Mid-Holocene regional reorganization of climate variability

Introduction Conclusions References


Introduction
There was hardly any period in Earth's history that experienced more stability in the climate system than did the Holocene.According to Richerson et al. (2001), even the onset of human civilizations is owed to the lack of large climate fluctuations, since change was the constant rule for all preceding glacial as well as interglacial periods within the Quaternary.The notion of a stable Holocene climate has been, meanwhile, questioned by many researchers (e.g.Fairbridge and Hillaire-Marcel, 1977;Bond et al., 1997c;Mayewski et al., 2004).Refined accuracy of paleorecords and of noise reduc- tion methods has disclosed a series of distinct shifts from most proxies for the last 11,500 years (Barber et al., 2004;Kim et al., 2007).These shifts attract further attention for at least four reasons.
1. Disruptions have -with the exceptions of the Saharan desertification at 5500 yr BP (Claussen et al., 1999) and the 8.2 kyr event (Renssen et al., 2001) -not yet been reproduced by numerical modeling.Mechanistic understanding of the complex interplay between ice, ocean, atmosphere and vegetation bundled in regional climate sub-systems is far from being complete (Steig, 1999).
2. We also still lack knowledge about the spatial extension of prominent disruptions (deMenocal et al., 2000b;Sirocko, 2003;Mayewski et al., 2004).So far, only the work of Rimbu et al. (2004) based on 18 records of alkenone sea surface temperature worldwide revealed consistent regional differences of climate variability.
3. Holocene climate fluctuations define the range of natural variability to which the signatures of anthropogenic interference with the Earth system should be compared (von Storch et al., 2004;Moberg et al., 2005).Attribution and analysis of past shifts is pivotal for assessing ongoing changes, as well as reproduction of recorded natural variability (of the Holocene) by models is a prerequisite for simulating future climates (in the Anthropocene).
4. Another motivation to seek for past, but not too remote alterations, especially in regional temperature or moisture is the "breeding" role of the Holocene for civilizations.While its stability has been claimed to be responsible for their rising, instabilities are first candidates for explaining their collapses.This idea is reflected by numerous studies where individual outbreaks in proxy time series have been correlated with eclipses of ancient empires (Fagan, 1999;Anderson, 2001;deMenocal, 2001;Binford, 2001).Provoked by the simplicity of this approach, a large group of historical scientists dislike such intrusion by geoscience into their discipline (Erickson, 1999;Coombes and Barber, 2005).Fragmented (model) knowledge and controversy are hence characterizing both the effects as well as causes of singular Holocene climate shifts.Evidence for the nearly regular cyclicity of events, however, seems to be firm.Predominant modes of punctuated disruptions on millennial time scales have since long been identified (e.g.Fairbridge and Hillaire-Marcel, 1977).For the North Atlantic, the 1450 yr periodicity was proposed by Bond et al. (1997c), and is relevant in tropical regions as well (deMenocal et al., 2000b;Thompson et al., 2003b).On the centennial time scale, recurrent climate anomalies were detected by McDermott et al. (2001) or Sarnthein et al. (2003b).Often, an external trigger was suspected behind the quasi-cyclicity -above all variations of solar activity (e.g., Hodell et al., 2001c;Bond et al., 2001c).
Spectral analysis can indicate system properties of the global climate (like stability, turn-over times, regularity of modes) and helps to identify active teleconnections which are generated by the coupling of sub-systems and their feed-backs.If extended to external forcings, analysis in the frequency domain will shed light on the possible origin of shifts.Sensitivity to variable forcing can, for example, be estimated using the quantitative relationship of spectral intensities (Debret et al., 2007).We here provide such a reconstruction of the spectral behavior inherent to climate proxy time series with global coverage and for the entire Holocene.Former data reviews were mostly centered around an eventual synchronicity or spatial intensity of shifts (e.g.Mayewski et al., 2004;Sepp ä et al., 2007), while spectral synthesis studies cover either a single climate zone or a shorter interval within the Holocene (e.g.Morrill et al., 2003;Moberg et al., 2005).
Here, a spectral analysis technique is introduced that is adapted to unravel the temporal wax and wane of dominant fluctuation modes.Operating with this tool on a global set of proxy time series, we not only aim to address the question of how quasi-cyclic modes were distributed worldwide and which factors could be tentatively claimed being responsible for them.Acknowledging the observation that millennial as well as centennial modes are discontinuous or even missing in a number of high-resolution records (e.g.Noren et al., 2002b;Moberg et al., 2005), our study focuses on temporal and spa-Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc

Printer-friendly Version
Interactive Discussion tial trends in the intensity of climate fluctuation modes.The study of variability trends in time and space reflects in particular the questions: Has the climate system of the Holocene continuously stayed in a unique state?Does its stability (sensu retaining of large shifts) as well as instability (disposition to minor fluctuations) reflect a global property or distinct regional features?

Selection of proxy data
In total 130 long-term high-resolution time series obtained at 109 globally distributed sites were collected from existing literature.Due to low sedimentation rate resulting in coarse temporal resolution, open ocean locations are underrepresented with respect to coasts and land masses (Fig. 1).77% of the records have temporal resolution better than 100 yr and 80% span more than 9000 yr within the period 12 kyr BP to the present (see Table 1 and 2).68 data sets are accessible from one of the Publishing Network for Geoscientific & Environmental Data (PANGEA, www.pangea.de)or the National Climate Data Center (NOAA NCDC, www.ncdc.noaa.gov).The remaining time series were digitized with an error of less than 2% from original publications .
We chose to analyze time series based on quantities that serve as proxies for temperature, effective moisture (i.e.precipitation -evaporation), and rarely also wind regime, not unlike Mayewski et al. (2004).Records were excluded which involve more complex relationships to climate, such as productivity or stable carbon isotopes.The types of records are broadly categorized in Table 1 into (1) isotope fractionation, mostly 18 O, (2) lithic composition, and (3) relative species abundance (tree pollen or algae).
In addition, solar activity was inferred from 10 Be abundance and 14 C flux (Bond et al., 2001c).Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version Interactive Discussion

Lomb-Scargle spectral analysis
We employ a spectral analysis since this approach is not sensitive against possible absolute dating errors.Spectral information can be interpreted independently from the different and sometimes uncertain correlations of individual variables with the climate state (like for isotope fractionation in ice cores, lake or marine sediments, or pollen composition) what allows to use a broader spectrum of proxy types.If one is in particular interested in the temporal change of power spectra, wavelet transformation is usually the method of choice (Moy et al., 2002;Moberg et al., 2005).However, wavelets require evenly sampled time-series, while time sequences of proxy records are mostly irregular.To avoid the introduction of a spectral bias by interpolation and, in addition, to make the method also applicable to records with relatively few samples, we base our approach on the procedure suggested by Schulz and Mudelsee (2002).This means to employ a Lomb-Scargle Fourier transform followed by a bias correction with correction factor obtained from a theoretical red-noise spectrum which, in turn, is rooted on a Monte-Carlo ensemble of 1000 first order autoregressive processes.We use version 3.5 of the software package REDFIT (www.ncdc.noaa.gov/paleo/softlib/redfit/redfit.html) with parameters ofac=4, hifac=0 and two Welch windows with an overlap of 50%, and assume a 95% confidence level for identifying significant spectral anomalies.For time series with a small fraction (n) of data points in each Welch window, we follow the recommendation by Thomson (1990) and take 1−1/n as the threshold for significance.
Acknowledging the notion of a globally visible Mid-Holocene climatic change (e.g.Steig, 1999;Morrill et al., 2003) we split the time series into two overlapping intervals; these intervals (12-5.5 kyr BP and 6.5-0 kyr BP) will be referred to as Lower and Upper Holocene, respectively.The initial age 12kyr BP compromises between the different starting points of the time-series which in some cases reflect the globally asynchronous onset of the Holocene.The exact choice of the starting age, however, was not found to be critical for our analysis.In order to diagnose the change in spectral intensity from Lower to Upper Holocene we employ a selective bootstrap.Randomly chosen data points were substituted with also randomly chosen values from the same part (Lower/Upper Holocene).Subsequently, the spectrum is examined for changes in significance.We obtained good results for 5000 realizations with substitution fraction of 33% for each time series.From this we statistically identified local long-term changes in the variability signal.Either for the entire spectrum or for a frequency window we checked whether its spectral significance is sensitive to bootstrapping.If a mode looses significance by bootstrapping in the upper interval, but endures changes in the lower part, this corresponds to a positive change in cyclicity (periodic signal originates from the Upper Holocene part of the timeseries).The opposite behavior (sensitivity in bootstrapping the lower and robustness in the upper time interval) is attributed to a negative temporal trend.
Spatial clustering of locations is performed by radially extrapolating the peak intensity I from each record location with exponential decrease (I• exp(−r/1500 km)).Peak intensity is a binary measure with I=1 if at site i the sum of all available local proxy information significantly indicates the presence/increase of frequencies, and I=−1 for negative trends.

Non-cyclic event frequency
The analysis in the frequency domain is cross-checked by a simple counting method relying on a straightforward definition of climate events.After removal of the 2 kyr running mean, the time series are normalized by their standard deviation.Frequency peaks are considered as a distinct event if (1) they exceed a threshold p a and (2) are separated by a zero-line crossing to the preceding event.By using in parallel a vector of thresholds p a =1.5 −1,0,1,2 the method is made robust with respect to the choice of p a .The non-cyclic event frequency is calculated as the average number of events for different thresholds p a , divided by the length of the time period.

Mid-Holocene change in climate modes
The way how changes in the spectral intensity are detected by our method is visualized in Fig. 2 for three selected records, i.e. temperature reconstruction for Southeast Europe, δ 18 O at Sajama, Bolivia, and solar activity from Greenland ice cores.Only those frequency peaks that are with 95% probability not compatible with red noise mark a significant mode.Random displacement of proxy values in one half of the Holocene dampens some of those modes, as, for example, obvious for the 1/340 yr cycles in Southeast Europe temperature during the Upper Holocene.For isotopic oxygen at Sajama, spectral changes are manifold.The 1/900 yr mode vanishes when either of the two halves is distorted by bootstrapping, and the two prominent centennial cycles (1/250 yr, 1/200 yr) appear in the Upper Holocene only.In this period, the well known 1000 yr mode of solar activity turns out to be absent.Just by visual inspection, one could suspect its presence in the Lower Holocene according to the relatively high spectral intensity, the value of which, however, is below the critical threshold for significance.Apart from the three example records, we detected in all 130 time series 223 significant modes in the spectral interval between 1/200 yr and 1/1800 yr (see also Fig. 3).When contrasting Lower with Upper Holocene, 62 of these peaks gain significance while 24 loose it.We also checked for prominent frequencies: neither the presence of modes nor their changes did accumulate in particular spectral windows.Instead, we found a relatively flat histogram of significant frequency bands (Fig. 3).

Regional clustering of spectral properties
Only a minority of sites reveal significant modes in the Lower Holocene.After applying the clustering algorithm, absence of oscillations is globally the norm except for small regions in the North Atlantic (tropical western and northern parts), East Africa and Antarctica (red areas in Fig. 4).In the Upper Holocene, North Atlantic and in Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version Interactive Discussion particular eastern American sites with dominant modes build large regional clusters.
The global distribution of locations with changes in spectral modes displayed in Fig. 5 demonstrates that these are remarkably homogeneous within these regional clusters.Clusters are generally made out of about 4-9 sites with independent proxy records which, in their large majority, exhibit the same spectral trend.
Obviously, damping or amplification of a climate fluctuation mode is only modestly affected by the heterogeneous quality of records, inherent random noise or other local phenomena.The spectral differences between the two selected temperature related records shown above, i.e. for Southeast Europe and Bolivia, can therefore be extrapolated to collections of large areas at sub-continental scale.Like for the two centennial cycles at Sajama, new modes appear during the Mid-Holocene in North and entire South-West America, East and North Atlantic and East Asia.Contrary, climate fluctuations fade out in eastern South America/West Atlantic, the Arctic, East Africa and, to some extent also South-East Asia.
The spatial organization of clusters persist after mapping the change for two frequency sub-domains.In Fig. 6, Mid-Holocene trends separated according centennial and millennial cycles still occur in great spatial uniformity.Since the total band-width is higher for all centennial modes, their global trend pattern also largely resemble the one for the entire frequency band (1/200 yr-1/1800 yr), with the exception that the disappearance of regular climate shifts in the East Atlantic is less pronounced.The clustering in Fig. 5 as well as the differentiated spectral map (Fig. 6) reveal a clear longitudinal pattern.For example, nearly all East Pacific sites become more oscillatory while the contrary happened for the entire Southwest Atlantic.Overall, cycles with periods smaller than 850 yr are more likely to be enhanced in the Upper Holocene compared to millenial modes.

Cross-validation with trends in non-cyclic event frequency
Our second indicator for the spatio-temporal organization of climate variability which is the density change in non-cyclic events from the first to the second half of the Holocene, Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version Interactive Discussion turns out to be spatially coherent as well (Fig. 7).Though, the longitudinal separation for non-cyclic frequency change is differently oriented compared to the Lomb-Scargle derived trends in Fig. 5. Abundance of climatic shifts is, for example, decreasing in East Asia where due to the spectral picture one would expect an increase.However, the pronounced Mid-Holocene enhancement of climate variability in a Pan-American corridor is equally reproduced, and the increase in periodic spectral power in the East Atlantic corridor is corroborated for the southern part.The analysis of non-cyclic events also confirms spatial patterns in centennial spectral trends (Fig. 6, left).

Spectral divorce at Mid-Holocene
From the literature on Holocene climate variability one could expect that at least four frequencies play a privileged role in our spectral analysis.The two millennial scale oscillations which comprise the 900 yr periodicity extracted in the GISP2 δ 18 O records by Schulz and Paul (2002); Rimbu et al. (2004) and the classical 1450 yr period (e.g.O'Brien et al., 1995) are points of reference for many paleo-proxy studies.In addition, a 320 yr and a 550 yr cycle have been repeatedly observed in proxy records as well as output of general circulation models (GCM) with reduced complexity (Weijer and Dijkstra, 2003;Weber et al., 2004;Te Raa and Dijkstra, 2003).However, none of the four modes arose more prominently than others within our global study.The most robust result, instead, derives from an apparent non-stationary of regular climate oscillations.Only about 10% of spectral peaks are stable, i.e. found over the entire period as well as after partial boostrapping.This substantiates our hypothesis that around the 5.5 kyr event (termination of the African Humid Period) the climate system may have undergone an irreversible re-organization.So far, non-stationary variability has only been reported for regional systems, like the Southern Pacific with its decadal to centennial cyclicity related to the El Ni ño Southern Oscillation (ENSO, Moy et al., 2002).Previous review studies, however, weren't aware of the global dimension of the re-organization between Lower and Upper Holocene.One reason for this may be the Introduction

Conclusions References
Tables Figures

Back Close
Full reference character of Greenland and the North Atlantic.Records from this area show persistent millennial cycles (Bond et al., 1997c), in contrast to nearly all other locations around the globe.There, a temporal switch of modes is more often found than their persistence.
A second major outcome of our study is the regional uniformity of temporal trends.
Although coverage of many regions is yet much below a value that allows for statistically firm conclusions, the similar, albeit not equal, global distribution of changes in periodic (Fig. 5 as well as non-periodic (Fig. 7) modes and the consistent clustering of adjacent sites which, in general, comprise different proxy types, should be taken as a strong indication for a regionalization of variability modes.Equal trends are even more homogeneously clustered for non-periodic climate variability.They remarkably differ from cyclic trends in Eurasia.Nevertheless, all modes (centennial, millennial and non-cyclic) intensified during the Holocene in a longitudinal zone crossing the Peruvian upwelling area and North-east America.A counterexample, though, against spatial uniformity is evident through the scatter of modal trends within the East Asian monsoon system.This may be due to its internal complexity and various active teleconnections to which the monsoon is sensitive.For example, it has been speculated that the atmospheric connection between the western Asian monsoon and the thermohaline circulation (THC) in the North Atlantic decreased in intensity from the Lower to the Upper Holocene (Morrill et al., 2003).This might explain a reduced correlation between trends in millenial and non-cyclic modes in the two climate sub-systems observed in our study.

Possible mechanisms for variability changes
In face of the fragmented understanding of the mechanisms producing climate shifts or quasi-cyclic fluctuations during the Holocene, it seems premature to ask for what may have caused their temporal change or their regional organization.We here only briefly discuss the possible role of ocean and atmospheric circulation, and of external Introduction

Conclusions References
Tables Figures

Back Close
Full forcings.The sun's influence on Holocene climate variability has been deduced from the synchronicity of climate anomalies and variations in solar activity (e.g.Bond et al., 2001c;Hodell et al., 2001c).Our analysis includes records of cosmogenic nuclide production ( 10 Be and 14 C flux) as well as reconstructed sunspot number of ?, and shows persistence of the 208 yr Suess cycle together with a damping of millennial modes after 6 kyr BP (Fig. 2 and yellow sun in Figs.5-6).A recent analysis of the sunspot number power spectrum (Dima and Lohmann, 2005) identified periods of 6,500, 2,500, 950 and 550 yr.Debret et al. (2007) already questioned the hypothesis of Bond et al. (2001a) that the 1500 yr cycles are due to variations in solar activity.Still, the possibility of solar variability being amplified by oceanic feedbacks can not be entirely excluded (Renssen et al., 2006).
Central in the literature discussion is the THC, which is connected to North Atlantic deep water formation.The THC is commonly considered the primary process for the generation of climate fluctuations on interdecadal or centennial (Broecker et al., 1985) as well as millennial time scales (Stocker et al., 1992;Weijer and Dijkstra, 2003).Schulz and Paul (2002), however, questioned its connection with the prominent 1450 yr oscillation.The Iceland-Scotland overflow water is an important component of the thermohaline circulation.Its record contains dominant periodicities of 1,400 and 700 years associated with THC variations over the Holocene (Bianchi and McCave, 1999).It is conceivable that ocean circulation changes, like those of the Atlantic multidecadal oscillations, modulate SST variability in the North Atlantic basin on centennial and longer timescales.Particular THC variations are detected for the Mid-Holocene in the ocean circulation through hydrographic changes (R ühlemann et al., 2004), which were more pronounced in the early Holocene (Kim et al., 2007).
To understand the atmospheric circulations related to Holocene variability, one can look for analogous patterns during the instrumental period.This procedure was used to explore the physical processes producing SST patterns and thereby to understand recorded Holocene SST changes (Marchal et al., 2002;Rimbu et al., 2004;Lorenz , 2006).The spatial pattern of Northern Hemisphere SST reconstructions for the Holocene emphasizes a North Atlantic Oscillation trend towards its negative phase, a feature which can be reproduced by a coupled climate model (Lorenz and Lohmann, 2004).This trend is overlaid by 2,400 and 900 yr cycles (Rimbu et al., 2004), consistent with other proxy data sets and solar variation.
Other mechanisms behind temporally changing modes include (1) high frequency "noise" (2) low frequency orbital forcing, and (3) particular events.The Pacific decadal oscillation and the El Ni ño-Southern Oscillation (ENSO) show long-term enhancement (Moy et al., 2002;Rimbu et al., 2004).The origin of high frequency fluctuations is controversially discussed but a combination of nonlinear oceanographic interactions in the tropical Pacific and orbital forcing well fits into model-based and instrumental evidence (Clement et al., 1999;Loubere et al., 2003).These fluctuations may have activated multicentennial modes (Simmonds and Walland, 1998).Phase transitions in climate sub-systems could also be triggered by a slow change in insolation.Given the wider distribution of millennial modes after the cessation of oscillations in solar activity at 6 kyr BP, we suggest that the Mid-Holocene reversal of insolation change differently affected regional subsystems.As a result, oceanic or atmospheric teleconnections between subsystems could have weakened or strengthened.Such shifts were possibly mediated by dislocations of convergence zones or trade winds, thereby modifying the damping and amplification forces of modes (Lohmann and Lorenz, 2007).
Intensification of ENSO in the Mid-Holocene, with a progressive increase in frequency to periodicities of 2 to 8.5 years, is believed to mark the onset of modern-day precipitation patterns in low-latitude regions (Rodbell et al., 1999).Analysis of archives in northern Africa indicate strong reductions in tropical trees and Sahelian grassland cover, which allowed large-scale dust mobilization after 4300 yr BP (Kropelin et al., 2008).Haug et al. (2001) found a large transition (in the frequency space) at Cariaco Basin ∼3800 years ago indicative for tropical reorganization of the hydrological cycle.Similar millennial-scale hydrologic variability is reported at other locations (Koutavas et al., 2002;Ekdahl et al., 2005).Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version

Interactive Discussion
Most of the transitions overstress the capabilities of concurrent climate models, but it is useful to compare the induced variability with internal oscillatory modes (without external trigger) which are seen in models of reduced complexity (Mikolajewicz and Maier-Reimer, 1990;Weijer and Dijkstra, 2003).Model perturbation experiments reveal eigenmodes on millennial time scales.These modes are generated by the advection of buoyancy anomalies around the overturning loop, both in a single-hemispheric basin leading to centennial modes or throughout the global ocean responsible for millennial cycles.The most negative eigenvalues (strongest damping) was found for centennial oscillations (Weijer and Dijkstra, 2003;Te Raa and Dijkstra, 2003).However, such modes could be activated if fluctuations in radiative energy input are included into simulation studies (Weber et al., 2004).

Consequences for cultural development?
The evidence that climate variability was not uniformly distributed over the globe, gains a new aspect to the dispute of whether or how Holocene climate affected the development of human societies.We propose that frequency of climate changes rather than average characteristics made an effective difference between inhabited world regions.In particular cultures situated in the Pan-American corridor have seen a significant rise in event frequency by about two additional disruptions more per millenium after 6 kyr BP.It is well imaginable that increased climate variability meant an elevated stress for early American societies.Repeated cultural discontinuities go well beyond the effects of singular events.In western Asia, like in many other world regions, the probability of climatic shifts (also visible as centennial modes) declined over the Holocene.Ancient civilizations in the Near East or later in southern Europe were challenged by abrupt climatic changes and resulting subsistence crises with much lower frequency.In his Pullitzer price winning work, J. Diamond related the lagged development between Old and New World before the colonization era to differences in biodiversity determined by the geographical orientation of the Americas with respect to Eurasia (Diamond, Introduction

Conclusions References
Tables Figures

Back Close
Full Screen / Esc Printer-friendly Version Interactive Discussion 1999).We here suggest that climate variability should be taken into consideration in these conceptual approaches and also in quantitative approaches to simulating the effect of climate on human prehistory (e.g., Wirtz and Lemmen, 2003).

Conclusions
Our analysis substantiates two hypotheses.First, the propensity of the climate system to amplify external fluctuations or to endogenously generate them is not a global property, but differs among regions.Secondly, most regional propensities underwent a irreversible change during the Mid-Holocene.Fluctuations especially intensified along a Pan-American corridor.This eventually led to an unequal probability of crisis for early human civilizations in the Old and New World.

Conclusions References
Tables Figures

Back Close
Full Table 1.Proxy variables used in this study.

Proxy type Description
Isotopic oxygen fractionation The interpretation of isotopic fractionation of oxygen (δ 18 O) highly depends on geographic location, environmental setting, and type of record.δ 18 O variations in Andean glaciers and polar ice caps represent temperature.Measured in biologic deposits from closed water bodies, δ 18 O usually indicates changes in the water balance, and thus effective moisture.However, the signal cannot be separated from temperature effects during times when the lake system is open.Speleothem δ 18 O records the amount and composition of cave water, mostly during the wet season, and may also be influenced by temperature and above-cave lithology.

Lithic composition
Mg/Ca, Charcoal, Clay, Bulk dens., Lightness, Grayscale, HSG, Lithic grains, GSD, LOI The ratio between Mg and Ca relates to salinity (from calcium precipitation and Mg dissolution) or inorganic material input (magnesium from weathering).Wick et al. (2003b) interpret the variable Mg at constant Ca in a closed lake as changes in effective moisture, similar to the Sr/Ca ratio in ostracod shells (Ricketts et al., 2001).Grayscale density (GSD) and loss on ignition (LOI) quantify the inflow of inorganic versus organic matter.From this, cold/humid and warm/dry climates can identified but the temperature signal not separated from precipitation.The shape of grains indicates the transport process: rounded quartz grains with frequent silica coating point to aeolian import pointing to dry phases in the origin region (Stanley and Deckker, 2002b).Coarse sediment layer with low organic content and many terrestrial plant macrofossils have been used by Noren et al. (2002b) to find exceptional runoff events and thus storminess.Occurence of allochtonous hematite-stained grains (HSG) and icelandic glass points to ice-rafting as a consequence of increased glacier calving (e.g.Bond et al., 2001c); The preservation status of aragonite at intermediate measures the bottom-water corrosiveness, and is therefore indicative for changes in the THC.ing in coarse temporal resolution, open ocean locations are underrepresented with respect to coasts and land masses (Fig. 1).77% of the records have temporal resolution better than 100 yr and 80% span more than 9000 yr within    (Davis et al., 2003b), isotopic oxygen at Sajama, Bolivia (Thompson et al., 2003b) and sunspot number from Greenland ice cores (Solanki et al., 2004b).Left: De-trended and normalized data.Non-cyclic events according our definition with pa=1.5 are marked with red triangles.Right: Spectral amplitudes after applying the Lomb-Scargle transformation (black: total record, red: bootstrapping of upper interval, thus indicating periodicity confined to the Lower Holocene, green: Upper Holocene spectrum).Dashed lines indicate significant frequencies.Holocene, North Atlantic and in particular eastern American sites with dominant modes build large regional clusters.The global distribution of locations with changes in spectral modes displayed in Fig. 5 demonstrates that these are remarkably homogeneous within these regional clusters.Clusters are generally made out of about 4-9 sites with independent proxy records which, in their large majority, exhibit the same spectral trend.
Obviously, damping or amplification of a climate fluctuation mode is only modestly affected by the heterogeneous quality of records, inherent random noise or other local phe-Fig.2. Three selected records: Temperature reconstruction for Southeast Europe, (Davis et al., 2003b), isotopic oxygen at Sajama, Bolivia (Thompson et al., 2003b) and sunspot number from Greenland ice cores (?).Left: De-trended and normalized data.Non-cyclic events according our definition with p a =1.5 are marked with red triangles.Right: Spectral amplitudes after applying the Lomb-Scargle transformation (black: total record, red: bootstrapping of upper interval, thus indicating periodicity confined to the Lower Holocene, green: Upper Holocene spectrum).Dashed lines indicate significant frequencies.4 Spectral divorce at Mid-Holocene From the literature on Holocene climate vari expect that at least four frequencies play a in our spectral analysis.The two millenn The spatial organization of clusters persist after mapping events also confirms s trends (Fig. 6, left).

Spectral divorce a
From the literature on H expect that at least fou in our spectral analysi tions which comprise t GISP2 δ 18 O records by ( 2004) and the classica 1995) are points of refe In addition, a 320 yr a edly observed in proxy circulation models (GC and Dijkstra, 2003;We 2003).However, none nently than others wit bust result, instead, der of regular climate osci peaks are stable, i.e. fo after partial boostrappi sis that around the 5.5k tive teleconnections to which the monsoon is sensitive.For example, it has been speculated that the atmospheric connection between the western Asian monsoon and the thermohaline circulation (THC) in the North Atlantic decreased in intensity from the Lower to the Upper Holocene (Morrill et al., 2003).This might explain a reduced correlation between trends in millenial and non-cyclic modes in the two climate sub-systems observed in our study.

Fig. 1 .
Fig. 1.Global distribution of high-resolution proxy time series used in this study.Numbers of the 130 records (109 sites) refer to Tab. 2.

Fig. 1 .
Fig. 1.Global distribution of high-resolution proxy time series used in this study.Numbers of the 130 records (109 sites) refer to Table 2.

Fig. 2 .
Fig.2.Three selected records: Temperature reconstruction for Southeast Europe,(Davis et al., 2003b), isotopic oxygen at Sajama, Bolivia(Thompson et al., 2003b)  and sunspot number from Greenland ice cores(Solanki et al., 2004b).Left: De-trended and normalized data.Non-cyclic events according our definition with pa=1.5 are marked with red triangles.Right: Spectral amplitudes after applying the Lomb-Scargle transformation (black: total record, red: bootstrapping of upper interval, thus indicating periodicity confined to the Lower Holocene, green: Upper Holocene spectrum).Dashed lines indicate significant frequencies.

Fig. 4 .Fig. 4 .
Fig. 4. Sites with significant periodicity (red, without: blue) in the Upper and Lower Holocene.For spatial extrapolation see Methods, results for proxies related to solar activity are displayed in the yellow polygon.

Fig. 5 .
Fig. 5. Change in periodic fluctuations from the Lower to the Upper Holocene (positive change: red, negative: blue, no significance: empty circle, significant mode without change: filled black).

5
Possible mechanisms for variability changesIn face of the fragmented understanding of the mechanisms producing climate shifts or quasi-cyclic fluctuations during the Holocene, it seems premature to ask for what may have

Fig. 6 .Fig. 7 .
Fig. 6.Change in periodic fluctuations from the Lower to the Upper Holocene separated according spectral interval (for symbols, see Fig. 5).

Table 2 .
Location, type, temporal coverage and reference of the climate proxy time series used in this study; the numbers in the table identify the position of proxy sites in the map (Fig.1).Abbreviations: SS=sea surface, T=temperature, S=salinity, P=precipitation, LG=lithic grains, SSN=sun spot number, Lk=Lake, Cv=Cave.

Table 2 .
Introduction Histogram of dominant modes in all 130 records.