You are on page 1of 15

Geophysical Journal International

Geophys. J. Int. (2014) 199, 829–843 doi: 10.1093/gji/ggu282


GJI Mineral physics, rheology, heat flow and volcanology

Vertical variation in heat flow on the Kola Peninsula: palaeoclimate


or fluid flow?

C. Vogt,1,2 D. Mottaghy,1,3 V. Rath,1,4,5 G. Marquart,1 L. Dijkshoorn,1,6 A. Wolf7,8


and C. Clauser1
1 Institute for Applied Geophysics and Geothermal Energy, E.ON Energy Research Center, RWTH Aachen University, Mathieustr. 10, D-52056 Aachen,
Germany. E-mail: cvogt@eonerc.rwth-aachen.de
2 Schlumberger GmbH, Niederlassung Aachen, Ritterstraße 23, D-52072 Aachen, Germany
3 Geophysica Beratungsgesellschaft mbH, Lütticher Str. 32, D-52064 Aachen, Germany
4 Department of Earth Sciences, Astronomy and Astrophysics,Universidad Complutense de Madrid, Ciudad Universitaria, E-28040 Madrid, Spain

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
5 Institute for Advanced Studies, 5 Merrion Square, Dublin 2, Ireland
6 Divisie NL Energie en Klimaat, Agentschap NL, Sittard, Netherlands
7 Scientific Computing, RWTH Aachen University, Seffenter Weg 23, D-52074 Aachen, Germany
8 Hochschulrechenzentrum, Technische Universität Darmstadt, Mornewegstraße 30, D-64293 Darmstadt, Germany

Accepted 2014 July 18. Received 2014 July 17; in original form 2014 February 5

SUMMARY
Following earlier studies, we present forward and inverse simulations of heat and fluid transport
of the upper crust using a local 3-D model of the Kola area. We provide best estimates for
palaeotemperatures and permeabilities, their errors and their dependencies. Our results allow
discriminating between the two mentioned processes to a certain extent, partly resolving the
non-uniqueness of the problem. We find clear indications for a significant contribution of
advective heat transport, which, in turn, imply only slightly lower ground surface temperatures
during the last glacial maximum relative to the present value. These findings are consistent
with the general background knowledge of (i) the fracture zones and the corresponding fluid
movements in the bedrock and (ii) the glacial history of the Kola area.
Key words: Numerical solutions; Inverse theory; Hydrology; Heat flow; Permeability and
porosity; Heat generation and transport.

to separate advective from conductive heat transport quantitatively


1 I N T RO D U C T I O N
(Rühaak et al. 2010), correction for palaeoclimate may be often
Subsurface temperatures are influenced by different factors, such useful or even required. Heat flow profiles derived from thermo-
as radiogenic heat generation, fluid flow, sedimentation and het- physical measurements in boreholes often show vertical variations,
erogeneity, that is, spatial variation of thermal and hydraulic rock such as those from the Kola super-deep borehole (Kukkonen &
properties. Additionally, borehole temperatures often show signa- Clauser 1994; Popov et al. 1999), from other deep boreholes in
tures of transient variations of the Earth’s ground surface temper- Russia (Kukkonen et al. 1998; Glaznev et al. 2004; Golovanova
ature history (GSTH). Therefore, these borehole temperatures may et al. 2008; Demezhko et al. 2013), and from the KTB boreholes
yield information for comparison with the results of large-scale cli- (Clauser et al. 1997). This vertical variation of heat flow with depth
mate models. In addition, accounting for past surface temperature may be used for identifying and quantifying the relative influences
change is essential for identifying the factors and processes listed of the various processes mentioned above. Based on a data set of
above. temperature and thermal rock properties from seven shallower bore-
Long-term palaeoclimatic effects have been known since the holes and the super-deep hole SG-3 from the Kola Peninsula, we
study of Hotchkiss & Ingersoll (1934), and corrections for palaeo- investigate the relative contributions of heat advected by ground-
climatic effects are required for obtaining meaningful estimates of water flow and transient variation of the vertical heat flow due to
geothermal heat flow (amongst others, Vasseur & Lucazeau 1983; changing palaeotemperatures at the surface to the observed profiles
Clauser 1984; Clauser et al. 1997; Fernandez et al. 1998; Golo- of borehole temperatures and heat flow.
vanova et al. 2008; Slagstad et al. 2009; Majorowicz & Wybraniec The super-deep borehole SG-3 is located on the Kola Peninsula,
2010; Westaway & Younger 2013). Climatic disturbance of subsur- at the northern rim of the Fennoscandian Shield near the Finnish-
face temperatures related to the warming after the last ice age may Russian border. It provides an unique opportunity for studying ad-
well modify the vertical temperature gradient by a few Kelvin per vective and conductive heat transport in the uppermost crust, with
metre (Kohl 1998; Rath et al. 2012). Such temperature anomalies glacial and interglacial climate changes superimposed. The area
may also be caused by slow groundwater flow. Thus, when trying was strongly influenced by a succession of glaciations, resulting in


C The Authors 2014. Published by Oxford University Press on behalf of the Royal Astronomical Society. 829
830 C. Vogt et al.

a gently rolling topography in the study region, with elevations be- volcanic rocks of the lower Proterozoic complex 2 down to 6842 m,
tween 300 and 400 m above sea level. The effect of the last glacial and gneisses and amphibolites of the Archean basement below. At
cycle (LGC) can also be found in subsurface temperatures. Domi- the surface, there is a thin layer of Quaternary glacial deposits,
nant effects originate from the conditions at the base of the ice sheet typically reaching a thickness of only a few metres. Most of the
during and before the period of the last glacial maximum (LGM), published data are summarized in a review article by Kremenetsky
and the subsequent retreat of the ice sheet, exposing the ground & Ovchinnikov (1986), the monograph edited by Kozlovsky (1987)
surface to atmospheric conditions. In contrast to earlier results (see, and the technical report delivered by NEDRA (1992), where many
e.g. Boulton et al. 2001; Hagdorn 2003; Forsström 2005), an en- details can be found. All this data originate from a joint research
semble of calibrated ice sheet models for the Fennoscandian Ice collaboration between Russia, Finland and Germany. During this
Sheet implies that the retreat of the ice sheet started near 20 ka BP. field campaign, 14 new temperature logs were recorded in the shal-
However, the region considered here remained covered by ice until low boreholes around the SG-3. Additionally, 22 existing logs were
about 11 ka BP (Tarasov 2013). re-valuated. Regarding petrophysical data, tensor components of
For the Kola super-deep borehole, earlier results already indi- thermal conductivity were determined on 1375 core samples from
cated an important influence of advection in addition to the effects 21 boreholes. Additionally, specific heat capacity, density, porosity,
of past ground temperature changes (Kukkonen & Clauser 1994; heat production rate and permeability were measured on a subset
Popov et al. 1999). However, studies in adjacent areas in Finland and of samples from 16 boreholes. Finally, this comprehensive data set

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
in the central Kola Peninsula (e.g. Kukkonen et al. 1998; Glaznev was complemented by determining the temperature dependence of
et al. 2004) hold palaeoclimate effects responsible for the vertical thermal conductivity and specific heat capacity on 35 representative
variation of heat flow. In the Kola region, the vertical variation of samples. In Mottaghy et al. (2005), the detailed presentation of all
thermal conductivity is well known from laboratory measurements data is given. In the present study, we make use of this data set,
on rock samples of the super-deep borehole (Popov et al. 1999; additionally to the data set available for the super-deep borehole
Mottaghy et al. 2005). This leaves advection and palaeoclimatic ef- SG-3 (Popov et al. 1999). A summary of the data is given in the
fects as candidates for explaining the observed heat flow variations. sections below.
Extensive modelling has shown that perturbations due to human ac-
tivity, for example, by the nearby open-pit mines of the Nikel area,
2.1 Thermal and hydraulic properties
cannot explain the temperature variations found in the boreholes
due to the short operation times involved (Mottaghy & Rath 2007). An early overview on the thermal properties of the Kola region was
Two studies by Mottaghy et al. (2005) and Mottaghy (2007) evalu- provided by Kremenetsky & Ovchinnikov (1986). For SG-3 as well
ated the influence of variable ground surface temperature histories as for most of the shallower surrounding boreholes (depth up to
and groundwater advection for the area around the SG-3 super-deep 1600 m), laboratory measurements of thermal conductivities and
borehole based on 3-D forward and 1-D inverse simulations. They heat capacities were available (1375 samples). For seven boreholes,
proposed advection as the main reason for the vertical variation on which are used in this study and are indicated by black dots in
heat flow observed both in deep and in shallow boreholes, but the Fig. 1, temperature-dependent properties were measured (Popov
quantitative contribution of palaeoclimate and fluid flow remained et al. 1999; Mottaghy et al. 2005).
open. Porosities of the rocks in SG-3 reported by Kozlovsky (1987)
Here, we present a follow-up study focusing on the relative con- are generally low (0.3–3.0 per cent). Unfortunately, information on
tributions of palaeoclimate and fluid flow effects using the large data permeability in the study region is sparse (Borevsky et al. 1987;
base on vertical heat flow compiled by Mottaghy et al. (2005) and Clauser et al. 1999). While the bulk permeability of metamorphic
a joint inverse 3-D modelling of transient diffusive and advective rocks is low (Mottaghy et al. 2005), relevant permeability due to
heat transport. For a 3-D model based on geological observation faults and fractures cannot be excluded and may have a significant
and data from the SG-3 borehole, we invert for the permeabili- impact on groundwater recharge and fluid flow. For the SG-3 bore-
ties of four geological units and the palaeotemperature step at the hole, higher cavernosity was detected over a short depth interval at
surface corresponding to a given glaciation model. Thus, besides 1800 m, and increasing cavernosity was found below 4700 m, with a
accounting for the influence of heat advection, our results also help maximum around 7 km (Gorbatsevich et al. 2010). The term ‘higher
constraining the GSTH better, particularly regarding the amplitude cavernosity’ was used by Gorbatsevich et al. (2010) for sections in
of the Pleistocene-Holocene warming. SG-3 with numerous major borehole breakouts. Here, we identified
these sections with rock units of increased fracture density. At the
depth of 4700 m, the borehole intersects the Luchlompolo fault, a
mylonitic shear zone (Vidal 1984) and at 7000 m to the transition
2 GEOLOGICAL SETTING,
of the metamorphic rocks to the Archean basement. At the depth of
P E T R O P H Y S I C A L P R O P E RT I E S A N D
1800 m, significant water influx was reported, too (Vidal 1984).
T E M P E R AT U R E O B S E RVAT I O N S
The Kola super-deep borehole SG-3 is situated in the Pechenga ore
2.2 Temperature data
district at 69◦ 23 N, 30◦ 36 E (see Figs 1 and 3). It is located in the
Penchenga graben-syncline and is the deepest borehole (12 262 m) Temperature profiles are available for all shallow boreholes shown
in the world to date. In the surroundings of the SG-3 borehole, in Fig. 1. We selected only those seven boreholes which rested undis-
26 additional boreholes with a maximum depth of 1600 m were turbed for several years or even decades before measurements were
considered for our study and are shown in Fig. 1. taken in 1994 and they are assumed to be in thermal equilibrium
Geological information derives mainly from mining exploration and also provide information on inclination for a depth correc-
in the Pechenga copper nickel mining district and the research re- tion. The subset used here is summarized in Table 1. Temperature
lated to the super-deep SG-3 borehole. The crustal rocks at the data were continuous logs and point measurements with a sampling
SG-3 drilling site consist of high-degree metamorphic sedimentary- interval of 5–10 m. The logging tool used an Analog Device AD
Fluid flow or palaeoclimate, Kola borehole 831

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 1. Topographic map of the study area with the location of the boreholes (from GPS measurements) indicated by their numbers. The boreholes used in
this study are marked by dots, the size of the dots refers to the heat flow at the site; boreholes used for the inversion in this study are indicated by black dots.
(Figure redrawn from Mottaghy et al. 2005, the contour interval is 10 m.) Additionally, the location of the studied region on the Kola Peninsula is shown at the
upper right corner.

Table 1. Data characterizing the shallow boreholes studied: name, longitude version approach based on temperature profiles from the boreholes
(lon), latitude (lat), depth range d for which temperature measurements shown in Fig. 1, confirming this result. Inversions for the scientific
exist, surface elevation h, current average ground surface temperature T0 borehole in Outokumpu (Kukkonen et al. 2011) yielded differences
and average specific heat flow q over the given depth range. of about 8 K relative to modern temperatures, or 10 K relative to
Name lon lat d h T0 q the Holocene temperature maximum. This agrees well with val-
(◦ E) (◦ N) (m) (m asl) (◦ C) (mW m−2 ) ues derived from pollen and other biogeoclimatic indicators from
2271 30.6422 69.4067 50–1364 340 1.28 41 ± 3 nearby sites (Heikkilä & Seppä 2003), or large-scale multiproxy
3359 30.6278 69.3989 50–959 340 0.95 37 ± 3 reconstructions (Luterbacher et al. 2004). In contrast, much larger
3356 30.6614 69.3964 50–1063 350 1 38 ± 4 temperature differences were concluded from borehole observations
3209 30.6197 69.4022 50–806 336 1.1 38 ± 4 in northern Poland (Majorowicz et al. 2007; Mottaghy 2007) and
3200 30.6253 69.395 50–1620 370 1.03 35 ± 3 northwestern Russia (Demezhko et al. 2013), suggesting values of
2915 30.6522 69.4069 50–670 350 0.46 39 ± 5 12–20 K lower than today. Again, this consistent with surface tem-
2400 30.6433 69.4036 50–689 367 −0.3 38 ± 3 perature reconstructions for this area derived from various proxies
(see, e.g. Kageyama et al. 2001; Jost et al. 2005; Wu et al. 2007).
Reconstructions of the late Weichselian glaciation (Hubberten
590 chip sensor (resolution: ±10 mK, and accuracy: ±100 mK). et al. 2004; Siegert & Dowdeswell 2004; Svendsen et al. 2004;
Residual temperature logs obtained from these high-quality tem- Hättestrand & Clark 2006a,b) as well as ice sheet models (e.g.
perature measurements showed no obvious signatures of flow into Siegert & Dowdeswell 1998; Hagdorn 2003; Forsström 2005;
or out of the boreholes. Tarasov 2013) suggest that the Kola region was covered by ice
The temperature profile for the SG-3 (Fig. 2, left) from Popov much longer than Northern Poland. The calibrated model ensemble
et al. (1999) is available in digital format only down to a depth of of Tarasov (2013) shows that the Kola area had been ice-covered ap-
5000 m. The upper part of the SG-3 profile, above 2000 m, deviates proximately from 30 to 10 ka BP. The simulations imply that the ice
significantly deviation from profiles of the shallow boreholes and sheet was comparatively thick (≈1500 m), with near-melting con-
indicates too high temperatures close to the surface. Since we have ditions at its base. Near melting conditions, that is, about 0 ◦ C at the
no information about the conditions inside the super-deep borehole rock surface could explain the warmer palaeotemperatures inferred
SG-3 at the recording time of the available temperature profile, we for the Kola region. In contrast, northern Poland remained exposed
assumed a larger uncertainty range in the inverse modelling studies to the atmosphere most of the time, and was covered only between
for these data compared to those from the shallow boreholes. 22 and 18 ka BP with much thinner ice sheets. However, the par-
ticularly low temperatures are only partly due to the much shorter
time of ice coverage and smaller thickness of the ice sheet in these
2.3 Palaeoclimate, surface temperature history
locations. They constitute a pertinent challenge for global circula-
and basal heat flow
tion models of this region (Ramstein et al. 2007; Kim et al. 2008).
The GSTH during the LGC is essential for interpretation of verti- Regional models, taking into account the full coupling between at-
cal temperature profiles. Ground surface temperatures at the time mosphere, ice sheets and sea ice indicate that a reorganization of the
of drilling for the Kola Peninsula are around 1 ◦ C (see Table 1), regional circulation patterns related to the late development of the
compared to a surface temperature at the LGM which was 4–7 K ice sheet may play a dominant role (Harrison et al. 1995; Krinner
lower than today, as estimated by Kukkonen & Joeleht (2003). et al. 2004; Bromwich et al. 2005; Peyaud et al. 2007; Wohlfarth
Mottaghy (2007) and Rath & Mottaghy (2007) used an 1-D in- et al. 2007).
832 C. Vogt et al.

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 2. Temperature depth profile (left) and heat flow with depth (right) for the SG-3 Kola deep borehole. The red curve in the left figure shows the observed
profile and the red curves in the right figure show the inferred heat flow profiles according to Borevsky et al. (1987) and Popov et al. (1999). The blue curves
show modelling results of Mottaghy et al. (2005) for purely conductive and conductive-advective (coupled) heat transport, using the simplified boxcar model
of the palaeoclimate for the last 80 ka with a temperature step T as shown in the inset; the modelled temperature profiles (left figure) are shown for T = 9 K,
the modelled heat flow profiles (right figure) are shown for different values of T.

3 MODELLING TOOLS where v is the specific discharge (m3 m−2 s−1 ), k the hydraulic
We use the 3-D finite difference code SHEMAT-Suite (Rath et al. permeability tensor (m2 ), μf the fluid dynamic viscosity (Pa s),
2006) for our forward and inverse numerical simulations. This ρf fluid density (kg m−3 ), g gravity (m s−2 ) and P the hydraulic
code, developed from the SHEMAT code (Clauser 2003), contains a pressure (Pa). Here, the unity vector is pointing upwards.
module for inverse estimation of model parameters and boundary In the following, hydraulic head h (m) is used. The corresponding
conditions. pore water pressure can simply be calculated from the distribution
of head and depth z d as (de Marsily 1986; Clauser 2003):
 zd
P(z d , h) = P0 + ρf (z˜d ) g(h − z˜d )dz˜d , (2)
0
3.1 Forward model
where P0 (z 0 ) ≈ 105 Pa is the pressure at the surface z d = 0. The
While various coupled non-linear differential equations are imple- equation for fluid flow implemented here is derived from eqs (1) and
mented in SHEMAT-Suite, we use only those for single-phase fluid (2), and the equation of continuity, using an extended Oberbeck–
flow and heat transport in this study. Fluid flow through a porous Boussinesq approximation (e.g. Kolditz et al. 1998; Diersch &
medium is commonly described by Darcy’s law (Darcy 1856): Kolditz 2002)
 
k ∂h ρf gk
v=− (∇ P + ρf g∇z) , (1) ρf g (α + φβ) =∇· (∇h + ρf ∇z) + W. (3)
μf ∂t μf
Fluid flow or palaeoclimate, Kola borehole 833

Table 2. Variation of matrix and fluid properties with tem- quantities. Under these conditions, the minimum of the functional in
perature and pressure implemented according to the given eq. (5) is called the maxmimum a posteriori (MAP) estimate, which
references. can, for instance, be found by a variety of methods. Our implementa-
Property Function description tion includes data and parameter space formulations (Rodgers 1976;
Tarantola & Valette 1982a,b), and direct minimization of eq. (5) by
λm (T ), cm (T ) Mottaghy (2007)
non-linear conjugate gradients, or a limited-memory quasi-Newton
λf (T, P) Phillips et al. (1981)
cf (T, P), ρf (T, P) Modified after Zylkovskij et al. (1994) technique (Nocedal & Wright 1999). Here, we use a quasi-Newton
μ(T, P) for T > 0 Modified after Zylkovskij et al. (1994) iteration,
μ(T, P) for T ≤ 0 Modified after Speedy (1987)
pk+1 = pk + μ · (JT C−1 −1
d,prior J + C p,prior )
−1
(6)
× {JT C−1 −1
d,prior [d − g(p )] − C p,prior [p − pprior ]}.
k k
Here, φ is porosity, while α and β denote the compressibility
(Pa−1 ) of the rock and the fluid phase, respectively. W corresponds The Jacobian J = ∂g/∂p refers to the matrix of partial deriva-
to a mass source term (kg m−3 s−1 ). tives of the output variables with respect to the input parameters,
In an analogue way, the heat transport equation follows from and μ is a damping parameter which is necessary for ensuring
conservation of energy (e.g. Beardsmore & Cull 2001; Clauser sufficient decline of the objective functional. The calculation of

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
2003) the Jacobian J employs automatic differentiation of the forward
∂T code (e.g. Rall 1981; Griewank & Walther 2008) in the forward
(ρc)e = ∇ · [λe ∇T ] + (ρc)f v∇T + H. (4) (tangent linear) mode, using the tools TAF (Giering & Kamin-
∂t
ski 1998, http://www.fastopt.com/), or TAPENADE (Hascoët &
In this equation, T is the temperature (◦ C), (ρc)e is the effec- Pascual 2013, http://www-sop.inria.fr/tropics).
tive thermal capacity of the saturated porous medium and the fluid Data to be inverted may comprise a subset of the nodal fields P
(J m−3 K−1 ), (ρc)f is the thermal capacity of the fluid (J m−3 K−1 ), and T calculated by the forward model, or linear combinations of
λe is the tensor of effective thermal conductivity (W m−1 K−1 ) and these, that is, point measurements, gradients or integrals of both
H a heat generation rate source term (W m−3 ). The effective thermal data types. Parameters to be inverted in SHEMAT-Suite may be
conductivity of the rock, λe , depends on porosity φ, and is given by rock properties such as porosity φ, permeability k, matrix thermal
the geometric mean of rock and fluid thermal conductivity λm and conductivity λm or boundary conditions. Hydraulic permeability k,
λf : λe = λφf λm
1−φ
. Due to the very low porosity (below 2 per cent), as well as matrix thermal conductivity λm may be anisotropic and
matrix thermal conductivity is approximately equal to effective ther- are defined by their vertical component and anisotropy factors in
mal conductivity. The physical properties of rock matrix and fluid, x- and y-directions. To guarantee positivity and to improve scaling,
and thus also the Darcy velocity or specific discharge in eq. (1), parameters may be replaced by their natural logarithm, implying a
depend on temperature and pressure. These functions for λ, c and corresponding transformation of the Jacobian. Physical properties
ρ form a non-linear coupling between both eqs (3) and (4). The are assigned to the grid cells by means of an index array. These
code SHEMAT-Suite allows to include the non-linear coupling due indices may represent, for instance, layers, isolated bodies and fault
to local (p,T)-conditions in a fully modular way. We implemented zones, or may be assigned to single cells. For keeping the size
site-specific temperature dependencies (Table 2) for the rock prop- of the inverse problem moderate, the parameters of interest may
erties, for example, porosity changes due to permafrost, and specific be assigned to geological units with homogeneous properties. The
permeability-depth functions (see Section 3.2). current code provides considerable freedom for the user to choose
For solving eqs (3) and (4), they are discretized in space by a finite the appropriate setup for a given problem. In this study, we only
difference/volume scheme, employing a general two-level stepping inverted for permeability, while porosity and thermal conductivity
scheme in time (see, e.g. Lynch 2005). The resulting non-linear were kept constant at values derived from core measurements.
systems of equations are solved by a simple alternating fixed-point
iteration (Huyakorn & Pinder 1983).
4 M O D E L L AY O U T
3.2 Inverse model Our model is a local refinement of an existing regional 3-D model
(Mottaghy 2007) and covers a domain of 12 km × 12 km × 10 km.
Inverse methods are a powerful tool to calibrate numerical models
The model is shown in Fig. 3. It covers essentially the area shown
on observed data (Cooley 2004; Hill & Tiedeman 2006). In addition
in Fig. 1 and is extended at the sides. It contains, in particular,
to finding a preferred model, most of the applied methods will also
SG-3 and the seven shallow boreholes mentioned in Section 2. The
provide information on the uncertainty of the derived parameters.
goal of our modelling is studying the influence of fluid flow and
A straightforward Bayesian inverse approach (Tarantola 2004) was
palaeoclimate on the temperature gradient in more detail as well
chosen here, which seeks a parameter set minimizing the non-linear
as investigating the possibility of separating and quantifying the
functional:
competing effects.
B = (d − g(p))T C−1
d,prior (d − g(p)) The model comprises six units: a top layer (unit 1), representing
 T −1   (5) air, so that the surface of unit 2 coincides with topography, and
+ p − pprior C p,prior p − pprior .
five different layers representing the geology as encountered in the
Here, g(p) is the vector of data simulated by the forward model super-deep borehole SG-3 (Table 3). We did not resolve the thin
based on the parameter vector p, and d is the observed data vector. Quaternary top cover since it is not significant for deep heat transport
Prior values for parameters pprior as well as the inverse covariance processes. The model comprises in particular the sections in SG-3
matrices C−1 −1
d,prior and C p,prior need to be specified beforehand. It is around 1800 m and below 4600 m depth (intersection with the
assumed that these are sufficient to characterize the prior probability Luchlompolo Mylonitic shear zone) where increased permeability
distributions of observations, that is, normality is assumed for both is plausible. Topography is included (Mottaghy et al. 2005), and
834 C. Vogt et al.

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 3. 3-D model around the super-deep borehole SG-3 and its setting on the Kola Peninsula. The black asterisk marks the location of the SG-3 on the
map and the model, respectively. The figure shows five different geological units: Proterozoic complex 1 (blue), Ultrabasic intrusions (light blue), Proterozoic
complex 2 (green), Mylonitic shear zone (yellow) and Archean basement (red). The constant properties are described in Table 3 and the permeabilities for
different models and inversion results in Tables 4 and 5.

Table 3. The five geological units of the 3-D model and corresponding depths in the super-deep borehole SG-3 and
average thermal and hydraulic parameters which are kept fixed for forward modelling and inversion: matrix thermal
conductivity λm at standard laboratory conditions, radiogenic heat generation rate A, porosity φ. For all layers, ρcp , the
product of density and specific heat capacity is 2.06 MJ K−1 m−3 . Colours refer to Fig. 3.
Code Geological unit Colour Depth in SG-3 (m) λm (W m−1 K−1 ) A (μW m−3 ) φ(−)
1 Air Not shown 0 0 1
2 Proterozoic complex 1 Blue 0–1500 3.0 0.67 0.007
3 Ultrabasic intrusions Light blue 1500–1800 3.3 0.26 0.001
4 Proterozoic complex 2 Green 1800–4500 3.4 0.4 0.02
5 Mylonitic shear zone Yellow 4500–5000 3.0 0.56 0.018
6 Archean basement Red 5000–10 000 2.9 1.4 0.01

the hydraulic head is set equal to topography. This has an important a following transient forward simulation. For the transient model,
effect on lateral pressure variations which drive fluid flow. we assumed 2000 equal-sized time steps as a time-dependent tem-
Regarding the model size, we chose dimensions where lateral perature boundary condition at the surface for a period of 100 ka,
boundaries have no influence on the flow field in the area of borehole which represents a highly simplified parameterization of the GSTH
data on the one hand. On the other hand, the model must not be too of LGC surface conditions. We used the forward models for finding
large in order to keep CPU times at an acceptable level for the at which depth range data are needed for distinguishing between
computational demanding inverse simulations. the influence of fluid flow advection and downward diffusion of
The model is girded in cells with dimensions of 100 m × 100 m in palaeotemperature. This is important for constraining the parame-
horizontal orientation. The thicknesses of the cells vary and increase ter space and error range for inversion.
with depth. At the top, a small thickness is required because the head
boundary condition and the surface temperature at the top of unit 2
4.1 Thermal properties and thermal conditions
follow the elevation. Thus, the smaller the cell thickness, the better
resolved is the topographic surface. In the present study, we followed Mottaghy et al. (2005) and use
Modelling proceeds in two steps: First, we run a set of forward thermal conductivities and radiogenic heat production rates reported
models for studying the influence of surface temperature history and by Popov et al. (1999) for SG-3 for the five geological units. For
fluid flow on the temperature profile; and secondly, we use inver- the initial steady-state modelling, we assume a mean annual ground
sion methods for estimating permeability and palaeotemperature. surface temperature of 1 ◦ C, which is larger than the mean annual
For the forward models, we started with a steady-state model with air temperature of −1 ◦ C in the area (Boettger et al. 2003), or
fixed surface temperature for controlling the initial conditions for the values slightly below 0 ◦ C which can be derived from recent
Fluid flow or palaeoclimate, Kola borehole 835

reconstructions (Luterbacher et al. 2004; Jones et al. 2012). The in- Therefore, when assigning the permeability, we chose the fol-
creased value with respect to these reconstructions can be explained lowing approaches: in a first, single unit approach, we used a model
by the insulating effects of snow cover (e.g. Kukkonen 1987; Bartlett with only one unit with an exponential decrease of permeability.
et al. 2004, 2005; Smerdon et al. 2006), which are commonly ob- Alternatively, we assume that permeability is distributed according
served in high latitude sites with tundra-like vegetation (Elina et al. to the same units as thermal conductivity (Table 3). We call this the
2010). multiunit approach.
For the transient simulation, we assumed a ground surface In the single-unit case, the decrease of permeability k with depth
history approximating the effect of the glaciation by a boxcar- d is fitted to permeability data reported by Borevsky et al. (1987)
shaped variation of temperature with time (see inlet Fig. 2) with and Clauser et al. (1999) with the functional relationship
T0 (t) = 1 ◦ C + T between 15 and 90 ka BP, and T0 (t) = 1 ◦ C
− dd
at all other times. Thus, palaeoclimate is characterized by only k = k0 · e 0 . (7)
one parameter, namely the temperature amplitude T. Following −14
Mottaghy (2007), we assume an prior value step of −6 K, but also Here, the zero-depth permeability k0 = 1.8 × 10 2
m and the
used inverse methods to constrain this value. attenuation length d0 = 950 m is estimated by fitting the available
Heat flow at the bottom of the model was set to 46 mW m−2 permeability data manually.
according to the value estimated from the temperatures at SG- For the multiunit case, we assumed for the forward model and for

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
3 in a depth of approximately 10 km (Fig. 2) and based on the the prior model for inversion a permeability of the upper Proterozoic
well-constrained petrophysical properties (Borevsky et al. 1987; complex 1 (unit 2) of 3 × 10−16 m2 corresponding to the (logarith-
Kukkonen & Clauser 1994; Popov et al. 1999). This value agrees mic) mean of the values by Borevsky et al. (1987) and Clauser et al.
with estimates given by Kukkonen & Clauser (1994). In addition, (1999). This value is slightly decreased to 10−16 m2 for the deeper
this estimate is probably not influenced by palaeoclimatic effects Proterozoic complex 2 (unit 4) considering pore or fracture closure
due to the great depth, and not perturbed by groundwater flow due with pressure. It is even more reduced to 10−18 m2 for the deep
to the low permeability of the metamorphic rocks. Zero heat flux Archean basement. This value corresponds to the permeability for
conditions were assumed at the lateral boundaries of the model. the deepest rocks studied by Borevsky et al. (1987) and Clauser
Numerical experiments indicated that perturbing the specific heat et al. (1999). For the two sections between 1500 and 1800 m, and
flow by ±3 mW m−2 does not result in significant changes of our 4500 and 5000 m, where Gorbatsevich et al. (2010) detected higher
results. However, it should be mentioned that larger variations of cavernosity, we increased the permeability to 2 × 10−15 m2 . Addi-
this value will probably impact on the estimation process. tionally, we superpose an exponential depth-dependent decrease of
permeability separately for each unit according to eq. (7), where k0
is the permeability at the top of each unit and set to the values of
4.2 Hydraulic properties and boundary conditions k1 according to Tables 4 and 5. The permeability depth distribution
Porosity values for the five units were assigned according Popov based on k1 is shown in Fig. 4. It is used for forward modelling and
et al. (1999). Without doubt, assigning permeability is a bigger as prior for inversion.
problem. As stated in Section 2.1, the porosity-related permeability For fluid flow, we assumed zero-flow conditions at all lateral
is very low and flow is mainly due to fracture networks in the boundaries. No fluid flow occurs at the upper and lower boundary
bedrock of the Kola region. However, the complexity of fracture due to the low permeability in the lowermost unit 6 and zero per-
systems generally requires an approximation in the modelling code, meability in the top unit 1 (air). However, at the outcrop of unit 2
here we do so by representing fracture flow with Darcy flow in (uppermost geological layer), a hydraulic head boundary condition
porous media. allows fluid to discharge or recharge into the model.
Very few reported values of permeability are available for SG-3
(Borevsky et al. 1987; Clauser et al. 1999), indicating a roughly
exponential decrease of permeability with depth with values of 6 ×
5 M O D E L L I N G R E S U LT S
10−15 m2 at 1 km to 10−17 m2 at 7 km depth. A near exponential de-
crease of permeability with depth d (m) according to log10 (k) = −14 As stated before, we first run a set of forward models for studying
− 3.2 × log10 (d) was also suggested in general for non-sedimentary the influence of surface temperature history and fluid flow on the
crustal rocks by Manning & Ingebritsen (1999) due to loss of poros- temperature profiles at the seven shallow boreholes and the deep
ity caused by the increase of pressure with depth. However, both borehole SG-3. Secondly, we use inversion methods for inferring
Vidal (1984) and Gorbatsevich et al. (2010) show evidence for the permeabilities of rock units 2–5 and the palaeotemperature step
increased permeability in SG-3 around 1800 m depth and below amplitude and study their correlation. Highlighting the importance
4600 m. of data from SG-3, we first present our modelling results based (i)

Table 4. The five geological units of the 3-D model and corresponding permeabilities k of the multiunit approach
for the models based on temperature profiles of only the shallow boreholes: k1 represents the values for forward
modelling and the initials and priors for inversion, k−3 K and k−6 K are inversion results for a surface temperature
decrease of −3 K and −6 K, respectively. Estimation errors are discussed in Section 5.2.2. Colours refer to Fig. 3.
Code Geological unit Colour Depth in SG-3 (m) k1 (m2 ) k−3 K (m2 ) k−6 K (m2 )
2 Proterozoic complex 1 Blue 0–1500 3 × 10−16 4 × 10−15 1 × 10−15
3 Ultrabasic intrusions Light blue 1500–1800 2 × 10−15 3 × 10−16 6 × 10−15
4 Proterozoic complex 2 Green 1800–4500 1 × 10−16 6 × 10−18 3 × 10−18
5 Mylonitic shear zone Yellow 4500–5000 2 × 10−15 2 × 10−12 2 × 10−12
6 Archean basement Red 5000–10 000 1 × 10−18 1 × 10−18 1 × 10−18
836 C. Vogt et al.

Table 5. The five geological units of the 3-D model and corresponding permeabilities k of the multiunit approach for the models based on
temperature profiles of the shallow boreholes and the deep SG-3 borehole: ki , kii and kiii are three different permeability profiles used in
forward models for studying the influence of a high permeability in unit 3 and unit 5; k1 represents the initial and prior values for inversion
(the same as in Table 4) and k2 the posterior joint inversion results. Estimation errors are discussed in Section 5.2.2. Colours refer to Fig. 3.
Code Geological unit Colour Depth in SG-3 (m) ki (m2 ) kii (m2 ) kiii (m2 ) k1 (m2 ) k2 (m2 )
2 Proterozoic complex 1 Blue 0–1500 1× 10−18 1× 10−18 1× 10−18 3× 10−16 9 × 10−15
3 Ultrabasic intrusions Light blue 1500–1800 1 × 10−18 1 × 10−15 5 × 10−16 2 × 10−15 4 × 10−17
4 Proterozoic complex 2 Green 1800–4500 1 × 10−18 1 × 10−18 1 × 10−18 1 × 10−16 6 × 10−17
5 Mylonitic shear zone Yellow 4500–5000 1 × 10−13 5 × 10−13 1 × 10−18 2 × 10−15 2 × 10−12
6 Archean basement Red 5000–10 000 1 × 10−18 1 × 10−18 1 × 10−18 1 × 10−18 1 × 10−18

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 5. Comparison between temperature data from borehole 3200 lo-
cated close to the SG-3 (crosses), and modelled temperatures based on a
steady-state purely conductive (circles) and a coupled diffusive-advective
(triangles) model.

Figure 4. Permeability depth profiles: single-unit case (dotted line), prior this to be the explanation for the too high purely conductive heat
multiunit case based on values for k1 , posterior multiunit case based on flow. The temperatures obtained for the coupled process are too
values for k2 (Table 5). low, which might be attributed to exceeding advective downward
directed transport in the unit 3 (just below the depth section shown
in Fig. 5) for which we assumed a permeability of 2 × 10−15 m2 .
only on temperature profiles from the shallow boreholes and (ii) on While the data might be fitted using a different basal heat flow, this
shallow temperature profiles and data from SG-3. value is relatively well constrained within a few mW m−2 (Popov
et al. 1999), as discussed above, and variations within this range do
not change the results significantly.
5.1 Results based on temperature profiles of the shallow
For a purely conductive model, fitting can simply be archived by
boreholes
applying a surface temperature history according to the boxcar func-
In a first step, we calculated by forward modelling steady-state tion (inlet to Fig. 2) and a temperature decrease T of −18 K (see
temperature fields. We assumed both, a purely conductive situa- Fig. 6). However, for demonstrating that an equally good fit may be
tion based only on the thermophysical properties, and a coupled obtained by considering a conductive-advective model, we assumed
diffusive–advective simulation based on the multiunit model in- the shown temperature profile, assumed values palaeoclimatic tem-
cluding also heat advection. For these first models, we show the perature decrease of −3 K and −6 K, respectively, and inverted for
temperature profiles only for the upper 1000 m, for which sufficient the permeabilities in units 2–5 (the very low permeability in unit
temperature data are available, but the numerical simulation was 6 was fixed) by the method described in Section 3.2. Note that, at
performed for the entire model. Fig. 5 shows the resulting simulated this stage, we inverted only for the permeabilities in the different
and observed temperatures at the location of one selected borehole units but kept the palaeotemperature step amplitude fixed during
(number 3200, see Fig. 1). Obviously, the steady-state solutions the inversion. Inversion requires specifying prior values and prior
do not explain the measured temperatures, neither for purely con- errors. Prior values (k1 ) are given in Table 4. Prior errors were set
ductive, nor for coupled diffusive-advective heat transport. This is such that possible optimal solutions of the inversion may be found
not surprising because transient downward diffusion of past subsur- within a range of two orders of magnitude (logarithmic standard
face low temperatures is not considered in this model. We suppose deviation) around the prior values, because very little information
Fluid flow or palaeoclimate, Kola borehole 837

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 7. Temperature profile observed in S-3 (Popov et al. 1999) and from
Figure 6. Comparison of measured and simulated temperatures for bore-
different simulated models reduced by a conductive steady-state reference
hole 3200. Measured temperature (crosses) and the result of a steady-state
model. Red lines show transient conductive models with different palaeotem-
simulation (circles) are the same as in Fig. 4. Tpalaeo (dotted line) is the re-
perature step amplitudes (Tpalaeo ) and green lines show conductive-advective
sult of a pure conductive transient simulation with a boxcar surface ground
models of different permeability distributions ( ki , kii and kiii ; see Table 5)
temperature history and a temperature drop of −18 K. Tmultiunit are the tem-
and a palaeotemperature step amplitude of −3 K.
perature profiles according to assumed temperature drops of −3 and −6 K,
respectively, and permeability depth structures resulting from inversion (see
k−3 K and k−6 K in Table 4). 5.2 Results based on shallow temperature profiles
and data from SG-3
For testing whether deep temperature data can help to separate the
respective effects and to obtain a better understanding of the role
of the deep rock units for advective heat transport, we simulated by
is available on this parameter. Thus, we avoid fixing permeability
forward modelling the temperature profiles also at the location of
to small and potentially erroneous ranges.
SG-3. This was done for purely conductive heat transport models
The resulting (inverted) permeabilities k−3 K and k−6 K are given
with different palaeoclimatic temperature changes as well as for
in Table 4, while the corresponding temperature profiles are shown
conductive-advective multiunit models with different permeability–
in Fig. 6. The observed temperature can be explained equally well
depth distributions and a fixed surface temperature decrease of −3 K
by a strong cooling of −18 K during the LGC or by an only moderate
during the LGC.
temperature decrease of −3 K and advective processes with small
We used three permeability–depth distributions (ki , kii and kiii ;
residuals below 1 K.
given in Table 5) to study in particular the effect of high perme-
As can be expected from an inversion procedure, the fit is very
abilites in unit 3 and unit 5. For all cases, we assumed a low per-
good. The main characteristic of these conductive-advective models
meability of 10−18 m2 in all units, except (i) 10−13 m2 in unit 5, (ii)
is a high permeability around 10−12 m2 in the mylonitic shear zone,
10−15 and 5 × 10−13 m2 in units 3 and 5, respectively and (iii) 5 ×
indicating the importance of the Luchlompolo fault zone for the
10−16 m2 in unit 3.
tectonic history of the Kola peninsula. We show here the findings
The resulting temperature profiles are shown in Fig. 7 together
for the shallow borehole 3200, but they are pretty much the same
with temperatures in SG-3 (Popov et al. 1999). In order to the
for all seven shallow boreholes (marked in Fig. 1).
effects of fluid flow and palaeoclimate more visible, we references
Fig. 6 clearly demonstrates the non-uniqueness of our problem
the temperatures to a purely conductive steady-state temperature
when performing the joint inversion. Numerical test showed that as-
profile at the respective site.
suming palaeotemperature amplitudes of −3 to −18 K at the surface
The similarity of the shapes of the plotted curves in Fig. 7 above
together with heat advection resulting from appropriate hydraulic
about 1000 m illustrates that only beneath a depth of 1000 m below
properties may explain the observations in the shallow boreholes.
sea level the two processes can be clearly distinguished. Conse-
Therefore, it is impossible to distinguish between the effects of heat
quently, a data set from a sufficient depth is essential for obtaining
advection due to groundwater flow and transient downward diffu-
unique estimates. Therefore, the temperature profile of the super-
sion of past subsurface temperature variations based on the shallow
deep borehole SG-3 has to be taken into account at least partially
boreholes alone. In particular, a palaeotemperature step amplitude
for a successful inversion—even though we have less control on the
of −18 K in a purely conductive model fits the shallow data quite
measurement conditions in this case.
well as shown in Fig. 6.
Therefore, in a next step, we accounted for temperature data from
the super-deep borehole SG-3 and from the other seven shallow
5.2.1 Joint inversion for permeability and post-glacial warming
boreholes (see Fig. 1) in a joint inversion for studying whether the
individual contributions of both processes may be distinguished. For providing best estimates and covariances, we performed a joint
That is, we inverted jointly for the permeabilities of the rock units inversion for permeability and palaeocooling based on a transient
and the palaeotemperature step. model for conductive-advective transport. To this end, we used the
838 C. Vogt et al.

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Figure 8. Temperature difference T between simulated and measured tem-
peratures for the super-deep borehole SG-3. Lines correspond to transient
simulations accounting for the downward diffusion of a sudden decrease of
surface temperature between 15 and 90 ka BP. Symbols represent tempera-
tures from purely conductive steady-state simulations (circles) and temper-
atures from joint inversions of coupled diffusion and advection for a single-
and a multiunit model (triangles and squares, respectively).
Figure 9. Temperature difference T between simulated and measured
temperatures for boreholes 2271, 3359, 3356, 3209, 3200, 2915 and 2400
data from the shallow boreholes and from a section of the SG-3 located nearby the Kola super-deep borehole SG-3 (see Fig. 1). Symbols are
temperature profile between 2000 and 5000 m (see Section 2.2). explained in the caption of Fig. 8.
Even though data from SG-3 suffer from a larger error (1 K), deep
data are vital for a joint inversion, as shown in the previous section.
This corroborates our result that temperature observations from
Prior values and errors for permeability were set in the same way
shallow boreholes—while indispensable for information of the up-
as described above. We assumed a prior value for the palaeotem-
per 1 km and also for providing an areal coverage—are not suf-
perature step amplitude of −6 K ± 2 K at the top model boundary,
ficient for discriminating between different GSTHs for the glacial
in accordance with our previous results based on shallow boreholes
period. Fig. 9 also shows that the residuals do not scatter arbitrarily
and results of Mottaghy (2007).
around zero but correlated with depth. This implies some additional
For the single-unit approach (i.e. exponential decrease of perme-
depth-dependent factors influencing the temperature, such as het-
ability with depth (see eq. 8), we estimated the zero-depth perme-
erogeneity not captured by the simplified model geometry or the
ability k0 by numerical Bayesian inversion (eq. 5 and eq. 6) to about
simplified step function used for the GSTH. Therefore, any uncer-
6 × 10−15 m2 , the corresponding ‘optimal’ palaeotemperature step
tainty estimate can only be an approximation.
amplitude was −0.6 K, which corresponds to a surface tempera-
Regarding the posterior permeability of the multiunit model
ture of 0.4 ◦ C. For the multiunit approach, Table 5 lists the prior
(Table 3), it turned out that in the mylonitic shear zone (unit 5),
permeability values k1 and the values k2 obtained by the numerical
it is considerably increased (by three orders of magnitude) com-
joint inversion for units 2–5. The corresponding palaeotemperature
pared to the prior assumption. In the top geological layer (unit 2),
step amplitude was −0.3 K, which corresponds to a temperature of
it is slightly increased, but reduced again in the thin layer beneath
0.7 ◦ C. Hence, a very small palaeotemperature step amplitude is re-
(unit 3).
quired for fitting the observed temperature profile down to 5 km for
both assumptions on the subsurface permeability distribution. For
the model with an exponential decrease of permeability with depth,
the temperature step amplitude is positive, implying a temperature 5.2.2 Errors and parameter resolution
that was even higher during glaciation than today, which seems to
be unlikely. For characterizing the information content of the inverse solution,
In Fig. 8, we plotted the temperature depth profile at SG-3 based we calculate the posterior covariance matrix (Tarantola 2004), which
on the two conduction-advection inversion results (single-unit and is meaningful in a linear neighbourhood around the MAP estimate.
multiunit model) and on the transient conductive model for various Keeping in mind this assumption, this symmetric matrix can be
palaeotemperature step amplitudes at the surface. written as
Clearly, the best fit of the SG-3 temperature data below 1500 m  −1
C p,post = JT C−1 −1
d,prior J + C p,prior . (8)
depth is obtained with the multiunit model. As suggested by the
inversion for the shallow temperature data in well 3200, the fit for In contrast to the prior covariance matrix C p,prior , which was
the temperature data of all seven shallow wells is equally good for assumed to be diagonal in this study, this new covariance matrix
the purely conductive model with T = −18 K and the single-unit, is fully occupied in general. While the diagonal elements can be
and multiunit models with the small temperature step amplitudes of used for estimating the errors of the corresponding parameters, the
−0.6 and −0.3 K, respectively (see Fig. 9). relations between different parameters induced by the particular
Fluid flow or palaeoclimate, Kola borehole 839

Here, T sim denotes the simulated temperature, T obs the observed


temperature, σ T its error, nd the number of data, and np  nd the
number of parameters. A rms of 1 indicates a fit within the given
prior errors.
For the multiunit approach, the rms equals about 1, while for
the single-unit approach, it equals about 1.5. The smaller rms for
the multiunit approach reflects the improved fit of the multiunit
approach as shown in Fig. 8. Since the rms for the single-unit
approach does not equal about 1, the posterior variance needs to be
multiplied with rms2 for obtaining a meaningful error estimate (e.g.
Hill & Tiedeman 2006). Resulting logarithmic errors are less than
one order of magnitude in unit 2 and unit 5 for both the multiunit
approach and the single-unit approach. Thus, the uncertainty is
drastically decreased by the inversion. Given the lack of data at
greater depth, the posterior uncertainty in units 2 and 5 seems
acceptable for identifying the influence of fluid flow. The uncertainty

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
for units 3 and 4 is not reduced by the inversion, remaining on the
Figure 10. Correlation matrix M (see eq. 9), illustrating the correlations be- order of two magnitudes. Again, the rms indicates the multiunit
tween the different parameters. Note the strong anti-correlation between unit approach as preferable compared to the single-unit approach. While
2 and unit 5. The temperature amplitude T seem reasonably well resolved generally more free parameters yield better fits but not necessarily
independently, showing no significant correlation with other parameters. more information, the multiunit approach also captures the overall
geology more appropriately.
Taking into account the covariance matrix information, the ac-
prior information can be studied with the help of the correlation tive flow zones represented by unit 2, and in particular by unit 5
matrix M (Tarantola 2004). It is calculated from the covariance as dominate the estimation of the optimal subsurface properties. At
Ci j the same time, they help fitting the observed temperatures even
Mi j = . (9) at great depths. Interestingly, unit 5 is supposed to be a high-
Ci j · Ci j
permeability zone according to prior geological information (Vidal
M inherits symmetry from C p,post and can be interpreted 1984), whereas unit 2 is not. Fluid flow in the other units, including
more easily than this matrix, as it is normalized to the range the supposed high permeability unit 3, does more or less not affect
−1 ≤ M(i, j) ≤ +1. A value of ±1 indicates perfect correlation the fit of the temperature field. However, the permeabilities of unit
or anti-correlation. 2 and unit 5 cannot be identified independently as they are strongly
Prior errors were assumed to be two orders of magnitude in per- anti-correlated. This is indicated clearly by the correlation matrix
meability for units 2–5 and 2 K for the palaeotemperature step M. Surprisingly, this matrix also indicates no significant correlation
amplitude. The final estimation errors are defined by the posterior between palaeotemperature and permeabilities.
covariance matrix in eq. (8). The inversion indicates that a sig-
nificant uncertainty reduction is obtained by our joint numerical
inversion only for unit 2 and unit 5 and for the palaeotemperature
5.2.3 Vertical variation of heat flow
step amplitude.
For the prior model, we assumed slightly increased permeabil- Fig. 11 shows the vertical component of heat flow at the Kola
ities in unit 3 and unit 5, but inversion yielded slightly increased super-deep borehole SG-3 reported by Borevsky et al. (1987) and
permeability in unit 2 and a high permeability in unit 5. While Popov et al. (1999) based on observed data and laboratory mea-
fluid circulation in the deep and thick unit 5 is essential for fitting surements, and on the results of our numerical models. The curve
the temperature profile, fluid flow in the thin unit 3 seems to be of Popov et al. (1999) based on high-resolution temperature mea-
insignificant with respect to the temperature profile at the given surements and thousands of thermal conductivity measurements on
resolution. However, the permeabilities of unit 2 and unit 5 can- rock samples shows large variations with depth with a rather small
not be estimated independently from each other, because they are uncertainty on the order of ±4 mW m−2 (Popov et al. 1999). In
strongly anti-correlated as shown by the correlation matrix (Fig. 10) contrast, the vertical profile of heat flow by Borevsky et al. (1987)
computed using eq. (9). represents an average value over greater depth intervals. The mod-
For both, the single-unit and the multiunits models, the poste- elled curves are shown for the case of a transient conductive model
rior error of the palaeotemperature step amplitude is 0.2 ◦ C, with with −18 K temperature decrease during the LGM, and for the mul-
corresponding estimated surface temperatures close to 0 ◦ C. Thus, tiunit and single-unit conductive-advective models with parameters
our results support the assumption of near-melting conditions at the for permeability and temperature step obtained by inversion. Be-
base of the ice sheet at this site. For the ice thicknesses simulated low, a depth of about 5 km heat flow for all three models converges,
by Tarasov (2013), melting would occur between −0.5 and −1 ◦ C since at that depth, advection is negligible and only conductive heat
(e.g. Marshall 2012). transport is relevant.
However, also, the goodness-of-fit as given by the root-mean- Comparison of the model results and the observed heat flow
square (rms) needs to be considered. It is defined as shows that none of our modelling results can exactly reproduce
 the data. This is not surprising considering the various unknowns


T sim −T obs 2 and simplifications in our study. In particular, heat flow might

σT be too small in our models in the mylonitic shear zone (unit
rms = .
nd − n p 5) at depths between 4500 and 5000 m. As Popov et al. (1999)
840 C. Vogt et al.

temperatures from shallow boreholes are insufficient for identifying


the glacial signal properly. Therefore, in a second step, we consid-
ered data from the deeper parts of the super-deep borehole SG-3.
Combining the data from the deep well with those from the shallow
wells is important, compared to only considering data from SG-3
alone, because they show a spatial change in vertical thermal gra-
dient which cannot be explained by a (constant) surface palaeotem-
perature change. Indeed, including data from the deeper parts of the
super-deep borehole SG-3 contradicts a purely conductive regime
and such a low palaeotemperature.
Therefore, we considered significant fluid flow consistent with the
supposed permeability distribution and known surface topography.
One may conclude that deep temperature data are best to estimate
the advective fluid flow in (deep) highly permeable layers, while
shallower data are best to estimate the palaeotemperature change
which is closer to the surface. However, as mentioned above, the

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
geometry of the local geology of the Kola peninsula is complicated
Figure 11. Vertical component of specific heat flow q at the Kola super- and units with considerable advective flow even outcrop in the north
deep borehole SG-3 according to observations (full and dotted black lines; and east of our model (see Fig. 3). Thus, the shallow boreholes are
the curve of Popov et al. 1999, is smoothed) and our numerical models (red not uniformly far away from the units with considerable advective
dotted line: result of a transient conductive model with −18 K temperature flow. Therefore, fluid flow in these units also contributes to the
drop during the last glacial; green squares and blue triangles: results of temperatures measured in the shallow wells.
multiunit and single-unit conductive-advective models with parameters for From a transient joint inversion of permeabilites and palaeotem-
permeability and temperature step obtained by inversion). perature, the best fit of the observed temperatures supposes
palaeotemperatures very near to 0 ◦ C and advective transport within
conclude, purely conductive heat flow cannot explain this increase a dipping high permeable structure. As the geometry of units is fixed
up to nearly 80 mW m−2 . They attribute this non-conductive regime (by geological prior knowledge) in our model, this process is located
to the presence of a fracture zone with fluid flow, which is, however, in unit 5 (yellow colour in Fig. 3) by the inversion algorithm. The
not separately resolved in our model. It is rather approximated by combined model leads to a consistent interpretation of both data
the bulk parameters of unit 5, yielding lower heat flow values. sets (SG-3 and shallow wells).
In general, the multiunit approach (green squares in Fig. 10) The result indicates near-melting conditions at the base of the ice
generally agrees with the trend of the observations. An important sheet. Independent palaeoclimatic observations as well as simula-
result is that we can explain this trend with parameters estimated tion results indicate a long duration (20 ka), large thickness of ice
only by using data from a maximum depth of ≈5000 m, since fluid cover and moderate basal velocities for the Kola region (Siegert &
flow and heat transport follow inclined structures from these shallow Dowdeswell 1998; Hagdorn 2003; Forsström 2005; Tarasov 2013;
depths to deeper regions of the model in this particular geological Kleman et al. 2013). Thus, large palaeotemperature step amplitudes
setting. This enhances the confidence in this estimated parameter (<−7 K) seem unlikely. We conclude that a combination of fluid
configuration. Geological units encountered in the shallow depths flow and palaeoclimate with a low to moderate postglacial tempera-
are similar to those at larger depths. This justifies the extrapolation ture step is the most plausible explanation of the observed variation
of rock properties to large depths to a certain degree, of course, in the vertical temperature gradient. It should also be pointed out
taking into account the P-T-dependence. that a temperature at the base of the ice sheet not much lower than
present has also been found by several studies for the Canadian
Ice sheet (Sass et al. 1971; Rolandone et al. 2003; Chouinard &
6 D I S C U S S I O N A N D C O N C LU S I O N
Mareschal 2009), and was inferred for the Fennoscandian Ice Sheet
In this study, we used temperature–depth profiles from a number of by Kukkonen & Joeleht (2003). However, no ice sheet is homo-
shallow wells (down to 1000 m), and from the Kola super-deep bore- geneously at melting conditions, neither in space not in time. At
hole SG-3 down to 5000 m for assessing the impact of past surface the Kola location state-of-the-art glacial system modelling (Tarasov
temperature changes during the last glacial period on the subsurface 2013, personal communications) shows that this result is consistent
temperature regime, and to evaluate the individual contributions of with ice sheet physics and climatic development, as it was a site of
conductive and advective heat transport. high ice velocities during most of the LGM.
We used forward and inverse modelling for estimating the per- The favoured hydrogeological model employs a small number of
meability of the various geological layers in the subsurface and the geologically motivated units. Permeability is assigned to each unit
increase of surface palaeotemperature at the end of the last glacial. individually, and is assumed to decrease exponentially with depth
We demonstrated that an analysis including only data from shal- within each unit (eq. 7). Small-scale variations of thermal properties
low boreholes is ambiguous and does not allow discriminating be- were not considered in this study and are of minor importance
tween conductive and advective processes affecting the temperature compared to the studied effects of advection and palaeotemperature
gradient in the Kola region. diffusion, as was shown by Mottaghy et al. (2005) and Mottaghy
In principle, palaeotemperatures of 18 K lower than today may (2007). All results are based on a fixed value of basal heat flow,
explain the observations from shallow boreholes without any con- which is a first assumption. The value of 46 mW m−2 at 10 km depth
tribution of heat advection. However, the majority of boreholes is used in this study might be wrong by ±(3–5) mW m−2 (Popov et al.
shallower than 1000 m. Compared to the depth of the maximum 1999). Additionally, it is likely to vary laterally at a depth of 12 km
peak of the reduced temperature (Fig. 7), this indicates already that due to the dip of the geological layers.
Fluid flow or palaeoclimate, Kola borehole 841

We tested our results from the shallow boreholes against the heat CL 121/4-4 to Christoph Clauser. We thank Yuri Popov for pro-
flow obtained for the SG-3 (Popov et al. 1999). This is, at least partly, viding digital temperature data for the Kola super-deep borehole.
an independent test, since thermal conductivities in the various units Volker Rath was funded by the Spanish Government by its Ramón
had not been adjusted by the inversion. Our favoured heat flow for y Cajal Program, project CCG10-UCM/ESP-5070 (Comunidad de
the multiunit configuration is consistent with the data, in particular Madrid/UCM) and project CGL2011-29672-C02-01 (Ministerio de
at depths below 5000 m in the super-deep borehole SG-3. The Economı́a y Competitividad).
improved fit of our modelled heat flow below 5000 m also represents
the main improvement of our modelling approach compared to the
results of Mottaghy et al. (2005) (see Fig. 2). Moreover, we used
a fully 3-D model including all available geological information, REFERENCES
while Mottaghy (2007) only performed 3-D forward modelling and Bartlett, M.G., Chapman, D.S. & Harris, R.N., 2004. Snow and the ground
1-D inverse modelling. temperature record of climate change, J. geophys. Res., 109, F04008,
In addition to best estimates for palaeotemperature and perme- doi:10.1029/2004JF000224.
ability, the inversion also provides parameter covariances and corre- Bartlett, M.G., Chapman, D.S. & Harris, R.N., 2005. Snow effect on
lations. A more detailed study of resolution also favours a multiunit north American ground temperatures, 1950–2002, J. geophys. Res., 110,
F03008, doi:10.1029/2005JF000293.
model but indicates that the permeabilities of unit 2 (top geological

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
Beardsmore, G.R. & Cull, J.P., 2001. Crustal Heat Flow, Cambridge Uni-
unit) and unit 5 (mylonitic shear zone below 4500 m) cannot be
versity Press.
identified independently from each other. Despite this, the observed Bense, V. & Person, M.A., 2008. Transient hydrodynamics within inter-
heat flow below 4500 m, of up to 80 mW m−2 as well as increased cratonic sedimentary basins during glacial cycles, J. geophys. Res., 113,
3
He/4 He ratios below 3500 m (Gorbatsevich et al. 2010) indicate a F04005, doi:10.1029/2007JF000969.
fractured region providing pathways for motion of fluid (including Boettger, T., Hiller, A., Kremenetski, C. & Friedrich, M., 2003. Interrelations
3
He) from greater depth. While our inverted permeability for unit between climatic changes and northern and alpine Holocene pine-limit
5 (between 5 × 10−13 m2 and 2 × 10−12 m2 ) is probably overesti- movements ? Deduced from stable isotope signals of 14C-dated subfossil
mated, however, an increased permeability in this unit is very likely pines (Pinus sylvestris L.) on the Kola Peninsula, northwestern Russia, in
and consistent with Gorbatsevich et al. (2010). Thus, within the TRACE—Tree Rings in Archaeology, Climatology and Ecology, Proceed-
ings of the DENDROSYMPOSIUM 2002, Bonn/Jülich, Germany, Vol. 33,
limitations of the approach, the individual contributions to the heat
pp. 52–59.
flow by fluid flow driven heat advection and transient downward
Borevsky, L.V., Vartanyan, G.S. & Kulikov, T.B., 1987. Hydrological essay,
diffusion of past surface temperature variations can be identified. in The Superdeep Well of the Kola Peninsula, pp. 271–287, ed. Kozlovsky,
Since we combined all available information with respect to the Y.A., Springer.
appropriate resolution of the multiunit model, the values obtained Boulton, G., Dongelmans, P., Punkari, M. & Broadgate, M., 2001.
for permeability and palaeotemperature may be considered best Palaeoglaciology of an ice sheet through a glacial cycle: the Euro-
estimates in view of data quality and estimation errors. pean ice sheet through the Weichselian, Quat. Sci. Rev., 20(4), 591–
This being said, our results still leave several questions unan- 625.
swered. Although we presumably captured the main geological fea- Bromwich, D.H., Toracinta, E.R., Oglesby, R.J., Fastook, J.L. & Hughes,
tures in our subsurface model, there is clearly room for model im- T.J., 2005. Lgm summer climate on the southern margin of the Laurentide
ice sheet: wet or dry?, J. Climate, 18(16), 3317–3338.
provement. The most important limitations of the current study are
Chouinard, C. & Mareschal, J.-C., 2009. Ground surface temperature history
related to the complex glaciation history of the area. The Fennoscan-
in southern Canada: temperatures at the base of the Laurentide ice sheet
dian ice sheet had important effects on subsurface conditions, in and during the Holocene, Earth planet. Sci. Lett., 277, 280–289.
addition to the changes in surface temperature involved. The ice Clauser, C., 1984. A climatic correction on temperature gradients us-
sheet of the LGM also produced a huge water pressure pulse at ing surface-temperature series of various periods, Tectonophysics, 103,
the surface which would conservatively amount to up to 10 MPa 33–46.
(corresponding to a hydraulic head of about 1000 m). Deep migra- Clauser, C., 2003. Numerical Simulation of Reactive Flow in Hot Aquifers,
tion of subglacial waters into the subsurface is well known in many Springer-Verlag.
sedimentary basins worldwide (e.g. Person et al. 2007; Bense & Clauser, C. et al., 1997. The thermal regime of the crystalline continental
Person 2008; Lemieux et al. 2008). Lately, it has been found im- crust: implications from the KTB, J. geophys. Res., 102(B8), 18 417–
18 441.
portant in crystalline areas in the context of nuclear waste disposal
Clauser, C., Popov, Y. & Kukkonen, I.T., 1999. Heat transfer processes
(Neuzil 2012; Person et al. 2012a,b; Provost et al. 2012; Vidstrand
in the upper crust — a detailed geothermal and hydrological study in
et al. 2013). Both the hydrogeological conditions at the base of the the area of the Kola deep hole, Russia, Final report, intas-93-273-ext,
ice sheet and the thermal consequences are not well known. In par- Archiv Nr. 119469, Leibniz Institute for Applied Geosciences, Hannover,
ticular, the interaction with existing hydrogeologically important Germany.
structures (e.g. fault zones) needs further research. Furthermore, Cooley, R.L., 2004. A theory for modeling ground-water flow in heteroge-
the deformation of the crust originating from the burden of the ice neous media, Professional Paper 1679, U.S. Geological Survey (USGS),
masses will probably have an effect on crustal permeability. With Reston VA, Available at: http://pubs.usgs.gov/pp/1679/report.pdf (last
regional ice sheet models (e.g. Tarasov 2013) and reconstructions accessed 19 August 2014).
developing rapidly (e.g. Gyllencreutz et al. 2007; Mangerud et al. Darcy, H., 1856. Les Fontaines Publiques de la Ville de Dijon, Dalmont.
de Marsily, G., 1986. Quantitative Hydrogeology, Academic Press.
2011), more sophisticated studies may be feasible in the future.
Demezhko, D.Y., Gornostaeva, A.A., Tarkhanov, G.V. & Esipko, O.A., 2013.
30,000 years of ground surface temperature and heat flux changes in
Karelia reconstructed from borehole temperature data, Bull. Geogr. Phys.
AC K N OW L E D G E M E N T S Geogr. Ser., 6, 7–25.
Diersch, H.J.-G. & Kolditz, O., 2002. Variable-density flow and transport
This project was funded by the National German Science Foun- in porous media: approaches and challenges, Adv. Water Resour., 25,
dation (Deutsche Forschungsgemeinschaft) under research grant 899–944.
842 C. Vogt et al.

Elina, G.A., Lukashov, A.D. & Yurkovskaya, T.K., 2010. Late Glacial Kageyama, M., Peyron, O., Pinot, S., Tarasov, P., Guiot, J., Joussaume,
and Holocene – Palaeovegetation and Palaeogeography of Eastern S. & Ramstein, G., 2001. The last glacial maximum climate over the
Fennoscandia, no. 4 in The Finnish Environment, The Finnish Environ- Europe and western Siberia: a PMIP comparison between models and
ment Institute, Natural Environment Centre. data, Climate Dyn., 17, 23–43.
Fernandez, M., Marzan, I. & Ramalho, A.C.E., 1998. Heat flow, heat produc- Kim, S.-J., Crowley, T.J., Erickson, D.J., Govindasamy, B., Duffy, P.B. &
tion, and lithospheric thermal regime in the Iberian Peninsula, Tectono- Lee, B.Y., 2008. High-resolution climate simulation of the last glacial
physics, 291, 29–53. maximum, Climate Dyn., 31, 1–16.
Forsström, P.-L., 2005. Through a glacial cycle: simulation of the Eurasian Kleman, J., Fastook, J., Ebert, K., Nilsson, J. & Caballero, R., 2013. Pre-lgm
ice sheet dynamics during the last glaciation, PhD thesis, Helsinki northern hemisphere ice sheet topography, Climate Past, 9(5), 2365–
University. 2378.
Giering, R. & Kaminski, T., 1998. Recipes for adjoint code construction, Kohl, T., 1998. Paleoclimatic temperature signals – can they be washed out?,
ACM Trans. Math Softw., 24(4), 437–474. Tectonophysics, 291, 225–234.
Glaznev, V.N., Kukkonen, I.T., Raevskii, A.B. & Jokinen, J., 2004. New data Kolditz, O., Ratke, R., Diersch, H.-J.G. & Zielke, W., 1998. Coupled ground-
on thermal flow in the central part of the Kola peninsula, Doklady Earth water flow and transport: 1. Verification of variable density flowand trans-
Sci., 396, 512–514. port models, Adv. Water Resour., 21, 21–46.
Golovanova, I.V., Puchkov, V.N., Salmanova, R.Y. & Demezhko, D.Y., 2008. Kozlovsky, Y.A. ed., 1987. The Superdeep Well of the Kola Peninsula,
A new version of the heat flow map of the Urals with paleoclimatic Springer-Verlag.

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
corrections, Doklady Earth Sci., 422(7), 1153–1156. Kremenetsky, A. & Ovchinnikov, L., 1986. The precambrian continental
Gorbatsevich, F.F., Ikorsky, S.V. & Zharikov, A.V., 2010. Structure and crust: its structure, composition and evolution as revealed by deep drilling
permeability of deep-seated rocks in the Kola superdeep borehole section in the USSR, Precambrian Res., 33, 11–43.
(SG-3), Acta Geodyn. Geomater., 7(2), 145–152. Krinner, G., Mangerud, J., Jakobsson, M., Crucifix, M., Ritz, C. & Svendsen,
Griewank, A. & Walther, A., 2008. Evaluating Derivatives: Principles and J.I., 2004. Enhanced ice sheet growth in Eurasia owing to adjacent ice-
Techniques of Algorithmic Differentiation, 2nd edn, no. 105 in Applied dammed lakes, Nature, 427, 427–432.
Mathematics, SIAM. Kukkonen, I.T., 1987. Vertical variation of apparent and palaeoclimatically
Gyllencreutz, R., Mangerud, J., Svendsen, J.-I. & Lohne, Ø., 2007. DATED – corrected heat flow densities in the central baltic shield, J. Geodyn., 8,
a GIS-based reconstruction and dating database of the Eurasian deglacia- 33–53.
tion, in Applied Quaternary Research in the Central Part of Glaciated Kukkonen, I.T. & Clauser, C., 1994. Simulation of heat transfer at the Kola
Terrain, vol. 46 of Special Paper, pp. 113–120, eds Johansson, P. & Sar- deep-hole site: implications for advection, heat refraction and paleocli-
ala, P., Geological Survey of Finland. matic effects, Geophys. J. Int., 116, 409–420.
Hagdorn, M.K.M., 2003. Reconstruction of the past and forecast of the Kukkonen, I.T., Gosnold, W.D. & Šafanda, J., 1998. Anomalously low heat
Future European and British Ice sheets and associated sealevel change, flow density in eastern Karelia, Baltic Shield: a possible palaeoclimatic
PhD thesis, University of Edinburgh. signature, Tectonophysics, 291(1–4), 235–249.
Harrison, S.P., Kutzbach, J.E., Prentice, C.E., Behling, P.J. & Sykes, M.T., Kukkonen, I.T. & Joeleht, A., 2003. Weichselian temperatures from
1995. The response of northern hemisphere extratropical climate and geothermal heat flow data, J. geophys. Res., 108(B3), 2163,
vegetation to orbitally induced changes in insolation during the last inter- doi:10.1029/2001JB001579.
glaciation, Quat. Res., 43, 174–184. Kukkonen, I.T., Rath, V., Kivekäs, L., Šafanda, J. & Čermak, V., 2011.
Hascoët, L. & Pascual, V., 2013. The Tapenade automatic differentiation Geothermal studies of the Outokumpu Deep Drill Hole, Finland: vertical
tool: Principles, model, and specification, ACM Tran. Math. Softw., 39(3), variation in heat flow and palaeoclimatic implications, Phys. Earth planet.
20:1–20:43. Inter., 188, 9–25.
Hättestrand, C. & Clark, C., 2006a. The glacial geomorphology of Kola Lemieux, J.-M., Sudicky, E.A., Peltier, W.R. & Tarasov, L., 2008. Dynam-
Peninsula and adjacent areas in the Murmansk region, Russia, J. Maps, ics of groundwater recharge and seepage over the Canadian landscape
2006, 30–42. during the Wisconsinian glaciation, J. geophys. Res., 113, F01011, doi:
Hättestrand, C. & Clark, C., 2006b. Reconstructing the pattern and style of 10.1029/2007JF000838.
deglaciation of Kola Peninsula, NE Fennoscandian Ice Sheet, in Glaciol- Luterbacher, J., Dietrich, D., Xoplaki, E., Grosjean, M. & Wanner, H.,
ogy and Earth’s Changing Environment, Chap. 39, pp. 199–201, ed. 2004. European seasonal and annual temperature variability, trends and
Knight, P., Blackwell. extremes since 1500, Science, 303, 1499–1503.
Heikkilä, M. & Seppä, H., 2003. A 11,000 yr palaeotemperature recon- Lynch, D.R., 2005. Numerical Partial Differential Equations for En-
struction from the southern boreal zone in Finland, Quat. Sci. Rev., 22, vironmental Scientists and Engineers: A First Practical Course,
541–554. Springer.
Hill, M.C. & Tiedeman, C.R., 2006. Effective Model Calibration: With Majorowicz, J. & Šafanda, J.S. Torun-1 Working Group, 2007. Heat
Analysis of Data, Sensitivities, Predictions, and Uncertainty, Wiley. flow variation with depth in Poland: evidence from equilibrium tem-
Hotchkiss, W. & Ingersoll, L., 1934. Post-glacial time calculations from perature logs in 2.9-km-deep well Torun-1, Int. J. Earth Sci., 97,
recent geothermal measurements in the Calumet copper mines, J. Geol., doi:10.1007/s00531-007-0210-2.
42, 113–142. Majorowicz, J. & Wybraniec, S., 2010. New terrestrial heat flow map of
Hubberten, H.W. et al., 2004. The periglacial climate and environment in europe after regional paleoclimatic correction application, Int. J. Earth
northern Eurasia during the last glaciation, Quat. Sci. Rev., 23(11–13), Sci., 100(4), 881–887.
1333–1357. Mangerud, J., Gyllencreutz, R., Lohne, O. & Svendsen, J.I., 2011. Glacial
Huyakorn, P.S. & Pinder, G.F., 1983. Computational Methods in Subsurface history of norway, in Quaternary Glaciations–Extent and Chronology: A
Flow, Academic Press. Closer Look, vol. 15 of Developments in Quaternary Sciences, pp. 279–
Jones, P.D., Lister, D.H., Osborn, T.J., Harpham, C., Salmon, M. & Morice, 298, eds. Ehlers, J., Gibbard, P.L. & Hughes, P.D., Elsevier.
C.P., 2012. Hemispheric and large-scale land-surface air temperature vari- Manning, C.E. & Ingebritsen, S.E., 1999. Permeability of the continental
ations: an extensive revision and an update to 2010, J. geophys. Res., 117, crust: implications of geothermal data and metamorphic systems, Rev.
D05127, doi:10.1029/2011JD017139. Geophys., 37, 127–150.
Jost, A., Lunt, D., Kageyama, M., Abe-Ouchi, A., Peyron, O., Valdes, P.J. & Marshall, S.J., 2012. The Cryosphere, Princeton University Press.
Ramstein, G., 2005. High-resolution simulations of the last glacial max- Mottaghy, D., 2007. Heat transfer processes in the upper crust: influence
imum climate over Europe: a solution to discrepancies with continental of structure, fluid flow, and paleoclimate, Doctoral dissertation, RWTH
palaeoclimatic reconstructions, Climate Dyn., 24, 577–590. Aachen University.
Fluid flow or palaeoclimate, Kola borehole 843

Mottaghy, D. & Rath, V., 2007. Paleoclimate from the surroundings of the Rühaak, W., Rath, V. & Clauser, C., 2010. Detecting thermal anomalies
Kola deep drilling site: influences of topography and fluid flow?, Geophys. within the Molasse Basin, Southern Germany, Hydrogeol. J., 18(8), 1897–
Res. Abstr., 9, EGU2007-A-02019. 1915.
Mottaghy, D., Popov, Y.A., Schellschmidt, R., Clauser, C., Kukkonen, I.T., Sass, J.H., Lachenbruch, A.H. & Jessop, A.M., 1971. Uniform heat flow
Nover, G., Milanovsky, S. & Romushkevich, R.A., 2005. New heat in a deep hole in the Canadian Shield and its paleoclimatic implications,
flow data from the immediate vicinity of the Kola superdeep borehole: J. geophys. Res., 76, 8586–8596.
vertical variation in heat flow density confirmed, Tectonophysics, Siegert, M.J. & Dowdeswell, J.A., 1998. Late Weichselian Glaciation of the
401(1–2), 119–142. Russian High Arctic, Quat. Res., 52, 273–285.
NEDRA, 1992. Characterization of crystalline rocks in deep boreholes. Siegert, M.J. & Dowdeswell, J.A., 2004. Numerical reconstructions of the
The Kola, Krivoy Rog and Tyrnauz boreholes, Tech. Rep. TR-92-39, Eurasian Ice Sheet and climate during the Late Weichselian, Quat. Sci.
SKB. Rev., 23(11–13), 1273–1283.
Neuzil, C.E., 2012. Hydromechanical effects of continental glaciation on Slagstad, T., Balling, N., Elvebakk, H., Midttømme, K., Olesen, O., Olsen,
groundwater systems, Geofluids, 12(1), 22–37. L. & Pascal, C., 2009. Heat-flow measurements in Late Palaeoprotero-
Nocedal, J. & Wright, S.J., 1999. Numerical Optimization, Springer. zoic to Permian geological provinces in south and central Norway and a
Person, M., McIntosh, J., Bense, V. & Remenda, V.H., 2007. Pleis- new heat-flow map of Fennoscandia and the Norwegian Greenland Sea,
tocene hydrology of North America: the role of ice sheets in reor- Tectonophysics, 473, 341–361.
ganizing groundwater flow systems, Rev. Geophys., 45, RG3007, doi: Smerdon, J.E., Pollack, H.N., Cermak, V., Enz, J.W., Kresl, M., Safanda,

Downloaded from http://gji.oxfordjournals.org/ at Dublin Institute for Advanced Studies on September 11, 2014
10.1029/2006RG000206. J. & Wehmiller, J.F., 2006. Daily, seasonal and annual relationships be-
Person, M., Bense, V., Cohen, D. & Banerjee, A., 2012a. Models of ice-sheet tween air and subsurface temperatures, J. geophys. Res., 111, D07101,
hydrogeologic interactions: a review, Geofluids, 12(1), 58–78. doi:10.1029/2004JD005578.
Person, M., McIntosh, J., Iverson, N., Neuzil, C.E. & Bense, V., 2012b. Speedy, R.J., 1987. Thermodynamic properties of supercooled water at 1
Geologic isolation of nuclear waste at high latitudes: the role of ice sheets, atm, J. Phys. Chem., 91(12), 3354–3358.
Geofluids, 12(1), 1–6. Svendsen, J.I. et al., 2004. Late quaternary ice sheet history of northern
Peyaud, V., Ritz, C. & Krinner, G., 2007. Modelling the Early Weichselian Eurasia, Quat. Sci. Rev., 23(11–13), 1229–1271.
Eurasian Ice Sheets: role of ice shelves and influence of ice-dammed Tarantola, A., 2004. Inverse Problem Theory. Methods for Model Pa-
lakes, Climate Past, 3, 375–386. rameter Estimation, Society for Industrial and Applied Mathematics
Phillips, S.L., Igbene, A., Fair, J.A., Ozbek, H. & Tavana, M., 1981. (SIAM).
A technical data book for geothermal energy utilization, Techni- Tarantola, A. & Valette, B., 1982a. Generalized nonlinear inverse problems
cal Report 12810 UC-66a, Lawrence Berkeley Laboratory, University solved using the least squares criterion, Rev. Geophys. Space Phys., 20(2),
of California, Berkeley, CA, Available at: http://adsabs.harvard.edu/ 219–232.
abs/1981STIN...8212614P (last accessed 19 August 2014). Tarantola, A. & Valette, B., 1982b. Inverse problems = quest for information,
Popov, Y.A., Pevzner, S.L., Pimenov, V.P. & Romushkevich, R.A., 1999. J. Geophys., 50, 159–170.
New geothermal data from the Kola superdeep well SG-3, Tectonophysics, Tarasov, L., 2013. GLAC-1b: a new data-constrained global deglacial ice
306, 345–366. sheet reconstruction from glaciological modelling and the challenge of
Provost, A.M., Voss, C.I. & Neuzil, C.E., 2012. Glaciation and re- missing ice, Geophys. Res. Abstr., 15, EGU2013-12342.
gional groundwater flow in the Fennoscandian shield, Geofluids, 12(1), Vasseur, G. & Lucazeau, F., 1983. Bounds on paleotemperatures and pale-
79–96. oclimatic corrections, Zentralbl. Geol. Paleontol., 1, 17–24.
Rall, L.B., 1981. Automatic Differentiation: Techniques and Applications, Vidal, H., 1984. The Kola super-deep borehole SG-3 – first look at the
vol. 120 of Lecture Notes in Computer Science, Springer. deepest hole of the world, GeoJournal, 9(4), 431–432.
Ramstein, G., Kageyama, M., Guiot, J., Wu, H., Hély, C., Krinner, G. & Vidstrand, P., Follin, S., Selroos, J.-O., Näslund, J.-O. & Rhén, I., 2013.
Brewer, S., 2007. How cold was Europe at the last glacial maximum? Modeling of groundwater flow at depth in crystalline rock beneath a mov-
A synthesis of the progress achieved since the first PMIP model-data ing ice-sheet margin, exemplified by the Fennoscandian Shield, Sweden,
comparison, Climate Past, 3, 331–339. Hydrogeol. J., 21(1), 239–255.
Rath, V. & Mottaghy, D., 2007. Smooth inversion for ground surface tem- Westaway, R. & Younger, P.L., 2013. Accounting for palaeoclimate and
perature histories: estimating the optimum regularization parameter by topography: a rigorous approach to correction of the British geothermal
generalised cross-validation, Geophys. J. Int., 171(3), 1440–1448. dataset, Geothermics, 48, 31–51.
Rath, V., González-Rouco, J. & Goosse, H., 2012. Impact of postglacial Wohlfarth, B., Lacourse, T., Bennike, O., Subetto, D., Tarasov, P., Demi-
warming on borehole reconstructions of last millennium temperatures, dov, I., Filimonova, L. & Sapelko, T., 2007. Climatic and environmental
Climate Past, 8, 1059–1066. changes in north-western Russia between 15,000 and 8000 yr BP: a re-
Rath, V., Wolf, A. & Bücker, M., 2006. Joint three-dimensional inversion view, Quat. Sci. Rev., 26(13–14), 1871–1883.
of coupled groundwater flow and heat transfer based on automatic dif- Wu, H., Guiot, J., Brewer, S. & Guo, Z., 2007. Climatic changes in Eurasia
ferentiation: sensitivity calculation, verification, and synthetic examples, and Africa at the last glacial maximum and mid-Holocene: reconstruction
Geophys. J. Int., 167, 453–466. from pollen data using inverse vegetation modelling, Climate Dyn., 29,
Rodgers, C.D., 1976. Retrieval of athmospheric temperature and composi- 311–329.
tion from remote measurements of thermal radiation, Rev. geophys. Space Zylkovskij, G.A., Robinson, B.A., Dash, V.B. & Trease, L.L., 1994. Models
Phys., 14, 609–624. and methods summary for the FEHMN application, Technical Report
Rolandone, F., Mareschal, J.-C. & Jaupart, C., 2003. Temperatures at the LA-UR-94-3787, Los Alamos Natioal Laboratory, Albuquerque NM,
base of the Laurentide Ice Sheet inferred from borehole temperature data, Available at: https://fehm.lanl.gov/pdfs/fehm_mms.pdf (last accessed
Geophys. Res. Let., 30, 1944, doi:10.1029/2003GL018046. 19 August 2014).

You might also like