Palaeotemperature reconstructions of the European permafrost zone during Oxygen Isotope Stage 3 compared with climate model results

JOURNAL OF QUATERNARY SCIENCE (2003) 18(5) 453–464Copyright 2003 John Wiley & Sons, Ltd.Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/jqs.766

Palaeotemperature reconstructions of theEuropean permafrost zone during marine oxygenisotope Stage 3 compared with climate modelresults†

KO VAN HUISSTEDEN,1* JEF VANDENBERGHE1 and DAVID POLLARD2

1 Faculty of Earth Sciences, Vrije Universiteit, Amsterdam, The Netherlands2 Earth Systems Science Center, Pennsylvania State University, USA

van Huissteden, K., Vandenberghe, J. and Pollard, D. 2003. Palaeotemperature reconstructions of the European permafrost zone during marine oxygen isotope Stage 3compared with climate model results. J. Quaternary Sci., Vol. 18 pp. 453–464. ISSN 0267-8179.

Received 26 October 2002; Revised 17 February 2003; Accepted 13 March 2003

ABSTRACT: A palaeotemperature reconstruction based on periglacial phenomena in Europe northof approximately 51 °N, is compared with high-resolution regional climate model simulations ofthe marine oxygen isotope Stage 3 (Stage 3) palaeoclimate. The experiments represent Stage 3warm (interstadial), Stage 3 cold (stadial) and Last Glacial Maximum climatic conditions. Thepalaeotemperature reconstruction deviates considerably for the Stage 3 cold climate experiments,with mismatches up to 11 °C for the mean annual air temperature and up to 15 °C for the wintertemperature. However, in this reconstruction various factors linking climate and permafrost havenot been taken into account. In particular a relatively thin snow cover and high climatic variabilityof the glacial climate could have influenced temperature limits for ice-wedge growth. Based onmodelling the 0 °C mean annual ground temperature proves to be an appropriate upper temperaturelimit. Using this limit, mismatches with the Stage 3 cold climate experiments have been reduced butstill remain. We therefore assume that the Stage 3 ice wedges were generated during short (decadaltime-scale) intervals of extreme cold climate, below the mean temperatures indicated by the Stage3 cold climate model simulations. Copyright 2003 John Wiley & Sons, Ltd.

Journal of Quaternary Science

KEYWORDS: periglacial features; ice wedges; palaeoclimate; climate model; marine oxygen isotope Stage 3; Europe.

Introduction

Palaeoclimate modelling has been attempted for several timewindows of the Last Glacial Maximum and its termination(e.g. Manabe and Broccoli, 1985; COHMAP Members,1988; Isarin and Renssen, 1999; Kageyama et al., 2001).The palaeoenvironment on the European continent duringmarine oxygen isotope stage 3 (Stage 3) is of special interestfor several reasons. The rapid climatic oscillations duringStage 3, expressed in the Greenland ice-core records bythe Dansgaard–Oeschger cycles (GRIP Members, 1993) areintriguing. This type of climatic variability appears to be anintrinsic property of glacial climates (McManus et al., 1999;Lu et al., 2000). Stage 3 may be a model for the climaticconditions that determined environmental processes duringmuch of the Quaternary. Also, during Stage 3 modern manreplaced the Neanderthals in Europe. The climate changes

* Correspondence to: Dr Ko van Huissteden, Faculty of Earth and Life Sciences,Vrije Universiteit, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands.E-mail: [email protected]†This article is part of a series of articles reporting the results of the Stage 3Project (Van Andel, 2002).

may have influenced human development and migrationpatterns (Van Andel and Tzedakis, 1996; Hublin, 1998).The ‘Stage 3 Project’ (Van Andel, 2002) aims to provide abetter understanding of Stage 3 environmental conditions inEurope by bringing together climate modellers and specialistsfrom various palaeoecological disciplines. Whereas previousclimate modelling studies use GCMs (global circulationmodels) with the accompanying low resolution grid, anRCM (regional circulation model) has been applied in theStage 3 project to resolve the effects of the Europeantopography.

This study concentrates on the periglacial zone in con-tinental Europe during Stage 3. Palaeotemperatures derivedfrom periglacial features, supported by botanical and fau-nal palaeoclimate proxies, have been compared with climatemodel results. Periglacial features (thermal contraction cracks,cryoturbations, cryogenic soil microfabrics and periglaciallandforms; Table 1 and Fig. 1) have been frequently used as anabiotic palaeotemperature proxy in continental environments(Maarleveld, 1976; Karte, 1983; Vandenberghe and Pissart,1993; Isarin and Renssen, 1997; Huijzer and Isarin, 1997;Huijzer and Vandenberghe, 1998).

454 JOURNAL OF QUATERNARY SCIENCE

A B

D

C

Figure 1 (A) Large ice-wedge cast of marine oxygen isotope stage 2 age in fluvial and aeolian sands (Nochten, Germany). (B) Smaller ice-wedgecast of marine oxygen isotope stage 3 (Stage 3) age in coarse fluvial sand and gravel (Nochten). (C) Large cryoturbations of Stage 3 age in fluvial silt(Hengelo, The Netherlands): arrow indicates a small ice wedge cast. (D) Ice-wedge cast (Stage 3, Hasselo stadial) developed in silt with peat beds,overlain by thaw lake deposits (Hengelo). Length of scale bar in all photographs: 1 m

Table 1 Palaeotemperature estimates derived from periglacial features (summary after Huijzer and Vandenberghe, 1998): MAAT, mean annual airtemperature; MTCM, mean temperature of the coldest month; MAGT, mean annual ground temperature (soil temperature in the root zone)

Periglacialphenomena

MAAT, °C(conservative

estimate)

MTCM, °C References Proposed upperlimits (this study)

Ice-wedge, sand-wedgeand composite-wedge casts

≤ −4 (silt and clay substrate) ≤ −8(coarse grained substrate) meanannual ground temperature ≤ −2

≤ −20 Lachenbruch (1962)Pewe (1966a,b)Romanovskij (1985)Burn (1990)Melnikov and Spesivtsev (2000)

MTCM of approximately−15 °CMAGT 0 °C (correspondingvery approximately to a MAATof −2 to −3 °C)

Large amplitude (>0.6 m)cryoturbations

Presence of continuous permafrostpreceding their formation

Vandenberghe (1988)Vandenberghe and Pissart (1993)Murton et al. (1995)Murton and French (1995)Melnikov and Spesivtsev (2000)

Similar palaeoclimate studies have been performed for thelast glacial maximum and the climatic oscillations at thetermination of the last glacial; e.g. Isarin and Renssen (1997),Huijzer and Vandenberghe (1998), Isarin and Renssen (1999)and Renssen and Isarin (2001) also used periglacial featuresin palaeodata–climate model comparisons. In addition to the

uncertainties of climate model simulations, palaeotemperaturereconstructions from periglacial features also have limitations(Renssen and Isarin, 2001). First, reconstructions are beingmade for palaeoenvironments and climates for which amodern analogue is not available. Second, most reconstructionmethods are unable to quantify uncertainty. Third, the areal

Copyright 2003 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 18(5) 453–464 (2003)

DATA–MODEL COMPARISONS OF STAGE 3 PALAEOTEMPERATURE 455

coverage of the palaeodata may be strongly dependent onpresent-day research effort. In this paper we include anevaluation of the first two problems.

Stage 3 climate model experiments

The Stage 3 climate over Europe was modelled by firstrunning a global climate model, then using its stored resultsto force a finer-grid regional climate model over Europe.The global climate model (GCM) used is GENESIS version2.0 (Thompson and Pollard, 1997), which has been designedparticularly for use in palaeoclimate experiments (e.g. Pollardand Thompson, 1997; Pollard et al., 1998; DeConto et al.,2000). The boundary conditions and other parameters forthese runs are described in detail by Barron and Pollard(2002), Arnold et al. (2002) and Van Andel (2002). Briefly,prescribed global fields of monthly sea-surface temperaturesand sea-ice were derived for 21 000 cal. yr BP from CLIMAP(1981), augmented by GLAMAP 21 000 cal. yr BP core datain the Atlantic, and further augmented for the Stage 3experiments by core data in the northern North Atlantic forwarm and cold Stage 3 periods (Barron and Pollard, 2002).Other boundary conditions differing from the present wereatmospheric CO2 amounts (200 ppm vol. for 21 000 and30 000 cal. yr BP), ice-sheet sizes and the Earth–Sun orbit.The GCM was connected interactively to the BIOME 3.5predictive vegetation model (Haxeltine and Prentice, 1996;Alfano et al., 2003), with the GCM climate responding to thephysical vegetation attributes, and the vegetation distributionsupdated annually by BIOME3.5 from the GCM climate of theprevious year(s). Each global simulation was run for severaldecades to allow the climate and the vegetation to equilibratefully.

The regional climate model (RCM) used is RegCM2 (Giorgiet al., 1993a,b), used successfully for present-day applicationsover many regions of the world, and for late Pleistocene studiesby Hostetler et al. (1994). The European domain and regionalboundary conditions used for the Stage 3 experiments aredescribed in detail in Barron and Pollard (2002). The horizontalresolution used was 60 km (compared with ca. 400 kmfor the GCM). Each RegCM2 run was 3 years in duration,forced at the boundaries by 6-h fields of air temperatures,humidities, wind speeds and surface pressure, interpolatedfrom stored GCM fields from the latter years of each globalsimulation.

The land-surface components of the models are of particularinterest in comparing with observed periglacial features. Boththe GCM and RCM include vertical-column models of snowand soil at each grid point. Each snow model predicts varyingsnow depth depending on snowfall, surface melting andsublimation, and the GCM model includes a parameterisationof fractional snow coverage for small amounts of snow. TheRCM includes a reduction of snow albedo with age. However,neither model accounts for refreezing of meltwater or snowcompaction; any melt or rain immediately is transferred tothe soil, and a constant snow density is assumed. The GCM’spresent-day snow results are included among other GCMsin the intercomparison study of Foster et al. (1996). The soilmodels extend down to 4.25 m (GCM) and 1–2 m (RCM),involve vertical heat conduction and vertical movement of soilmoisture, and include the latent heat of soil moisture as thesoil freezes or thaws. The soil thermal properties are based onan average soil with a loamy texture. However, the RCM soil

depths (1–2 m) may not be deep enough to extend throughthe active layer and reach permafrost depths.

In this paper, we show regional-model results from fourStage 3 experiments:

1 simulating a typical ‘Stage 3 cold’ interval (ST3Cold1hereafter);

2 Stage 3 cold ‘ad hoc’ (ST3Cold2 hereafter), as previous butwith North Atlantic sea surface temperatures lowered, andsea ice increased, to the extreme limits allowed by thedeep-sea core data (Barron and Pollard, 2002);

3 simulating a typical ‘Stage 3 warm’ interval (ST3Warm);4 simulating Last Glacial Maximum conditions (LGM),

21 000 yr.

The differences between the ST3Warm and ST3Cold exper-iments were the augmentation of sea-surface temperatureswith several degrees C, and a smaller Scandinavian ice-sheet for the ST3Warm case. The difference between theST3Warm and ST3Cold1 experiments proves to be relativelysmall, therefore the ST3Cold2 has been done to evaluatethe effect of sea-surface temperatures. The LGM experimentserves as a reference experiment (Barron and Pollard, 2002;Pollard and Barron, 2003). Next to these experiments, acontrol experiment to simulate the present-day climate hasbeen run.

The primary field examined below is surface air (2 m) tem-perature (Fig. 2). Summer and winter surface air temperaturescool significantly at 30 000 yr compared with the present, withgreater cooling in the winter (by 7 to >10 °C north of ca.55 °N; ca. 4–7 °C to the south) than in the summer (ca. 4 to7 °C north of ca. 55 °N; 2–4 °C to the south). The cooling isgreater in winter primarily because of the much more exten-sive sea-ice cover in the North Atlantic compared with thepresent. There are only a few degrees difference between theST3Cold and ST3Warm temperatures, which suggests that thedifferences in the prescribed SSTs between these two periodsare too small (Barron and Pollard, 2002). In the LGM simu-lation, the cooling from the present is considerably greater,as expected from the much larger prescribed ice sheets andcooler north Atlantic SSTs (cf. Pollard et al. (1998) for theGCM).

Despite generally reduced annual precipitation rates inST3Cold compared with the present, owing to the reduc-tion in the overall hydrological cycle at colder tempera-tures, the autumn snowfall season begins earlier and thedepth of winter snow cover is significantly greater than atpresent over northern and central Europe in the RCM, byfactors of two or more. This trend is accentuated in theLGM. However, compared with present-day subarctic cli-mates with surface air temperatures similar to those of themodelled ice-age climates, the snow cover is similar orthinner.

Generally, the wintertime winds are more intense (byca. 30%) at ST3Cold compared with modern, owing tointensification of westerly onshore winds forced by greaterthermal north–south gradients at the sea-ice edge in theNorth Atlantic (also for the LGM; e.g., Kageyama et al.,1999). Probably as a consequence, both the synoptic andinterannual variability of temperatures in the RCM andGCM at ST3Cold and LGM are significantly greater (by ca.50–100%) than present over Europe, indicating considerablymore storminess.



LGM

ST3Cold1

ST3Cold2

ST3Warm

-20

0

-10

-15

5

20

10

15

-5

MAAT

Ice & sand wedges

On sand or gravel

On fine-grained substratum

°C

Ice-cap

Glacial coast linein climate model

N

0 °C mean annual groundtemperature isotherm

-15 °C January airtemperature isotherm

Permafrost according toTTOP model

Figure 2 Comparison of modelled mean annual air temperatures for the ST3Warm, ST3Cold1 and 2, and LGM experiments, and locations of ice-and sand-wedge casts of Stage 3 age

Palaeotemperature reconstruction andcomparison with climate model results

Data collection

The data on relict Stage 3 periglacial features have beenderived from a multiproxy palaeoclimate data base (MPDB)of palaeoclimate data obtained from published sources inEurope (Huijzer and Isarin, 1997; Isarin et al., 1998). The datahave been checked on the following criteria, based on thesource material: (i) a correct identification of the features (forinstance, an ice-wedge cast should show certain diagnosticcriteria to distinguish it from other types of soil structures, seee.g. Vandenberghe, 1983; Harry and Gozdzik, 1988); (ii) aclear site description allowing consideration of the effects oflocal site conditions; (iii) reliable dating control, consisting ofradiocarbon or luminescence dates. The ages of the wedgesare based on all available stratigraphical evidence includingbio and lithostratigraphy. In Table 2, the age control is markedwith ‘A’ if the age is based on radiocarbon or luminescencedates above and below the wedges, or ‘B’ if either dates aboveor below are lacking and litho- or biostratigraphical evidenceis used, or the ‘below’ dates are radiocarbon dates > 45 kyrBP. In particular for the A sites, the assumption that a Stage4 or Stage 2 wedge has been erroneously dated as a Stage 3wedge is unlikely.

Stage 3 corresponds broadly to the British Middle Deven-sian Substage and the northwest European Weichselian Middle

Pleniglacial (Vandenberghe, 1985). Radiocarbon dates indi-cating the age of periglacial features have been calibratedusing CALPAL (Joris and Weninger, 2000), which is based on acombination of different data sets relating radiocarbon yearsto calendar years (Bard et al., 1998; Kitegawa and Van derPlicht, 1998; Voelker et al., 1998). The approach by Jorisand Weninger (2000) allows an evaluation of the uncertaintyinvolved in conversion of radiocarbon to calendar ages.

Palaeotemperature reconstruction from periglacialfeatures

Thermal contraction cracks are generated by rapid tempera-ture drops at low winter temperatures (Lachenbruch, 1962;Mackay, 2000). When cracks occur repeatedly at the samelocation and become filled with ice or blown-in sand, an icewedge or sand-wedge results. These wedges are related to con-tinuous permafrost. Their palaeoclimatic significance has beenreviewed by, for example, Pewe (1966a,b), Harry and Gozdzik(1988) and French (1996). Commonly cited palaeotemperaturevalues are a mean annual air temperature (MAAT) ≤ −8 °Cor ≤ −4 °C depending on the substrate of the wedge, and amean temperature of the coldest month (MTCM) ≤ −20 °C(Table 1). Harry and Gozdzik (1988) state that ice-wedge castsmerely indicate the former existence of continuous permafrostand are only crudely related to air temperatures. Smaller ther-mal contraction cracks (e.g. seasonal frost fissures, sand veins)occur abundantly in Stage 3 deposits (e.g. Van Huissteden,



Table 2 Dated Stage 3 ice-wedge casts in northern Europe. Wedge type: I, ice wedge; S, sand wedge, C, composite wedge. Substratum column:L, loam/silt; S, sand; G, Gravel, T, Till; P, peat. Age control column: A, dates above and below wedge or TL dated sand in sand wedge; B, datingpartly based on other stratigraphical correlation

Site name Longitude Latitude Wedge Type Substratum Approximate age(cal kyr)

Agecontrol

References

Kesselt 5.62 50.8 I L 33–34.5 A Huijzer, 1993I L 34.5–41 A

Hostrup 8.67 55.7 C S 39 A Kolstrup and Mejdahl, 1986Kepno 17.93 51.3 I L >35.1 B Rotnicki and Tobolski, 1969Kalinko 19.8 51.5 I S >32.2 B Manikowska, 1993Hengelo 6.82 52.3 I L,S 37–41 A Van Huissteden, 1990;

I L,P 41–46 A Ran et al., 1990Grouw 5.83 53.1 I S,P >46.7 B Kasse et al., 1995

I L,S 43.5–46.7 AI L,S 41.1–43.5 A

Four Ashes −2.12 52.7 I G 34–43 A Morgan, 1973Konigsaue 11.47 51.8 I L 36.5–28.5 A Mania and Toepfer, 1973

I L 45–36.5 AI L 46–45 A

Amersfoort 5.4 52.2 I L,P >47 B Zagwijn and Paepe, 1968Scheibe 14.32 51.5 I S >47 A Mol, 1997

I L,S >47 BBitterfeld 12.33 51.6 I S,G <31.4 B Hiller et al., 1991Poledno 18.42 53.4 S T >27.3 B Drozdowski and Fedorowicz,

1987Beckford −2.03 52 I S,G >31.3 B Briggs et al., 1975Nochten 14.34 51.3 S 29.7–39.3 A Bos et al., 2001

S,G 39.9–46.5 A

1990; Mol, 1997). Although these have been used in palaeo-climatic reconstruction (Maarleveld, 1976) their significanceis uncertain.

The palaeotemperature estimates refer to surface air temper-atures, but Romanovskij (1985) stresses the importance of soiltemperatures. The relationship between soil and air tempera-tures depends on the soil heat balance, which is determinedby local site conditions such as thermal properties of the soil,snow cover and vegetation (e.g. Smith and Riseborough, 2002,and references therein). A difference of +2 to +3 °C betweenair and soil temperature has been used as a rule of thumb inpalaeotemperature inferences (Ran et al., 1990; Huijzer andVandenberghe, 1998).

Periglacial involutions are attributed to cryostatic pressuresin the soil (e.g. Van Vliet-Lanoe, 1988) or to soft-sedimentdeformation promoted by high pore-water pressures causedby melting ice lenses in the soil (e.g. Vandenberghe, 1988).Larger forms (involutions with amplitudes in the order of 0.5 mand more) are generally attributed to differential loading duringdegradation of ice-rich permafrost (Vandenberghe and Van denBroek, 1982; Vandenberghe, 1988; Murton and French, 1995;Murton et al., 1995), and therefore imply the former presenceof permafrost. Other permafrost degradation features in Stage3 deposits are thaw lakes (Ran et al., 1990; Van Huissteden,1990; Kasse et al., 1995). However, permafrost degradationmay also be caused by changes in local conditions and neednot be related to climate warming (e.g. Burn and Smith, 1988).A climatic origin of thermokarst features requires widespreadand synchronous occurrence, preferably in conjunction withother evidence of climate change.

Temporal evolution of climate in the Stage 3periglacial zone

Ice-wedge casts dating from Stage 4 have been reportedfrequently (Huijzer and Vandenberghe, 1998). Between thestart of Stage 3 up to 50 000 yr BP no indications for permafrost

have been preserved. After 50 000 cal. yr BP ice wedgesformed in eastern Germany (Mol, 1997). Palaeobotanicaldata from peat deposits point to mean temperatures of thewarmest month (MTWM) ranging episodically from 8–10 °Cto ≥12–13 °C (Kasse et al., 1995; Bos et al., 2001). Indicationsfor warm intervals are the Upton Warren Interstadial in theUK (Coope, 2002; MTWM 15–18 °C) and the Riel Interstadial,dated around 50 000 yr BP (Vandenberghe, 1985). Periodswith cold conditions including ice-wedge formation (MTWMaround 10°) alternated with considerably warmer intervals(MTWM >13° or higher). These warm intervals are probablyshort-lived ‘thermal spikes’, which are recorded primarily inrapidly reacting species (Huijzer and Vandenberghe, 1998;Coope, 2002).

A marked cold interval between 44 000 and 43 000 cal.yr BP (Hasselo Stadial) is known from palynological recordsand ice-wedge casts (Kolstrup and Mejdahl, 1986; Ran et al.,1990; Van Huissteden, 1990). The presence of ice-wedge castsis limited to fine-grained soils in The Netherlands and Belgium,whereas they seem to be absent in France. A composite wedgecast on coarse-grained substratum in Denmark (Kolstrup andMejdahl, 1986) indicates permafrost. From plant and insectspecies a MTWM around 10–11 °C and a MTCM (meantemperature of the coldest month) between −27° and −20 °C isinferred. This is followed by the Hengelo Interstadial between43 000 and 40 000 cal. yr BP (Zagwijn, 1974), accompaniedby permafrost thaw phenomena (Van Huissteden, 1990; Kasseet al., 1995). The widespread occurrence of these permafrostdegradation features indicates a climatic origin, suggestingMAAT above 0 °C; pollen data in The Netherlands indicate aMTWM of ≥13 °C (Kasse et al., 1995).

Summer temperature dropped at least 3° after the HengeloInterstadial (Kasse et al., 1995). Ice-wedge casts in coarse-grained deposits indicate continuous permafrost at least duringcolder intervals. The MTWM of cold periods was around10° (Kolstrup and Wijmstra, 1977), in warm intervals theMTWM was at least higher than 11.5° and the permafrost



degraded (MAAT > −4°) at least locally, as testified by largecryoturbations.

This record of Stage 3 periglacial features indicates episodicdegradation and regrowth of permafrost. However, the smallnumber of sites in each time interval precludes a reconstructionof the extension of permafrost within particular time intervals.The distribution of the wedges (Table 2 and Fig. 2) is thereforeassumed to indicate their maximum extent during a typicalStage 3 stadial. The wedges in Table 2 have been classifiedaccording to type (ice, sand or composite wedge) and substratetype (fine-grained or coarse-grained). Generally ice-wedgesof Stage 3 age are relatively small compared with theircounterparts of Stage 2 age (Fig. 1). This smaller size suggeststhat during Stage 3 conditions for ice-wedge formation mayhave been less optimal compared with Stage 2 (shallowerwedges, a thicker active layer and/or shorter or more infrequentactivity).

Uncertainty in the palaeotemperaturereconstruction from ice-wedge casts

Our temperature reconstruction is based mainly on ice- andsand-wedge casts, as these are relatively widespread andwell dated. Also evaluation of the uncertainty of the MAATand MTCM limits for ice wedge casts in Table 1 is nec-essary. Although these figures are generally accepted forpalaeotemperature reconstruction (e.g. Huijzer and Vanden-berghe, 1998), the ‘error bars’ are unknown. Error sourcesare: (i) uncertainty in the present-day relationship betweenice wedges and temperature; (ii) uncertainty introduced byfavourable local site conditions; (iii) uncertainty owing the factthat we try to reconstruct palaeotemperatures for environmentsand climates for which a modern analogue no longer exists.

There is uncertainty in the present-day temperature limitsfor ice-wedge formation. Pewe (1966a,b) mentions uppertemperature limits between −6° to −8°. Romanovskij (1985)mentions MAGT (mean annual ground temperature, taken hereas the mean annual temperature of the root zone) up to −2 °Cas an upper limit corresponding to MAAT of approximately−4° to −5°. However, the difference between MAGT andMAAT is strongly influenced by local site conditions such asvegetation and snow cover (e.g. Yershov, 1998; Smith andRiseborough, 2002). Burn (1990) reports frequent ice-wedgecracking in an area with a mean annual air temperature of−4.0 °C. The MTCM limit of −20 °C in Table 1 is a theoreticalone, based on a physical model by Lachenbruch (1962) thatindicates that cracking is most likely at rapid temperature dropsbelow −20°, and observations by Pewe (1966b) and Mackay(1974, 1993).

The distinction between active and inactive ice wedgeshas been made on ground surface morphology (active, low-centered polygons versus inactive high-centered polygons,Pewe, 1966a,b) or physical indications of recent cracking (e.g.Mackay, 1974; Burn, 1990). However, a range of crackingfrequencies exists between ‘active’ and ‘inactive’ wedges(Mackay, 1991, 1993, 1999). Thermal contraction crackingis related to the incidence of rapid temperature drops in winterrather than directly to MTCM or MAAT (Mackay, 1991, 1993).Even this relationship strongly depends on local site conditionsand antecedent conditions of the permafrost. Mackay (1993)and Mackay and Burn (2002) show that young, recentlyformed ice wedges crack at higher temperatures than old,mature wedges. Thermal contraction coefficients of soils varywidely both in space and time owing to spatial differences insoil physical properties, time-dependent frozen water content

of the soil and thermal history (review by Mackay, 2000).As shown by Mackay (1991, 1993) ice-wedge cracking is ahighly probabilistic process, and cracking frequencies varywidely in the same area. The temperature limit betweenactive and inactive ice wedges is at best a fuzzy boundary.Associated with the temperature limits in Table 1, a rangeof temperatures should exist within which the probability ofice-wedge formation decreases with increasing temperature.

Local site conditions may promote or suppress the forma-tion of ice or sand wedges (e.g. Brown, 1973; Mackay, 1993).Ice-wedge casts in fine-grained soils occur at higher air tem-peratures than in coarse-grained soils, because of differencesin thermal properties of the substrates (Romanovskij, 1985). InTable 1, this effect has been incorporated by using differentMAAT values for coarse-grained and fine-grained substrates.Also snow cover is important because of its insulating proper-ties (Mackay, 1993). As the wedges have been found in fluvialor loess sequences they may have been located partly onwind-exposed river terraces or hill slopes. The interaction withthe vegetation cover also plays a role. Vegetation traps snowin winter causing insulation from low temperatures, althoughthe vegetation also insulates the soil from high temperatures insummer (Brown, 1973). The net effect depends on the durationof the seasons and the continentality of the climate (Yershov,1998). All these site characteristics behave—at the spatialscale of the climate model grid cells—as stochastic variables,justifying variance around the temperature limits in Table 1.

Present-day high-latitude climates differ from Pleistocenemid-latitude glacial climates, in particular with respect toinsolation (French, 1996), but also other climatic factors (snowcover, wind, temperature variability) may have been importantin the Stage 3 climate. Albeit that these features have beenlargely derived from the climate models themselves, it is stilljustified to account for their effects, because these features areplausible from the boundary conditions and theory on whichthe models are based.

Snow cover

Snow cover is the critical factor in determining the southernlimit of continuous permafrost. Smith and Riseborough (2002)modelled the influence of MAAT, snow cover, vegetation andsoil thermal properties on the permafrost top temperature withthe TTOP (temperature at the top of permafrost) model. Theirresults indicate that below −6° to −8° continuous permafrostoccurs at practically any snow depth. At higher temperaturespermafrost distribution is strongly dependent on snow cover,the MAAT at which permafrost occurs in mineral soils rises to−3 °C if snow cover decreases to 0.25 m, and with absence ofsnow cover permafrost may occur even at a MAAT of +2 °C inmineral soils.

The snow cover thickness and thermal properties areinfluenced by snowfall amount, losses through melting andsublimation, compaction and melt–refreeze cycles, and windthat may strip snow from exposed surfaces. The strongerdiurnal temperature cycle caused by the lower latitude positionof the Stage 3 periglacial zone may have caused moremelting–refreezing cycles, decreasing the thickness of thesnow cover and modifying its thermal properties. The modelresults also indicate low winter precipitation for the ST3Coldand LGM simulations and stronger winter winds, comparedwith modern high-latitude climates. The low precipitationshould have decreased the snow cover in general, and thewind may have created at least locally areas with greatlyreduced snow cover.



Climate variability

The climate model results show a higher variability of theStage 3 temperatures compared with the present, as shownby the high standard deviation of instantaneous temperaturesof the Stage 3 climates (the instantaneous variance of thetemperature is ca. 24 for Stage 3 Cold climates versus ca.21 in the modern control experiment for all grid cells with a‘periglacial’ climate with MAAT < 0°). Similar variability hasbeen found in other climate model simulations (Renssen andBogaart, 2003). As mentioned above, ice-wedge cracks aregenerated by rapid temperature drops in winter, rather thanlow temperatures alone. The incidence of these temperaturedrops is directly related to the variance of the temperature time-series. A high synoptic variability therefore will enhance thefrequency of rapid temperature drops and thereby the rate ofthermal contraction cracking. Burn (1990) reports cracking ata site that has a marginal temperature for ice-wedge formationaccompanied by a high temperature variability (standarddeviation about MAAT ±2 °C).

We modelled the influence of climatic variability on ice-wedge activity with Monte Carlo simulations of synthetictime-series of daily temperatures with a specified varianceand correlation structure. The time series are 50 yr long.Temperature drops at a rate ≥1.8 °C day−1 over more than3 days, at a minimum temperature below −20 °C, arefavourable for ice-wedge cracking (Mackay, 1993). To accountfor spatial and temporal variations in soil resistance tocracking, a random resistance factor determined by a Weibulldistribution is added. If the temperature-drop requirementsincluding the resistance factor are exceeded, the event iscounted as an ice-wedge cracking event. Thermal contractioncracking activity is computed as a cracking ratio, which is theproportion of years in which a crack occurs. The experimentsare repeated with 100 synthetic time-series to determine thevariance of the cracking ratio.

Experiments with different winter temperatures and a Julytemperature of 10 °C indicate that the cracking ratio drops fromnearly 1.0 at a January temperature of <−20 °C (= MAAT of−5 °C) to a low value at a MTCM of −10 °C (Fig. 3A). Modelruns using different daily temperature variability demonstratethat the daily synoptic variability indeed strongly influencesthe cracking ratio positively (Fig. 3B). By contrast, interannualvariability appears to have a negligible effect (Fig. 3C), asa high interannual variability does not increase the actualnumber of winter extremes, but only the magnitude of extremewinter temperatures.

Mid-latitude insolation cycle

Increased insolation at middle latitudes, compared with thepresent-day Arctic with its long arctic night, will increase sum-mer temperatures, increasing the active layer depth. However,if thermal contraction cracking penetrates sufficiently deep inwinter, the cracks will still be preserved although the resultingice wedges will be thinner and shallower.

Based on the above, we assume the existence of a varianceband around temperature limits for ice-wedge formation inTable 1. This variance requires exploration of its magnitudeand hence possible upper temperature extremes for the growthof ice-wedge casts. This upper limit, based on the assumptionof extremely favourable conditions for ice-wedge growth, isbased on the following reasoning. A MAGT of 0 °C is theupper physical limit for the existence of permafrost (Harryand Gozdzik, 1988) for palaeotemperature interpretation.This corresponds to a MAAT of approximately −2 °C to−3 °C (Huijzer and Vandenberghe, 1998). However, this offsetbetween MAAT and MAGT depends strongly on snow cover.With a paleobotany-based MTWM of approximately +10 °Cduring Stage 3 stadials (Kolstrup and Wijmstra, 1977; Coope,2002), this indicates an approximate MTCM of −15 °C insteadof −20 °C. Based on the TTOP model of Smith and Riseborough(2002) this may even be higher in the case of a very thin snowcover.

On the other hand, the MTCM limit of −20 °C has beenconfirmed by palaeobotany and coleopteran faunal evidencecited above. However, biological palaeotemperature estimatesare equally subject to uncertainty. The error bars on thepalaeobotanical temperature estimates used by Kageyamaet al. (2001) are of the same order as the discrepancies betweendata and model. Also the error bars on the temperatureestimates of the coleopteran data of Coope (2002) indicatewinter temperatures up to −10 °C in most cases. Elias et al.(1999) have shown that the mutual climatic range methodapplied to present-day coleopteran faunal assemblages on theAlaskan coast may underestimate the MTCM by more than20 °C (on average 12.8 °C), owing to synoptic variability of theclimate in the shape of extreme cold spells in winter.

In conclusion, a comparison of ice-wedge sites withthe climate model should be based on the modelled soiltemperatures rather than air temperatures, with a temperatureof 0 °C as an extreme upper limit for ice-wedge formation. Thewinter temperature limit of −20 °C is subject to considerableuncertainty and is probably too low. It is based largely ontheoretical considerations (Lachenbruch, 1962) and thermalcontraction coefficients of soils vary widely in space and

0

-10

1.0

0.8

0.6

0.4

0.2

crac

king

rat

io

mean annual air temperature °C

mean January temperature °C

-10 -8 -6 -4 -2

-30 -25 -20 -15

A

5 6 7 8 9

standard deviation daily temperature

1.0

0.8

0.6

0.4

0.2

crac

king

rat

io

B

5 6 7 8 9

standard deviation yearly temperature

1.0

0.8

0.6

0.4

0.2

crac

king

rat

io

C

Figure 3 Effect of winter temperature variability on ice-wedge activity resulting from Monte Carlo experiments using randomly generatedtemperature time-series. (A) Effect of mean annual and winter temperature on cracking ratio, with fixed July temperature (10°) and varying Januarytemperatures (−30° to −10°). (B) Effect of synoptic variation (standard deviation of daily temperature). (C) Effect of annual variability (standarddeviation of the temperature of the coldest month). Experiment series B and C use a mean July temperature of 10° and mean January temperatureof −15°



time (Mackay, 2000). An upper limit of −15 °C is acceptable.Nevertheless the temperature limits in Table 1 are alreadyupper limits in present-day climates. Using the limits givenhere is only valid under the assumption of otherwise favourableclimatic conditions (thin snow cover and high synopticvariability) at the Stage 3 ice-wedge sites.

Data-model comparison

A comparison between the palaeodata and modelled tem-peratures is shown in Table 3. For the ST3Warm simulation,the permafrost distribution is unknown. However, thaw lakedeposits and large cryoturbations dating from the HengeloInterstadial in conjunction with palaeobotanical evidence sug-gest that the continuous permafrost boundary could haveshifted at least 3° of latitude northward in The Netherlands(Table 4; Ran et al., 1990; Van Huissteden, 1990; Kasse et al.,1995). Hence, the modelled MAAT for the ST3Warm exper-iment, ranging between 1 and 4 °C may be consistent withthe periglacial evidence, although an upper temperature limitcannot be given.

The modelled ST3Cold and LGM experiments are toowarm based on the conservative limits in Table 1. TheMAAT of both ST3Cold experiments deviate on average

≥7.5 to 7.6 °C, with a maximum of 11 °C; the differencesbetween ST3Cold1 and ST3Cold2 are marginal (Table 3). TheLGM experiment matches the Stage 3 data more closely(MAAT deviation on average ≥3.8 °C, range 0–7 °C) butthis experiment needs boundary conditions (ice-cap and sea-surface temperatures) that do not match Stage 3 conditions.The discrepancies in temperature are mainly determined by thewinter temperatures (on average ca. 13 °C, up to15 °C for theST3Cold experiments, and on average 8.3 °C, up to10.8 °C forthe LGM experiment). The discrepancy in winter temperatureshows a clear west–east gradient, with highest disagreementin western Europe.

If it is assumed, based on the modelled climate, that duringStage 3 favourable climatic conditions for ice-wedge formationprevailed and the upper limits indicated in Table 1 are used,the discrepancies are reduced. Using the 0 °C MAGT isothermas the limit for ice-wedge formation, the discrepancies rangebetween 0 and 2.9 °C for the ST3Cold simulations, whereasfor the LGM all wedge locations are within the temperaturelimit. The deviations for the ST3Cold experiments are withinthe error range of the model temperatures and the varianceintroduced by spatial variation of soil characteristics withinthe climate model grid cells. Applying a winter temperaturelimit of −15 °C instead of −20 °C, all simulations remain toowarm, by ≥10 °C for the ST3Cold simulations and ≥5.8 °Cfor the LGM simulation. However, for most sites the LGM

Table 3 Comparison of palaeotemperature estimated from ice wedges with climate model results. Model temperature columns: ST3W, ST3Warmexperiment; ST3C, ST3Cold experiments 1 and 2; LGM, Last Glacial maximum (‘21k’ experiment). The first column refers to the site name inTable 2. Each row attributed to a site represents a stratigraphical level with ice wedges. The second column shows the estimated palaeotemperatureaccording to Table 1, column 2. The other columns show the modelled temperatures for each climate model simulation. The bottom three rowscontain minimum, maximum and average differences between palaeotemperature estimate and model

Site name Max MAATTable 1, col 2

Model MAAT Modelled Januarytemperature

Model MAGT TTOP modeltemperature

ST3W ST3C1

ST3C2

LGM ST3W ST3C1

ST3C2

LGM ST3W ST3C1

ST3C2

LGM ST3C1

ST3C2

LGM

Kesselt −4 3.8 2.6 2.6 0 −3.1 −6.0 −5.0 −9.6 3.6 2.3 2.4 −0.3 2.5 2.6 −0.1−4 3.8 2.6 2.6 0 −3.1 −6.0 −5.0 −9.6 3.6 2.3 2.4 −0.3 2.5 2.6 −0.1

Hostrup −8 1 −0.4 −0.5 −6.6 −5.9 −11.0 −10.5 −15.4 0.7 −0.8 −0.8 −8.6 −0.5 −0.4 −2.4Kepno −4 3.8 2.9 3.2 −1.7 −5.1 −7.7 −7.4 −14.1 3.5 2.6 2.9 −1.8 2 2 −1.9Kalinko −8 3.5 2.6 2.7 −1.8 −5.8 −8.2 −8.5 −14.9 3.2 2.3 2.3 −2.0 2 1.9 −1.3Hengelo −4 4 2.8 2.8 −0.4 −2.9 −6.4 −5.4 −10.0 3.8 2.6 2.6 −1.2 1.9 1.6 −0.4

−4 4 2.8 2.8 −0.4 −2.9 −6.4 −5.4 −10.0 3.8 2.6 2.6 −1.2 1.9 1.6 −0.4Grouw −4 3.6 2.4 2.3 −0.8 −3.3 −6.9 −5.9 −10.2 3.3 2.2 2.1 −1.6 1.4 1.3 −0.8

−4 3.6 2.4 2.3 −0.8 −3.3 −6.9 −5.9 −10.2 3.3 2.2 2.1 −1.6 1.4 1.3 −0.8−4 3.6 2.4 2.3 −0.8 −3.3 −6.9 −5.9 −10.2 3.3 2.2 2.1 −1.6 1.4 1.3 −0.8

FourAshes −8 3.2 1.5 1.1 −2.9 −2.3 −6.2 −5.7 −9.9 3 1.3 0.9 −3.4 0.5 0.4 −1.4Konigsaue −4 3.5 2.4 2.5 −0.9 −3.9 −7.1 −6.3 −11.2 3.2 2.2 2.3 −1.1 2 2.5 −1.0

−4 3.5 2.4 2.5 −0.9 −3.9 −7.1 −6.3 −11.2 3.2 2.2 2.3 −1.1 2 2.5 −1.0−4 3.5 2.4 2.5 −0.9 −3.9 −7.1 −6.3 −11.2 3.2 2.2 2.3 −1.1 2 2.5 −1.0

Amersfoort −4 4 2.9 2.8 −0.2 −2.8 −6.2 −5.2 −9.7 3.8 2.6 2.5 −1.1 2 2 −0.5Scheibe −4 3.3 2.4 2.5 −1.6 −4.5 −7.4 −6.9 −12.8 3.1 2.1 2.4 −1.8 2.2 2.4 −1.9

−4 3.3 2.4 2.5 −1.6 −4.5 −7.4 −6.9 −12.8 3.1 2.1 2.4 −1.8 2.2 2.4 −1.9Bitterfeld −8 3.9 2.9 3 −1.3 −3.6 −6.8 −6.0 −11.7 3.7 2.7 2.4 −1.5 2.3 2.5 −1.3Poledno −4 2 1 1.1 −6 −6.4 −9.3 −9.5 −16.8 2 0.9 2.9 −7.6 0.2 0.2 −3.8Beckford −8 3.6 2 1.5 −2 −1.9 −5.6 −5.3 −9.2 3.5 1.9 0.9 −2.5 0.7 0.5 −1.4Nochten −8 3.3 2.4 2.5 −1.6 −4.5 −7.4 −6.9 −12.8 3.1 2.1 1.3 −1.8 2.2 2.4 −1.9

−8 3.3 2.4 2.5 −1.6 −4.5 −7.4 −6.9 −12.8 3.1 2.1 2.4 −1.8 2.2 2.4 −1.9

Model–palaeotemperature differences

Comparisoncriterion

Temperaturesin column 2

−20 °C (Table 1,column 2)

MAGT <0 °C TTOPtemperature <0 °C

Smallest difference 5.0 5.1 None 9 9.5 3.2 None None None None None NoneAverage difference 7.6 7.6 3.7 12.8 13.5 8.4 2.1 2.1 None 1.7 1.8 NoneLargest difference 10.9 11 6.7 14.4 15.0 10.8 2.7 2.9 None 2.5 2.6 None



Table 4 Dated permafrost degradation features recorded in Stage 3 successions

Site name Longitude Latitude Approximateage (cal. kyr)

Event References

Amersfoort 5.38 52.15 <39.1 Large cryoturbations Van der Hammen et al.,1967

Hengelo 6.78 52.23 40.6–42.1 Large cryoturbations Zagwijn, 1974Hengelo 6.8 52.28 41.4–47.0 Large cryoturbations Zagwijn, 1974; Van

Huissteden, 1990Hengelo A1 6.82 52.27 ca. 37 Large cryoturbations Van Huissteden, 1990Hengelo A1 6.82 52.27 44–43(38) Thaw lake Van Huissteden, 1990Scheibe 14.32 51.45 ca. 33.3, ca. 35.1, Large cryoturbations Mol, 1997

ca. 38.1, >47Grouw 5.83 53.07 41.1–39.7, Large cryoturbations Kasse et al., 1995

43–41, 49–43Nochten 14.34 51.27 38–39 Large cryoturbations Bos et al., 2001

temperature discrepancy is only 2–3 °C, within the error rangeof the climate model.

We applied the TTOP model of Smith and Riseborough(2002) to the climate model output to integrate the effects ofMAAT, snow cover and vegetation, with thermal propertiesfor mineral soils as soil physical parameters. A negativetemperature at the top of the permafrost modelled by TTOPindicates continuous permafrost. Input parameters for theTTOP model are the thawing index (number of degree-days above 0 °C), the MAAT and the average snow depth,derived from the climate model output. Vegetation is includedby a scaling factor for the difference between summer airand ground surface temperature, set at a value of 0.9 foropen, treeless vegetation. The resulting continuous permafrostboundary follows the 0 °C MAGT isotherm closely (Fig. 2).Again, for the LGM simulation all wedge locations are withinthe area where continuous permafrost occurred according theclimate model. The TTOP temperatures calculated from theST3Cold simulations are up to 3 °C too warm (Table 3). TheTTOP model supports our approach of taking the 0 °C MAGTisotherm as the continuous permafrost limit for mineral soilsunder the modelled Stage 3 thin snow cover conditions.

Discussion

The discrepancies between the ice-wedge-based palaeotem-perature and the climate model results are not restricted to ourStage 3 simulations. Similar discrepancies have been foundby Isarin and Renssen (1997, 1999) in Younger Dryas climatesimulations and Kageyama et al. (2001) in LGM simulations.Kageyama et al. (2001) compared outputs of 17 models withclimate reconstructions from palaeobotanical data. Between40° and 47 °N latitude the models produce temperatures ofthe coldest month that are approximately 10 °C too warm. Ineastern Europe, the discrepancies are smaller and in westernSiberia the models produce temperatures that agree with thereconstructions. Isarin and Renssen (1997, 1999) also reporttemperature discrepancies ranging from of ≥15 °C in Irelandto ≤5 °C in Finland, between an AGCM model simulationand reconstructed MTCM based on periglacial features andcoleopteran data. These discrepancies are also based on thetemperature limits listed in column 2 in Table 1. It is clearthat many climate model simulations of glacial cold stagestend to overestimate winter temperatures in western Europe.The authors list several possible causes of the discrepancies,related to model structure or boundary conditions. In partic-ular, the SSTs of the northern Atlantic Ocean are considered

as inadequate because of the east–west gradient in the wintertemperature deviations (Renssen et al., 2000; Kageyama et al.,2001).

Although we attempt here to reduce the discrepancybetween the model and the data by stretching the interpretationof the data to its limits, this still does not bridge the gapentirely for the ST3Cold simulations. Pollard and Barron (2003)and Alfano et al. (2003) evaluate the causes of mismatchesbetween the Stage 3 climate model results extensively from theperspective of model boundary conditions and structure, andsuch an evaluation is outside the scope of this paper. However,we suggest the following aspects, based on the periglacialenvironment we studied. First, the vegetation (boreal forest)derived from the BIOME model deviates from the treelessvegetation indicated by palaeobotanical data (e.g. Ran, 1990).The low winter albedo of boreal forest may cause 4° –5 °Chigher winter temperatures in climate model simulations (e.g.Bonan et al., 1992; Foley et al., 1994). If the modelled borealforest had been forced to tundra or steppe vegetation innorthern Europe, this could have reduced the mismatch forthe ST3Cold simulation. Second, existing permafrost results inlower summer temperatures owing to the latent heat losses byenhanced evaporation from moist soils (Renssen et al., 2000).Third, the model results indicate strong winter winds thatshould have affected snow cover and induced wind stress onthe vegetation (Kolstrup and Wijmstra, 1977). A synergy mayhave existed between these vegetation, permafrost and windeffects, resulting in lower summer and winter temperaturesand preventing forest migration.

A very important factor is the time-scale represented by ourpalaeodata. Recent field research indicates that ice wedgesare initiated in a very short time and thus may representshort-lived climatic extremes. Most Stage 3 wedges are rathersmall compared with ice-wedge casts of Stage 2 age at the samesites (Fig. 1), indicating a short time of development. Initial ice-wedge growth is fastest, in particular on freshly exposed sitessuch as drained lake bottoms (Mackay, 1993, 1999; Mackayand Burn, 2002). Mackay and Burn (2002) report ice-wedgeinitiation within 12 yr, with rapid initial ice-wedge growth ofup to 3 cm yr−1 at a newly exposed site. Afterwards, ice-wedgegrowth may decrease or stop when developing vegetation startstrapping more snow in winter (Mackay, 1999). Similar ratesmay have applied to the dynamic alluvial plain environmentsin which most of the Stage 3 wedges in Table 2 have beendeveloped. In these environments, freshly exposed bars, leveesand crevasse splays should have occurred abundantly and havebeen the preferential sites for ice-wedge development (VanHuissteden, 1990; Mol, 1997). Hence, the Stage 3 ice-wedgesmay represent short periods of extreme climate, in the order of



one to a few decades, in combination with favourable local siteconditions. This time range is below the temporal resolution ofthe marine palaeoclimate records on which the climate modelboundary conditions (in particular sea-surface temperatures)are based. These boundary conditions represent longer termaverages, resulting in a failure of the climate model to simulatethese extreme climates. Because of the better agreement of theLGM simulation with the palaeotemperatures derived from theice wedges, the climatic conditions during these short coldintervals should have been close to those prevailing during theLast Glacial Maximum.

Conclusion

Based on conventional palaeotemperature estimates from icewedges the modelled mean annual air temperatures of theST3Cold climate simulation are up to 11 °C too warm inEurope between 50° and 56 °N latitude. The LGM experimentdeviates 2–6.7 °C. The discrepancies are largely attributableto a too high temperature of the coldest month (up to15 °Ctoo high for the ST3Cold experiments and up to 11 °C forthe LGM experiment). The winter temperature deviations arelargest in western Europe. For the ST3Warm simulation thelimited amount of data suggest an at least 3 °N shift of thepermafrost boundary, which is in agreement with the modelledtemperatures.

This study shows that for palaeoclimatic interpretation ofperiglacial features, as with other palaeoclimatic proxies:(1) the interaction of different climatic and environmentalfactors must be taken into account (2) the effects of climaticvariability should not be ignored and (3) the time-scalerepresented by the proxies is important.

1 The large mismatches above do not account for uncertaintyin the palaeotemperature interpretation introduced by siteand climate characteristics. Although still hypothetical,compared with present-day Arctic climates the glacialclimates may have been favourable to ice-wedge growth:(i) a reduced snow cover thickness by lower precipitation,strong winds and more melt–refreeze cycles and (ii) a strongsynoptic variability enhancing the incidence of rapid soiltemperature drops. Based on modelling of the temperaturesof the top of the permafrost (TTOP model) we developedupper temperature limits for ice-wedge growth applying toextremely favourable conditions: a MAGT of <0 °C andMTCM of < −15 °C. Using these limits, the discrepancybetween modelled and inferred MAGT or the TTOP modeltemperatures are <3 °C for the ST3Cold experiments andnone for the LGM experiment. The modelled MTCMdiscrepancy is up to 10 °C for the ST3Cold experimentsand up to 6 °C for the LGM experiment.Local site conditions also may have been favourable for ice-wedge growth. Most wedges have been recorded in fluvialsuccessions. Ice-wedge formation occurs rapidly in newlyexposed sites, and in the dynamic fluvial environment thesesites may have been abundant.

2 Climatic variability, both synoptic and year-to-year variabil-ity, may strongly influence palaeotemperature estimation.We have demonstrated its potential influence, based onstochastic modelling of the effect of synoptic variabilityon ice-wedge growth. More knowledge is required of theinfluence of these types of variability.

3 A further explanation for the mismatches between data andmodel is the time-scale over which palaeoclimatic proxies

average temperature estimates. The Stage 3 ice-wedges mayhave formed in short periods of a few decades with extremecold climate, with characteristics close to those of the LGMclimate, well below the ‘average’ Stage 3 stadial conditionsmodelled by the ST3Cold simulations.From the modeling side, the mismatches between dataand model may be explained by the effects of thevegetation generated by BIOME (boreal forest versusopen vegetation) and the influence of permafrost onsoil–atmosphere interactions.

Acknowledgements This paper relies on the work of Bert Huijzerand Rene Isarin, who compiled the Multi-proxy DataBase. We thankProfessor T. van Andel and his co-workers for the organisation of theStage 3 Project with its lively, stimulating meetings in Cambridge.We acknowledge Hans Renssen for reviewing an earlier version ofthis manuscript. Professor H. M. French and Professor A. Pissart arethanked for their constructive reviews.

References

Alfano MJ, Barron EJ, Pollard D, Huntley B, Allen J. 2003. Comparisonof climate model results with European vegetation and permafrostduring Oxygen Isotope Stage Three. Quaternary Research 59:97–107.

Arnold N, Van Andel TH, Valen V. 2002. Extent and dynamicsof the Scandinavian ice sheet during Oxygen Isotope Stage 3(65,000–25,000 cal years B.P.). Quaternary Research 57: 37–47.

Bard E, Arnold M, Hamelin B, Tisnerat-Laborde N, Cabioch G. 1998.Radiocarbon calibration by means of mass spectrometric 230Th/234Uand 14C ages of corals: an updated database including samples fromBarbados, Mururoa and Tahiti. Radiocarbon 40: 1085–1092.

Barron EJ, Pollard D. 2002. High-resolution climate simulations ofOxygen Isotope Stage 3 in Europe. Quaternary Research 58:296–309.

Bonan GB, Pollard D, Thompson SL. 1992. Effects of boreal forestvegetation on global climate. Nature 359: 716–718.

Bos JAA, Bohncke SJP, Kasse C, Vandenberghe J. 2001. Vegetationand climate during the Weichselian Early Glacial and Pleniglacialin the Niederlausitz, eastern Germany—macrofossil and pollenevidence. Journal of Quaternary Science 16: 269–289.

Briggs DJ, Coope GR, Gilbertson DD. 1975. Late Pleistocene terracedeposits at Beckford, Worcestershire, England. Geological Journal10: 1–16.

Brown RJE. 1973. Influence of climatic and terrain factors on groundtemperatures at three locations in the permafrost region of Canada.In Permafrost; North American Contribution, Second InternationalPermafrost Conference, Yakutsk, USSR. Publication 2115, NationalAcademy of Science: Washington, DC; 27–34.

Burn CR. 1990. Implications for palaeoenvironmental reconstructionof recent ice-wedge development at Mayo, Yukon Territory.Permafrost and Periglacial Processes 1: 3–14.

Burn CR, Smith MW. 1988. Thermokarst lakes at Mayo, YukonTerritory, Canada. Proceedings Vth International Conference onPermafrost, Trondheim, Norway; 700–705.

CLIMAP (Climate: Long-Range Investigation, Mapping and PredictionProject) Members. 1981. Seasonal Reconstructions of the Earth’sSurface at the Last Glacial Maximum. Map Chart Series, MC-36,Geological Society of America: Boulder, CO.

COHMAP Members. 1988. Climatic changes of the last 18,000 years:observations and model simulations. Science 241: 1043–1052.

Coope GR. 2002. Changes in the thermal climate in NorthwesternEurope during Marine Oxygen Isotope Stage 3, estimated from fossilinsect assemblages. Quaternary Research 57: 401–408.

DeConto RM, Thompson SL, Pollard D. 2000. Recent advancesin paleoclimate modeling: toward better simulations of warmpaleoclimates. In Warm Climates in Earth History, Huber BT,MacLeod KG, Wing SL (eds). Cambridge University Press:Cambridge; 21–49.



Drozdowski E, Fedorowicz S. 1987. Stratigraphy of Vistulianglaciogenic deposits and corresponding thermoluminescence datesin the lower Vistula region, northern Poland. Boreas 16: 139–153.

Elias SA, Andrews JT, Anderson KH. 1999. Insights on the climaticconstraints on the beetle fauna of coastal Alaska, U.S.A.,derived from the Mutual Climatic Range Method of palaeoclimatereconstruction. Arctic, Antarctic and Alpine Research 31: 94–98.

Foley JA, Kutzbach JE, Coe MT, Levis S. 1994. Feedbacks betweenclimate and boreal forests during the Holocene epoch. Nature 371:52–54.

Foster J, Liston G, Koster R, Essery R, Behr H, Dumenil L, Verseghy D,Thompson S, Pollard D, Cohen J. 1996. Snow cover and snow massintercomparisons of general circulation models and remotely senseddatasets. Journal of Climate 9: 409–426.

French HM. 1996. The Periglacial Environment. Longman: Harlow.Giorgi F, Marinucci MR, Bates GT. 1993a. Development of a second-

generation regional climate model (RegCM2). Part I: Boundary-layerand radiative transfer processes. Monthly Weather Review 121:2794–2813.

Giorgi F, Marinucci MR, De Canio G, Bates GT. 1993b. Developmentof a second-generation regional climate model (RegCM2). PartII: convective processes and assimilation of lateral boundaryconditions. Monthly Weather Review 121: 2814–2832.

GRIP Members 1993. Climate instability during the last interglacialperiod recorded in the GRIP ice core. Nature 364: 203–207.

Harry DG, Gozdzik JS. 1988. Ice wedges: growth, thaw transformationand palaeoenvironmental significance. Journal of QuaternaryScience 3: 39–55.

Haxeltine A, Prentice IC. 1996. BIOME3: an equilibrium terrestrialbiosphere model based on ecophysiological constraints, resourceavailability, and competition among plant functional types. GlobalBiogeochemical Cycles 10: 693–709.

Hiller A, Litt T, Eissmann L. 1991. Zur Entwicklung desJungquartaren Tieflandstaler im Saale-Elbe-Raum unter besondererBerucksichtigung von C-14 daten. Eiszeitalter und Gegenwart 41:26–46.

Hostetler SW, Giorgi F, Bates GT, Bartlein PJ. 1994. Lake–atmospherefeedbacks associated with paleolakes Bonneville and Lahontan.Science 263: 665–668.

Hublin J-J. 1998. Climatic changes, palaeogeography and theevolution of the Neandertals. In Neandertals and Modern Humansin Western Asia, Akazawa T, Aoki K, Bar-Yosef O (eds). PlenumPress: London; 295–310.

Huijzer AS. 1993. Cryogenic macrofabrics and macrostructures:interrelations, processes, and environmental significance. Thesis,Vrije Universiteit, Amsterdam.

Huijzer AS, Isarin R. 1997. The multi-proxy approach to thereconstruction of past climates with an example of the WeichselianPleniglacial in northwestern and central Europe. Quaternary ScienceReviews 16: 513–533.

Huijzer AS, Vandenberghe J. 1998. Climatic reconstruction of theWeichselian Pleniglacial in northwestern and central Europe.Journal of Quaternary Science 13: 391–417.

Isarin RFB, Renssen H. 1997. Surface temperatures in north-westernEurope during the Younger Dryas: AGCM simulation comparedwith temperature reconstructions. In The climate in north-westernEurope during the Younger Dryas. A comparison of multi-proxyclimate reconstructions with simulation experiments. Isarin, RFB.Thesis, Vrije Universiteit. Amsterdam; 75–92.

Isarin RFB, Renssen H. 1999. Reconstructing and modelling LateWeichselian climates: the Younger Dryas in Europe as a case study.Earth Science Reviews 48: 1–38.

Isarin RFB, Huijzer AS, Van Huissteden J. 1998. Time-slice orientedmulti-proxy database (MPDB) for palaeoclimate reconstruction.In CAPS (Circumpolar Active-Layer Permafrost System) CD-ROM,Version 1.0, National Snow and Ice Data Center. CIRES, Universityof Colorado.

Joris O, Weninger B. 2000. Radiocarbon calibration and theabsolute chronology of the Late Glacial. In L’Europe Centrale etSeptentrionale au Tardiglaciaire (Table-ronde de Nemours, 13–16mai 1997). Memoires du Musee de Prehistoire d’Ile de France 7:19–59.

Karte J. 1983. Periglacial phenomena and their significance as climaticand edaphic indicators. Geo- Journal 7: 329–340.

Kageyama M, Valdes PJ, Ramstein G, Hewitt C, Wyputta U. 1999.Northern hemispheric storm tracks in present day and Last GlacialMaximum climate and simulations: a comparison of the EuropeanPMIP models. Journal of Climate 12: 742–760.

Kageyama M, Peyron O, Pinot S, Tarasov P, Guiot J, Joussaume S,Ramstein G. 2001. The Last Glacial Maximum climate over Europeand western Siberia: a PMIP comparison between models and data.Climate Dynamics 17: 23–43.

Kasse C, Bohncke SJP, Vandenberghe J. 1995. Fluvial periglacialenvironments, climate and vegetation during the MiddleWeichselian in the Northern Netherlands with special referenceto the Hengelo Interstadial. Mededelingen Rijks Geologische Dienst52: 387–414.

Kitegawa H, Van der Plicht J. 1998. A 40,000 year varve chronologyfrom lake Suigetsu, Japan: extension of the 14C calibration curve.Radiocarbon 40: 505–515.

Kolstrup E, Mejdahl V. 1986. Three frost wedge casts from Jutland(Denmark) and TL-dating of their infill. Boreas 15: 311–321.

Kolstrup E, Wijmstra TA. 1977. A palynological investigation ofthe Moershoofd, Hengelo and Denekamp Interstadials in theNetherlands. Geologie en Mijnbouw 56: 85–102.

Lachenbruch AH. 1962. Mechanics of thermal contraction cracks andice-wedge polygons in permafrost. Geological Society of America,Special Paper 70: 69 pp.

Lu H, Van Huissteden J, Zhou J, Vandenberghe J, Liu X, An Z. 2000.Variability of the East Asian winter monsoon in Quaternary climaticextremes in North China. Quaternary Research 54: 321–327.

Maarleveld GC. 1976. Periglacial phenomena and the mean annualtemperature during the last glacial time in the Netherlands. BiuletynPeryglacjalny 26: 57–78.

Mackay JR. 1974. Ice-wedge cracks, Garry Island, NorthwestTerritories. Canadian Journal of Earth Sciences 11: 1366–1383.

Mackay JR. 1991. The frequency of ice-wedge cracking (1967–1987)at Garry Island, western Arctic coast, Canada. Canadian Journal ofEarth Sciences 29: 236–248.

Mackay JR. 1993. Air temperature, snow cover, creep of frozenground, and the time of ice-wedge cracking, western Arctic Coast.Canadian Journal of Earth Sciences 30: 1720–1729.

Mackay JR. 1999. Periglacial features developed on the exposed lakebottoms of seven lakes that drained rapidly after 1950, Tuktoyaktukpeninsula area, Western arctic coast, Canada. Permafrost andPeriglacial Processes 10: 39–63.

Mackay JR. 2000. Thermally induced movements in ice-wedgepolygons, western Arctic coast: a long-term study. Geographiephysique et Quaternaire 54: 41–68.

Mackay JR, Burn CR. 2002. The first 20 years (1978–1979 to1998–1999) of ice-wedge growth at the Illisarvik experimentaldrained lake site, western Arctic coast, Canada. Canadian Journalof Earth Sciences 39: 95–111.

Manabe S, Broccoli AJ. 1985. A comparison of climate modelsensitivity with data from the Last Glacial Maximum. Journal ofAtmospheric Science 42: 2643–2651.

Mania D, Toepfer V. 1973. Konigsaue. Gliederung, Okologie undmittelpalaolithische Funde der letzten Eiszeit. Veroffentlichungendes Landesmuseums fur Vorgeschichte in Halle 26: 1–165.

Manikowska B. 1993. Mineralogy and abrasion of sand grains dueto Vistulian (Late Pleistocene) aeolian processes in central Poland.Geologie en Mijnbouw 72: 167–177.

McManus JF, Oppo DW, Cullen JL. 1999. A 0.5-million-year recordof millennial-scale climate variability in the North Atlantic. Science283: 971–975.

Melnikov VP, Spesivtsev VI. 2000. Cryogenic Formations in the Earth’sLithosphere, Konischev VN (ed.). Scientific Publishing Center of theUIGGM, Publishing House, Siberian Branch of the Russian Academyof Sciences: Novosibirsk; 343 pp.

Mol JA. 1997. Fluvial response to climate variations. The LastGlaciation in Eastern Germany. Thesis Vrije Universiteit,Amsterdam.

Morgan V. 1973. Pleistocene geology around Wolverhampton.Philosophical Transactions of the Royal Society, Series B London,265: 233–297.

Murton JB, French HM. 1995. Thermokarst involutions, SummerIsland, Pleistocene McKenzie delta, Western Canadian Arctic.Permafrost and Periglacial Processes 4: 217–229.



Murton JB, Whiteman CA, Allen P. 1995. Involutions in the MiddlePleistocene (Anglian) Barham Soil, eastern England: a comparisonwith thermokarst involutions from Arctic Canada. Boreas 24:269–280.

Pewe TL. 1966a. Paleoclimatic significance of fossil ice wedges.Biuletyn Periglacjalny 15: 65–73.

Pewe TL. 1966b. Ice wedges in Alaska—classification, distributionand climatic significance. In Proceedings, 1st InternationalConference on Permafrost. Publication 1287, National Academyof Science, National Research Council of Canada: Ottawa; 76–81.

Pollard D, Barron EJ. 2003. Causes of model-data discrepancies inEuropean climate during oxygen isotope Stage 3 with insights fromthe Last Glacial Maximum. Quaternary Research 59: 108–113.

Pollard D, Thompson SL. 1997. Climate and ice-sheet mass balance atthe last glacial maximum from the GENESIS version 2 global climatemodel. Quaternary Science Reviews 16: 841–864.

Pollard D, Bergengren JC, Stillwell-Soller LM, Felzer B, Thompson SL.1998. Climate simulations for 10 000 and 6000 years BP usingthe GENESIS global climate model. Palaeoclimates—Data andModelling 2: 183–218.

Ran ETH. 1990. Dynamics of vegetation and environment duringthe Middle Pleniglacial in the Dinkel Valley (The Netherlands).Mededelingen Rijks Geologische Dienst 44: 141–205.

Ran ETH, Bohncke SJP, Van Huissteden J, Vandenberghe J. 1990.Evidence of episodic permafrost conditions during the WeichselianMiddle Pleniglacial in the Hengelo basin (The Netherlands).Geologie en Mijnbouw 44: 207–220.

Renssen H, Isarin RFB. 2001. The two major warming phases of thelast deglaciation at ∼14.7 and ∼11.5 ka cal BP in Europe: climatereconstructions and AGCM experiments. Global and PlanetaryChange 30: 117–153.

Renssen H, Isarin RFB, Vandenberghe J, Lautenschlager M, Schlese U.2000. Permafrost as a critical factor in paleoclimate modelling; theYounger Dryas case in Europe. Earth and Planetary Science Letters176: 1–5.

Renssen H, Bogaart PW. 2003. Atmospheric variability over the∼14.7 kyr BP stadial–interstadial transition in the North Atlanticregion as simulated by an AGCM. Climate Dynamics 20: 301–313.

Romanovskij NN. 1985. Distribution of recently active ice and soilwedges in the USSR. In Field and Theory: Lectures in Geocryology,Church M, Slaymaker O (eds). University of British Columbia Press:Vancouver; 154–165.

Rotnicki K, Tobolski K. 1969. Stanowisko Interstadiału Paudorfw Kepnie (połodnowo - zachodnia Wielkopolska). BadaniaFizjograficzne nad Polska Zachodnia 23: 119–127.

Smith MW, Riseborough DW. 2002. Climate and the limit ofpermafrost: a zonal analysis. Permafrost and Periglacial Processes13: 1–15.

Thompson SL, Pollard D. 1997. Greenland and Antarctic massbalances for present and doubled CO2 from the GENESIS version-2global climate model. Journal of Climate 10: 871–900.

Van Andel TH. 2002. The climate and landscape of the middlepart of the Weichselian Glaciation in Europe: the Stage 3 Project.Quaternary Research 57: 2–8.

Van Andel TH, Tzedakis PC. 1996. Palaeolithic landscape of Europeand environs, 150,000–25,000 years ago: an overview. QuaternaryScience Reviews 15: 481–500.

Vandenberghe J. 1983. Ice-wedge casts and involutions as permafrostindicators and their stratigraphic position in the Weichselian.Proceedings IVth International Conference on Permafrost, Fairbanks;1298–1302.

Vandenberghe J. 1985. Paleoenvironment and stratigraphy duringthe Last Glacial in the Belgian-Dutch border region. QuaternaryResearch 24: 23–38.

Vandenberghe J. 1988. Cryoturbations. In Advances in PeriglacialGeomorphology, Clark MJ (ed.). Wiley: Chichester; 179–200.

Vandenberghe J, Pissart A. 1993. Permafrost changes in Europe duringthe last glacial. Permafrost and Periglacial Processes 4: 121–135.

Vandenberghe J, Van den Broek P. 1982. Weichselian convolutionphenomena and processes in fine sediments. Boreas 11: 299–315.

Van der Hammen T, Maarleveld GC, Vogel JC, Zagwijn WH. 1967.Stratigraphy, climatic succession and radiocarbon dating of the lastglacial in the Netherlands. Geologie en Mijnbouw 46: 79–95.

Van Huissteden J. 1990. Tundra rivers of the Last Glacial:sedimentation and geomorphological processes during theMiddle Pleniglacial in the Dinkel valley (eastern Netherlands).Mededelingen Rijks Geologische Dienst 44(3): 3–138.

Van Vliet-Lanoe B. 1988. The genesis of cryoturbations and theirsignificance in environmental reconstruction. Journal of QuaternaryScience 3: 85–96.

Voelker AHL, Sarnthein M, Grootes PM, Erlenkeuser H, Laj C,Mazaud A, Nadeau M-J, Schleicher M. 1998. Correlation ofmarine 14C ages from the Nordic seas with the GISP2 isotoperecord: implications for radiocarbon calibration beyond 25 ka BP.Radiocarbon 40: 517–534.

Yershov ED. 1998. General Geocryology. Cambridge University Press:Cambridge.

Zagwijn WH. 1974. Vegetation, climate and radiocarbon datings inthe Late Pleistocene of the Netherlands, Part II: Middle Weichselian.Mededelingen Rijks Geologische Dienst N.S. 25(3): 101–110.

Zagwijn WH, Paepe R. 1968. Die Stratigraphie der weichselzeitlichenAblagerungen der Niederlande und Belgiens. Eiszeitalter undGegenwart 19: 129–146.


Palaeotemperature reconstructions of the European permafrost zone during Oxygen Isotope Stage 3 compared with climate model results

Documents