation image0,equation image1) minimizes

equation image

where Kh(·) = (1/h)K(·/h), and K is a kernel function chosen as a unimodal probability density function symmetric about zero. The parameter h, frequently called bandwidth, controls the degree of smoothness in the estimation equation imageh. A description of this estimator is given by Fan [1992].

[16] The SiZer approach is motivated by scale-space ideas from computer vision [Lindberg, 1994]. It departs in two important ways from the classical approach of nonparametric smoothing, where focus is on inference about the true underlying curve. First, SiZer examines a wide range of bandwidths and thereby avoids the choice of one bandwidth. Second, the bias problem in doing inference is avoided by shifting the focus from the true underlying curve to the true curve, viewed at varying levels of resolution. In SiZer, the notion of scale is controlled through the bandwidth in the kernel estimator. For each scale and location of the signal, a test is performed to see whether the smooth has a derivative significantly different from zero. In the local linear kernel estimator, this test is based on the simple and appealing slope, equation image1, of the local line. A key idea in SiZer is that significant features are found at different scales (i.e., at different levels of smoothing). Wavelet Analysis

[17] We applied wavelet transform to the reconstructed SST time series (WA-PLS and I&K). Wavelet transform decomposes a time series into wavelets and allows highlighting of the variability features at different timescales. The wavelet analysis provides thereby a useful extension to the common spectral estimates showing how the frequency content of the analyzed time series varies in time [Torrence and Compo, 1998; Percival and Walden, 2000]. Given a discrete sequence yi, i = 1,…,N with a uniform time increment δt, a wavelet transform is defined as a convolution of yi with a scaled and translated versions of the time-localized “mother wavelet” ψ which forms a basis of the transform. We write

equation image

where s denotes the wavelet scale. For improving the visual representation of results we convert the wavelet scale to a more convenient timescale using the notion of the wavelet pseudo-frequency.

[18] The wavelet power spectrum (also called the wavelet variance) [Maraun and Kurths, 2004], which is defined as Pn (s) = (∣Wn (s)∣2)/σ2, decomposes the time-dependent variance of the process yi on a scale-by-scale basis. Here σ2 denotes the signal variance, and is used as a normalization for the wavelet spectrum. Our analysis is based on a real-valued Morlet wavelet, which is believed to provide the most optimal trade-off between the resolution in scale and time. Since the wavelet function is real, the analysis allows isolating the peaks in the wavelet spectra. The initially unevenly sampled time series were rebinned with a 10-year increment with the gaps filled using spline interpolation. The wavelet transform routine applied in the present work is based on the cwt function of the Wavelet Toolbox extension package for Matlab.

[19] The statistical significance of wavelet power at each particular point (t,s) of the wavelet variance was assessed at 95% confidence levels against a red noise background (see Torrence and Compo [1998] for further details). The appropriateness of the AR1 (red noise) model to describe the analyzed time series was tested using a nonparametric runs test [Bendat and Piersol, 1986]. This also provided a sufficient condition for weak stationarity of the time series, which is necessary for the wavelet spectrum to be well defined [Percival and Walden, 2000]. The respective lag-1 autocorrelations were estimated directly from the original time series using the method devised by Mudelsee [2002] and Schulz and Mudelsee [2002]. This avoided the potential overestimating of the autocorrelation coefficient due to data resampling and made the analysis less conservative. For the latter two procedures we used RedFit package, which has them embedded [Schulz and Mudelsee, 2002].

4. Results

4.1. Surface Ocean Productivity

[20] Holocene sediment from Reykjanes Ridge has a high abundance of diatoms (Figure 2). From 11.2 to 1.4 ka BP, average diatom abundance is about 30 million diatoms per gram sediment. From 1.4 to 1.0 ka BP, diatom abundances exceed 1.5 billion diatoms per gram sediment.

Figure 2.

Total diatom abundance shown as diatom valves/g sediment and fluxes at the Reykjanes Ridge site through the Holocene. The flux is given by: number of valves/g dry sediment × AR[number-of-valves/cm2/ka].

[21] Diatom fluxes to the seafloor probably relate to past surface ocean productivity [Broecker and Peng, 1982; Pokras and Molfino, 1986]. We have converted diatom abundances to fluxes as number of valves per gram dry sediment/cm2/ka. Generally, fluxes are higher during the early and late Holocene than during the middle Holocene Climate Optimum period (7−5 ka BP) (Figure 2). We attribute the high surface productivity observed during the early and late Holocene to the proximity of the SAF to the site during these periods [Andersen et al., 2004a]. The SAF is a zone of stronger convergence where nutritious deep water is brought to the surface, and therefore an area of increased surface nutrients. During the early Holocene, melting ice sheets might provide additional nutrients by transporting material eroded from land.

[22] Large and irregular pulses of diatoms also characterize the early and late Holocene. Several intervals between 11 and 9.2, 3.3 and 2.3, and 1.3 and 0.5 ka BP are dominated by Rhizosoleniaborealis, which can reach between 50 and 90% of the total floral assemblage. High abundances of Rhizosolenia borealis have previously been recorded from the eastern equatorial Pacific Ocean and the Mediterranean sapropels [Kemp and Baldauf, 1993; Yoder et al., 1994; Kemp et al., 1999]. At the Reykjanes Ridge this is interpreted to represent elevated nutrient concentrations in surface water masses caused by a strong convergence zone between warm and cold waters. As such, we can conclude that during these periods a stronger convergence of surface waters took place possibly in relation with the SAF [Andersen et al., 2004a].

4.2. Factor Analysis

[23] The diatom assemblages are closely connected to different water masses in the area [Andersen et al., 2004b], and changes in the abundance of these assemblages through the Holocene will reflect changes in the dominance of different water masses over the study site. High communalities (between 0.7 and 0.9) indicate a tight correspondence between the modern assemblages and the Holocene assemblages at the study site, which also provides confidence in the estimated SST.

[24] The sub-Arctic assemblage (Factor 3) dominates the study site throughout the Holocene (Figure 3). However, the North Atlantic Assemblage (Factor 2) and the Greenland Current Assemblage (Factor 7) periodically played significant roles. The Greenland Current Assemblage was most influential during the early Holocene (11−7 ka BP), although its strength was highly variable. In fact, during this period, all three factors show high variability, suggesting highly dynamic oceanic conditions. The North Atlantic Assemblage was most influential during the mid-Holocene (7−5 ka BP). This suggests increased transport of warm Atlantic waters to the site and a stronger IC, although again with some variability. Less dramatic changes in the abundance of different assemblages are recorded through the late Holocene (5 ka BP to present), supporting generally more stable surface water conditions compared to the earlier periods. The dominance of the Sub-Arctic Assemblage (50−80%) suggests a weaker influx of warm waters diverging westward from the NAD during the late Holocene.

Figure 3.

Diatom assemblages (as factors loadings × 100) and reconstructed sea surface temperatures (by weighted averaging partial least square) from Core LO09-14 at the Reykjanes Ridge. Holocene climate optimum is noted as HCO.

4.3. Sea Surface Temperature (SST) Records

[25] All three SST records give generally similar patterns (Figure 4a). Surface waters were relatively cool and highly variable in temperature during the early Holocene (11−7 ka BP), relatively warm during a mid Holocene Climate Optimum (7−5 ka BP), and relatively cool and more stable during the late Holocene (5 ka BP to present). There are, however, differences in the SST reconstructions on the order of about ±1°C in most intervals and up to ±2°C in some intervals (Figures 4a and 4b). A regression plot of observed versus predicted SST for the core top calibration set [Andersen et al., 2004b] show that the ML method do not perform well in reconstructing temperatures below 5°C, the I&K method have problems in reconstruction of temperatures above 14°C, while the WA-PLS method seems to provide the best fit (Figure 4c).

Figure 4.

Difference between the I&K, ML and WA-PLS for LO09-14 sea surface temperature (°C) reconstructions. (a) Sea surface temperature (°C) (I&K, ML, and WA-PLS) from the LO09-14 core at the Reykjanes Ridge. (b) Difference between the I&K and WA-PLS for LO09-14 sea surface temperature reconstructions. (c) Regression plot of the three transfer function methods, showing observed sea surface temperature versus predicted sea surface temperature.

[26] Significant trends and features appear when SiZer is applied to the I&K and WA-PLS SST records (Figure 5). At the highest level of smoothing, SiZer shows a general increase in SST through the Holocene for both records. With less smoothing several millennial- to century-scale cooling and warming events are observed. During the early Holocene, two significant cooling periods are recorded (Figure 4a). These are from 10 to 9.4 ka BP and from 8 to 7 ka BP. Significant warming events are recorded between 11 and 10 ka BP, 9.4 and 8 ka BP, and 7 and 5 ka BP. Consequently, high-amplitude changes of 1°−3°C are observed across this interval.

Figure 5.

(a, b) SiZer reconstruction of SST based on I&K method, and (c, d) SiZer reconstruction of SST based on WA-PLS method. Figures 5a and 5c show a family plot of smooths, denoted as a family plot. The green dots represent the raw data of the diatom countings, and the red line illustrates the general smooth of the record. Figures 5b and 5d show a SiZer map is given as a function of location of scale. A significant decrease is flagged as blue while a significant increase is flagged as red. The color purple is used at locations where the derivative is not found to be significantly different from zero. The color gray is used to indicate the too few data are available to do inference. Typically the color gray occurs at very small scales for SiZer. The black horizontal line illustrates the significant trend given in 1000-year resolution (indicated by the dashed curved line). The red solid curve in the family plot corresponds to a choice of h that typically would be chosen if only one scale were to be used. This particular h is a data driven bandwidth and a best choice from a purely statistical point of view [Ruppert et al., 1995]. In Figures 5b and 5d feature map as a function of scale (controlled by h) and location (given by t) for the signal are given. In agreement with common practice in the geological community, we interpret curves from right to left. With this interpretation in mind, a significant increase in the SiZer map is flagged as red, while a significant decrease is flagged as blue. At locations where there is no significant change, purple is used. The horizontal black line (Figures 5b and 5d) corresponds to the smoothing obtained by the solid red line in the family plot. Dark grey is used to indicate that too few data points are available to do inference. Typically, dark grey color occurs at very small scales for SiZer.

[27] The mid Holocene Climate Optimum is punctuated by two abrupt cooling events at 5.9 and 5.3 ka BP, where temperatures drop 3°C and 1.5°C, respectively (Figure 4a). At intermediate scales (log10(h) of 3) significant features support the observation of HCO between 7 and 5 ka BP (Figures 5b and 5d). At smaller scales (log10(h) of ∼2) the SiZer shows an increasing feature just after 5.9 ka BP, which corresponds to the 5.9 cooling event. However, the 5.3 ka BP cooling event is not well detected by SiZer.

[28] Although the late Holocene was a time of generally stable SST, a series of significant cooling events are observed, especially in the WA-PLS record (Figure 4a). Small but significant features are also found in the SiZer maps: one at 1.8 ka BP and another at 2.7 ka BP, both with an abrupt temperature drop of 1.5°C (Figures 5b and 5d). At smaller scales (log10(h) of ∼2) the 2.7 ka BP event is found in both SiZer maps, indicated by a significant increase. The 1.8 ka BP event is precisely detected only in the WA-PLS SiZer.

[29] Factor analysis and SST reconstructions give an indication for the evolution of the past current system in midlatitude North Atlantic. The presence of the Sub-Arctic Assemblage together with the Greenland Current Assemblage, the Arctic Greenland Assemblage, and the Arctic Assemblage during the early Holocene reflects surface conditions cooler than today and proximity of the SAF to the site. This implies a weaker IC than at present over the site. The mid Holocene period show increased influence of the North Atlantic Assemblage, which reflects more Atlantic water reaching the site and a stronger Irminger Current. The decreased influence of the North Atlantic Assemblage in late Holocene Period suggests oceanographic conditions characterized by a weaker IC compared to mid Holocene Period.

4.4. Wavelet Power

[30] For estimating the wavelet power spectrum, we used data from present to 9742 years BP, which fits the AR1 model at a 95% confidence level. A visual comparison of the reconstructions (Figure 4b) reveals the tendency for the I&K transfer function to underestimate SST from 3 ka BP to present and to overestimate SST from 8 to 5 ka BP relative to the WA-PLS reconstruction. The RedFit estimates show that the magnitudes of the time series variance differ in approximately two times: 1.6°C for I&K and 0.77°C for WA-PLS reconstructions. The lag-1 autocorrelations coefficients estimates show somewhat higher dependency in the I&K reconstruction, as demonstrated by the value of the coefficient of about 0.8 for I&K against 0.6 for WA-PLS transfer function. The increased variance for the I&K transfer function therefore can be attributed to the low-frequency component of the variability, which is also demonstrated by the wavelet decomposition of the time series.

[31] Figure 6 (left) shows the normalized wavelet power spectra for the I&K and WA-PLS reconstructions for core LO09-14. Both wavelet spectra demonstrate well-pronounced millennial-scale variability, which is, however, located mainly within the “cone of influence” and therefore cannot be identified reliably. Multicentennial variability on the timescales from 600 to 1000, ∼1500 and 2500 years is prominent in the analyzed time series, but statistically significant only in the reconstruction obtained using the WA-PLS transfer function. Integrating the wavelet spectra along the time axis provides the consistent estimates of the spectral density function [Percival and Walden, 2000]. These global wavelet spectra are shown for both transfer functions in the right panels of the Figure 6 and may additionally provide an inference about the typical amplitude of variability in the SST inherent to a particular timescale. The results lie in a general agreement with what was inferred from the wavelet spectra. This confirms the conclusion made earlier from the visual consideration of the plots that the more pronounced millennial changes in the I&K-derived reconstruction tend to conceal the variability on the finer timescales.

Figure 6.

(left) Normalized wavelet power spectra for the LO09-14 sea surface temperature proxy time series for the (top) I&K and (bottom) WA-PLS transfer functions. Solid contours enclose the regions where the wavelet power is above the red noise background with the 95% confidence level. The semitransparent areas highlight the “cone of influence” where the edge effects of the wavelet transform become important [Torrence and Compo, 1998]. (right) Black solid lines show the respective global wavelet spectra, while the red dashed lines indicate the 95% confidence level assuming testing against the red noise background.

5. Discussion

5.1. General Climate Change During the Holocene

[32] Diatom based SST reconstructions from the Nordic Seas show general cooling over the Holocene in step with decreasing Northern Hemisphere insolation since 11 ka BP [Koç et al., 1993; Koç and Jansen, 1994; Andersen et al., 2004b]. In contrast, similarly generated SST records from Reykjanes Ridge exhibit variable temperatures throughout the Holocene, with a general pattern showing a relatively cool and highly variable early Holocene period (11 to 7 ka BP), a relatively warm and variable mid Holocene Climate Optimum (7 to 5 ka BP), and a generally cooler and stable late Holocene period (5 ka to present) (Figure 4a). Superimposed on this general Holocene climate development is high-amplitude millennial to century-scale variability in surface water conditions.

5.2. Mode Shift in Climate Variability

[33] Even though climate oscillations with 600- to 1000-, ∼1500- and 2500-year periodicities are evident in the LO09-14 SST records, there is also a change in the mode of different periodicities from about 5 ka BP (Figure 6). Before 5 ka BP, strong 2500- and ∼1500-year variations are present. After 5 ka BP, lower amplitude variations with periods of 900 and 1000−1500 years are found.

[34] The mode shift in the variability of surface ocean conditions may relate to cessation of meltwater supply. After 7 ka BP, the decaying Laurentide ice sheet and shallowing of the Canadian Arctic channels due to melting of the ice sheets caused dramatic changes in the hydrographical regime [Dyke and Prest, 1987; Solignac et al., 2004]. Model experiments by Schulz et al. [2007] show that a continuous freshwater perturbation in the Labrador Sea pushes the overturning circulation of the Atlantic Ocean into a bistable regime, characterized by phases of active and inactive deep-water formation in the Labrador Sea. This can explain the SST variability recorded in core LO09-14 during the Early Holocene.

[35] The mode shift might also record the passing of a climate threshold. The early Holocene climate was affected by high insolation values, which caused a different atmospheric circulation pattern than later in the Holocene [Renssen et al., 2005]. Through the Holocene the meridional pressure gradient decreased and the midlatitude westerlies experienced a marked weakening. The reduced zonality of the atmospheric circulation caused less effective heat transport from the oceans to downwind continents. Following the summer insolation curve, the sea ice cover expanded, which caused a substantially higher surface albedo, which in turn enhanced the cooling. Consistent with the expansion of sea-ice cover, the strength of meridional overturning circulation in the Nordic Seas decreased.

[36] Previous investigations of marine sediment cores and ice cores have found evidence of millennial-scale fluctuations in ocean temperature and strength of the overturning, in ice-rafting episodes and atmospheric temperatures during the Holocene; identified frequencies for these changes include 2500, 1500, 1000, 950 and 500 years [Stuiver and Braziunas, 1989; O’Brien et al., 1995; Bond et al., 1993, 1997; Campbell et al., 1998; Bianchi and McCave, 1999; Chapman and Shackleton, 2000; Giraudeau et al., 2000; Schulz and Paul, 2002; Oppo et al., 2003; Solignac et al., 2004; Hall et al., 2004], which are interpreted to be caused by internal and/or external forcings or a combination of the two.

[37] According to Schulz and Paul [2002] the 950-year climate oscillations between 6.6 and 4 ka BP may be linked to multicentennial variations in strength and location of the Icelandic Low. The ∼1000-year cycle variations in the North Atlantic Deep Water (NADW) circulation are coherent with fluctuations in atmospheric conditions over Greenland [Chapman and Shackleton, 2000], and thereby related to solar forcing. The relative NADW contribution began to decrease at 6.6 ka BP. At 5.4 ka BP a transition in the hydrological cycle occurred in the North Atlantic [Schulz and Paul, 2002], reaching a minimum in the NADW contribution at 5 ka BP [Oppo et al., 2003]. Significant variations in the NADW contribution can occur in the absence of forcing by large ice sheets, and may be more sensitive to surface forcing than was previously imagined [Oppo et al., 2003]. This rapid zonal shift from ∼6.5 ka BP is seen as part of a general reorganization of the surface hydrology affecting synchronously at least the middle and high latitude of the North Atlantic in response to decreased solar insolation [Giraudeau et al., 2000].

5.3. Holocene Cooling Events and Correlation to Other Climate Signals

[38] The LO09-14 SST records document several Holocene cooling events (HCEs) (Figure 7 and Table 2). Bond et al. [1997, 1999] documented a series of IRD events in the North Atlantic, recorded by increases in hematite stained grains in cores VM28-14 and VM29-191 and their stacked record (Figure 1). The IRD record show evidence of 1500-year periodicity, and an issue is whether the cooling and IRD events are related.

Figure 7.

Comparison of WA-PLS SST (°C) reconstructions from Core LO09-14 to IRD events from core V28-14, V29-292 and stacked record [Bond et al., 1997]. Holocene cooling events (HCE 1-18) for core LO09-14 are noted as shaded areas, and HCE correlative to core V28-14 and V29-292 are marked with red crosses. Triangles represent AMS14C dates.

Table 2. Holocene Climate Event (HCE), Core LO09-14
HCECalendar Years BP

[39] However, the temporal resolution and the stratigraphic control for the SST record from Core LO09-14 are higher than those for the IRD records from cores VM28-14 and VM29-191. There are also discrepancies in the timing of IRD events between cores VM28-14 and VM29-191. These factors limit assessment of relationships between HCEs and IRD events in the subpolar North Atlantic. Nonetheless, three HCEs (6. 9 and 13) recorded in core LO09-14 appear to correlate to IRD events in core VM29-191, and five HCEs (1, 2, 3, 11, 13) seem to correlate to IRD events in core VM28-14 (Figure 7). We can conclude that in cases where there is a positive correlation between the IRD events and HCE the advances of icebergs into the North Atlantic were accompanied by 2°−3°C cooling of the surface waters. Solignac et al. [2004] finds good correlation between IRD events from VM29-191 and sea-surface salinity changes from core MD99-2254 east of the Reykjanes Ridge between 6 and 3 ka BP. However, they find no relationship between the IRD events and the salinity changes in the Labrador Sea and, therefore, conclude that the western and eastern North Atlantic show different behaviors with respect to Holocene high-frequency climate fluctuations.

[40] Some of the major HCEs recorded in LO09-14 are also well known from terrestrial and other marine environments. The first distinct cooling event in core LO09-14 is centered at 10.3−10.2 ka BP (HCE 1) and is characterized by 3°C cooling of the SST over the Reykjanes Ridge (Figure 7). This event is also recorded in lacustrine, tree ring, ice core and marine records from the Northern Hemisphere, and is correlated to a disturbance in the North Atlantic Ocean thermohaline circulation [Björck et al., 2001].

[41] Two significant cooling events at 10−9.9 ka (HCE 2) and 9.9−9.7 ka BP (HCE 3) are recorded in core LO09-14 with 4°C and 5°C decrease in SST, respectively. Timing of these events corresponds to glacier advances from southern Norway, called the Erdalen Event 1 and 2, respectively [Bakke et al., 2005].

[42] The 3.5°C SST decrease recorded at 8.5−8.4 ka BP (HCE 6) correlates to a distinct advance of glaciers in southern part of Norway called the Finse Event [Dahl and Nesje, 1996]. Associated with this event a peak in hematite stained grains is recorded in core V29-191 indicating advances of icebergs to the North Atlantic [Bond et al., 1997, 2001] (Figure 7).

[43] At 4.5−4.2 ka BP a distinct HCE (HCE 11) is recorded in the SST record. This event has been previously recorded by several other proxies [Bond et al., 1997, 1999; Schultz and Paul, 2002; Giraudeau et al., 2000; Oppo et al., 2003; Hall et al., 2004].

[44] The late Holocene LO09-14 record shows a distinct cooling from 2.8 to 2.6 ka BP (HCE 14), which has been documented by several proxies indicating a change in the nature of arctic intermediate water (AIW), a major contributor to the formation of Intermediate Icelandic overflow water (ISOW), over the Vøring Plateau [Risebrobakken et al., 2003], change in deep-water flow and ventilation [Bianchi and McCave, 1999; Oppo et al., 2003; Hall et al., 2004], and as an IRD [Bond et al., 1997, 1999] and EH events [Giraudeau et al., 2000].

[45] It is suggested that the oceanic changes associated with HCE 14 appear to be part of at least a regional and possibly global signal occurring at this climatic transition that includes a step in the atmospheric σ14C concentration, an equatorward relocation of the midlatitude storm tracks and glacier advanced in south America and northwestern Europe [van Geel et al., 1996, 2000; Hall et al., 2004]. Hall et al. [2004] found the most pronounced and consistent deep water perturbation present to take place ∼2.7 ka ago, when the transition, starting at ∼4.5 years BP, from strongly stratified to well mixed surface water. Previous work also suggests a decrease in the IC in the late Holocene, and more specifically an expansion of the late EGC and a shift of the IC toward the south occurred around 3−2 ka BP [Rasmussen et al., 2002].

5.4. Forcing Mechanisms

[46] The cause(s) of the Holocene millennial-scale climate changes in the North Atlantic remain the source of debate. External forces like the Sun`s radiative output [Bond et al., 1997, 2001; Björck et al., 2001; van Geel et al., 1996, 2000] and internal oscillations in both the ocean circulation system [Campbell et al., 1998; Schulz et al., 2007] and atmospheric processes have been suggested [Broecker, 2000; Giraudeau et al., 2000; Kim et al., 2004; Knight et al., 2005; Rimbu et al., 2003]. Most of the HCEs recorded in core LO09-14 show anticorrelation with 14C production rate and 10Be flux, implying solar-related changes as an important underlying mechanism for the observed ocean climate variability (Figure 8). An atmospheric general circulation (GCM) model imply that times of reduced solar irradiance could produce a change in surface climate through the atmospheres dynamic response to changes in stratospheric ozone, a slight southward shift of the northern subtropical jet, and a decrease in the Northern Hadley circulation and temperatures [Haigh, 1996].

Figure 8.

Comparison of WA-PLS SST (°C) reconstructions from core LO09-14 to cosmogenic 10Be and 14C production rate (normalized) based on IntCal04 [Finkel and Nishiizumi, 1997; Yiou et al., 1997; Stuiver et al., 1998]. Holocene Cooling events (HCE 1-18) are noted as shaded areas. Erdal Event 1, EE1; Erdal Event 2, EE2; and Finse Event, FE [Bakke et al., 2005; Björck et al., 2001; Dahl and Nesje, 1996].

[47] Also changes in the NADW formation and deep-water flow during the Holocene [Bianchi and McCave, 1999; Oppo et al., 2003; Hall et al., 2004] correlate well with the timing of the HCEs recorded at the LO09-14 site. This indicates a close coupling between the deepwater masses and surface ocean variability through the Holocene. Further, comparison of atmospheric temperature data with sortable silt record of NEAP4K highlights a strong similarity in the long-term trends as well as in some of the submillennial features through the Holocene [Hall et al., 2004]. These evidences point to a strong coupling of atmospheric changes with deep and surface ocean through the Holocene.

6. Conclusions

[48] A sediment record from Reykjanes Ridge shows that the subpolar North Atlantic was strongly influenced by decaying remnants of the Laurentide ice sheet in the early Holocene. This influence was strong enough to override the influence of the solar insolation, which played a key role in the development of the Holocene climate in the eastern North Atlantic and the Nordic Seas. There are also evidences of a strong coupling of atmospheric changes with deep and surface ocean through the Holocene.

[49] Superimposed on the general Holocene climate development, there is high-frequency SST variability, which is in order of 1°−3°C. We find that the SST changes during the Holocene document a climate oscillations with 600- to 1000-, ∼1500- and 2500-year periodicities. There is also a time-dependent dominance of different periodicities through the Holocene indicating a clear change of mode from about 5 ka BP. Most of the cooling events recorded in core LO09-14 show anticorrelation with 14C production rate and 10Be flux. This implies solar-related changes as an important underlying mechanism for the observed ocean climate variability.

Appendix A

[50] Table A1 provides a Varimax factor score matrix from factor analysis of surface sediment samples.

Table A1. Varimax Factor Score Matrix From Factor Analysis of Surface Sediment Samples
VariableFactor 1Factor 2Factor 3Factor 4Factor 5Factor 6Factor 7Factor 8
Thalassiotrix longissima0.05420.05440.16320.0272−0.0068−0.0011−0.0020−0.0426
Thalassionema nitzschioides−0.04240.1752−0.10450.80290.0173−0.0083−0.1515−0.2185
Rhizosolenia hebetate f. hebetate0.0489−0.00800.04010.0158−0.02030.0679−0.0224−0.0924
Rhizosolenia hebetata f. Semispina−0.0112−0.2900.8861−0.00160.02780.0764−0.1574−0.3651
Rhizosolenia borealis−0.03750.01180.37200.18650.0263−0.858−0.09080.8737
Proboscia alata0.0392−0.0370−0.00660.39430.0008−0.0463−0.07380.0415
Rhizosolenia bergonii0.00160.0744−0.0004−0.0176−0.00100.00170.00460.0056
Bacterosira fragilis0.01720.0014−0.0040−0.00410.07180.01130.00060.0096
Roperia tesselata0.00020.0822−0.01530.04950.0019−0.0007−0.00700.0035
Prosira glacialis0.01920.0008−0.00040.00170.02810.0017−0.0021−0.0038
Actinocyclus curvatulus0.12360.0102−0.01070.0476−0.01440.1322−0.11350.0302
Asteromphalus robustus0.1074−0.00120.00490.0057−0.0161−0.0345−0.00610.0169
Hemidiscus cuneiformis0.00150.0358−0.0053−0.0076−0.00090.00200.0035−0.0049
Thalassiosira gravida spore0.01180.0308−0.0836−0.02670.10180.9384−0.22590.1159
Thalassiosira gravida veg.0.1752−0.05800.14640.2055−0.02170.24750.91180.0166
Thalassiosira angus-lineata0.87420.0031−0.0131−0.0163−0.0575−0.0541−0.14570.0029
Thalassiosira eccentica−0.00620.0188−0.02090.10080.00140.0035−0.0258−0.0315
Thalassiosira trifulta0.37400.0041−0.00560.0155−0.0575−0.0233−0.06560.0254
Thalassiosira lineata−0.00050.0302−0.01480.0248−0.0022−0.00760.0364−0.0114
Thalassiosira nordenskioeldii0.0028−0.02330.03620.09430.10060.0369−0.0415−0.1371
Thalassiosira oesterupii0.00860.95500.0447−0.1045−0.0074−0.01110.08150.0178
Thalassiosira hyalina0.1103−0.0005−0.0080−0.00340.2090−0.0402−0.00320.0031
Thalassiosira angulata−0.0225−0.631−0.04350.21540.02880.0038−0.01640.0112
Thalassiosira pacifica−0.0091−0.0112−0.00590.06390.0220−0.0019−0.0267−0.0121
Thalassiosira decipiens−0.00100.0050−0.00060.00470.00010.0029−0.0052−0.0023
Coscinodiscus radiatus−0.0076−0.0159−0.02730.17740.0053−0.00530.00580.0660
Coscinodiscus marginatus−0.01450.00480.0042−0.00230.00020.0356−0.02430.0160
Coscinodiscus nodulifer0.00040.0071−0.0013−0.0011−0.00020.00040.0005−0.0018
Coscinodiscus oculus-iridis−0.0014−0.00740.00180.01300.00050.0049−0.00270.0169
Coscinodiscus asterophalus−0.0011−0.0072−0.00010.0178−0.0002−0.00210.00040.0196
Nitzschia marina0.00230.05920.0120−0.0136−0.0001−0.00550.0045−0.0119
Nitzschia bicapitata0.00350.1159−0.00870.0055−0.00030.0030−0.0083−0.0054
Fragilariopsis cylindrus0.11410.0011−0.0130−0.01140.2647−0.0730−0.00380.0301
Nitzschia angularis0.0189−0.00330.00500.00680.0053−0.01000.00280.0044
Fragilariopsis oceanica0.02910.0059−0.0182−0.02790.9212−0.07430.0492−0.0175
Bacteriastrum hyalinum0.00190.0480−0.0034−0.0059−0.0009−0.00110.0057−0.0040
Synedra ssp.0.00280.04200.0119−0.00010.0048−0.0065−0.01270.0173
Nitzschia atlantica0.0025−0.0006−0.00010.0004−0.00070.0029−0.00110.0035
Nitzschia sp. 1−0.00850.00620.0228−0.01460.0052−0.00550.0348−0.0172
Nitzschia sp. 20.00030.0094−0.0011−0.0035−0.00020.00030.0023−0.0002
Nitzschia capitata0.00020.0071−0.0008−0.0034−0.00010.00030.00230.0004
Nitzschia kolaczeckii0.00020.0087−0.0011−0.0028−0.00010.00040.0019−0.0011
Pseudonotia doliolus0.00070.03224−0.0031−0.0059−0.00040.00100.0037−0.0035
Actinocyclus ehrenbergeii0.00010.0129−0.0014−0.0058−0.00030.00120.00520.0082
Thalassiosira nodulineata−0.0005−0.0025−0.00280.00510.00220.00460.00490.0116
Sp. Y.0.00040.02020.0023−0.00620.0001−0.00120.00320.0047
Coscinodiscus linearis0.00020.0107−0.0010−0.0037−0.00030.00050.00190.0017
Coscinodiscus crenulatus0.00120.0434−0.0063−0.00175−0.00070.00200.0130−0.0021
Coscinodiscus kutzingii0.00100.040−0.0031−0.0049−0.00050.00080.0022−0.0015
Coscinodiscus stellaris0.00020.0064−0.0003−0.0017−0.00010.00010.0008−0.0014
Coscinodiscus africanus0.00010.0018−0.0003−0.00090.00000.00000.0008−0.0001
Thalassionema nitzschiodes v. parva0.00150.0498−0.0059−0.0219−0.00150.00110.01850.0101


[51] This study is financed by the Research Council of Norway and the European Commission through the NORPAST-II and the PACLIVA projects, and the Norwegian Polar Institute. We are grateful to G. Dickens, J. Barron, and an anonymous reviewer for constructive comments to the manuscript.