New Estimates of Permafrost Evolution during the Last 21k Years in Eurasia

PERMAFROST AND PERIGLACIAL PROCESSESPermafrost and Periglac. Process. (2013)Published online in Wiley Online Library(wileyonlinelibrary.com) DOI: 10.1002/ppp.1787

New Estimates of Permafrost Evolution during the Last 21 k Years in Eurasia usingNumerical Modelling

D. C. Kitover,* R. T. van Balen, D. M. Roche, J. Vandenberghe and H. Renssen

Department of Earth Sciences, VU Universiteit Amsterdam, Amsterdam, The Netherlands

* CoUnivThe

Copy

ABSTRACT

The evolution of past permafrost since the Last Glacial Maximum (LGM) is simulated using the Vrije UniversiteitAmsterdam Permafrost (VAMPER) model. This method is different from a proxy-based approach which translatesreconstructed air temperatures to estimate past permafrost extent and thickness. First, a sensitivity analysis wasperformed to assess the behaviour of the model. Then five case studies within Eurasia were performed using meanannual ground surface temperatures derived from an Earth system model as the surface forcing. In Central and WestSiberia, the simulated LGM permafrost thicknesses of 730–940m and 365–445m, respectively, agree well withprevious estimates. The LGM and present-day estimates for South Russia (9–15m) are underestimated, which islikely due to a highly simplified land-atmosphere coupling. In West and Central Europe, however, the VAMPERmodel was not able to produce permafrost during LGM conditions, which is due to previously recognised biasesof the Earth system model. A supplementary simulation was then performed, resulting in an LGM permafrost thick-ness estimate of 260–320m. Average thawing rates are on the order of 1 to 3 cm/yr except for Central Siberia, wherepermafrost thawed at rates of 0.3 to 0.4 cm/yr. Overall results of these simulations provide a basis for future improve-ment in modelling the permafrost-climate relationship over millennia. Copyright © 2013 John Wiley & Sons, Ltd.

KEY WORDS: climate model; deglaciation; LGM; permafrost

INTRODUCTION

Cryostratigraphic data such as pingo remnants, ice-wedgepseudomorphs, large cryoturbations and sand wedges indi-cate that during the Last Glacial Maximum (LGM, 21 kaBP) permafrost in Eurasia extended southwards in frontof the Fennoscandian Ice Sheet to approximately 50°N(Vandenberghe et al., 2012; Figure 1). These areas includesouthern Siberia (Baulin et al., 1984), the Russian Plain(Kondratjeva et al., 1993), central Europe (Gozdzik andFrench, 2004; Kovács et al., 2007; Ewertowski, 2009)and Western Europe (Van Vliet-Lanoë, 1989; Huijzerand Vandenberghe, 1998). However, reconstructing thedevelopment of permafrost through time is difficult be-cause estimates of past permafrost extent are made empir-ically using present-day analogues, which assume a simplerelationship between climate parameters and permafrostfeatures such as thermal-contraction-crack polygons andpingos (Vandenberghe and Pissart, 1993; French, 2007;

rrespondence to: D. Kitover, Department of Earth Sciences, VUersiteit Amsterdam, De Boelelaan 1085, NL-1081 HV, Amsterdam,Netherlands. E-mail: [email protected]

right © 2013 John Wiley & Sons, Ltd.

Matsuoka, 2011). These indicators, however, only formunder certain geological and climatic conditions (Murtonand Kolstrup, 2003; French, 2007), adding uncertainty toreconstructing distributions of past permafrost. For exam-ple, thermal contraction cracking occurs at higher tempera-tures in a fine-grained soil than in a sandy soil(Romanovskii, 1985; Williams and Smith, 1989), and icewedges form not necessarily as a result of meeting a certainair temperature condition but rather as a function of the rateof cooling (Lachenbruch, 1962). In addition, local surfacefactors such as snow cover, surface organic content, vege-tation and topography not only influence permafrost devel-opment but also the ‘self-destroying nature’ of thermokarst(French, 2007, p. 302).

Given these limitations, we propose that physicallyconsistent and meaningful estimates of past permafrostthickness can be achieved by numerical modelling. Here,we provide the first estimates of the evolution of permafrostthickness for the last 21 k years at selected locations inEurasia. Using the Vrije Universiteit Amsterdam Permafrost(VAMPER) model, permafrost response to climate warmingduring the last deglaciation is simulated using surfacetemperature anomaly forcings provided by LOVECLIM, a

Received 6 September 2012Revised 24 June 2013Accepted 3 July 2013

Figure 1 Map of Eurasia showing VAMPER model simulation areas (bolded cells) with their associated climate stations (red star), modern extent ofpermafrost (Brown et al., 1997), LGM (21 ka BP) reconstructed permafrost boundaries (Vandenberghe et al., 2012), LOVECLIM grid in light grey andheat flow measurements (Jessop et al., 1976).See text for abbreviations.This figure is available in colour online at wileyonlinelibrary.com/journal/ppp

D. C. Kitover et al.

three-dimensional Earth system Model of IntermediateComplexity (EMIC). An EMIC is generally defined as atype of climate model designed to be partially comprehen-sive and partially conceptual, where certain interactionsare parameterised depending on the research question(Claussen et al., 2002; Weber, 2010). The VAMPER modelwas designed to examine millennial-scale permafrost be-haviour (Kitover et al., 2012), while LOVECLIM is suitedfor studying past climates and provides long temperatureseries (Goosse et al., 2010). This approach to modellingpermafrost evolution over the last deglaciation differs in anumber of ways from previous studies: (1) we use meanannual ground surface temperatures (MAGST) derived froma well-established EMIC to generate forcing data rather thanusing reconstructed palaeo-values; (2) we employ anefficient subsurface parameterisation that is simple enoughnot to overburden the model with site-specific details butstill represents freeze-thaw processes over millennia; and(3) we apply an averaged geothermal flux based on mea-sured values. Our study is not intended to produce highlyaccurate simulations of past permafrost but rather todescribe somewhat conceptual permafrost conditions; sinceboth the VAMPER model and LOVECLIM include alimited number of adjustable parameters, which in turn allowsimulation over geological timescales and broad spatial scales.This study aims to constrain when and how much perma-

frost has degraded since the LGM in certain parts of Eurasia.These estimates, in turn, can improve the simulation of past

Copyright © 2013 John Wiley & Sons, Ltd.

ground surface temperatures, ground heat flux and frozenorganic carbon storage, allowing a better assessment of thefeedback mechanisms between permafrost and climate(Renssen et al., 2000; Zimov et al., 2006).

Numerous permafrost models that are not specific togeological timescales have been either integrated into landsurface models or employed as stand-alone versions. Theformer include the Terrestrial Ecosystem Model (Zhuanget al., 2001), the Simple Biosphere Model (Li and Koike,2003; Schaefer et al., 2011) and the Northern EcosystemSoil Temperature Model (Zhang et al., 2003). In fact, theProject for Intercomparison of Land-Surface Parameteriza-tion Schemes phase 2(d) (Luo et al., 2003) has consideredthe benefit of including freezing soil in land surfaceschemes. This assessment included 21 different land surfaceschemes and found that soil temperatures are more realisti-cally modelled due to the damping effect of latent heat onseasonal temperature changes. In terms of the stand-alonedynamical permafrost models (e.g. Kane et al., 1991; Zhanget al., 1997; Oelke et al., 2003; Ling and Zhang, 2004;Marchenko et al., 2008), most were designed for closeexamination of near-surface and active-layer processes overrelatively short time periods while using small time steps(daily or sub-daily), unlike the VAMPER model

Although there are numerous modelled future projections ofpermafrost extent (Anisimov and Nelson, 1996; Malevsky-Malevich et al., 2001; Lawrence and Slater, 2005; Cheng andWu, 2007; Zhang et al., 2008a), far less work has simulated

Permafrost and Periglac. Process., (2013)

New Estimates of Permafrost Evolution during the Last 21 k Years

changes in permafrost over millennia. Delisle (1998) andDelisle et al. (2003) used a numerical method to examine per-mafrost growth and decay during theWeichselian. Osterkampand Gosink (1991) and Lunardini (1995) used reconstructedpalaeotemperatures of the last 120 k years for modellingpermafrost dynamics on the North Slope, Alaska. Lebretet al. (1994) simulated permafrost thicknesses for the last120 k years for regions in France. Kukkonen and Šafanda(2001) modelled permafrost thickness variation in the north-ern Fennoscandian region as a response to the major climaticshifts during the Holocene. Also in this region, Hartikainen(2006) produced a one-dimensional numerical permafrostmodel for the next 200 000 years, and Hartikainen et al.(2010) reported on their two-dimensional model, whichdemonstrated specific enhancements such as water bodiesand lateral subsurface heat flow. Using temperature histories(via borehole temperature inversion techniques), Rath andMottaghy (2007) applied reconstructed temperatures to theirforward numerical model for case studies in Russia and Po-land. In a newer methodology, Levavasseur et al. (2011)employed a statistical downscaling technique to achieve afiner resolution of estimated permafrost extent during theLGM. More recently, Vandenberghe et al. (2012) estimatedLGM and future permafrost distribution using LOVECLIM-modelled mean annual air temperatures (MAAT).In the following sections, we first describe the VAMPER

model methods, including a summary of the generalassumptions and limitations. Then results are shown for sensi-tivity testing of the different parameters within the VAMPERmodel setup. This is followed by application of the VAMPERmodel to five case study areas in Eurasia, which were selectedbased on previously published observations that indicate theexistence of past permafrost. In the Case Studies section, thecomputational setup of the experiments is described, followedby the results and subsequent discussion where comparisonsare made between the model results and estimates based ongeological reconstructions.

METHODS

VAMPER Model Description

Kitover et al. (2012) validated the VAMPER model by com-paring simulation results (under similar parameter settings)with other numerical heat conduction models that modelpermafrost thickness over millennial time periods. Given theunknowns and uncertainties in the setup of the publishedmodels, the results can be considered reasonable. Theseauthors also showed simulation results for a present-daythermal profile at Barrow, Alaska, where the modelled perma-frost thickness is close to reported values.In the VAMPER model, heat transfer is assumed to be

through conduction and is calculated using the one-dimensionalheat transport equation (Carslaw and Jaeger, 1959):

ρCp∂T∂t

¼ ∂∂x

Ke∂T∂x

� �(1)

where ρ is the density (kg/m3), Cp is the heat capacity


(J/m3K), T is the temperature (K) and Ke is the thermalconductivity (W/mK). The ‘e’ subscript indicates a computedeffective thermal conductivity (Equation 6), described below.The equation is solved numerically using the fully implicitmethod at an annual time step. This time step was selected af-ter experimenting with smaller time steps, which gave similarresults. The bottom boundary is set to a constant geothermalheat flux (Gfx) and the upper boundary is MAGST (Tsur).

Phase change is handled using the apparent heat capacitymethod (Williams and Smith, 1989; Zhang et al., 2008b),where the heat capacity is replaced by an effective heatcapacity meant to account for the additional heat released/absorbed during freezing/thawing. We follow a version ofthis method utilised by Mottaghy and Rath (2006), takenfrom Lunardini (1988). In their approach, the amount ofadded heat capacity is a function of the temperature-watercontent relationship where the phase change occurs over anarrow temperature range (w) between the thawing temper-ature (TL) and the freezing temperature (Tf). This means thatwhen the temperature is between TL and Tf, a proportionatemixture of ice and liquid water will be present.

Assuming saturated conditions, the unfrozen fraction(Θ) (expressed as a portion (0.0–1.0) of the porosity (n))is determined for each subsurface layer using the follow-ing equation (Mottaghy and Rath, 2006, taken fromLunardini, 1988):

Θ ¼ exp � T � TL

w

� �2" # !

; T < TL

1 T≥TL

8><>: (2)

The per cent of porosity which is unfrozen (nw) is then cal-culated as n ∙Θ and the remaining frozen portion (nf) is n - nw.

If T is above TL, Cp is determined using a weighted aver-age of the mass heat capacities of dry soil (Cm) and liquidwater (Cw):

Cp ¼ 1� nð ÞρmCm þ nwρwCw (3)

If T falls below TL, freezing/thawing occurs, Cap replacesCp and is calculated in Equation 4:

Cap ¼ 1� nð ÞρmCm þ nwρwCw þ nf ρf Cf þ nρwLdΘdT

(4)

where L is the latent heat of fusion (kJ/kg), Cap is theapparent heat capacity and the subscript f stands for ice.The last two terms represent the heat capacity of ice andthe latent heat released (absorbed) as each subsurfacelayer freezes (thaws).

Typically, dΘ/dT is determined empirically from soilfreezing curves, which vary based on the subsurface mate-rials (Williams and Smith, 1989; Davis, 2001). However,Lunardini (1988) and Mottaghy and Rath (2006) found thatby using the derivative of Equation 2, a smoother functioncan be applied:



dΘdT

¼ � 2 T � TLð Þw2

exp � T � TL

w

� �2" #

; T < TL

0 T≥TL

8><>:

(5)

In the VAMPER model, the thermal conductivity (Ke) istemperature- dependent since it is determined from eachlayer’s proportion of solid material and frozen/unfrozenwater. Any portion of air is considered negligible. Assum-ing a saturated subsurface material and using a geometricmean from Farouki (1981), the effective thermal conductiv-ity (Ke) is calculated as follows:

Ke ¼ K1�nm K

nff K

nww (6)

where Km, Kf and Kw are the thermal conductivities of soilminerals, frozen water and unfrozen water, respectively.Farouki (1981) reported that Equation 6 gives similar results

to the De Vries’method, a commonly used parameterisation inestimating thermal conductivity for frozen soil models (Zhanget al., 2008b). Use of the geometric mean to estimate thermalconductivity has been employed in studies such as those byLachenbruch et al. (1982) and Poutou et al. (2004).To calculate the thermal conductivity of the earth materials

(Km), Johansen (1975) proposed using a geometric mean:

Km ¼ KqqKo

1�q (7)

where Kq is the thermal conductivity of quartz (7.7W/mK)and Ko is the thermal conductivity of the other mineral types,which is assumed to be on average 2.0W/mK. For thisformula, the percentage of crystalline quartz (q) is assumedto be 80 per cent since we assume a sandy subsurface mate-rial, which is typically found to have a high quartz content(Peters-Lidard et al., 1998).The VAMPER model assumes porosity, n, decreases

with depth using an empirically based depth-porosityequation (e.g. Athy, 1930; Sclater and Christie, 1980):

n ¼ φe�0:000395d (8)

where d is the depth (m) and φ is the porosity of the toplayer. This equation assumes an equilibrium state of com-paction and is parameterised only for a subsurface describedas clayey sand. The surface porosity for this lithology is inthe range of 0.3 to 0.5.The spacing of the subsurface layers follows an efficient

approach where each layer thickness Δx increases logarith-mically with depth. A sensitivity analysis performed earlyin the model design supported this spacing scheme andshowed comparable results to the same model run but withdifferently discretised nodal spacing (Δx = 1m). Any differ-ence in the thermal profiles was small enough to be consid-ered negligible.


VAMPER Model Limitations and Assumptions

As mentioned above, the VAMPER model is forced byMAGST. This is different from the commonly referencedmean annual air surface temperature or simplyMAAT, whichrefers to the temperature slightly above the ground or landsurface. Smith and Riseborough (2002) provided a cleardepiction of air and ground surface temperatures, with thedifference between them designated as the ‘surface offset’.This offset, or the land-air coupling, is not represented inthe VAMPER model, which instead uses the MAGST. Thistemperature, as shown in Smith and Riseborough (2002), isbelow any influencing surface features such as snow, vegeta-tion, or water bodies. As such, the VAMPER model does notdynamically integrate surface features, most notably snowcover, into the ground temperature regime. Instead, it isassumed that these influences are reflected in the MAGST.

The VAMPER model was designed to be forced withvalues representative of large areal extents since it willeventually be coupled with LOVECLIM. Such a limita-tion is both an advantage – because it needs few site-specific details, thus making it widely applicable – anda disadvantage – because it cannot resolve local influ-ences. This may also be a limitation for the lower bound-ary of permafrost because continental heat flow can varydue to tectonic and volcanic activity.

Another limitation of the VAMPERmodel is that it does notconsider other heat transport mechanisms, particularly by waterflowing through unfrozen ground. Although subsurface heattransport in continuous permafrost regions is primarily conduc-tive (Lachenbruch and Marshall, 1969), heat transfer may bepartially convective (Kane et al., 2001). The latter effects arenear surface, mostly occurring in the organic layer, and stron-gest in the spring and fall periods, which are seasonal effectsnot relevant to the present work.

The VAMPER model treats the substrate as homogenous,except for the decreasing porosity with depth (Equation 8).This equation is applied assuming a sandy material, whereasother depth-porosity functions would be needed for other li-thologies. Although it would be interesting to apply and testadditional depth-porosity functions and their influence onthe thickness of permafrost, Equation 8 has been shown towork well in a previous calibration and validation of theVAMPER model (Kitover et al., 2012). With aheterogeneous substrate, more attention would be necessaryto examine the varying structure of the material and its asso-ciated hydraulic and thermal properties.

The thickness of permafrost is calculated as the totaldepth of subsurface layers that have an annual averagetemperature at or below �1 °C. If the traditional designa-tion of permafrost at or below 0 °C were to be used, itcould imply seasonal temperatures above 0 °C and assuch would not classify as perennially frozen. Phasetransition is modelled to occur over a temperature rangebetween complete liquid water (TL = 0 °C) and completeice (Tf =�2 °C). Therefore, -1 °C was considered a rea-sonable indicator of permafrost, so as not to overestimateor underestimate the permafrost thickness.


Figure 2 Comparison of permafrost thickness for values of surface po-rosity φ = 0.3, 0.4 and 0.5 after 100 k model years. The grey vertical linerepresents freezing at �1°C. The blue lines represent porosity asindicated by lower x-axis values. This figure is available in colour onlineat wileyonlinelibrary.com/journal/ppp


Finally, the use of Equation 5 assumes idealised subsur-face conditions: a homogenous saturated substrate. Thisdoes not account for the seasonal and diurnal variationswhich cause dynamic near-surface conditions and processessuch as an unsaturated zone and active-layer development.In addition, unfrozen water content within the active layerstrongly depends on the soil type, organic layer and thermalhistory of the soil and cannot be characterised by a two-de-gree temperature range as calculated by Equation 5.Although accurate simulation of the active layer is impor-tant at finer temporal and spatial scales, the active layermay play a less significant role in permafrost evolution overmillennia. In fact, during VAMPER model development, asensitivity analysis of the chosen time step showed thatthe difference between a monthly and annual time stepresulted in a 20-m difference in equilibrium permafrostdepth after 50k model years (MAGST -8 °C). This isrelatively small considering a larger sensitivity in theparameters analysed for the sensitivity analysis presentedin this work. However, a further assessment andimproved near-surface parameterisation would be neededto assess the role of the active layer in millennial-scalepermafrost dynamics.

Figure 3 Temperature-depth profile and porosity-depth profile of experimentswith and without the depth-porosity function (Equation 8) applied. The greyvertical line represents freezing at �1°C. See text for symbols.This figure isavailable in colour online at wileyonlinelibrary.com/journal/ppp

VAMPER Model Sensitivity

A sensitivity analysis was performed using the VAMPERmodel to describe the effect on simulated permafrost forma-tion caused by variations in surface porosity (φ), geothermalflux (Gfx), thermal conductivity (Ke) and MAGST. Modelresults are interpreted by examining the temperature profilethroughout the column and the thickness of permafrost after100 000 model years. Every run was started from a no-permafrost state. The base experiment, to which all othervariations were compared, was run with the following param-eters set: {φ=0.4, Gfx=60mW/m2, MAGST=�6 °C}. Alsoassumed in the base case was variable thermal conductivity(Equation 6) and porosity as a function of depth (Equation8). This run is always indicated by the black solid line inFigures 2–6.

Figure 4 Comparison of permafrost thickness after 100 k model years forconstant values of Ke = 2.0 W/mK and 3.0 W/mK (dashed black lines).Also shown is the profile using Equation 6 (solid black line). The greyvertical line represents freezing at �1°C. See text for symbol.This figureis available in colour online at wileyonlinelibrary.com/journal/ppp

PorosityOne major parameter used in the VAMPER model is

surface porosity, which, based on the assumptions ofEquation 8, may vary between 0.3 and 0.5. Therefore, thesurface porosity parameter (φ) was tested at values of 0.3,0.4 and 0.5 (Figure 2).After 100 k years, the thickest permafrost (300m) formed

at φ= 0.3 and the shallowest (250m) at φ= 0.5. In a satu-rated substrate, a higher porosity allows for greater watercontent, in turn requiring a greater latent heat exchange dur-ing phase transition. These results match well with those ofMottaghy and Rath (2006) and Lunardini (1995), who alsohave explained slower permafrost growth due to greater watercontent from higher porosity values. This effect, often referredto as the ‘zero curtain effect’, slows thermal diffusion rates.

Copyright © 2013 John Wiley & Sons, Ltd. Permafrost and Periglac. Process., (2013)

Figure 5 Comparison of permafrost thickness after 100 k model years fordifferent values of the geothermal heat flux (Gfx) as the lower boundary.The grey vertical line represents freezing at �1°C.

Figure 6 Comparison of permafrost thickness after 100 k model years fordifferent values of mean annual ground surface temperatures (MAGST).The grey vertical line represents freezing at �1°C.


A comparison run was made between applying thedepth-porosity relationship (applied via Equation 8) andkeeping porosity constant (n = 0.4) throughout the sub-strate (Figure 3). When the depth-porosity function isapplied, permafrost develops about 200m deeper thanwhen porosity remains constant with depth. This isbecause the decreased porosity, and hence the loweredsubsurface water content, lowers latent heat demand,allowing deeper propagation of the freezing front.

Thermal ConductivityAs explained above, the thermal conductivity is calcu-

lated using Equation 6 at each subsurface layer, whereKe is a function of the proportional ice and liquid watercontent. Therefore, the thermal characteristics of the sub-surface are dynamic as freezing/thawing occurs. Thisapproach is different from previous modelling studies of,


for example, Delisle (1998), who used a constant ther-mal conductivity. To illustrate the sensitivity betweenusing Equation 6 versus a constant thermal conductivity,two experiments were performed with prescribed valuesof 2.0 and 3.0W/mK, which are common subsurfacevalues for unfrozen and frozen subsurface materials, re-spectively (Williams and Smith, 1989). This resulted ina difference of about 80m of permafrost (Figure 4).These results essentially show two thermal gradientsthroughout the profile, with the shift occurring at thebottom of the permafrost. Such behaviour in the geother-mal profile is also seen in observed underground profilesat, for example, Prudhoe Bay, Alaska (Lachenbruchet al., 1982) and the West Siberian platform (Duchovand Devyatkin, 1992).

Geothermal Heat FluxAnother major parameter found to affect the

VAMPER model is the choice of terrestrial heat flux asthe lower boundary condition. Although the estimatedaverage continental heat flow of 65mW/m2 may be as-sumed (Pollack et al., 1993), local fluxes can deviate.Lachenbruch et al. (1982) estimated the basal heat flowto be 55mW/m2 (± 8mW/m2) at Prudhoe Bay, Alaska.In East Siberia, near Lake Baikal, a series of deepboreholes showed heat flow to range between 32 and50mW/m2 (Dorofeeva et al., 2002). Observations of heatflow depend to some extent on the depth at which it ismeasured and other observation effects (i.e. drilling distur-bance) (Luijendijk et al., 2011), but these are not discussedhere. Although these factors of uncertainty are recognised,permafrost modelling studies usually just apply values nearthe worldwide average; for instance, Delisle (1998); Oelkeet al. (2003) and Zhang et al. (2003) assumed a value of60mW/m2, 53mW/m2 and 65mW/m2, respectively. Forthis study, we used the largest database of geothermal heatflow measurements (Jessop et al., 1976). Similar to Pollacket al. (1993), they have reported an average continentalheat flux measurement of 62mW/m2. Given that this da-tabase has a wide range of measurements, a reasonablescale for sensitivity testing is considered between 30and 80mW/m2.

The effect on permafrost thickness within the range ofapplied heat flows was about 350m (Figure 5). Asexpected, a higher heat flow counteracts the freezing front,limiting permafrost growth. Results show that with thebase case conditions (MAGST =�6 °C, φ= 0.4) and an80mW/m2 heat flux, 210m of permafrost is sustained,whereas with 30mW/m2, permafrost can develop to athickness of approximately 550m. Within the given rangeof testing, the sensitivity of permafrost thickness togeothermal heat flow averages 70m per 10mW/m2. Thisrepresents about 16m of permafrost per 1 °C change inthe geothermal gradient (°C/km). However, the influenceof geothermal heat flow on permafrost thickness appearsto be nonlinear since there is a greater difference in per-mafrost thickness ( Δ 130m) between 30 and 40mW/m2



than between 70 and 80mW/m2 (Δ 30m). It is thereforepossible that the trend relating permafrost thickness togeothermal heat flux also depends on the surface temper-ature regime; but additional testing would be required tofurther describe the expected relationships between thetwo boundary conditions.

MAGSTQuaternary palaeoclimate reconstructions provide a wide

range of temperatures thought to be necessary for perma-frost development (Washburn, 1980). For example, Brownand Péwé (1973) placed the southern extent of active icewedges in Alaska and Canada along the �6 °C mean annualair isotherm, although French (2007) has suggested that thisis too general. After a thorough review of investigationslinking past temperatures to different periglacial features,Washburn (1980) has concluded that soil- and ice-wedgepolygons imply an annual average air temperature of �5 °C,if not colder. Other estimates of �8 °C (MAAT) imply thedevelopment of continuous permafrost (Osterkamp and Burn,2002). However, isolated patches of permafrost may formwhere the MAAT is above 0 °C (Shur and Jorgenson,2007). For the sensitivity analysis, the VAMPER model wasforced with MAGST values of �10, -8, -6, -4 and �2 °C(Figure 6). This series of experiments resulted in a varyingpermafrost thickness of 450m, where a one-degree changein MAGST in the VAMPER model causes about a 55-mchange in equilibrium permafrost thicknessResults from all the sensitivity experiments (Table 1)

show, not surprisingly, that geothermal heat flux and groundsurface temperature were the primary factors in permafrostdevelopment among the parameters tested for the VAMPERmodel. The latter relationship has been widely substantiated(Williams and Smith, 1989; Andersland and Ladanyi, 2004;French, 2007). But the other sensitivity tests also showsubstantial effects on permafrost thickness over a millennialtimescale.

CASE STUDIES

A series of experiments was run with the VAMPER modelto simulate permafrost thickness during the last deglacia-

able 1 Summary of sensitivity analysis for major modelarameters.

arameter Range

Maximumvariation

in permafrostthickness

orosity 0.3 to 0.5 50mhermal conductivity 2.0 to 3.0W/mK 80meothermal flux 30 to 80mW / m2 340mean annualround surfacemperature

�10 to �2 °C 450m

Tp

P

PTGMgte


tion. Five LOVECLIM grid cells in Eurasia were selected ascase studies.

Computational Setup

Case Study Selection.Each case study is represented by a grid cell approxi-

mately 560 km x 560 km, which is the resolution ofECBilt (the land-atmosphere component of LOVECLIM).This is illustrated as the overlaid grid in Figure 1. Thelocations were selected based on data availability andproximity to referenced periglacial indicators such as icewedges and relict permafrost. Further, all of the areaswere not covered by ice sheets but in relative proximityand thus held the periglacial conditions necessary fordevelopment and/or maintenance of permafrost duringthe LGM (Hubberten et al., 2004). Although coordinates(or ‘locations’) are referenced to the case studies, theresults are intended to represent the average permafrostevolution over this grid cell rather than what may haveoccurred at a specific location.

LOVECLIM MAAT (LC-MAAT).To obtain the MAAT series from LOVECLIM, we used

two suites of transient coupled simulations performed withthe LOVECLIM 1.2 model which jointly cover the periodfrom 21 ka to 0 ka BP. The first part of the deglaciationcomes from a transient run that covers the period 21 ka to9 ka BP using imposed varying ice sheets, greenhouse gasesand orbital parameters as described in Roche et al. (2011).The Holocene part of the deglaciation run is performedusing similar imposed boundary conditions between 9 and0 ka BP, as published by Renssen et al. (2009). Becauseof land-ocean mask discrepancies, the 9 ka BP stateobtained from the deglaciation and the 9 ka BP used as astarting point for the Holocene transient evolution are notidentical. We thus pasted the two simulations togetherassuming that the last (first) 100 years’ air temperature meanaround 9 ka BP were identical in the 21 to 9 ka BP (9 to 0 kaBP) simulations, obtaining a complete 21 ka BP transientglobal air temperature evolution for the last deglaciation,referred to hereafter as LC-MAAT.

Rather than using the absolute LC-MAAT, anomalies ofthe LC-MAAT were determined and subsequently appliedto the present-day average annual air temperature observedfrom a weather station within the respective case study gridcell. Using an anomaly forcing avoided any potential biasfrom the LOVECLIM 0 ka BP state. As only the anomalyis used, the end of each VAMPER model simulation isconsistent with the nearby weather station.

The modern-day annual mean air temperature for eachweather station was derived from the Global SurfaceSummary of the Day dataset (1980–2011) made availablefrom the US National Climatic Data Center. More informa-tion on the datasets can be found at: www.ncdc.noaa.gov/oa/gsod.html.


http://www.ncdc.noaa.gov/oa/gsod.html

http://www.ncdc.noaa.gov/oa/gsod.html


Upper Boundary: LOVECLIM MAGST (LC-MAGST).Because the VAMPER model is forced by MAGST

rather than MAAT, an offset to determine the LC-MAGSTfrom the LC-MAAT series needed to be defined. This isoften done with a transfer function such as n-factors,which characterise the ratio between air and groundtemperatures by considering surface conditions such asvegetation, organic layers, water bodies and snow cover.However, in these experiments, the large areal coverageand millennial timescale make these locally varying andtemporally sensitive details impossible to integrate. There-fore, we assumed an average offset of 2 °C for all casestudy locations, which is meant to represent an averageover the LOVECLIM grid cell. This simplified air-landcoupling is based on the dominating surface offset of thenival regime (Smith and Riseborough, 2002). The effectof snow cover is for these case studies kept conservativesince it is believed that much of Eurasia during the LGMwas quite dry (Hubberten et al., 2004). Additionally,Osterkamp and Burn (2002) have reported the typicalwarming effect of snow cover to be between 2 and 4 °C.Other examples of this range of surface offset in perma-frost environments include Romanovsky and Osterkamp(1995); Smith et al. (1998) and Goodrich (1982).The final forcing temperatures (LC-MAGST), a function

of both the present-day average temperature for the regionand the surface offset of 2 °C, are shown in Figure 7.

Lower BoundaryThe lower boundary of the VAMPER model was

defined at a depth of 2000m. This is based on the tran-sient time of 21 k years, where the model depth shouldcorrespond with the timescale of interest (Alexeev et al.,2007). In addition, Chouinard and Mareschal (2009) andmore recently Rath et al. (2012) have asserted that a bore-hole depth of at least 1800–2000m is necessary to infer

Figure 7 LOVECLIM-derived mean annual ground surface temperatures(LC-MAGST). These series are produced from applying LOVECLIMmonthly air temperature anomalies to the associated modern annual averageair temperature and then adding a 2°C surface offset. The figure showssmoothed values using a 100-year running mean.This figure is availablein colour online at wileyonlinelibrary.com/journal/ppp


the ground surface temperature history. This implies thatlow-frequency ground surface temperature perturbations,such as those occurring over glacial cycles, penetratedeeply into the ground. For the VAMPER model, a sensi-tivity analysis showed some difference between the resultsof 1000-m model depth versus 2000m but nothing signif-icant between 2000m and 3000m.

At the lower boundary, a specific geothermal heat fluxwas applied for each case study. This was determined usinggrid cell-averaged measurements from the Global HeatFlow Database made available by the University of NorthDakota (Jessop et al., 1976). Only heat flux measurementsdeeper than 1000m were used in the averaging to reducepossible effects from surface influences. However, in theWest Europe cases, the value of Jessop et al. (1976) wasnot considered representative according to Lucazeau andVasseur (1989) and Luijendijk et al. (2011) and wasreplaced by the value in parentheses in Table 2.

Experiment SetupFor each of the five case study locations, three transient

experiments were performed, each with different porosityvalues (φ= {0.3, 0.4, 0.5}), resulting in 15 total runs. Inaddition to these 15, three experiments (16–18) wereperformed using a palaeoclimate reconstruction, referred toas R-MAGST. But since R-MAGST data primarily coverThe Netherlands/Belgium region, the grid cell referenced(Figure 1) is actually one grid cell north and one grid celleast of the original West Europe grid cell. Table 2 detailsthe parameter settings.

For each of the experiments, the equilibrium permafroststate at 21 ka BP was found using consecutive cycles ofthe first 100 years of the LC-MAGST or R-MAGST series,where both are assumed to be representative of LGM condi-tions. Each of these solutions was also taken as the initialcondition in the respective transient experiment. Therefore,the results consist of two parts: the estimated range of per-mafrost thickness during the LGM and the subsequent per-mafrost evolution through the deglaciation.

Results

Central Siberia (Experiments 1–3).To apply a modern average annual air temperature to the

LC-MAAT anomalies, we used the climate data fromYakutsk, Russia (62.02° N, 129.72° E), where the MAATis �8.8 °C. In this area, Federov (1971) has reportedQuaternary-age clay and sand overlying Jurassic mud-stones and sandstones with coal, minerals, silty clay andsand lying beneath.

The centre of Siberia is currently in a zone of continuouspermafrost and has been perennially frozen since at least thelast glacial (Gavrilova, 1993). The persistent deep perma-frost is primarily due to the dominating effect of winteranticyclones, resulting in low atmospheric moisture andthus little snow cover to insulate the ground surface.Rozenbaum and Shpolyanskaya (1998) have estimated that


Table

2Detailsof

each

experimentperformed

atthefive

differentcase

studyareas.

CaseStudy

Coordinates

Nearbyclim

ate

station

Corodinates

Lith

ologya

VAMPERmodel

parameters

Experim

ent

Latitu

deLongitude

Latitu

deLongitude

Terrestrial

heat

flux

(mW/m

b)

Surface

porosity

CentrtalSiberia

63.7°N

129.4°

EYakutsk,RS

62.0°N

129.7°

ESiliciclastic

sedimentary

consolidated

rocks

440.3

10.4

20.5

3WestSiberia

63.7°N

73.1°E

Surgut,RS

61.3°N

73.5°E

Allu

vial

deposits

550.3

40.4

50.5

6South

Russia

52.6°N

50.1°E

Uralsk,

KZ

51.3°N

51.4°E

Sem

i-Consolid

ated

and

Unconsolid

ated

Sedim

entary

Rocks

/Carbonate

Rocks

550.3

70.4

80.5

9Central

Europe

47.1°N

16.9°E

Budapest,HU

47.4°N

19.2°E

Sem

i-Consolid

ated

and

Unconsolid

ated

Sedim

entary

Rocks

/Carbonate

Rocks

/Allu

vial

Deposits

310.3

100.4

110.5

12

WestEuropeA

47.1°N

0.0°

EParis,FR

49.0°N

2.5°

ESem

i-Consolid

ated

and

Unconsolid

ated

Sedim

entary

Rocks

103(88)

b0.3

130.4

140.5

15WestEuropeB

52.6°N

5.6°

EPalaeoreconstruction

N/A

N/A

Sem

i-Consolid

ated

and

Unconsolid

ated

Sedim

entary

Rocks

/Allu

vial

Deposits

77(58)

c0.3

160.4

170.5

18

a Lith

ologytakenfrom

amap

inDürret

al.(2005

)bAverage

was

considered

biased

andreplaced

byavaluefrom

LucazeauandVasseur

(1989)

c Average

was

considered

high

andreplaced

byavaluefrom

Luijendijk

etal.(2011).RS=Russia;

KZ=Kazakhstan;

HU=Hungary;FR=France.


Copyright © 2013 John Wiley & Sons, Ltd. Permafrost and Periglac. Process., (2013)


the LGM permafrost thickness in the northern (arctic)region to have been 700–800m. In this area, the existenceof palaeo-permafrost is supported by radiocarbon-dated icewedges (Kondratjeva et al., 1993).During the LGM, permafrost thickness in Central Siberia

could have ranged between 730 and 940m according toresults of the VAMPER model (Figure 8). Although theLOVECLIM simulations show about an 8 °C warming sincethe LGM, the annual average ground surface temperaturesremained well below freezing and therefore allowed rela-tively little permafrost degradation. Over the last 21 k years,permafrost thawed between 85 and 60m to leave an esti-mated 670 to 855m at present day.

West Siberia (Experiments 4–6).The climate station used for the West Siberia experiments

is in Surgut, Russia (61.25° N, 73.5° E), on the West Sibe-rian Plain and has a current MAAT of �1.5 °C. Rozenbaumand Shpolyanskaya (1998) have reported that the groundtemperature during the last glacial was probably 10 °Ccolder than today, which is slightly greater than the LC-MAGST temperature series (Figure 7).An investigation by Velichko et al. (2011) has described

the lithology as a result of Late Pleistocene aeolian processes

Figure 8 Results of the Central Siberia transient experiments 1–3 showing(a) change in permafrost depth over the last 21 k years and (b) progressionof temperature-depth profiles for experiment 2 (φ= 0.4). The grey verticalline represents freezing at �1 °C.This figure is available in colour onlineat wileyonlinelibrary.com/journal/ppp


with a series of boreholes showing mostly sandy sedimentsbeneath a top peat layer. A morphoscopic examination ofthese sand sediments has shown evidence of an expansivecold desert occurring in the northern half of West Siberiaduring the last glacial (Velichko et al., 2011). During theHolocene warming, this area changed into the currentwetland environment.

A well-documented feature in Western Siberia is relict per-mafrost, typically found at depths hundreds of metres below athawed subsurface. Kondratjeva et al. (1993) have reportedrelict permafrost in Western Siberia to be between 59 and60°N and 300–400m deep. From more moderninvestigations, Ananjeva et al. (2003) have expanded thiszone to be between 60 and 64°N. They found the top ofrelict permafrost to be as shallow as 100m deep and toextend down to 400m deep. Beyond the evidence of relict for-mations, the widest extent of past permafrost in West Siberia,occurring in the Late Pleistocene, is observed by pseudo-morphs found as far south as 52°N (Kondratjeva et al., 1993).

According to the results of the VAMPERmodel (Figure 9a),permafrost thickness during the LGM inWestern Siberia couldhave ranged between approximately 445 and 365m basedon varying porosity. The transient response to changingsurface temperatures through the last deglaciation resulted

Figure 9 Results of the West Siberia transient experiments 4–6 showing(a) change in permafrost depth over the last 21 k years and (b) progressionof temperature-depth profiles for experiment 5 (φ= 0.4). The grey verticalline represents freezing at �1 °C.This figure is available in colour onlineat wileyonlinelibrary.com/journal/ppp



in degradation between 445 and 245m of permafrost occur-ring primarily from the bottom (Figure 9b). During the last10 k years, the top few metres were warm enough to beginthaw from the top.

Figure 10 Results of the South Russia transient experiments 7–9 showing(a) change in permafrost depth over the last 21 k years and (b) progressionof temperature-depth profiles for experiment 8 (φ= 0.4). The grey verticalline represents freezing at �1 °C.This figure is available in colour onlineat wileyonlinelibrary.com/journal/ppp

South Russia (Experiments 7–9).The South Russia case is in a region well studied for

reconstructing glacial-interglacial periods using loess-palaeosol sequences (Velichko and Nechayev, 1984;Little et al., 2002; Bolikhovskaya and Molodkov,2006). For example, at the research site Korstylievo,which is within the bounds for the South Russia casestudy grid cell, Rutter et al. (2003) have described thetop metre to be a modern Chernozem soil, followed bya 1-m thick loess bed (dated marine oxygen isotopestage (OIS) 2), which then overlies another few metresof older (OIS 3 and 4) loess-palaeosol sequences. For thisset of experiments, the annual average air temperature of5.9 °C from Uralsk, Kazakhstan was applied to theLOVECLIM anomalies.There is evidence supporting permafrost existence in

this South Russia area during the LGM. Although loessdeposits are not a definite indicator of past permafrostconditions, the record does allude to a cold and dryperiglacial environment (Bradley, 1999), suggesting alikely scenario for permafrost development. Indicationsof perennial frost action can also be found within such de-posits. Examples include areas in Alaska (Muhs et al.,2003), Belgium (Vandenberghe et al., 1998) and Poland(Jary, 2009). Within loess deposits, many pseudomorphformations have been discovered as far south as 47–49°Nin the western part of the former USSR (Kondratjevaet al., 1993).Results from the VAMPER model show that permafrost in

South Russia could have been between 9 and 15m thick dur-ing the LGM and disappeared around 18 ka BP (Figure 10).

Figure 11 Progression of temperature-depth profiles for experiment 11(φ= 0.4) in Central Europe. The grey vertical line represents freezing at�1 °C.This figure is available in colour online at wileyonlinelibrary.com/journal/ppp

Central Europe (Experiments 10–12).The climate station used for Central Europe is in

Budapest, Hungary (47. 4 °N, 19.2 °E) with a modern aver-age annual air temperature of 10.7 °C. At a relic sand-wedgesite 15 km northeast of Budapest, Kovács et al. (2007) havedescribed the lithology as an Upper Pliocene alluvial fansediment originating from the Palaeo-Danube River coveredby aeolian sandy deposits.Relict periglacial features found in Hungary have been pre-

viously described by Pécsi (1997) andKovács et al. (2007). Ingeneral, it is believed that during the Pleistocene, Hungaryhad a cold, dry climate similar to the current area of northernSiberia, with temperatures 13 °C colder than today(Washburn, 1980).Although the average geothermalheat flux for the Central

Europe case study region was lower than for all the othercasestudies, no permafrost developed here (Figure 11). This islikely due to the LOVECLIM-derived temperatures, whichare simply too warm to produce permafrost.


West Europe A (Experiments 13–15).The averaged geothermal heat flow using measurements

from the heat flow database was 103mW/m2. This highvalue is likely due to an uneven distribution of



measurements where some geologic regions such as theMassif Central have anomalously higher values thansurrounding areas in France (Lucazeau and Vasseur,1989). To apply a more representative value, we took theaverage flux of 88mW/m2 from Lucazeau and Vasseur(1989), who processed and corrected the original Frenchdatasets. The weather station used for this case study is inParis, France, with a MAAT of 10.8 °C.Although periglacial phenomena have been well docu-

mented through observations in The Netherlands, Belgiumand France (Vandenberghe and Pissart, 1993; Huijzer andVandenberghe, 1998), the VAMPER model was unable togenerate permafrost (Figure 12). Just as in the CentralEurope case study, the LC-MAGST series for this regionis well above freezing (Figure 7).

Figure 13 A simplified mean annual ground surface temperature recon-struction (R-MAGST) for the West Europe B grid cell used as forcing inthe VAMPER model (Bohncke and Vandenberghe, 1991; Vandenbergheet al., 2004). See text for abbreviations.This figure is available in colouronline at wileyonlinelibrary.com/journal/ppp

West Europe B (Experiments 16–18).As an alternative approach to using LC-MAGST, the

VAMPER model was forced with a simplified reconstructedtemperature series (R-MAGST), interpreted from Bohnckeand Vandenberghe (1991) and Vandenberghe et al. (2004)with the assumed 2 °C offset between MAAT and MAGST(Figure 13). Unfortunately, the current T21 grid used inLOVECLIM is too coarse to differentiate relatively smallareas such as The Netherlands and Belgium and is hencepart of the grid cell that is covered by the LGM ice sheet.This explains the reasoning behind the chosen grid cell forthe West Europe A experiment. But for this experiment,the R-MAGST series is applicable for The Netherlands.Hence, a different grid cell represents the West Europe Bregion (Figure 1).With a MAAT of �8 °C at 21 ka BP, The Netherlands

developed permafrost between about 320 and 260m deep,according to the VAMPER model results. Permafrostthen completely disappeared between 11 and 9 ka BP(Figure 14).

Figure 12 Progression of temperature-depth profiles for experiment 14(φ= 0.4) in West Europe A. The grey vertical line represents freezing at�1 °C.This figure is available in colour online at wileyonlinelibrary.com/journal/ppp

Figure 14 Results of the West Europe B transient experiments 16–18 show-ing (a) change in permafrost depth over the last 21 kyears and (b) progressionof temperature-depth profiles for experiment 17 (φ=0.4).This figure is avail-able in colour online at wileyonlinelibrary.com/journal/ppp


Discussion

Table 3 presents an overview of the case study results.Unlike many modelling studies which can validate


Table 3 Summarised results of the five case studies.

Case study LGM permafrost thickness Time of disappearance Present-day permafrost thickness Linearised rate of thaw

Central Siberia 730-940 m Still present 670-855 m 0.3-0.4 cm yr-1

West Siberia 365-445 m 2 kaBP/still present 0-120 m 1.2-2.1 cm yr-1

South Russia 9-15 m 18 kaBP 0 m 0.3-0.5 cm yr-1

Central Europe 0 m n/a 0 m n/aWest Europe A 0 m n/a 0 m n/aWest Europe B 260-320 m 11-9 kaBP 0 m 2.2-3.2 cm yr-1

LGM=Last Glacial Maximum.


simulations based on some corresponding set of observa-tions, palaeoclimate simulations are not directly comparableto modern-day values. Instead, we must rely on previousestimates of LGM permafrost thicknesses, bearing in mindthat these are primarily based on reconstructed air tempera-tures. However, what can be directly compared is the resultof the transient simulations at the present day with modern-day observations of permafrost thickness. In addition,permafrost thaw rates are one more way to compare resultssince other permafrost modelling studies have reported suchrates as well. Together, these three elements can be used toevaluate the model results.Among the five case studies, the results from Central

and West Europe A, and to a lesser extent those fromSouth Russia, do not confirm established beliefs of LGMpermafrost distribution. These discrepancies are unlikelyto be due to the VAMPER model itself, since it has beenvalidated and shown to produce reasonable results(Kitover et al., 2012), but rather from the LC-MAGSTforcings (Figure 7), which are too warm to produce perma-frost levels comparable with those in earlier studies. Al-though it is possible for permafrost to occur in areas witha MAAT greater than 0 °C (Shur and Jorgenson, 2007),this would require a finer spatial resolution than the currentgrid cell size used in LOVECLIM. Instead, the lack ofVAMPER model-simulated permafrost existence at 21 kaBP results from the unduly warm temperatures providedby LOVECLIM. This discrepancy in simulated LGM airtemperatures originates from a known bias in LOVECLIM,as discussed by Roche et al. (2007).In South Russia, LGM permafrost was about 9–15m

thick. Although shallow permafrost is not common in thecontinuous permafrost zone, it has been observed. A pres-ent-day example comes from Alaska, where at the southernboundary of the continuous zone, permafrost is measured tobe only 12m deep (Jorgenson et al., 2008). However, inRussia, permafrost tends to be thicker due to lowergeothermal gradients (French, 2007). Baulin et al. (1984) alsogave a higher value of 200m, based on proxy records. It istherefore likely that these experiments result in anunderestimation of permafrost depth. This could be due to anunduly warm MAGST series. But since LOVECLIM has sim-ulated reasonable air temperatures for this region, as opposed tothe European cases, some amount of error may have been intro-duced via the surface offset. According to Simakova (2006),the Russian Plain consisted of steppe vegetation during the


LGM, which implies a dry, cold environment that could in turnreduce the offset between MAAT and MAGST. As the sensi-tivity tests have shown, even a one-degree change in annualground surface temperature could lead to roughly 50m greaterpermafrost thickness under a given set of model parameters.

For the other case studies, the 21ka BP estimates ofpermafrost depth are consistent with earlier estimates. InCentral Siberia, the VAMPER model LGM permafrost depthcorresponds well with the 800–1000-m estimate from Baulinet al. (1984). In West Siberia, the VAMPER model 21-ka BPpermafrost thickness range (365–445m) is just under theestimate from Rozenbaum and Shpolyanskaya (1998), whohave reported that the maximum LGM value was probably500–600m. ForWest Europe B, the VAMPERmodel estimateexceeds the values between 100 and 150m from Delisle(1998), but such a difference can be easily explained by theparameter model settings: the choice of using a constant ther-mal conductivity, which as shown from our sensitivity testing,can account for a 100-m difference in permafrost thickness.

The VAMPER model results for the present day in WestSiberia show permafrost thickness between 0 and 120m.These values are considered reasonable since discontinuouszones commonly show this range (French, 2007). Further,the present-day profile in Figure 9b shows a relict perma-frost layer about 100m deep, which also corresponds withobserved borehole data (Ananjeva et al., 2003).

A closer look at the present-day temperature-depth profilefor West Europe B (Figure 14b), which should resemble pres-ent subsurface thermal conditions, indicates a substrate that isstill in transition from a colder state. This is unexpected sinceobserved deep borehole profiles in The Netherlands reveal ageneral subsurface thermal equilibrium (Luijendijk et al.,2011). This mismatch suggests more time is needed to recoverfrom the prior frozen state, which in turn implies that thesimplified air surface temperature reconstruction does notaccurately capture temperature trends over the last 21 k years.The linearised thaw rate for West Europe B is the highest ofall the experiments, which additionally suggests an insufficientrate of warming estimated over the deglaciation (Figure 13).The relatively high thaw rate may also point to anoverestimation of the initial LGM permafrost thickness.

In Central Siberia, the resulting present-day (855–670m)thickness may be overestimated since Sazonova et al.(2004) has measured thickness to be at a maximum 500mwithin their studied East Siberian transect, which partially falls



within the LOVECLIM grid cell. This shallower permafrostdepth of 500mwould mean that over the last 21 kyears perma-frost degraded, for example, between 230 and 440m, produc-ing a thaw rate of about 1 to 2 cm/yr. This closely matchesthe linearised thaw rates for the other case study locations(Table 3). The discrepancy, therefore, between the observedand simulated present-day permafrost thickness may be dueto this relatively slower thawing rate, which in turn can bedue to changes in the annual surface offset via the air-landcoupling. Velichko and Nechayev (1984) suggested that theLGM climate in West and East Siberia was quite dry due tothe presence of anticyclones, as opposed to present-day condi-tions where Sazonova et al. (2004) have reported relativelythick snow cover (0.6 to 0.8m) in southern Yakutia. If thiswere the case, the surface offset would need to be dynamic,reflecting the low-frequency changes in precipitation over thelast 21 k years. Without improved MAAT-MAGST coupling,it is difficult to know how the changing air circulation patternsmay have affected permafrost response and thaw rates.The linearised rate of thaw is a general calculation of how

permafrost disappeared over the last 21 k years. Based on thethaw rates in Table 3, the transient simulations produced com-parable thawing rates with other millennial-scale modellingstudies (e.g. Lebret et al., 1994; Lunardini, 1995; Delisle,1998). The thaw rate of South Russia is probably not correctsince the VAMPER model results did not produce reasonableresults for this area.As explained above, one possible reason to explain the

difference between the thaw rates of Central Siberia andthe other case studies is changing of precipitation patternsover the last 21 k years, in turn modifying the air-land tem-perature coupling. However, it may also be possible thatthese slower thaw rates did indeed occur in Central Siberia,which has remained in the continuous permafrost zonethroughout the last deglaciation. Meanwhile, according toVAMPER model results, the regions of West Siberia andCentral Europe underwent higher rates of permafrost thawas they changed from a continuous zone to either a discon-tinuous zone or complete disappearance. Similarly, modern-day observations of basal thawing in the discontinuous zoneare on the order of 0.04 to 0.01m/yr (Osterkamp, 2003).This exemplifies a possible distinction between thawingrates in discontinuous versus continuous zones.

CONCLUSIONS

The sensitivity analysis of the VAMPER model gaverelationships between the major subsurface parameters(porosity, thermal conductivity, geothermal heat flux) andmillennial-scale permafrost growth. Sensitivity experimentsshowed that for porosity in the range of 0.3 to 0.5 permafrostthickness is affected by 50m and for geothermal heat fluxranging from 30 to 80mW/m2 by 340m. If a constant ther-mal conductivity is assumed between 2.0 and 3.0W/mK,permafrost can vary by 80m. In addition, varying the groundsurface temperature between �12 °C and �2 °C yields a


corresponding 450-m difference in permafrost thickness.Results of these sensitivity experiments show that whendoing regional studies, it is worthwhile to consider a rangeof possible parameter values if they are not well known.

The five Eurasian case study simulations were forced usingground surface temperatures derived from past experimentswith the LOVECLIM earth system model. Results giveestimated LGM permafrost thicknesses and thawing ratesduring the subsequent deglaciation. In Central Siberia andWestSiberia, the simulated LGM thicknesses agreed well withestimates from proxy-based palaeoreconstructions, which werebetween 800 and 1000m and 500 and 600m, respectively. InCentral Siberia, the LGM permafrost thickness was between730 and 940m, which is only slightly deeper than what existstoday. In West Siberia, which is underlain by discontinuouspermafrost, LGM thickness was between 365 and 445m anddegraded by about 120m over the last deglaciation. Thosesimulations compared particularly well with observationswhere deep (> 100m) relict permafrost currently exists. TheLGM and present-day estimates for South Russia wereunderestimated, which is likely due to our highly simplifiedrelation between air and ground surface temperatures and isprobably not justified in this region. In Central Europe andWest Europe A, however, the VAMPER model was not ableto produce any permafrost during LGM conditions, which isdue to previously recognised biases within the LOVECLIMresults. A supplementary simulation, West Europe B, was thenperformed using palaeoreconstructed temperatures as theforcing. This experiment gave an estimate of permafrost inWest Europe during the LGM to be between 260 and 320mthick, with complete disappearance by 10ka BP.

Overall, some challenges exist with attempting to model thepast permafrost response, particularly when trying to capturerepresentative behaviour for a large geographic extent. Boththe spatial and temporal scale of these experiments make apply-ing static relationships between MAAT and MAGST difficult.The application of a 2 °C air-ground temperature offset seemsto be underestimated in Central Siberia (at least for the later partof the transient experiment), while in West Russia, it wasperhaps an overestimate. This different response exemplifiesthe difficulty in coupling air-ground temperatures, particu-larly over a large representative area and subject to majorclimatic changes. This challenge is well recognised inclimate reconstruction using deep borehole measurements.González-Rouco et al. (2009) have presented a thoroughdiscussion on the effect of snow, among other biases, onthe air-ground temperature coupling. It becomes compli-cated to capture since both snow thickness and timingdepend on varying climatic conditions. Further, they havementioned, it is possible that changes of the nival regimeare not detectable in the land-air temperature coupling overlong timescales.

Similar challenges exist with finding one geothermalheat flux value to represent a large grid cell, which wascalculated using an arithmetic mean. Continental heat flowcan be quite variable due to tectonic and volcanic activity,in addition to the regional geologic history, while localphenomena can bias the statistic. This was the case in



Hungary where the Pannonian Basin has an averaged highheat flow (100mW/m2) (Lenkey et al., 2002) despite thegrid cell average of 31mW/m2. An inverse situationoccurs, for example, at an anorthosite intrusion in Poland,which has an observed low geothermal flux (40mW/m2)and as a result, despite its somewhat southern location(54°N), had an estimated 500m of permafrost during thelast glacial (Šafanda et al., 2004). A nonparametric statis-tical technique (i.e. weighted or geometric average) wouldprobably improve the representative value obtained foreach grid cell based on measurements. Another approachwould be to employ a statistical downscaling technique(Vrac et al., 2007).With the discussed reasoning behind certain VAMPER

model results, such as the biased air temperatures over West


Europe and the simplified MAAT-MAGST coupling, theexperiments here exemplify a new but still simple methodof capturing broad permafrost behaviour over millennia. Itis anticipated that with future coupling to LOVECLIM,the representation of permafrost and climate interactions,notably the air-land coupling, will improve even further.

ACKNOWLEDGEMENTS

This work was supported by funds provided from TheNetherlands Organisation for Scientific Research, projectno: 817.01.013. We strongly appreciate discussions withDr Volker Rath for providing constructive guidance inmodel building.

REFERENCES

Alexeev VA, Nicolsky DJ, Romanovsky VE,Lawrence DM. 2007. An evaluation of deepsoil configurations in the CLM3 for improvedrepresentation of permafrost. GeophysicalResearch Letters Sanger FJ, Hyde PJ (eds).34: L09502. DOI: 10.1029/2007GL029536

Ananjeva (Malkova) GV, Melnikov ES,Ponomareva OE. 2003. Relict permafrostin the central part of Western Siberia. InProceedings of the 8th International Confer-ence on Permafrost, Zurich, Switzerland.Phillips M, Springman SM, Arenson LU(eds). Belkama Publishers: Lisse; 5–8.

Andersland O, Ladanyi B. 2004. FrozenGround Engineering, Second Edition. JohnWiley & Sons: New York.

Anisimov OA, Nelson FE. 1996. Permafrostdistribution in the Northern Hemisphere un-der scenarios of climate change. Global andPlanetary Change 15: 59–72.

Athy LF.1930. Density, porosity, andcompaction of sedimentary rocks.American Association Petroleum Geolo-gists Bulletin 14: 1–24.

Baulin VV, Belopukhova B, Danilova NS.1984. Holocene permafrost in the USSR.In Late Quaternary Environments of theSoviet Union, Velichko AA (ed). Univer-sity of Minnesota Press: Minneapolis;87–91.

Bohncke SJP, Vandenberghe J. 1991.Paleohydrological development in thesouthern Netherlands during the last15,000 years. In Temperate Palaeohy-drology: Fluvial Processes in the TemperateZone during the Last 15,000 Years, StarkelL, Gregory KJ, Thornes JB (eds). John Wi-ley and Sons: Chichester; 254–281.

Bolikhovskaya NS, Molodkov AN. 2006. EastEuropean loess–palaeosol sequences: Paly-nology, stratigraphy and correlation. Qua-ternary International 149: 24–36. DOI: 10.1016/j.quaint.2005.11.015

Bradley RS. 1999. Paleoclimatology: Recons-tructing Climates of the Quaternary.Academic Press: San Diego.

Brown RJE, Péwé TL. 1973. Distribution ofpermafrost in North America and itsrelationship to the environment: A review1963–1973. In Proceedings of the 2ndInternational Conference on Permafrost,Yakutsk, USSR, North American Contribu-tion. Sanger FJ, Hyde PJ (eds). NationalAcademy of Science: Washington, DC;71–100.

Brown J, Ferrians OJ Jr, Heginbottom JA,Melnikov ES. 1997. Revised February2001. Circum-Arctic Map of Permafrostand Ground-Ice Conditions. National Snowand Ice Data Center/World Data Center forGlaciology: Boulder, CO. Digital media.

Carslaw HS, Jaeger JC. 1959. Conduction ofHeat in Solids, Second Edition. OxfordUniversity Press: London.

Cheng G, Wu T. 2007. Responses of climatechange and their environmental signifi-cance, Qinghai-Tibet Plateau. Journal ofGeophysical Research 112: FO2S03. DOI:10.1029/2006JF000631

Chouinard C, Mareschal J-C. 2009. Groundsurface temperature history in southernCanada: Temperatures at the base of theLaurentide ice sheet and during theHolocene. Earth and Planetary ScienceLetters, 277: 280–289

Claussen M, Mysak LA, Weaver AJ, CrucifixM, Fichefet T, Loutre M-F, Weber SL,Alcamo J, Alexeev VA, Berger A, CalovR, Ganopolski A, Goosse H, Lohmann G,Lunkeit F, Mokhov II, Petoukhov V, StoneP, Wang Z. 2002. Earth system models ofintermediate complexity: closing the gap inthe spectrum of climate system models,Climate Dynamics, 18: 579–586. DOI: 10.1007/s00382-001-0200-1

Davis N. 2001. Permafrost: A Guide to FrozenGround in Transition. Fairbanks: Universityof Alaska Press.

Delisle G. 1998. Numerical simulation ofpermafrost growth and decay. Journal ofQuaternary Science 13(4): 325–333.

Delisle G, Caspers G, Freund H. 2003.Permafrost in north-central Europe duringthe Weichselian: how deep? In Proceed-ings of the 8th International Conferenceon Permafrost, Zurich, Switzerland.Phillips M, Springman SM, ArensonLU (eds). Belkama Publishers: Lisse;187–191.

Dorofeeva RP, Shen PY, Shapova MV.2002. Ground surface temperature histo-ries inferred from deep borehole tempera-ture depth data in Eastern Siberia. Earthand Planetary Science Letters 203:1059–1071.

Duchov AD, Devyatkin DV. 1992. Reducedgeothermal gradients in the shallow West-Siberian Platform. Palaeogeography, Palae-oclimatology, Palaeoecology (Global andPlanetary Change Section) 98: 245–250.

Dürr HH, Maybeck M, Dürr SH. 2005.Lithologic composition of the Earth’s conti-nental surfaces derived from a new digitalmap emphasizing riverine material transfer.Global Biogeochemical Cycles 19:GB4S10. DOI: 10.1029/2005GB002515

Ewertowski M. 2009. Ice-wedge pseudomorphsand frost-cracking structures in Weichseliansediments, Central-West Poland. Permafrostand Periglacial Processes 20: 316–330.DOI: 10.1002/ppp.657

Farouki OT. 1981. Thermal properties ofsoils. Special Report 81–1. Cold RegionsResearch and Engineering Laboratory(CRREL), Hanover, NH.

Federov AN. 1971. The first experience ofoperation of underground water ofunderlake talik (Central Yakutia). Geocr-yologic Investigations-Yakutia 1971:123–127 (in Russian).

French HM. 2007. The Periglacial Environ-ment, Third Edition. John Wiley & SonsLtd: Chichester.



Gavrilova MK. 1993. Climate and permafrost.Permafrost and Periglacial Processes 4:99–111.

Gozdzik JS, French HM. 2004. Apparentupfreezing of stones in Late-Pleistocenecoversand, Belchatów Vicinity, CentralPoland. Permafrost and Periglacial Processes15: 359–366. DOI: 10.1002/ppp.491

González-Rouco JF, Beltrami H, Zorita E,Stevens MB. 2009. Borehole climatology:a discussion based on contributions fromclimate modeling. Climate of the Past 5:97–127.

Goodrich LE. 1982. The influence of snowcover on the ground thermal regime.Canadian Geotechnical Journal 19:421–432.

Goosse H, Brovkin V, Fichefet T, Haarsma R,Huybrechts P, Jongma J, Mouchet A, SeltenF, Barriat PY, Campin J-M, DeleersnijderE, Driesschaert E, Goelzer H, Janssens I,Loutre M-F, Morales Maqueda MA,Opsteegh T, Mathieu P-P, Munhoven G,Pettersson EJ, Renssen H, Roche DM,Schaeffer M, Tartinville B, TimmermannA, Weber SL. 2010. Description of the earthsystem model of intermediate complexityLOVECLIM version 1.2. GeoscientificModel Development 3: 603–633. DOI: 10.5194/gmd-3-603-2010

Hartikainen J. 2006. Numerical simulation ofpermafrost depth at Olkiluoto. Posiva2006–52, Posiva Oy.

Hartikainen J, Kouhia R, Wallroth T. 2010.Permafrost Simulations at Forsmar using aNumerical 2D Thermo-Hydro-ChemicalModel. TR 09 – 17. Swedish Nuclear Fueland Waste Management Company,Stockholm, Sweden.

Hubberten, HW, Andreev A, Astakhov VI,Demidov I, Dowdeswell JA, Henrikson M,Hjort C, Houmark-Nielsen M, JakobssonM, Kuzmina S, Larsen E, Lunkka JP, LysåA, Mangerud J, Möller P, Saarnisto M,Schirrmeister L, Sher AV, Siegert C,Siegert MJ, Svendsen JI. 2004. Theperiglacial climate and environment innorthern Eurasia during the LastGlaciation. Quaternary Science Reviews 23:1333–1357. DOI: 10.1016/j.quascirev.2003.12.012

Huijzer B, Vandenberghe J. 1998. Climaticreconstruction of the WeichselianPleniglacial in northwestern and centralEurope. Journal of Quaternary Science13(5): 391–417.

Jary Z. 2009. Periglacial markers within theLate Pleistocene loess-paleosol sequencesin Poland and Western Ukraine. QuaternaryInternational 198: 124–135. DOI: 10.1016/j.quaint.2008.01.008

Jessop AM, Hobart MA, Sclater JG. 1976. Theworld heat flow data collection - 1975.


Geothermal Series Number 5. Earth PhysicsBranch, Department of Energy, Mines, andResources.: Ottawa, Canada; 125pp.

Johansen O. 1975. Thermal conductivity ofsoils. PhD thesis, Trondheim, Norway.(CRREL Draft Translation 637, 1977.)

Jorgenson MT, Yoshikawa K, Kanevskiy M,Shur Y. 2008. Permafrost Characteristics ofAlaska. Institute of Northern Engineering,University of Alaska Fairbanks, 1 sheet,scale 1: 7,200,000.

Kane DL, Hinzman LD, Zarling JP. 1991.Thermal response of the active layer toclimatic warming in a permafrost environ-ment. Cold Regions Science and Technol-ogy 19: 111–122.

Kane DL, Hinkel KM, Goering DJ, HinzmanLD, Outcalt SI. 2001. Non-conductive heattransfer associated with frozen soils. Globaland Planetary Change 29: 275–292.

Kitover DC, Renssen H, Vandenberghe J, VanBalen RT. 2012. Modeling PermafrostResponse of the Last Glacial Termination:First results of the VAMPER model. InProceedings of the 10th InternationalConference on Permafrost. The NorthernPublisher: Salekhard; 209–214.

Kondratjeva KA, Khrutzky SF, RomanovskyNN. 1993. Changes in the extent of perma-frost during the late Quaternary period in theterritory of the former Soviet Union. Perma-frost and Periglacial Processes 4: 113–119.

Kovács J, Fábián SÁ, Schweitzer F, Varga G.2007. A relict sand-wedge polygon site innorth-central Hungary. Permafrost andPeriglacial Processes 18: 379–384. DOI:10.1002/ppp.600

Kukkonen IT, Šafanda J. 2001. Numericalmodeling of permafrost in bedrock in northernFennoscandia during the Holocene. Globaland Planetary Change 29: 259–273.

Lachenbruch AH. 1962. Mechanics of thermal-contraction cracks and ice-wedge polygonsin permafrost.Geological Society of America,Special Paper 70, 69pp.

Lachenbruch AH, Marshall BV. 1969. Heatflow in the Arctic. Arctic 22: 300–311.

Lachenbruch AH, Sass JH, Marshall BV,Moses TH Jr. 1982. Permafrost, heat flow,and the geothermal regime at PrudhoeBay, Alaska. Journal of GeophysicalResearch 87(B11): 9301–9316.

Lawrence DM, Slater AG. 2005. A projectionof severe near-surface permafrost degrada-tion during the 21st century. GeophysicalResearch Letters 32: L24401. DOI: 10.1029/2005GL025080

Lebret P, Dupas A, Clet M, Coutard J-P,Lautridou J-P, Courbouleix S, Garcin M,Levy M, Van Vliet-Lanoë B. 1994. Model-ling of permafrost thickness during the lateglacial stage in France: preliminary results.

Canadian Journal of Earth Sciences 31:959–968.

Lenkey L, Dövényi P, Horváth F, CloetinghSAPL. 2002. Geothermics of the Pannonianbasin and its bearing on the neotectonics.EGU Stephan Mueller Special PublicationSeries 3: 29–40.

Levavasseur G, Vrac M, Roche DM, PaillardD, Martin A, Vandenberghe J. 2011.Present and LGM permafrost from climatesimulations: contribution of statistical down-scaling. Climate of the Past 7: 1225–1246.DOI: 10.5194/cp-7-1225-2011

Li X, Koike T. 2003. Frozen soil parame-terization in SiB2 and its validation withGAME-Tibet observations. Cold RegionsScience and Technology 36: 165–182.

Ling F, Zhang T. 2004. A numerical model forsurface energy balance and thermal regimeof the active layer and permafrost cont-aining unfrozen water. Cold Regions Sci-ence and Technology 38: 1–15.

Little EC, Lian OB, Velichko AA, MorozovaTD, Nechaev VP, Dlussky KG, RutterNW. 2002. Quaternary stratigraphy andoptical dating of loess from the eastEuropean Plain (Russia). Quaternary Sci-ence Reviews 21: 1745–1762.

Lucazeau F, Vasseur G. 1989. Heat FlowDensity Data from France and SurroundingMargins. Tectonophysics 164: 251–258.

Luijendijk EM, ter Voorde M, Van Balen R,Verweij H, Simmelink E. 2011. Thermalstate of the Roer Valley Graben, part ofthe European Cenozoic Rift System. BasinResearch 23: 65–82. DOI: 10.1111/j.1365-2117.2010.00466.x

Lunardini VJ. 1988. Freezing of soil with anunfrozen water content and variable thermalproperties. CRREL Report 88–2.

Lunardini VJ. 1995. Permafrost formationtime. CRREL Report 95–8.

Luo L, Roback A, Vinnikov KY, SchlosserAC, Slater AG, Boone A, Braden H, CoxP, De Rosnay P, Dickinson RE, Dai Y,Duan Q, Etchevers P, Henderson-SellersA, Gedney N, Gusev YM, Habets F, KimJ, Kowalczyk E, Mitchell K, NasanovaON, Noilhan J, Pitman AJ, Schaake J,Shmakin AB, Smirnova TG, Wetzel P,Xue Y, Yang Z, Zeng Q. 2003. Effects ofFrozen Soil on Soil Temperature, SpringInfiltration, and Runoff: Results from thePILPS 2(d) Experiment at Valdai. Russia.Journal of Hydrometeorology 4(2):334–351.

Malevsky-Malevich SP, Molkentin EK,Nadyozhina ED, Shklyarevich OB. 2001.Numerical simulation of permafrost param-eters distribution in Russia. Cold RegionsScience and Technology 32: 1–11.

Marchenko S, Romanovsky V, Tipenko G.2008. Numerical modeling of spatial



permafrost dynamics in Alaska. In Proceed-ings of the 9th International Conference onPermafrost, Fairbanks, Alaska, USA. KaneDL, Hinkel KM (eds). Institute of NorthernEngineering, University of Alaska Fair-banks: Fairbanks, AK. 1125–1130.

Matsuoka N. 2011. Climate and material con-trols on periglacial soil processes: Towardimproving periglacial climate indicators.Quaternary Research 75: 356–365. DOI:10.1016/j.yqres.2010.12.014

Mottaghy D, Rath V. 2006. Latent heateffects in subsurface heat transport model-ling and their impact on palaeotemperaturereconstructions. Geophysical Journal In-ternational 164: 236–245. DOI: 10.1111/j.1365-246X.2005.02843.x

Muhs DR, Ager TA, Bettis EA III, McGeehin J,Been JM, Begét JE, Pavich MJ, StaffordTW Jr, Stevens DSP. 2003 Stratigraphyand palaeoclimatic significance of LateQuaternary loess–palaeosol sequences ofthe Last Interglacial–Glacial cycle incentral Alaska. Quaternary ScienceReviews 22: 1947–1986. DOI: 10.1016/S0277-3791(03)00167-7

Murton JB, Kolstrup E. 2003. Ice –wedgecasts as indicators of palaeotemperatures:precise proxy or wishful thinking?. Prog-ress in Physical Geography 27(2):155–170. DOI: 10.1191/0309133303pp365ra

Oelke C, Zhang T, Serreze MC, ArmstrongRL. 2003. Regional-scale modeling of soilfreeze/thaw over the Arctic drainage basin.Journal of Geophysical Research 108:4314. DOI: 10.1029/2002JD002722

Osterkamp TE. 2003. A thermal history of per-mafrost in Alaska. In Proceedings of the 8thInternational Conference on Permafrost,Zurich, Switzerland, Phillips M, SpringmanS, Anderson LU (eds). Belkama Publishers:Lisse. 863–868.

Osterkamp TE, Burn CR. 2002. Permafrost,Encyclopedia of Atmospheric Sciences,Osterkamp TE, Burn CR, Holton JR, PyleJ, Curry JA (eds). Academic Press: NewYork, NY; 1717–1729.

Osterkamp TE, Gosink JP. 1991. Variations inPermafrost Thickness in Response toChanges in Paleoclimate. Journal ofGeophysical Research 96(B3): 4423–4434.

Pécsi M. 1997. Szerkezeti és váztalajképződésMagyarországon. [Effect of the Quaternaryperiglatial processes on the relief and thestructural soil formation.] Elmélet-Módszer-Gyakorlat 57. MTA FKI:Budapest; 296pp.

Peters-Lidard CD, Blackburn E, Liang X,Wood EF. 1998. The Effect of Soil Ther-mal Conductivity Parameterization onSurface Energy Fluxes and Temperatures.Journal of the Atmospheric Sciences 55:1209–1224.


Pollack HN, Hurter SJ, Johnson JR. 1993.Heat flow from the earth’s Interior: analysisof the global data set. Review of Geophysics31(3): 267–280.

Poutou E, Krinner G, Genthon C, de Noblet-Ducoudré N. 2004. Role of soil freezing infuture boreal climate change. ClimateDynamics 23: 621–639. DOI: 10.1007/s00382-004-0459-0

Rath V, Mottaghy D. 2007. Smooth inversionfor ground temperature histories: estimatingthe optimum regularization parameter bygeneralized cross-validation. GeophysicalJournal International 171: 1440–1448.

Rath V, González-Rouco JF, Goosse H. 2012.Impact of postglacial warming on boreholereconstructions of last millennium tempera-tures. Climate of the Past 8: 1059–1066.DOI: 10.5194/cp-8-1059-2012

Renssen H, Isarin RFB, Vandenberghe J,Lautenschlager M, Schlese U. 2000. Perma-frost as a critical factor in paleoclimatemodelling: the Younger Dryas case inEurope. Earth and Planetary Science Letters176: 1–5. DOI: 10.1016/S0012-821X(99)00322-2

Renssen H, Seppä H, Heiri O, Roche DM,Goosse H, Fichefet T. 2009. The temporaland spatial complexity of the HoloceneThermal Maximum. Nature Geoscience 2:411–414. DOI: 10.1038/NGEO513

Roche DM, Dokken TM, Goosse H, Renssen H,Weber SL. 2007. Climate of the Last GlacialMaximum: sensitivity studies andmodel-datacomparison with the LOVECLIM coupledmodel. Climate of the Past 3: 205–224.

Roche DM, Renssen H, Paillard D, LevavasseurG. 2011. Deciphering the spatio-temporalcomplexity of climate change of the last degla-ciation: a model analysis. Climate of the Past7: 591–602. DOI: 10.5194/cp-3-205-2007

Romanovskii NN. 1985. Distribution of recentlyactive ice and soil wedges in the U.S.S.R. InField and theory: lectures in geocryology,Church M, Slaymaker O (eds). University ofBritish Columbia Press: Vancouver; 154–165.

Romanovsky VE, Osterkamp TE. 1995.Interannual variations of the thermal re-gime of the active layer and near-surfacepermafrost in northern Alaska. Permafrostand Periglacial Processes 6: 313–335.

Rozenbaum GE, Shpolyanskaya NA. 1998.Late Cenozoic Permafrost History of theRussian Arctic. Permafrost and PeriglacialProcesses 9: 247–273.

Rutter NW, Rokosh D, Evans ME, Little EC,Chlachula J, Velichko A. 2003. Correlationand interpretation of paleosols and loessacross European Russia and Asia over thelast interglacial-glacial cycle. QuaternaryResearch 60: 101–109.

Šafanda J, Szewczyk J, Majorowicz J. 2004.Geothermal evidence of very low glacial

temperatures on a rim of the FennoscandianIce Sheet. Geophysical Research Letters 31:L07211. DOI: 10.1029/2004GL019547

Sazonova TS, Romanovksy VE, Walsh JE,Sergueev DO. 2004. Permafrost dynamicsin the 20th and 21st centuries along the EastSiberian transect. Journal of GeophysicalResearch 109: D01108. DOI: 10.1029/2003JD003680

Schaefer K, Zhang T, Bruhwiler L, Barrett AP.2011. Amount and timing of permafrostcarbon release in response to climatewarming. Tellus 63B: 165–180.

Sclater JG, Christie PAF. 1980. ContinentalStretching: An explanation of the Post-Mid-Cretaceous subsidence of the centralNorth Sea Basin. Journal of GeophysicalResearch 85(B7): 3711–3739.

Shur YL, Jorgenson MT. 2007. Patterns ofpermafrost formation and degradation in re-lation to climate and ecosystems. Perma-frost and Periglacial Processes 18: 7–19.DOI: 10.1002/ppp.582

Simakova AN. 2006. The vegetation of theRussian Plain during the second part of theLate Pleistocene (33–18 ka). QuaternaryInternational 149: 110–114.

Smith MW, Riseborough DW. 2002. Climateand the limits of permafrost: A zonal analy-sis. Permafrost and Periglacial Processes13: 1–15. DOI: 10.1002/ppp.410

Smith CAS, Burn CR, Tarnocai C, Sproule B.1998. Air and soil temperature relationsalong an ecological transect through thepermafrost zones of northwestern Canada.In Proceedings of the Seventh InternationalConference on Permafrost Yellowknife,Canada. Lewkowicz AG, Allard M (eds).Centre d’études nordiques, UniversitéLaval: Québec. 1009–1015.

Van Vliet-Lanoë B. 1989. Dynamics andExtent of the Weichselian Permafrost inWestern Europe (Substage 5E to Stage 1).Quaternary International 3(4): 109–113.

Vandenberghe J, Pissart A. 1993. PermafrostChanges in Europe during the Last Glacial.Per-mafrost and Periglacial Processes 4: 121–135.

Vandenberghe J, Huijzer SB, Mücher H, LaanW. 1998. Short climatic oscillations in awestern European loess sequence (Kesselt,Belgium). Journal of Quaternary Science13(5): 471–485.

Vandenberghe J, Lowe J, Coope R, Litt T, ZöllerL. 2004. Climatic and environmentalvariability in the mid-latitude Europe sectorduring the last interglacial-glacial cycle. InPast Climate Variability through Europe andAfrica, Battarbee RW, Gasse F, Stickley CE(eds). Springer: Dordrecht, The Netherlands;393–416.

Vandenberghe J, Renssen H, Roche DM, GoosseH, Velichko AA, Gorbunov A, LevavasseurG. 2012. Eurasian permafrost instability



constrained by reduced sea-ice cover.Quaternary Science Reviews 34: 16–23.DOI: 10.1016/j.quascirev.2011.12.001

Velichko AA, Nechayev VP. 1984. LatePleistocene permafrost in EuropeanUSSR. In Late Quaternary Environmentsof the Soviet Union, Velichko AA,Wright HE, Barnosky CW (eds). Univer-sity of Minnesota Press: Minneapolis;79–86.

Velichko AA, Timireva SN, KremenetskiKV, MacDonald GM, Smith LC. 2011.West Siberian Plain as a late glacial desert.Quaternary International 237: 45–53.DOI: 10.1016/j.quaint.2011.01.013

Vrac M, Marbaix P, Paillard D, Naveau P.2007. Non-linear statistical downscalingof present and LGM precipitation andtemperatures over Europe. Climate ofthe Past 3: 669–682. DOI: 10.5194/cp-3-669-2007


Washburn AL. 1980. Permafrost Features asEvidence of Climatic Change. Earth-Science Reviews 15: 327–402.

Weber SL. 2010. The utility of Earth SystemModels of Intermediate Complexity (EMICs).Wiley Interdisciplinary Reviews: ClimateChange 1(2): 243–252.

Williams PJ, SmithMW. 1989. The Frozen Earth:Fundamentals of Geocryology. CambridgeUniversity Press: Cambridge; 306pp.

ZhangT,OsterkampTE, StamnesK. 1997. Effectsof climate on the active layer and permafrost onthe North Slope of Alaska, USA. Permafrostand Periglacial Processes 8: 45–67.

Zhang Y, Chen W, Cihlar J. 2003. A process-based model for quantifying the impact ofclimate change on permafrost thermal re-gimes. Journal of Geophysical Research108: D22. DOI: 10.1029/2002JD003354

Zhang Y, Chen W, Riseborough DW. 2008a.Transient projections of permafrost

distribution in Canada during the 21stcentury under scenarios of climate change.Global and Planetary Change 60: 443–456.DOI: 10.1016/j.gloplacha.2007.05.003

Zhang Y, Carey SK, Quinton WL. 2008b.Evaluation of the algorithms and param-eterizations for ground thawing and freezingsimulation in permafrost regions. Journal ofGeophysical Research 113: D17116) DOI:10.1029/2007JD009343

Zhuang Q, Romanovsky VE, McGuire AD.2001. Incorporation of a permafrost modelinto a large-scale ecosystem model:Evaluation of temporal and spatial scalingissues in simulating soil thermal dyna-mics. Journal of Geophysical Research106(D24): 33649–33670. DOI: 10.1029/2001JD900151

Zimov SA, Schuur EAG, Chapin FS III. 2006.Permafrost and the Global Carbon Budget.Science 312: 1612–1613.


New Estimates of Permafrost Evolution during the Last 21k Years in Eurasia

Documents