Available online at www.sciencedirect.

com

Geochimica et Cosmochimica Acta 72 (2008) 5009–5021

www.elsevier.com/locate/gca

Calcite precipitation instability under laminar, open-channel flow

Ø. Hammer a,*, D.K. Dysthe a, B. Lelu a,b, H. Lund a,c, P. Meakin a,d, B. Jamtveit a
a
Physics of Geological Processes, University of Oslo, P.O. Box 1048, 0316 Blindern, Oslo, Norway
b
Ecole Normale Supérieure de Lyon, 46, allée d’Italie, 69364 Lyon Cedex 07, France
c
Norwegian University of Science and Technology, 7491 Trondheim, Norway
d
Idaho National Laboratory, P.O. Box 1625, Idaho Falls, ID 83415-2211, USA

Received 2 October 2007; accepted in revised form 22 July 2008; available online 22 August 2008

Abstract

We present a 2D numerical model for the growth of calcite from supersaturated aqueous solutions under laminar, open-
channel flow conditions. The model couples solution chemistry, precipitation at solution/calcite interfaces, hydrodynamics,
diffusion and degassing. The model output is compared with experimental results obtained using an oversaturated calcite solu-
tion produced by mixing CaCl2 and Na2CO3. The precipitation rate is observed to increase when the supersaturated solution
flows over an obstruction, leading to a growth instability that causes the formation of terraces. At relatively high flow rates,
the most important mechanism for this behaviour seems to be hydrodynamic advection of dissolved species either towards or
away from the calcite surface, depending on location relative to the obstruction, which deforms the concentration gradients.
At lower flow rates, steepening of diffusion gradients around protrusions becomes important. Enhanced degassing over the
obstruction due to shallowing and pressure drop is not important on small scales. Diffusion controlled transport close to
the calcite surface can lead to a fingering-type growth instability, which generates porous textures. Our results are consistent
with existing diffusive boundary layer theory, but for flow over non-smooth surfaces, simple calcite precipitation models that
include empirical correlations between fluid flow rate and calcite precipitation rate are inaccurate.
Ó 2008 Elsevier Ltd. All rights reserved.

1. INTRODUCTION interesting example which involves not only transport of

Precipitation of minerals from aqueous solution is often
studied primarily in terms of chemical kinetics and diffusion
(e.g. Buhmann and Dreybrodt, 1985). However, advective,
hydrodynamically controlled transport will also influence
precipitation rates in systems with relatively high-velocity
flow such as springs, streams, high-velocity hydrothermal
conduits and water pipes. For flow in the complicated
geometries found in nature, the coupling between hydrody-
namics and precipitation become complex, and the causal
relationships difficult to establish. A modeling approach is
therefore useful. Calcite precipitation in open channels,
forming travertine, is a particularly ubiquitous and
*
Corresponding author.
E-mail addresses: ohammer@nhm.uio.no (Ø. Hammer),
d.k.dysthe@fys.uio.no (D.K. Dysthe).
the precipitating solutes but also transport of CO2 and its
exchange with the atmosphere.
The increase in the calcite precipitation rate on protru-
sions under open-channel flow is responsible for a number
of pattern formation phenomena in hot springs, karst
caves and similar systems. Travertine terracing is the
prime example of this positive feedback instability, where
precipitation causes faster growth of the step. Several
mechanisms have been proposed for this phenomenon.
Buhmann and Dreybrodt (1985) developed a model in
which higher flow rates under turbulent conditions, which
may be found near an obstruction, lead to thinning of the
laminar boundary layer. This results in faster diffusion
across the boundary layer, which is rate limiting during
diffusion-limited precipitation. This phenomenon, con-
firmed experimentally by Liu and Dreybrodt (1997), was
also used by Wooding (1991) to explain travertine terrac-
ing. The dynamics of travertine precipitation generally

0016-7037/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.gca.2008.07.028
5010 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021
leads to an expansion of the flow channel (rather than
localization as is the case for erosion), producing a wide
and shallow sheet of water with low Reynolds numbers
and little turbulence (Baker and Smart, 1995). The equa-
tions of Buhmann and Dreybrodt (1985) are still valid,
but complete mixing in a turbulent core cannot be as-
sumed. Other authors (Chen et al., 2004) have suggested
that the reduced volume/surface ratio, caused by shallow-
ing, itself enhances CO2 degassing, and therefore leads to
increased precipitation. Agitation of the surface at and
downstream from the obstruction can also facilitate degas-
sing. Finally, pressure drop in areas of fast flow (the Ber-
noulli effect) could conceivably enhance degassing by
reducing gas solubility (Chen et al., 2004).
Some authors (Goldenfeld et al., 2006; Hammer et al.,
2007) have developed models that simply assume a linear
relationship between flow velocity and precipitation rate,
as observed in nature, without addressing the mechanism
for this correlation. The pattern formation demonstrated
in these models is primarily due to this relationship.
In this paper, we develop a full 2D model of carbonate
precipitation under laminar, open-channel flow over an
obstruction, without any a priori assumptions with regard
to the relationship between flow rate and precipitation rate.
The model couples fluid dynamics with a detailed simula-
tion of the solution chemistry of the carbonate system, cal-
cite precipitation kinetics, species transport (diffusion and
advection) and non-equilibrium degassing.
The results indicate that precipitation is enhanced at an
obstruction, even without turbulence. The mechanism in-
volves compression and deformation of concentration gra-
dients due to advection. The fluid flow rate seems less
important in controlling the precipitation rate under lami-
nar flow conditions, and the field observations of a correla-
tion between the two may be due in part to the negative
correlation between depth and the flow rate imposed by
hydrodynamics.
The calcium carbonate system in aqueous solution (e.g.
Buhmann and Dreybrodt, 1985) involves the conversion be-
tween CO2, H2CO3, H+ and HCO3  , where the latter can
further be assumed to be in equilibrium with CO3 2 . This
conversion involves two reaction paths:
CO2 þ H2 O $ H2 CO03 $ Hþ þ HCO3  ð1aÞ
and
CO2 þ OH $ HCO3  ð1bÞ
Other reactions taking place in solution are fast and can be
assumed to be in equilibrium.
At the water–calcite surface, precipitation and dissolu-
tion involve the three reactions:
CaCO3 þ Hþ $ Ca2þ þ HCO3  ð2aÞ
CaCO3 þ H2 CO03 $ Ca2þ þ 2HCO3  ð2bÞ
and
CaCO3 þ H2 O $ Ca2þ þ HCO  bon concentration excluding [CO2] and ½H2 CO3 0 ). The
3 þ OH ð2cÞ
Under typical conditions for calcite precipitation, with
pH > 6 and P CO2 < 0:01 atm, Reaction (2c) is dominant.
The OH used up by precipitation according to this equa-
tion is balanced by loss of H+ through the right-to-left reac-
tion path of Eq. (1a), or supplied directly through Eq. (1b),
producing CO2 in solution.
The focus of this study is not calcite precipitation
kinetics in isolation, but development of a better under-
standing of the detailed processes responsible for the
growth instability that leads to travertine terracing. We
therefore need to carry out modeling and experiments in
geometry and at a scale comparable to natural terracing
conditions, and to be able to map out the spatial and tem-
poral distribution of precipitation at this scale. Well devel-
oped travertine terraces in nature have minimum heights
of about 5 mm and lateral dimensions of 1 to several
100 cm. Travertine growth surfaces have RMS roughness
of the order 10 lm to 1 mm measured over 4 mm2 areas.
Open channel flow over well developed travertine terraces
has three main characteristics: (i) slow, laminar flow, (ii)
shallow water depth over terrace edges (2–5 mm), and
(iii) the flow direction, water flux and water depth up-
stream of, and over terrace edges vary slowly and contin-
uously due to the growth of the travertine. A wealth of
experimental studies of growth and dissolution of carbon-
ates and other minerals focus on controlling the hydrody-
namic conditions at the mineral surface. Examples include
rotating disk experiments (e.g. Compton and Daly, 1987)
and experiments performed with channel flow cell (CFC)
setups (Compton and Unwin, 1990). The CFC is similar
to that used in our work, but all studies that we are aware
of attempt to achieve homogeneous flow conditions on the
mineral surface. In this paper, we describe a study of the
growth instability caused by flow inhomogeneity. A wide
range of different techniques (with or without lateral spa-
tial resolution) have been used to quantify growth/dissolu-
tion kinetics in such cells. Since the early 90s the methods
of choice to study growth and dissolution with spatial res-
olution have been atomic force microscopy (AFM)
(Hillner et al., 1992), X-ray reflectivity (Chiarello et al.,
1993), and interferometry (van Enckevort, 1992). The geo-
chemical studies with these techniques have focused on the
nanometer scale. Of these methods, only low coherence
interferometry can be used to characterize surface rough-
ness on the millimeter scale and growth inhomogeneities
on lateral scales of the order of 1 cm, as found on
travertine surfaces.
2. METHODS
2.1. Carbonate system model
The solution chemistry of the carbonate system is simu-
lated with a coupled reaction–diffusion–advection model
based on the three molar concentrations [CO2], [Ca2+]
and the sum of the carbonate and bicarbonate concentra-
tions ½c ¼ ½HCO3   þ ½CO3 2  (that is, total dissolved car-
rate of CO2 concentration change due to conversion of
CO2 into H+ and HCO3  (Eq. (1)) and the transport of
CO2 in bulk solution are given by
 Calcite precipitation instability under open-channel flow 5011

o½CO2  ½c ¼ ½HCO3   þ ½CO3 2  together with Eq. (7e) of

¼ ðk 1 þ k 2 ½OH Þ½CO2  þ ðk a ½Hþ  þ k 2 Þ
ot Buhmann and Dreybrodt (1985), we obtain the following
 ½HCO3   þ DCO2 r2 ½CO2   ur½CO2 ; ð3Þ equations:

where u is the fluid velocity [modified from Buhmann and aHþ cCO3 ½c
½HCO3   ¼ ; ð7aÞ
Dreybrodt (1985, Eq. (5a))]. The rate constants k1 and k2 K 2 cHCO3 þ aHþ cCO3
(forward rate of the first part of Eq. (1a), and forward rate K 2 aHCO3
of Eq. (1b)), k2 (reverse rate of Eq. (1b)) and the diffusion ½CO3 2  ¼ ; ð7bÞ
aHþ cCO3
constant DCO2 are given in Table 1. The rate constant ka ac-
counts for the fast dissociation of carbonic acid (second where K2 is the equilibrium constant for the dissociation of
part of Eq. (1a)) and is calculated according to Kaufmann bicarbonate (Table 1), and
and Dreybrodt (2007): ½OH  ¼ K W =½Hþ : ð7cÞ
k 1 cH cHCO3 In these equations, a are activities, calculated from concen-
ka ¼ ; ð4Þ
K5 trations using the corresponding activity coefficients. We
with the values of k1 (reverse rate of first part of Eq. (1a)) then compute the total charge concentration q:
and K5 (equilibrium constant for the fast second reaction of q ¼ 2½Ca2þ  þ ½Hþ   ½HCO3    2½CO3 2   ½OH : ð7dÞ
Eq. (1a)) given in Table 1. The activity coefficients c are cal-
culated from ionic strengths according to the extended For the simulations of the experimental model (see be-
Debye–Hückel theory (Dreybrodt, 1988). We also have low), the concentrations [Na+] and [Cl] are also in-
cluded in Eq. (7d). At near-neutral pH, Eq. (7d) can be
o½Ca2þ  approximated by
¼ Dr2 ½Ca2þ   ur½Ca2þ  ð5Þ
ot
q ¼ 2½Ca2þ  þ ½HCO3  : ð7eÞ
and
2+
Since [Ca ] and [c] are given by Eqs. (5) and (6), Eqs. (7a)–
o½c
¼ ðk 1 þ k 2 ½OH Þ½CO2   ðk a ½Hþ  þ k 2 Þ½HCO
3 (7d) can be combined to produce an explicit function
ot
q([H+]):
þ Dr2 ½c  ur½c ð6Þ
q ¼ 2½Ca2þ  þ ½Hþ 
 2 !
with ½c ¼ ½HCO3  þ ½CO3  as defined above. Eq. (6) is
the sum of Eqs. (5b) and (5c) in Buhmann and Dreybrodt cH ½Hþ cCO3 ½c K 2 cHCO3
 1þ2
(1985). K 2 cHCO3 þ cH ½Hþ cCO3 cH ½Hþ cCO3
The initial concentrations [CO2]0, [Ca2+]0 and [c]0 are set
 K W =½Hþ : ð7fÞ
to values known from the chemistry of the starting solution
+
at the upstream source. Subsequent compositions are then Charge balance requires that q([H ]) = 0, and we therefore
calculated from Eqs. (3)–(6). For each time step, the con- search for a root of this equation using a bisection algo-
centrations of the charged species H+, OH, rithm (Press et al., 1992). Inserting the resulting [H+] back
HCO3  and CO3 2 to be inserted into Eqs. (3)–(6) are into Eqs. (7a)–(7c), we then finally have the concentrations
found by assuming that they are in equilibrium with each [H+], ½HCO3  ; ½CO3 2  and [OH] to be used in Eqs. (3)–
other (due to fast reactions). This computation uses the fol- (6). After the bisection iteration has been used to determine
lowing method. First, ½HCO3  ; ½CO3 2  and [OH] are ex- [H+], the carbonic acid concentration can also be calculated
pressed as explicit functions of [H+] and [c]. Using using

Table 1
Chemical and physical parameters used in the model
Parameter Unit Value Reference
1
log10 k1 s 329.850  110.54 log10 T  17265.4/T Usdowski (1982)
log10 k M1 s1 329.850  110.54 log10 T  17265.4/T Usdowski (1982)
log10 k2 M1 s1 13.635  2895/T Usdowski (1982)
log10 k2 s1 14.09  5308/T Usdowski (1982)
log10 KW M2 22.801  4787.3/T  0.010365T  7.1321 log10 T Harned and Hamer (1933)
log10 K2 M 107.8871 + 5151.79/T  0.03252849T + 38.92561 log10 T  563713.9T2 Plummer and Busenberg (1982)
K5 M 1.707  104 Plummer and Busenberg (1982)
P atm
CO2 atm 3.5  104
log10 KH M atm1 108.3865  6919.53/T + 0.01985076T  40.45154 log10 T + 669,365T2 Plummer and Busenberg (1982)
k cm h1 10 Liss and Merlivat (1986)
log10 j1 cm s1 0.198  444/T Plummer et al. (1978)
log10 j2 cm s1 2.84  2177/T Plummer et al. (1978)
log10 j3 mM cm2 s1 5.86  317/T (valid for TC 6 25) Plummer et al. (1978)
DCO2 cm2/s 105(0.56 + 0.058TC) Buhmann and Dreybrodt (1985)
D cm2/s 0:7DCO2 Buhmann and Dreybrodt (1985)
Values are converted to SI units by the program. T is temperature in K, TC in centigrades.
5012 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021
 
cH ½Hþ  F CO2 ¼ k K H P atm
½H2 CO3 0  ¼ : ð8Þ CO2  ½CO2  ; ð17Þ
K 5 cH2 CO3 0
where KH is Henry’s constant. An estimated gas transfer
At the calcite surface, the precipitation (negative) or disso- velocity of k = 10 cm/h = 2.8  103 cm/s (Liss and
lution (positive) flux into the fluid, in mmol/cm2/s, is mod- Merlivat, 1986) was used. This gives the following bound-
eled using the PWP equation (Plummer et al., 1978; ary conditions at the air–water interface Caw:
Buhmann and Dreybrodt, 1985) 
o½CO2 
F ¼ j1 aHþ þ j2 aH2 CO3 þ j3  j4 aCa2þ aHCO3  ; ð9Þ ¼ F CO2 ; ð18Þ
ot Caw

where o½Ca2þ 
¼ 0; ð19Þ
aH2 CO3 ¼ aH2 CO3 0 þ aCO2 : ð10Þ ot Caw

The constant j4 depends on T and P CO2 . Kaufmann and and

Dreybrodt (2007) used the following expression, derived 
o½c
by fitting to the empirical data of Plummer et al. (1978): ¼ 0: ð20Þ
ot Caw

2:375 þ 0:025T C P CO2 P 0:05 atm;
log10 j4 ¼ These boundary conditions are different from those given
2:375 þ 0:025T C þ 0:56ð log P CO2  1:3Þ P CO2 < 0:05 atm;
ð11Þ
by e.g. Buhmann and Dreybrodt (1985). These authors used
F CO2 ¼ F , assuming that the amount of CO2 released by
where TC is the temperature in degrees centigrade and P CO2 precipitation through the reactions of Eqs. (1) and (2c) is
is in atmospheres. balanced by the amount of CO2 lost to the atmosphere by
This approximation is valid for temperatures up to degassing. This balance is a reasonable assumption in the
35 °C. The equation for j4 given by Buhmann and case of plug flow over a large, flat surface, but it cannot
Dreybrodt (1985) contains a printing error (Dreybrodt, pri- be used in the general case in which advection and concen-
vate communication). We also tested a simpler equation tration fields vary in space. No physical mechanism can
due to Compton and Pritchard (1990, Eq. (53)): maintain equal fluxes between any two given points on
F CP ¼ 9:5  1011  7:14  103 ½Ca2þ ½CO3 2 ; ð12Þ the water–calcite and the water–air interfaces in the pres-
ence of complex flow and evolving water chemistry. We
with FCP in mol/cm2/s. This gave similar precipitation pat- must therefore resort to basic physical principles such as
terns, but with lower overall precipitation rates than the the solution chemical kinetics and fluid dynamics (Eqs.
PWP model. Finally, we implemented the equation of (3)–(7)), the kinetics of precipitation/dissolution (Eq. (9))
Brown et al. (1993, Eq. (18)): and the transport of CO2 across the aqueous/air interface
(Eq. (17)).
K sp  ½Ca2þ ½CO3 2 
F B ¼ k b K Ca K CO3 ; ð13Þ The constants used in these equations are summarized in
ð1 þ K Ca ½Ca2þ Þð1 þ K CO3 ½CO3 2 Þ Table 1.
again with FB in mol/cm2/s, and with To calculate the velocity vector field u, the Navier–
k b K Ca K CO3 ¼ 990 cm4 =mol=s, KCa = 106 cm3/mol and Stokes equations for open-channel (free-surface) flow in a
K CO3 ¼ 3  107 cm3 =mol (Brown et al., 1993). KCa and vertical 2D slice are solved using the code NaSt2D (Griebel
K CO3 are Langmuirian adsorption constants. This equation et al., 1997). The reactive transport equations (3)–(6) are
also gave precipitation patterns similar to those obtained approximated using explicit finite difference methods, and
from the PWP model, but with precipitation rates an order the reaction and transport terms are calculated in the same
of magnitude lower. The results reported below were ob- time step (one-step method). The advection term is imple-
tained using the PWP equation. mented with a second-order Lax–Wendroff difference
The precipitation of one mole of CaCO3 removes of one scheme (Lax and Wendroff, 1960).
mole of Ca2+ and one mole of c from the solution at the
water–calcite interface Cwc (and conversely for dissolution). 2.2. Experiments
This gives the following boundary conditions at the water–
calcite interface: The main barrier to reproducing natural carbonate
 growth patterns in the laboratory is the difficulty in main-
o½CO2 
¼ 0; ð14Þ taining control over the concentrations of Ca2+ and
ot Cwc CO3 2 in aqueous solutions over long periods of time. An

o½Ca2þ  attempt was made to reproduce the conditions typical of
¼ F; ð15Þ
ot Cwc hydrothermal systems, by saturating water with CO2 at
high pressure, passing the fluid through crushed limestone,
and releasing it onto a surface and allowing CO2 to escape to
 the atmosphere. Although a significant amount of calcium
o½c
¼ F: ð16Þ carbonate was dissolved, the precipitation rate was never
ot Cwc
sufficient to observe growth patterns of interest. We there-
The CO2 flux across the air–water interface, in mmol/cm2/s fore turned to a different method for producing calcite
(negative for degassing) is given by supersaturation: mixing Na2CO3 and CaCl2 solutions at
 Calcite precipitation instability under open-channel flow 5013

the inlet of a long channel flow cell. As mentioned in the
introduction, open channel flow over well developed traver-
tine terraces is slow and laminar with a water depth of 2–
5 mm over terrace edges. The fact that the water flux and
water depth upstream of terrace edges vary slowly and con-
tinuously (due to the growth of the travertine) indicates that
the exact flow patterns upstream of the terrace edges are not
important for the instability mechanism. A significant
observation that has bearing on the construction of our
flow channel is that travertine terraces do not form imme-
diately adjacent to the fluid source. The flow channel was
therefore chosen to be almost one meter long.
Na2CO3 and CaCl2 solutions were brought together and
mixed at the top of a glass ramp with side walls, and the
supersaturated solution flowed down the ramp and over a
glass obstacle near its lower end (Fig. 1a). The solutions,
each approximately 4 mM, were prepared by injecting con-
centrated solutions into tubes with pure water flow rates of
0.5–4 ml/s and the two tubes containing diluted solutions of
Na2CO3 and CaCl2 were joined at the end of a common
tube where the solutions mixed. It was therefore possible
to maintain a steady, continuous flow on the ramp while
the concentrations of Na2CO3 and CaCl2 were controlled.
The ramp had a width of 59 mm with 1 cm high side walls,
and it was inclined at 0.8–1.7°. Various obstacles were
placed in the flow, 0.7 m down the ramp from the inlet
point. The initial velocity down the ramp was 3–5 mm/s.
Small amounts of grease or other impurities caused the
water to wet the surface incompletely and made the flow
pattern inhomogeneous. This did not perturb the instability
under investigation, but in order to get quantitative results
it is simpler to use a flow that is homogeneous along the
step edge. To obtain uniform wetting properties everywhere
along the ramp we removed the deposited carbonate with
hydrochloric acid, and sand blasted the ramp before each
experiment.
The flow obstacle had a rectangular cross-section (in the
x–z-plane in Fig. 1) with a length of 6–9 mm and a height of
2.9 mm. Both corners facing up towards the water flow
were ground to a 45° bevel of length 0.5 mm. Different sur-
face treatments to obtain homogeneous flow and good opti-
cal contrast between a dark background and the light
precipitate were tried. These included polished glass, sand
blasted glass and black matte lacquered surfaces.
We used two methods to measure the thickness of the
calcite precipitate on the solid surfaces on and around the
flow obstacle: white light interferometry (WLI) and imaging
of scattered light intensity. WLI is a standard technique for

a
camera

z y
10 cm
x

b z image c
x

lamp lens
d

air
e
water

glass
glass

Fig. 1. Sketch of the experimental setup for calcium carbonate growth. (a) 0.04 M solutions of Na2CO3 and CaCl2 were injected at the top of
a ramp, which was inclined at 0.8–1.7°. Flow obstacles were placed 0.6–0.7 m downslope from the top of the ramp, and a camera (C) was
mounted to follow the carbonate growth near and on the obstacle. (b) Geometry of illumination and imaging such that specularly reflected
light does not enter the imaging lens. (c)–(e) Different contributions to imaged light intensities. (c) The three planes reflecting light before
precipitation. (d) Forward dominated scattering (reflected and refracted) of first and second generation. (e) Varying contribution of specular
reflection and scattered intensity in different directions as function of thickness of precipitate.
5014 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021
measuring surface topography with nanometer resolution.
We used a Wyco NT1100 white light interferometer with
5 and 50 Wyco objectives. Most mineral growth/disso-
lution studies depend on the high resolution capability of
this technique (e.g. Arvidson et al., 2003). There are three
potential problems when this technique is applied to the
measurement of porous, microcrystalline precipitates with
height differences reaching the millimeter scale. The tech-
nique measures only the topography of the outer surface
and gives no information about the porosity of the mate-
rial. Secondly, in order to measure the thickness the posi-
tion of the bottom or substrate must be known. This is
simply achieved by using a flat substrate which facilitates
determination of the height of the substrate at all lateral
positions. Finally, small precipitated crystals are randomly
oriented and many parts of the surfaces are therefore at
high angles to the substrate plane (the plane normal to
the objective axis). High inclination angle surfaces may only
be measured by high numerical aperture (NA) objectives,
and even with a 50, 0.55NA objective many parts of the
surfaces will not be registered. After the measurement, the
topographic map is interpolated to fill the entire surface
(to estimate the surface heights in the high inclination angle
parts of the surface) and then the average height relative to
the substrate is calculated. In most cases we believe the
interpolation underestimates the average precipitate thick-
ness, and we estimate the accuracy of the calculated precip-
itate thickness to be ±30%. Since each single topography
measurement covers only 186  248 lm several measure-
ments are needed to obtain a representative average of
the precipitate thickness. For large and relatively homoge-
nous samples the standard deviation of repeated measure-
ments is 10–20%. In addition to these sources of error,
rough substrates (lacquered, ground or sand blasted sur-
faces) make the determination of the reference surface less
accurate.
Outside of the flat top surface of the flow obstacles the
WLI technique is not practical. Although the WLI can be
used in a fluid cell it is not practical to use it in situ because
any perturbation of the water level would change the mea-
sured topography. In order to estimate the precipitate
thickness all along the flow direction on and around the
flow obstacle a non-intrusive measurement technique that
is sensitive in the 1–100 lm vertical range with 10–100 lm
lateral resolution is needed. The natural choice was an opti-
cal technique, and we chose to analyze the scattered light
intensity image obtained from the precipitate. This in situ
measurement of precipitate thickness in the x–y-plane is
based on scattering of light by small, randomly oriented
calcite crystals. Fig. 1b shows how the lamps, flow obstacle
and imaging system were configured. The dashed lines indi-
cate the field of view (FOV) and the arrows indicate that
light rays from the lamps do not contribute to the recorded
intensity in the image when reflected specularly from the
surfaces inside the FOV. Fig. 1c and d indicates typical dis-
tributions of reflected light for two of the surface treatments
used. The lengths of the arrows indicate the reflected (re-
fracted) intensity from different interfaces and in different
directions. When the flow obstacle is polished (Fig. 1c)
the glass–water interface reflects specularly and the glass–
glass interface, which was dyed black, scatters forward with
a very low intensity. When the obstacle is sand blasted (see
Fig. 1d) the glass–water interface scatters light with a strong
forward intensity distribution. The lacquered surface also
scatters forward, but with a much lower intensity. When
small, randomly oriented crystals grow on the glass–water
interface they become centers of light scattering (see
Fig. 1e). A single crystal scatters light in a few, discrete
directions depending on orientation, but the refracted light
is in turn reflected from the original surface of the obstacle.
Averaged over random orientations and including the con-
tribution from the original surface the scattering intensity
has a strong forward distribution in the specular direction.
As an increasing number of small crystals grow ‘randomly’
on top of each other, the scattering distribution broadens
and the intensity in the direction of the imaging lens in-
creases. When the local volume of precipitate exceeds the
volume of scatterers contributing to the imaged intensity
will no longer increase, because the recording system is sat-
urated. Different flow obstacle surface treatments were tried
out to test the robustness of the technique.
A Nikon D100 with a 60 mm Nikkormat lens was used
and the images were recorded in the Tiff 8 bit format. The
2000  3000 pixel red–green–blue (RGB) images were aver-
aged to 2000  3000 grayscale images and treated as fol-
lows: The images were rotated, typically by a few degrees,
to exactly match the y-axis of the image with the down-
stream step edge. The recorded intensities, I (measured in
8 bits: 0 = no light recorded, 255 = maximum intensity)
were averaged along the y direction to obtain a vector of
3000 intensity values corresponding to about 20 mm in
the x-direction. These vectors were smoothed using a 10
pixel moving average yielding a 300 pixel intensity profile
in the x-direction. Then the intensity profile at time 0 was
subtracted from the intensity profiles at all later times to
obtain the diffuse scatter intensity profile in the direction
of the lens.
In order to obtain precipitate thickness from the inten-
sity profiles a series of calibration experiments was per-
formed. The calibration experiments confirmed our prior
observation that precipitation occurs as small crystallites
that scatter light and not as more-or-less uniform layer
growth on a substrate. The very first crystallites have a pre-
ferred orientation dictated by the substrate, but long before
the substrate surface is covered the crystallites have a high
degree of randomness in their orientations. Fig. 2 shows the
results of the calibration experiments that simply compare
measurement using the two techniques (WLI and scattered
light intensity) for the same precipitates. There is a region
over which the scattered light intensity depends linearly
on the precipitate thickness. At low thickness (<1 lm) the
reason for the deviation is not clear. At high thickness
(>50 lm) the scattered intensity becomes saturated as ex-
plained above. In the intermediate region the relation be-
tween thickness and intensity is characterized by a single
constant. Each set of experiments with equal imaging con-
ditions (surface quality, dry/wet, illumination, shutter
speed, camera lens, camera distance, etc.) has a distinct cal-
ibration constant a that must be determined by a few ex situ
topography measurements with the WLI.
 Calcite precipitation instability under open-channel flow
150
100
50
0 obstruction was also evident.
0 20 40 60 80
Fig. 2. Calibration of precipitation thickness from imaged scatter-
ing intensity. The thickness in lm was measured by WLI. The
intensity was measured with 8 bits (0–255) resolution, I0 is the
intensity before precipitation and a is a proportionality constant
for each experiment series with equal imaging conditions. Data
points with squares, asterisks and circles are results of different
calibration experiments and the solid line is the fitted linear
calibration curve. Squares: Ex situ simultaneous imaging of four
distinct experiments, each with homogeneous distribution of
precipitate and averages of 35 topography measurements with
a = 1. Asterisks: the same experiments using in situ imaging with
a = .5. Circles: in situ imaging of one experiment with inhomoge-
neous precipitation and 16 topography measurements at different
sites on the sample with a = 1.
3. RESULTS
3.1. Model results
The model was run with water chemistry close to that
determined near the source of the Troll hydrothermal
spring, Svalbard (Hammer et al., 2005). The model was ini-
tially run with a low flow velocity of 2 mm/s in a smooth
channel of depth 0.5 cm and length 100 cm, until a steady
state was reached after 500 s. In this initialization phase,
the diffusion constants were set to 50 times their nominal
values, to simulate the effects of vertical mixing due to a
rough substrate (in the following runs the diffusion con-
stants are not increased in this way). At the source, the
log of the calcite saturation index SI = 0.04, pH 6.50 and
the calcite precipitation rate was F0 = 0.05 mg/cm2/day,
while at the lower end of the channel, degassing led to val-
ues of SI = 0.57, pH 7.15 and F0 = 2.07 mg/cm2/day. We
use
ðCa2þ ÞðCO3 2 Þ
SI ¼ log10 ; ð21Þ
K sp
with the calcite solubility product Ksp = 3.35  109 (molar
units).
The profiles for [Ca2+], [CO2] and [c] were extracted
from the lower end of the modeled channel and used as in-
put for a second run. The depth-averaged concentrations at
this point are [Ca2+] = 3.9 mM (130 mg/l), [CO2] =
0.77 mM and [c] = 7.8 mM. In the second run, the modeled
domain was 3.0 cm long and 0.6 cm deep, and a 150  30
grid was used. A rectangular protrusion obstructed the
5015
flow. The computed velocity field (absolute values) is shown
in Fig. 3. Clearly, the protrusion has a number of effects on
the flow, including shallowing and consequently an increase
in velocity above and especially immediately downstream
from the protrusion. Both of these parameters have been
proposed as possible causes of increased calcite precipita-
tion – shallowing by the mechanism of enhanced degassing
(Chen et al., 2004), and velocity through the thinning of a
boundary layer through which solutes must diffuse in the
case of turbulent flow (Buhmann and Dreybrodt, 1985).
A standing wave (hydraulic jump) downstream from the
The carbon dioxide concentration field and the calcite
precipitation rate along the water–calcite interface are
shown in Figs. 4a and 5a. The most important feature is
the compression of the CO2 concentration profile and
resulting steepening of the CO2 concentration gradient over
the obstruction. This enhances CO2 removal from the cal-
cite surface, leading to higher precipitation rates. Interest-
ingly, high precipitation rates are found not only in the
area of highest flow velocity around the downstream corner
of the step, but also at the upstream corner and in the area
some distance downstream from the obstacle. The eddy be-
hind the step produces a stagnant zone and a reduction in
precipitation rate. The highest rates are at the corners of
the step, where the concentration gradients near to the sur-
face are steep. Also, the precipitation rate is elevated down-
stream because of the mixing. This precipitation
distribution would lead to a downstream migration of the
step, together with outwards and upwards growth of the
step edge.
Jamtveit et al. (2006) used stable isotopes to infer a cal-
cite precipitation rate of 14 mm/year at terrace edges in the
Troll system. Using a density of 2.5 g/cm3 for travertine,
this corresponds to a precipitation rate of about 9.6 mg/
cm2/day, or about four times the modeled rate. On the
other hand, direct field measurement taken in a microter-
raced location with a pH similar to that used in the model,
indicated a rate of 0.72 mg/cm2/day, or about one fourth of
the modeled rate (Hammer et al., 2005, sample 5). The
modeled precipitation rate is therefore within the range of
field measurements. Of course, the natural system has high
spatial variability due to its complicated geometry, biolog-
ical activity and large-scale trends in water chemistry.
In Fig. 4b, the flow rate has been reduced to zero while
keeping the water surface topography the same as that in
Fig. 4a, to facilitate comparison. In this simulation, the
concentration fields were initially homogenous, and diffu-
sive gradients evolved over time. To the left of the obstruc-
tion, CO2 concentrations are elevated because of a greater
water depth. In this case of pure diffusion, the elevation
of the precipitation rate on the flat top of the obstruction
is slight. This result shows that although reduction in water
depth may cause some increase in precipitation rate due to
enhanced degassing, this effect is relatively small compared
with the increase in precipitation caused by advection, at
least under the modeled conditions.
Figs. 4c and 5c show the results using a rounder obstruc-
tion. Here the connection between flow rate and precipita-
tion rate is even less marked, with high precipitation rates
5016 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021

Fig. 3. Absolute values of flow velocity (mm/s) computed using the hydrodynamic model. Flow from left to right.

Fig. 4. CO2 concentrations in the model. (a) Flow past a rectangular obstruction with parameters approximating natural hydrothermal spring
conditions. (b) Flow rate forced to zero (pure diffusion). (c) Flow past a step with a rough surface. (d) Parameters approximating experimental
conditions.

on the upstream flank of the obstruction. In addition, pre-
cipitation rates are highly variable on both flanks, where
the surface is rough (here the spatial discretization grid is
used to advantage by using it to represent a real, stepped
morphology). Each peak corresponds to a small, protrud-
ing corner of the surface, where the diffusive concentration
gradients are steep.
Figs. 4d and 5d show flow over a step with geometry and
chemical concentrations closer to those used in the experi-
ment. The depth-averaged concentrations at the inflow
boundary were [Ca2+] = 2.0 mM (79 mg/l),
[CO2] = 3.1  105 mM and [c] = 7.1 mM. In this alkaline
(pH 10–11) regime, CO2 concentrations were very low
and there was, in fact, ingassing from the atmosphere.
The CO2 distribution is likely to be a less important control
on calcite precipitation rate under such conditions. In this
solute, there will be some binding of ions into species such
as NaCO3  . Modeling using the Phreeqc ver. 2.15 software
(Parkhurst, 1995) shows that ½NaCO3   will be on the order
of 10% of ½CO3 2 , and this was ignored in the simulation.
Other species involving Na+ and Cl are negligible. The
precipitation rate distribution is comparable to that of
Fig. 5a, but with a more marked peak in the well-mixed
area downstream from the obstruction.
A modeling experiment was also carried out to study ef-
fects of local degassing and possibly high gas transfer rates
(Eq. (17)) in areas of high flow velocity, due to e.g. agitation
(Liss and Merlivat, 1986). In this experiment, transfer
velocities were made dependent on surface flow rate with
a large factor of proportionality, causing transfer velocities
to increase by a factor 100 in regions of fastest flow. The
resulting precipitation pattern was indistinguishable from
that for a constant transfer velocity, and therefore the re-
sults are not shown in the figures. Increasing the flow rate
 Calcite precipitation instability under open-channel flow 5017

4
a
3

1
4
b
Precipitation (mg/cm2/day)

1
4
c
3

1
100
d
50

0
0 10 20 30
mm
Fig. 5. Calcite precipitation rates, in mg/cm2/day, corresponding to the model runs shown in Fig. 4.

to 20 mm/s gave patterns that were similar to those ob- travertine (Guo and Riding, 1998). It should be emphasized
tained at low flow rates, except that the precipitation max- that no matter the surface treatment (polished glass, sand
ima at the corners of the protrusion were relatively reduced blasted glass, organic spray paint) there is no measurable
(not shown). barrier to precipitation of the first calcite crystal seeds.
Crystals form readily anywhere on any available interface.
3.2. Experimental results Fig. 6 shows rough spheroidal calcite growth (cf. Jettest-
uen et al., 2006) in the stagnant water upstream from the
3.2.1. Qualitative features of calcite growth in experiments injection point of the fluid. The porous crystal agglomerates
While the fluid concentrations were optimized for fast suggest a high rate of crystal nucleation. On the upstream
calcite growth we observed that at high concentrations side of the step edge, the precipitate is similarly porous
(>4 mM) there was homogeneous nucleation with a varying and with rounded growth shapes. This is consistent with
induction time and no heterogeneous crystal growth on the natural travertine steps that have a hard, smooth down-
substrate. When the concentration was reduced below stream ‘‘lip” and a more porous upstream character.
4 mM heterogeneous calcite growth on the ramp would
commence as soon as the homogeneous nucleation ceased.
This suggests that travertine growth by agglomeration of
homogeneously nucleated calcite particles in the fluid flow
is not likely. This result deserves further study to check un-
der which conditions this may be true in nature. On the
other hand, calcite would grow on any water interface
(water–solid, water–air). Under some conditions ‘‘calcite
ice” was formed on the water–air interface (as can be ob-
served for example in the Mammoth Hot Springs in Yel-
lowstone). Calcite was also observed to grow on the gas–
water interface of gas bubbles produced from degassing
of N2 or CO2. Such bubbles increase the area of the gas–
water interface and thereby the total precipitation rate in
that region, creating an additional feedback mechanism
for localization of precipitation. Spherical holes, due to car- Fig. 6. Rough spheroidal growth of calcite in stagnant water, after
bonate growth on gas bubbles, are very common in natural 2 days.
5018 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021
Fig. 7 illustrates the feedback between fluid flow and cal-
cite growth, as fluid pathways change continuously due to
growth on the rim. Upstream from the obstruction (left
side) calcite grows as small clusters. This may be attributed
to a high nucleation barrier (the fast nucleation on the
obstruction on the other hand makes this an unlikely expla-
nation) or a fingering instability upstream into the concen-
tration gradient.
3.2.2. Quantitative experimental measurements of growth
rate distribution
In the experiments reported here, the ramp was inclined
at 1.7° and the flow rate was 1.8 ml/s. A pure water flow
was turned on to wet the ramp, and the lighting and camera
were adjusted before Na2CO3 and CaCl2 were injected into
the flow. Images focused on the rectangular flow obstacle
(e.g. Fig. 8d) were taken every 30 s during experiments that
lasted for 1–5 h.
Fig. 8e indicates the regions shown in Fig. 8a and f, and
shows that the depth of focus is sufficient to simultaneously
map localized precipitation at the top of the obstacle and
immediately to its right. The images in Fig. 8a (ex situ white
light interferometer) and Fig. 8f (ex situ optical image)
show that the precipitate consists of small crystals with a
narrow size distribution (typically 10–30 lm) and fairly ran-
dom orientations. Thickness profiles at 20, 40, 60 and
80 min for two experiments with different surface treatment
are displayed in Fig. 8b and c. We concentrated on the first
hours of carbonate growth for three reasons: (i) The nonlin-
ear relationship between precipitate thickness and intensity
at later times; (ii) extreme difficulty in measuring the distri-
bution of precipitate at later times; and (iii) the growth of a
rim protruding into the flow and perturbing it.
The vertical axis in Fig. 8b and c shows the precipitate
thickness in micrometers. The linear mapping from light
intensity to precipitate thickness was obtained by removing
the rectangular obstacle from the ramp after the experiment
and measuring the topography by WLI ex situ. It is clear
Fig. 7. Fluid flow (left to right) pattern over terrace rim showing
how precipitation builds up to block the flow. The black traces are
due to ink poured into the flow. Upstream the ink is fairly evenly
distributed over the ramp. Across the precipitated travertine rim
there are only two openings through which the water can flow.
from Fig. 8a and f, showing a porous agglomerate of small
particles, that it is difficult to measure the thickness of the
precipitate locally. Fig. 8a shows a WLI topography image,
obtained with a 5 objective, of a region of the obstacle
indicated in Fig. 8e. Inset Fig. 8ai is a higher resolution im-
age (using a 50, 0.55NA objective to capture light re-
flected from surfaces at larger angles) of the part of the
large image. Inset 8aii was obtained from the dataset shown
in inset Fig. 8ai, but with two-dimensional interpolation to
fill in the unresolved pixels (black regions in the images). In-
set Fig. 8aii was used to determine the calibration factor a
of the experiment, which was used to calculate the precipi-
tate thickness profiles shown in Fig. 8b and c. Fig. 8b shows
precipitate thickness profiles at different times for a black
lacquered surface and Fig. 8c shows the same for the sand
blasted surface.
After 80 min, the average thickness at the top of the
obstacle was about 5 lm (a growth rate of 90 lm per
day). Considering the porous texture of the experimental
precipitate, this compares fairly well with the computer
model result of about 40 mg/cm2/day of dense precipitate
(Fig. 5d), or about 16 lm/day. The history of precipitation
studies demonstrates (e.g. Compton and Daly, 1987;
Compton and Unwin, 1990; Wilkins et al., 2001, 2002) that
exact match of precipitation rates from experiment and
modeling would require a much higher degree of control
of the following features: fluid and substrate purity, effect
of surface tension at the air–fluid interface, detailed com-
parison of the diffusion boundary layer (in our model the
solid is atomically flat, in the experiments there are crystal-
lites of 10–30 lm with randomly oriented surfaces).
More important for this study, we observe that the
shapes of the precipitation profiles are very similar in both
the model and experiment. The peaks on the corners of the
obstruction are stronger in the experiment, perhaps due to
the effect of surface tension and the limitations of our mea-
surement technique. We consider these results to be a rea-
sonable experimental validation of the model.
4. DISCUSSION
In general, advective solute transport seems to be the
overriding factor in controlling calcite precipitation rates
in open-channel flow, except when the surface relief is small
compared with the water depth, or when the flow rate is
very low. If the calcite substrate is not parallel to the flow
velocity, advection of dissolved species towards and away
from the surface is orders of magnitude faster than molec-
ular diffusion, and advection controls the steepness of the
concentration gradients. We propose that hydrodynami-
cally induced deformation of concentration gradients can
have a major impact on calcite precipitation patterns under
laminar flow conditions. Also, under turbulent flow, the
gradients in the laminar boundary layer will be similarly de-
formed by hydrodynamics.
In most natural systems, faster flow is generally found in
the vicinity of regions in which the water depth is small, and
it can be difficult to spatially separate regions with high flow
rates from those with small depths. In our model, the loca-
tion of maximum flow velocity is slightly displaced from the
 Calcite precipitation instability under open-channel flow 5019

Fig. 8. (a) White light interferometer topography of a 0.85 mm  0.65 mm area at a step edge after 80 min (exp. 06.12.06). Black areas are not
resolved by the method. The two inset topography images are the same data set with (lower) and without (upper) data interpolation, taken
with a magnification objective that resolves larger reflection angles. The upper inset covers the corresponding region in the large image. The
interpolated data set (lower inset) is used to estimate the average thickness of precipitate. (b) and (c) Precipitate thickness from imaged
scattering intensity in situ after 20 (black circles), 40 (red diamonds), 60 (green squares) and 80 (blue triangles) minutes. (b) Exp. 11.12.06
(black lacquered surface, a = 0.5), (c) 08.12.06 (glass surface with black colour underneath as in exp. 06.12.06, a = 1). The proportionality
constants of calibration were based on measurement of average precipitate thickness at different positions. We estimate the peak growth rate
to be 0.8 ± 0.4 lm/min. (d) In situ image of step in flow ramp after 412 h of growth. (e) Closeup of image (d) on top of and right behind step. (f)
Ex situ image of a region on the top of the step from experiment 06.12.06 which was stopped after 80 min growth. A similar region of this
experiment is also displayed in (a).

location of maximum precipitation rate, indicating that the
progressive shallowing and resulting compression of con-
centration gradients established upstream from the position
of maximum flow velocity facilitates precipitation, rather
than the increased flow velocity. The model demonstrates
how shallowing over an obstruction enhances precipitation,
causing an instability in the growth process under purely
laminar flow. The flow velocity in itself is not a causal fac-
tor under these conditions. Instead, it is another result of
shallowing. However, this result does not exclude the possi-
bility that other mechanisms act to strengthen the instabil-
ity. Such mechanisms include thinning of a boundary layer
at high flow rates (Buhmann and Dreybrodt, 1985), en-
hanced degassing under accelerating flow (Banerjee et al.,
2004) and agitation of the surface film, and sticking of par-
ticles to the obstruction.
The inhomogeneous precipitation pattern observed on
the flanks of the obstruction in Fig. 5c, which we interpret
as a consequence of diffusion control, would lead to the
growth of calcite ‘fingers’ due to a Mullins–Sekerka insta-
bility (Mullins and Sekerka, 1963, 1964). This could be
partly responsible for formation of the porous fabric that
is often observed in natural travertine. At high flow rates,
this diffusive effect is relatively small compared with advec-
tion, and more compact travertine would result. This effect
may be responsible for the morphologies shown in Fig. 8e,
with porous, spheroidal growth on the low-velocity up-
stream side and a more compact, smooth ‘‘lip” on the high-
er-velocity downstream side of the rim.
Both the model and experiment also demonstrate the
process that produces overhangs on travertine steps. The
precipitation rate is high on the downstream upper corner
of the obstruction, but very low at the base of the step.
Chen et al. (2004) invoked the Bernoulli effect as one of
their explanations for higher precipitation rates over
obstructions. Regions of higher flow velocities have reduced
pressures, which would also reduce the partial pressure of
CO2 on the water–air interface. This would increase degas-
sing according to Eq. (17). However, it can be easily shown
that this effect is minor, at least under the conditions mod-
eled here. Bernoulli’s law states that
1
P þ qv2 þ qgh ¼ c; ð22Þ
2
where P is the pressure in a fluid of density q and velocity v,
at a height habove a reference surface. Choosing the local
height of the air–water interface as our reference height
(h = 0), we find that c = Patm and
5020 Ø. Hammer et al. / Geochimica et Cosmochimica Acta 72 (2008) 5009–5021
1 Griebel, M., Dornseider, T. and Neunhoeffer, T. (1997) Numerical
P ¼ P atm  qv2 : ð23Þ
2 simulation in fluid dynamics. SIAM monographs on mathe-
It follows from Eq. (23) that even flow velocities as high as
1 m/s would only reduce the pressure by about 0.5%. In
addition, precipitation patterns were not changed when
the degassing across the obstruction was increased by a fac-
tor of 100 in the simulations, showing that locally enhanced
degassing near the water surface, whether due to agitation
or the Bernoulli effect, does not appreciably affect the car-
bon dioxide distributions near the solid–liquid interface.
This is partly because of loss of lateral contrasts in CO2 con-
centration from the air–water interface down to the water–
calcite interfaces due to advection and diffusion, and in the
case of low flow rates because the dissolved CO2 near the
air–water interface is in near equilibrium with atmospheric
CO2 (cf. Fig. 4b, remembering that [CO2] at equilibrium is
0.014 mM for the assumed atmospheric P CO2 ) so that degas-
sing is not locally rate limiting. Degassing effects may be-
come locally important at larger scales however.
ACKNOWLEDGMENTS
This study was supported by a Center of Excellence grant to
PGP from the Norwegian Research Council.
REFERENCES
Arvidson R. S., Ertan I. E., Amonette J. E. and Lüttge A. (2003)
Variation in calcite dissolution rates: a fundamental problem?
Geochim. Cosmochim. Acta 67, 1623–1634.
Baker A. and Smart P. L. (1995) Recent flowstone growth rates:
field measurements in comparison to theoretical predictions.
Chem. Geol. 122, 121–128.
Banerjee S., Lakehal D. and Fulgosi M. (2004) Surface divergence
models for scalar exchange between turbulent streams. Int. J.
Multiphase Flow 30, 963–977.
Brown C. A., Compton R. G. and Narramore C. A. (1993) The
kinetics of calcite dissolution/precipitation. J. Colloid Interf.
Sci. 160, 372–379.
Buhmann D. and Dreybrodt W. (1985) The kinetics of calcite
dissolution and precipitation in geologically relevant situations
of karst areas. 1. Open system. Chem. Geol. 48, 189–211.
Chen J., Zhang D. D., Wang S., Xiao T. and Huang R. (2004)
Factors controlling tufa deposition in natural waters at
waterfall sites. Sediment. Geol. 166, 353–366.
Chiarello R. P., Wogelius R. A. and Sturchio N. C. (1993) In situ
synchrotron X-ray reflectivity measurements at the calcite–
water interface. Geochim. Cosmochim. Acta 57, 4103–4110.
Compton R. G. and Daly P. J. (1987) The dissolution/precipitation
kinetics of calcium carbonate: an assessment of various kinetic
equations using a rotating disk method. J. Colloid Interf. Sci.
115, 493–498.
Compton R. G. and Pritchard K. L. (1990) Calcite dissolution
kinetics at pH > 7: kinetics and mechanism. Phil. Trans. Roy.
Soc. Lond. A 330, 47–70.
Compton R. G. and Unwin P. R. (1990) The dissolution of calcite
in aqueous solution at pH < 4: kinetics and mechanism. Phil.
Trans. Roy. Soc. Lond. A 330, 1–45.
Dreybrodt W. (1988) Processes in Karst Systems. Springer, Berlin.
Goldenfeld N., Chan P. Y. and Veysey J. (2006) Dynamics of
precipitation pattern formation at geothermal hot springs.
Phys. Rev. Lett. 96, 254501.
matical modeling and computation 3. Society for Industrial and
Applied Mathematics, Philadelphia.
Guo L. and Riding R. (1998) Hot-spring travertine facies and
sequences, Late Pleistocene, Rapolano Terme, Italy. Sedimen-
tology 45, 163–180.
Hammer, Ø., Jamtveit, B., Benning, L. and Dysthe, D. K. (2005)
Evolution of fluid chemistry during travertine formation in the
Troll thermal springs, Svalbard, Norway. Geofluids 5, pp. 140–
150.
Hammer Ø., Dysthe D. K. and Jamtveit B. (2007) The dynamics of
travertine dams. Earth Planet. Sci. Lett. 256, 258–
263.
Harned H. S. and Hamer W. J. (1933) The ionization constant of
water. J. Am. Chem. Soc. 55, 693–695.
Hillner P. E., Gratz A. J., Manne S. and Hansma P. K. (1992)
Atomic-scale imaging of calcite growth and dissolution in real
time. Geology 20, 359–362.
Jamtveit B., Hammer Ø., Anderson C., Dysthe D. K., Heldmann J.
and Vogel M. (2006) Travertines from the Troll thermal
springs, Svalbard. Norw. J. Geol. 86, 387–395.
Jettestuen E., Jamtveit B., Podladchikov Y. Y., deVillier S.,
Amundsen H. E. F. and Meakin P. (2006) Growth and
characterization of complex mineral surfaces. Earth Planet.
Sci. Lett. 249, 108–118.
Kaufmann G. and Dreybrodt W. (2007) Calcite dissolution kinetics
in the system CaCO3–H2O–CO2 at high undersaturation.
Geochim. Cosmochim. Acta 71, 1398–1410.
Lax P. D. and Wendroff B. (1960) Systems of conservation laws.
Commun. Pure Appl. Math. 13, 217–237.
Liss P. S. and Merlivat L. (1986) Air–sea gas exchange rates:
introduction and synthesis. In The Role of Air–Sea Gas
Exchange in Geochemical Cycling (ed. P. Buat-Menard). D.
Reidel Publishing Company, Dordrecht, pp. 113–127.
Liu Z. and Dreybrodt W. (1997) Dissolution kinetics of calcium
carbonate minerals in H2O-CO2 solutions in turbulent flow: the
role of the diffusion boundary layer and the slow reaction
H2 O þ CO2 ! Hþ þ HCO 3 . Geochim. Cosmochim. Acta 61,
2879–2889.
Mullins W. W. and Sekerka R. F. (1963) Morphological stability of
a particle growing by diffusion or heat flow. J. Appl. Phys. 34,
323–329.
Mullins W. W. and Sekerka R. F. (1964) Stability of a planar
interface during solidification of a dilute binary alloy. J. Appl.
Phys. 35, 444–451.
Parkhurst, D. (1995) User’s guide to PHREEQC: a computer
program for speciation, reaction-path, advective transport, and
inverse geochemical calculations. USGS Water Resources
Investigations Report 95-4227.
Plummer L. N. and Busenberg E. (1982) The solubilities of calcite,
aragonite and vaterite in CO2–H2O solutions between 0 and
90 °C, and an evaluation of the aqueous model of the system
CaCO3–CO2–H2O. Geochim. Cosmochim. Acta 46, 1011–1040.
Plummer L. N., Wigley T. M. L. and Parkhurst D. L. (1978) The
kinetics of calcite dissolution in CO2–water systems at 5–60 °C
and 0.0–1.0 atm CO2. Am. J. Sci. 278, 179–216.
Press W. H., Flannery B. P., Teukolsky S. A. and Vetterling W. T.
(1992) Numerical Recipes in C. Cambridge University Press,
Cambridge.
Usdowski E. (1982) Reactions and equilibria in the systems CO2–
H2O and CaCO3–CO2–H2O (0–50 °C). Neues Jahrb. Mineral.
Abhand. 144, 148–171.
van Enckevort W. J. P. (1992) Phase shifting interferometry of
growth-patterns on the octahedral faces of natural diamonds. J.
Cryst. Growth 119, 177–194.
 Calcite precipitation instability under open-channel flow
Wilkins S. J., Compton R. G., Taylor M. A. and Viles H. A. (2001)
Channel flow studies of the inhibiting action of gypsum on the
dissolution kinetics of calcite: a laboratory approach with
implications for field monitoring. J. Colloid Interf. Sci. 236,
354–361.
Wilkins S. J., Compton R. G., Viles H. A. and Taylor M. P. (2002)
A new technique to evaluate and quantify modified solution
View publication stats
5021
kinetics of calcareous materials after sulphuric acid pre-treat-
ment and urban exposure. Stud. Conserv. 47, 88–94.
Wooding R. A. (1991) Growth of natural dams by deposition from
steady supersaturated shallow flow. J. Geophys. Res. 96, 667–
682.
Associate editor: Roy A. Wogelius