Geophys. J . R. astr. SOC. (1987)89,549-577 Analysis of the cooling of a variable-viscosityfluid with applications to the Earth w . Roger Buck Lamont-Doherty Geological Observatory of Columbia University, Palisades, New York 10964, USA Accepted 1986September 24. Received 1986July 7; in original form 1985June 22 Summary. Analysis shows that the convective cooling of a fluid with a temperature-dependent viscosity should exhibit simple behaviour which is consistent with geophysical data on the cooling of the oceanic lithosphere and asthenosphere. The cooling rate of convecting fluid is predicted to vary approximately linearly with (time)-'12. This relationship has also been demonstrated for numerical solutions of the full governing equations of convection. It is found that a similarity solution can describe the evolution of temperature within the stagnant high-viscosity lid which forms over an actively convecting region. The rate of thickening of the lid (or lithosphere) is directly proportional to a single parameter, A; and the overall cooling rate is also a function of A. Expressions are derived which relate changes in the parameter (A) to the initial average viscosity (or temperature of the fluid) and to the temperature and pressure dependence of viscosity. With assumptions about how the cooling of variable-viscosity fluids can be used to describe the cooling of the oceanic lithosphere and asthenosphere, values of sea-floor depth and the isostatic geoid height can be predicted as a function of time. Sea-floor depth is predicted to increase linearly with and geoid height to vary nearly linearly with time in agreement with observations. The rate of change of the depth or geoid height is a strong function of the initial temperature of the convecting region. The mantle is taken to have physical properties appropriate for olivine. A change in the mantle temperature under the ridge of 150K causes only about a 20 per cent change in the rate of subsidence but causes nearly a 100 per cent change in the rate of change of the geoid height with time. The observed variations in these rates over different sections of oceanic ridges are of about this magnitude. Simple conductive cooling would require more than l000K variation in mantle temperature to account for the geoid height observations. This analysis can also be used to estimate the cooling rate of magma sills. Key words: convection, viscosity, lithosphere, mid-ocean ridge, mantle heterogeneity, geoid-age relation 550 W. R. Buck Introduction The problem of convection driven by the cooling of a fluid from above has many applications in physics and engineering (eg. Curlet 1976; Ingersoll 1966). Geophysicists are particular interested in the case when the viscosity of the fluid is a strong function of temperature. The Earth’s mantle is thought to be a temperaturedependent viscous fluid (Weertman & Weertman 1975). The temperature dependence of the viscosity of the mantle is the reason why the lithospheric plates behave rigidly and the hotter asthenosphere below can flow to allow the plates to move. In a numerical study of the cooling of a temperaturedependent fluid, Buck & Parmentier (1986) find a very simple relationship between the rate of convective cooling and time. Namely, the heat flux varies approximately linearly with (time)-”?. This simple behaviour suggests that it may be possible to describe analytically the rate of cooling of the fluid and the temperature profile and thickness of a stagnant highviscosity lid which forms at the top of the cooling fluid. The cooling of the lithosphere has usually been treated as a purely conductive phenomenon (Turcotte & Oxburgh 1967), at least up to cooling ages of about 80Myr (Parsons & Sclater 1977). Furthermore, it has been assumed that if a small scale of convection did develop beneath cooling lithosphere then there would be a sudden and large change in the rate of cooling and thickening of the lithosphere (Parsons & McKenzie 1978; Houseman & McKenzie 1982). Recent work on the interaction of convection in the asthenosphere and the lithosphere has indicated that such convection is likely to begin only a few million years after the lithosphere begins to cool and that there is not an abrupt change in the rate of cooling of the lithosphere (Buck 1983, 1984; Fleitout & Yuen 1984; Buck & Parmentier 1986). There are also data whch may require that convection begins under very young lithosphere. Haxby & Weissel (1986) have suggested that short wavelength (- 200km) gravity undulations derived from SEASAT altimetry data may be due to small-scale convection beneath oceanic lithosphere. They observe these features over ocean floor formed as recently as 6 Myr. Numerical experiments on the effects of small-scale convection described here, and in greater detail in Buck & Parmentier (1986), have shown that convection can produce the observed magnitude of the SEASAT gravity signals and not lead to unreasonable rates of subsidence of the ocean floor (Buck 1985). The subsidence rate is changed when convection is active beneath the lithosphere, but subsidence and the thickness of the lithosphere vary linearly with (time)”2 as they do for purely conductive cooling of a half-space. Likewise the isostatic geoid height varies linearly with time as it does for conductive cooling. The slope of the geoid height-time relation is strongly dependent on the viscosity parameters. The aim of this work is to derive an approximate analytic relationship between the rate of cooling and the physical properties which control the strength of convection in a variableviscosity fluid cooled from above. This relationship is then used to predict changes in geophysical observables due to variations in physical properties of the convecting fluid. This approach is fusdamentally different from the standard parameterized convection treatment which has been applied to the thermal evolution of whole planets (e.g. Sharpe & Peltier, 1978; Schubert, Stevenson & Cassen 1980; Turcotte, Cook & Willeman 1979) where the lithospheric thickness is taken to be in equilibrium with the heat flux advected to its base. This is not the case in the transient cooling problem described here. The analysis is based on the assumption of a strongly temperaturedependent viscosity and of a Nusselt number (Nu) which depends on the Rayleigh number (Ra) to a fvred power. Care is given to the definitions of the Rayleigh number and the Nusselt number since this problem is a transient (non-steady state) one and viscosity variations limit convection to only part of a layer of fluid. Detailed analysis of 2-D numerical calculations (Buck & Cooling of a variable-viscosity fluid 551 Parmentier 1986) shows that the Ra-Nu relation holds for all model parameters considered. Next, it is shown that, to leading order, the convective heat flux through a layer of variableviscosity fluid cooled from above should have the same functional dependence on time as the heat flux in simpler cooling problems which can be described using a similarity solution. The numerical experiments are shown to exhibit this behaviour. Therefore, the formalism of the simple moving boundary problem (e.g. where a phase change occurs when the temperature drops below a given value) can be applied to this problem. In this case, the interface between the conductive lid and the convecting region, including the flow boundary layer, is analogous t o the moving boundary. Parker & Oldenburg (1973) have also treated the cooling of the lithosphere as a moving boundary problem, but the heat supplied to their boundary was due to the latent heat of solidification not a convective heat flux, as in the present case. The rate of thickening of the conductive lid and the average temperature of the lid are both decreased by the convective heat flux. Both effects can be described by a single parameter, A, which is related to the viscosity parameters of the system assuming the derived relations. The numerical experiments are shown to agree with the estimated values of h for a given set of model parameters. Since it is difficult to predict the convective heat flux at a reference set of conditions, the expressions that are derived constitute a parameterization of the effect of changes in viscosity or fluid temperature on the rate of cooling of the fluid. With certain assumptions, changes in h are related to data that depend on the thickness and thermal structure of the oceanic lithosphere and asthenosphere: the rate of subsidence, the heat flux, and the rate of change of the geoid height with age. The most important geophysical application of this parameterization should be in estimating the effect of lateral heterogeneity of asthenospheric temperatures along the ridge crest on the rate of subsidence of the ocean floor and on the rate of change of geoid height with age. Numerical calculation results The rate of cooling of a fluid with viscosity which depends on temperature and pressure will be affected by the vigour of the convection in that fluid. When a box of temperaturedependent fluid is cooled from above, a region may develop at the top of that box where viscosities are so h g h as to preclude any significant convective flow there; this has been shown in the laboratory experiments of Richter, Nataf & Daly (1983) and the theoretical work of Jaupart (1981). Schematic boundaries separating the regions where advective heat flow dominates over conduction (the convecting region), the region where it is insignificant (the conductive lid), and the boundary layer are shown in Fig. 1. For boxes of various widths (w> and depths (D),the 2-D Navier-Stokes equations of energy, mass and momentum conservation in a variable viscosity, infinite Prandtl-number fluid (Turcotte, Torrance & Hsui 1973) are studied using standard finite difference methods (Parmentier 1975). The flow boundary conditions are free slip on the bottom and sides of the box and fixed at the top. The viscosity (p) is taken to depend on absolute temperature ( T ) and pressure (P) through a relation appropriate for olivine (Weertman & Weertman 1975) given by: where E is the activation energy; V is the activation volume, R is the universal gas constant, and A . controls the average viscosity. Values of E and V are in the range of those experimentally determined for creep in olivine (Goetze 1978; Kohlstedt, Nicholas & Hornack 1980; Reddy et al. 1980; Sammis, Smith & Schubert 1981). A . is based on geophysical estimates of shallow mantle viscosity which are discussed in Buck & Parmentier (1986). 552 W.R . Buck Temperature Tcr Flow Boundary Layer Depth Z - "-rI L 1 Figure 1. Schematic temperature profile and its relation to the regions of cooling fluid. In the conductive lid heat is only transported by diffusion. The bottom of the conductive lid is defined to be a point of constant temperature ( T P )at depth ZL which moves down with time. The temperature of the convecting region (T,) also decreases with time from an initial value of T,. The surface temperature is T o . Zb is the bottom of the flow boundary layer. The temperature at the top of the box is fixed to be 273 K (or 0°C). Ideally, the interior of the box would initially be set at a constant temperature, but this leads to numerical difficulties since temperature gradients are then infinite at time zero. Therefore, the initial temperature profile for all the numerical calculations is that resulting from 5 Myr of conductive cooling of a half space initially at temperature T,. Small temperature perturbations (< 1K) are given to the finite difference grid points to allow convection instabilities to develop. Table 1 gives the physical parameters which are common to all the cases for the fluid in the Table 1. Parameters set for numerical models. Symbol K Q g P K CP R E V Cls Name Diffusivity Thermal expansion Acceleration of gravity Mantle density Conductivity Specific heat Gas constant Activation energy Activation volume Reference viscosity at depth 150 km Value 3.OXlO' 9.8 3500 3.2 900 8.31 Varied Varied Varied Units mZs-' K-' m sT2 kg J (m-s-K)-' J (kg - K)-' J (mol K)-' kcal mol" cm3mol-' Pa SK' Cooling of a variable-viscosity fluid 553 box. Table 2 shows defined parameters which describe the calculation results, and Table 3 gives the parameters which are varied from case to case. Instead of tabulating the values of A . for the cases, the value of the viscosity (p,) at a standard temperature and pressure (1 573 K and a pressure corresponding to a depth of 150 km) is given. Typically, in a model calculation, convection begins immediately under a stagnant highviscosity lid. The flow pattern for one calculation is shown in relation to the horizontally averaged temperature, viscosity and advective heat flux in Fig. 2 . The boundary layer which separates the convecting and conducting regions can be defined in terms of the horizontal average of the vertical advective heat flux [qc(z)] where w is the vertical velocity at a point and W is the width of the box. Below the region where the convective heat flux is negligible, 4c varies nearly linearly with depth. Fig. 2 shows the relation between q C ( z ) and other quantities at one time in one of the numerical cooling calculations. This shows that when the viscosity is lowest, the stream function values are high and the advective heat flux is significant. A straight line is fit through two points of this curve at 0.2qcmax and 0.8qCmaxas illustrated in Fig. 3 where 4cmaxis the maximum advective heat flux for all depths in the box. The depth at which the value of this linear function is zero is considered to be the top of the boundary layer (23.The bottom of the boundary layer (zb) is defined as the point where this line intersects qcmax(also in Fig. 3). The region below z b is taken to be the convecting region and the average temperature defined there as Tcr. The depth of the bottom of the conductive lid (zL) is defined as the depth where the horizontally averaged temperature is 90 per cent of the temperature in the convecting region (Tcr- To). Although the depth of the bottom of the conducting region (zL) is not exactly Table 2. Defined parameters. Length scale for the flow weighted by advective heat flux Dissipation weighted average viscosity Nusselt number Rayleigh number Vertical component of velocity Horizontally averaged advective heat flux Initial temperature inside box Average temperature of the convecting region Temperature at bottom of conductive lid Average nondimensional temperature of conductive lid Temperature drop across convecting region Depth of top of thermal boundary layer Depth of bottom of thermal boundary layer Depth of the base of conductive lid (or lithosphere) Ratio of ito p (zb) Slope of Nu versus temperature Factor relating Nu and dT,/dt Proportionality between In (Nu) and time Parameter which defines the rate of movement of the base of the lithosphere; depends on convective vigour Total heat flux out of the convecting region Subsidence Isostatic geoid anomaly 12 14 15 17 18 19 20 22 23 Case no. 10"Pa s-') 5 .O 1.o 1.o 1.o 1.o 1.o 1.o 5 .o 1.o (X kS 110 110 110 102 80 110 110 110 110 (m) kcal E 7.5 75 7.5 0 .o 7.5 I .5 7.5 7.5 7.5 V 180 150 190 190 160 170 180 150 120 I 0.96 1.17 1.07 0.87 0.93 0.94 0.97 0.96 1.02 h 19.2 11.6 14.0 21.9 20.3 20.3 19.2 20.3 16.4 t o= 64 (5) 6 .O 4.4 4.2 6 .I 5.2 6.8 6 .O 7.6 7.8 4.6 5.3 5 .O 0 .o 4.4 45 4.6 4.6 4.8 170 88 110 160 150 180 170 180* 200 174 92 105 131 165 200 174 218 181 2 4 -5 -17 9 10 2 18 10 Percentage difference Table 3. The f i s t five columns give the model case numbers and the parameters which define them. The reference viscosity L ( ~is the viscosity at the initial temperature, T,, and at a pressure corresponding to 150 krn depth. It is determined by parameter A, in (1). Next, the parameter h defined by (22) for the model case. qcmax is the value of the maximum horizontally averaged heat flux at 64 Myr model time. The values of c,, c, and c, are defined in the text and are calculated using the model parameters given in the table. The combinations of these parameters are those that go into equation (25) and are used to get the predicted value of d (qc max)/d ( t - ' / z )The . graphical value of d (qc max)/d ( t P Z )is estimated from Fig. 11 at time = 70 Myr. Case 22* was not shown in Fig. 11, but was determined the same way. $ R -$I .f P wl vl Cooling of a variable-viscosityfluid 555 i (Pa-s) AVERAGE TEMP 1 I (OK) HEAT FLUX STREAM FUNCTION - 400. 0. WIDTH (km) ' DIFF ADV 273. 400. 1573 (rnW/rn'l 17. 21. -10. 40. -100 100. Figure 2. Equally spaced contours of the stream function between maximum and minimum values plus four profiles of horizontally averaged quantities for a model time 10Myr after the start of numerical cooling calculation 15. The parameters which characterize the model are given in Table 3. The contours show that the flow does not penetrate into the region at the top of the box where the temperatures are low and the viscosities are high. The profile of advective heat flux [nondimensionalized as given by (211 also indicates that the flow has not penetrated to the region of high viscosity at the top of the box nor into the region at the bottom where the pressure dependence of viscosity has led to higher viscosities. The difference between the temperature of the model case and the temperature resulting from conductive cooling of a halfspace (COND DIFF) shows that convection has cooled a broad region while temperatures at shallow depth are relatively higher. 0. Qc max Convective Heat Flux Figure 3. Illustration of the method of estimation of the boundary layer thickness ( a Z ) and the position of the top (zt) and bottom (zb) of the boundary layer from the variation of the convective heat flux (qc). Since the q c is nearly a linear function of depth in the range where it changes most rapidly a straight line can be fit through that region. The two points which are somewhat arbitrarily chosen to define that line are the points where q c equals 20 and 80 per cent of qcmax. Varying those arbitrary values does no1 greatly alter the resulting estimate of the boundary layer position, but a consistent way of doing the estimate must be chosen. The depth at which this line has a value of zero for q c is defined as zt and where it equals qmax is the bottom (Zb). 556 W. R. Buck equal to the top of the boundary layer (zt), this definition allows easy comparison of the lid temperature structure between different cases. The non-dimensional average temperature in the lid (TL) is defined as Dividing the average lid temperature by 0.9 (Tcr - T o )allows T L to be used as a measure of the shape of the average temperature profile in the lid. A value of TL of 0.5 indicates a linear temperature profile and a value of 0.6 corresponds to the error function profile which results from conductive cooling of a half-space. Fig. 4 shows that for the numerical cases TL decreases from an initial value of 0.6 (due to initial conditions) to a lower value which is relatively constant in time. (Time) (m.y.), D=400 km 20 40 60 80 100 120 . I I I I I I 5 10 15 20 25 30 (Time) (m.y.), D=200 km Figure 4. Values of the average lid temperature ( T L )defined by equation (3) are plotted against time for the cases indicated. Table 3 gives the parameters used in each case. A value of nondimensionalized temperature of 0.60 corresponds t o the conductive solution and a value of 0.50 indicates a linear temperature profile. Note that after an initial period of adjustment the values of T L are nearly constant. To dimensionalize the time two length scales for the depth of the box are used: 400 and 200 km. The larger depth gives the larger times. Cooling of a variable-viscosityfluid 557 Another feature of the numerical calculations which shows a regular behaviour in time is zL, the depth of the base of the conductive lid. When plotted versus (time)’l2 it is nearly linear (Fig. 5 ) , but the plots have slopes which are less than that predicted for conductive cooling of a half-space. It is the constancy of TL in time and the linear trend versus (time)”2 for lid thickness for the numerical calculations which have induced the following analysis. Two standard parameters are calculated through time for the convecting regions. They are the Rayleigh number (Ra) and the Nusselt number (Nu). The physically meaningful definition of the Rayleigh number for this problem is pgol A T1 Ra = (4) K P where p (density), g (acceleration of gravity), and (Y (coefficient of thermal expansion) are, the same for all calculations (Table 1). Because this problem is transient and viscosities vary across the convecting region, the values of AT (temperature drop), I (length scale), and p (average viscosity) must be defined in a self-consistent and physically reasonable way. These values change with time in the calculations. To define the temperature difference for the / t/ 0 ZLS x / 2 z t w 1. 0 ( Ti 2. 3. 4. 5. (m.yV2, D = 200 km. Figure 5. The nondimensional depth to the bottom of the conductive lid, ZL, calculated for case 15 plotted versus (time)”’. Depth is nondimensionalized by dividing by the depth of the box, D. 558 W.R. Buck convecting region (AT), the horizontally averaged temperature [Th(z)] is used. The temperature difference between the top and bottom of the boundary layer AT = Th(zb) - Th(z& The top and bottom of the boundary layer are defined in terms of the vertical advective heat flux, as illustrated in Fig. 3. The vertical advective heat flux is also used to define the thickness of the region over which the flow is vigorous as The viscosity to be used in the calculation of the Rayleigh number is an average weighted by the second invariant of the strain rate tensor ( E ) , suggested by Parmentier (1978), where A is the area of the box. Finally, the Nusselt number is defied as the maximum horizontally averaged vertical heat flux (qcma.J divided by the steady state conductive heat flux over the convecting region qc max 1 NU=-. KAT (7) This is a slight variation on the normal definition of the Nusselt number, which is the ratio of the total heat flux in the presence of steady-state convection to the heat flux across the same region when there is only conductive transport of heat (McKenzie, Roberts & Weiss 1974). The convection here is quasi steady state, that is, the heat flux is changing slowly with time, but at a given time the convection is in equilibrium with the heat flux it is transporting. The use of the maximum convective heat flux (qcmax),instead of the total heat flux, in the definition of the Nusselt number is justified because at the level where the convective heat flux is a maximum the conductive heat transport is negligible. Theory predicts that for steady-state convection the relation between Ra and Nu should be given by Nu = a Rab, (8) where b is a constant between 1/5 and 1/3 (Roberts 1979) for convective flow between fixed boundaries. Laboratory experiments by Booker (1976) and Richter et al. (1983) have shown that this relation holds even for fluids which have a viscosity which strongly depends on temperature for Rayleigh numbers up to lo6. Christensen (1984) has called this relation into question for variable viscosity convection, but his numerical calculations assume different boundary conditions (i.e. he considers the top boundary to be free to slip) and the viscosity he used to calculate his Rayleigh number is different from that used here. The definition of the terms given here that go into the Raleigh number do not scale simply with the temperature and pressure dependence of viscosity. However, the change in Nu for a given change in the average viscosity or the temperature of the fluid can be predicted through the Ra-Nu relation, once the value of the Nusselt number at a reference temperature is prescribed. Thus, this is a parameterization of the change of heat flux of a convecting system given a reference state. Eventually, the physics of variable-viscosity flow Cooling of a variable-viscosity fluid 559 may be sufficiently well understood that the effects of changes in E and V on the Rayleigh number can be more rigorously quantified. Then for a given set of viscosity parameters, a Rayleigh number and a Nusselt number can be predicted for a reference state. Fig. 6 shows the relation between log(Nu) and log(Ra) for several cases. For a range of Rayleigh numbers for each-case considered there is a linear relation between these quantities. The slope of this plot is generally between 0.25 and 0.30 [= b in equation (S)]. The points which do not lie on linear trends are all for times early in the calculation, when the average lid temperature (TL) was varying. For the points which do not fit, the Nusselt number is lower than expected from the linear relationship. The times when TL is constant and equation (7) holds, which shall be referred to as the interval of ‘transient equilibrium’ of the lid with the convecting region, shall be concentrated upon. The heat flux out of the convecting region affects the rate of cooling of the lid and the value of TL.The Nusselt number describes that heat flux, and it shall be investigated how it varies with time during cooling. Since the Nusselt number depends on the Rayleigh number [via (8)], it is necessary to find the parameters that are causing the variation of the Rayleigh number with time. The average viscosity changes by a large amount through a calculation, 1.2 1.1 n z‘ W 1 Case 14 A Case 18 rn Case20 0 0 t Q lmO I Case / ‘ /1 0.9 Slope = .30 /? H0X-= 0.7 0.6 ” 890’a’ a Case 14 + Case 20 Slope = .25 l0 3.2 3.6 4.0 4.4 4.8 5.2 Log ( R 4 Figure 6. A plot of log(Nu) versus log(Ra) where the Nusselt number (Nu) is given by (7) and the Rayleigh number (Ra) is given by equations (4)-(6) for several cases. All of the cases can be fit with straight lines with slopes, which equals b in equation (S), between about 0.2 and 0.3. The offset of the points for case 18 relative to the other cases indicate that Nu and Ra may not be perfectly defined for all different possible viscosity parameters, but given the large variation in parameters used in the calculations the correspondence of the results is very good. W. R. Buck 560 while the other parameters that define the Rayleigh number do not. The area where the viscosity is minimum is the area where the strain rate is greatest and, therefore, 6 is highest, as illustrated in Fig. 6. Therefore, the average viscosity (p) is weighted most heavily toward the minimum viscosity. The minimum viscosity occurs at the base of the thermal boundary layer (zb) because this is the shallowest depth at which the nearly isothermal convecting region occurs. The viscosities deeper than zb are higher if the pressure dependence of viscosity [V in (l)] is greater than zero. When V = 0 the convecting region is nearly isoviscous, except in the narrow boundary layer. Fig. 7 shows a plot of the ratio of the average viscosity (p) defined by ( 5 ) versus the viscosity at the base of the boundary layer [p(zb)] for two numerical cooling calculations through time. In case 20, v = 7.5 cm3mol-' and in case 17, V =0. In both cases p was linearly related to p(zb), except at the start of the calculation, but the ratio of the relation is different. The ratio was nearly 1.O for case 17 (no pressure dependence on viscosity) and about 5.0 for V = 7.5 cm3mol-' in case 20. In all the numerical cases, the ratio is nearly constant for a given model. Thus throughout a calculation, the effective average viscosity can be related to 1-1 (zb) as (n) p =c1I-1 (zb) (9) 9 where c1 is a constant that depends on V. Three factors affect the value of the minimum viscosity [p(zb)]. One is the average viscosity determined by A,, in (1). The other two factors 61 I I 1 I 0 @ O I h 0 0 Case 20 0 Average = 5.40 Std. Dev.= 0.16 II W N v 0 v) a .+ .cn 3- Case 17 El Average = 0.82 Std. Dev.= 0.08 .-0 > cc 0 0 .+ cd CT 2- r --la- -a- ,T 1- 1 0 & b I I I I 20 40 60 80 9 -L *- 1 100 Time (m.y.) Figure 7. The time variation of the ratio of the average viscosity (d, as defined by ( 6 ) and the horizontally averaged viscosity at t h e base of the boundary layer [c((zb)] is shown for the two cases which differ in activation volume. For case 17, the activation volume is effectively zero while for case 20 it is 7.5 cm3mol-'. Over the time of the calculations the ratio is nearly constant. Cooling of a variable-viscosity fluid 561 are more interesting because they change with time. One is the average temperature of the convecting region (Tcr) which affects viscosity in proportion to the magnitude of the activation energy ( E ) in (1). The other is the depth to the bottom of the thermal boundary layer (zb), since pressure (P)in (1) depends on depth. The effect of variations in the average temperature (Tcr) and in the depth (zb) on the Nusselt number will be considered. Furthermore, it will be shown that both effects are consistent with the Nusselt number being proportional over a significant range of time. to P 2 Theoretical treatment of the cooling problem If only the temperature dependence of the viscosity is considered, then (1) reduces to Combining (10) with (4) and (8) gives a relation between the Nusselt number and temperature where ). A , = a ( CrpgATl3 KClAO Fig. 8 shows that the relation between ,,eat flux and temperature is nearly linear. Fig. 9 shows that the relation between the heat flux and the temperature of the convecting region I 1 I I I I I 1260 1280 1300 Theoretical Heat Transporting 12 11 c 101 1200 1220 1240 (Tc,-To, ( O K ) Figure 8. The dashed line is a plot of the Nusselt number (Nu) versus temperature (TI given by (11). Here E = 100 kcal mol-', b = 0.25 and Tref= 1523 K and Nu (T,,f) = 10. There is a change in Tof 100 K for this plot. The solid line is the best fitting straight line approximation to this relation over the same range. W. R. Buck 562 (Tc,- To) 1220 1240 (OK) 1260 1280 1300 14 12 10 NU 8 6 4 2 .94 .96 .98 1.o Figure 9. The variation of the Nusselt number (Nu) with changing temperature of the convecting region (Tcr) is shown for two of the numerical cases considered here by plotting Nu against (Tcr - To)/(Tm- To). The trends are approximately linear; however, the change in Nu is greater than is predicted solely as a consequence of cooling a temperature-dependent fluid. The pressure dependence of viscosity also contributes as discussed in the text. is also linear for the numerical calculations. The temperature of the convecting region (Tcr) is simply related to the heat flux out of the region or Nu by where c2 =KAT/(l2pc,). Differentiating (1 1) with T = T,, and combining with (12) gives where Cooling of a variable-viscosityjluid / I I I I I 0.12 0.14 0.16 0.18 563 i I (m.y.)-’I2 Figure 10. This shows the dependence of Nu ( t ) for (14) and the linear approximation to that relation given by (15) on ( t ) - l I 2 . The two plots were made to coincide at time f = 50 Myr, which corresponds to (t)-”*= 0.141. The time range for the plots is 50 Myr. where Gef is a reference temperature. Integrating equation (13), assuming t = t o when Nu (0 = Nu(Tref),gives Nu(r) = Nu(to) exp [-c2c3(t - t o ) ] . (14) Expanding t around t oin termsof t-”’ gives Nu ( t ) = Nu ( t o )[ 1 t 2cZc3(f’” - f;’”) $‘I. To show that (15) is a good approximation to (14) for the range of parameters considered here, values of Nu(t)/Nu (to)were calculated using both formulas. Using E = 100 kcal mol-I, pc,= 3 x Jm-3K, Nu(to) = 20, 1 = 200km, A T = 100K, K = 3.2 J ( m s K)-’ and Tref= 1550K, it is found that the average error due to the approximation was less than 2 per cent over a period of 50Myr with t o centred on 50Myr (see Fig. 10). Therefore, for a cooling convecting fluid with temperature dependent viscosity given by (l), the Nusselt number, or the convective heat flux, can be shown to vary, to first order, with f ’ j 2 . A similar analysis can be applied to the variation of the Nusselt number with the depth to the isothermal convecting region or the pressure there, which is linearly related to depth. However, a further assumption must be made, namely that zb varies as t”’, which is the case for the numerical results. I t will then be shown that this causes the Nusselt number to vary as t-”’. Holding temperature constant so that viscosity only varies with Zb, (I) can be rewritten where v‘ = Vpg and A2 = A oexp (E/RT,,f). Since the Nusselt number (Nu) depends on the W.R . Buck 564 Time (M.Y) (a) 100 80 70 60 24 - - 1 1 1 1 50 40 30 I 1 I - A - 20 - w E -2 E = 16 X cs! 12 o C A S E 14 v CASE 17 0 CASE 18 0 A C A S E 20 * I I .I0 I I .I2 I I .I4 I I .I6 I I I .I8 Figure 11. The variation of q c m a x versus (time)-"* for the numerical calculations. qcmax is directly proportional to the Nusselt number (Nu). The relation is essentially linear in the time interval when T is relatively constant for the model calculations. Table 3 gives the parameters which define the model cases. (a) shows four cases which have the same box geometry, and initial conditions, but vary in the parameters which control the viscosity. (b) shows that the cases have the same viscosity parameters, but differ in the box geometry or the initial conditions. average viscosity through (Z), ( 7 ) and ( 8 ) , it will depend on zb as Nu (zb) = A (A2)-*exp (S) with A l defined in (1 1). The boundary layer thickness changes little during the numerical calculations compared to the thickness to the conductive lid (zL), and it is small compared with zL,so zb is approximately equal to zL.zLis assumed to be equal to 2 h (Kf)'12, where is a parameter which will be discussed in the next section and K is the thermal diffusivity. Thus by changing (17) the Nusselt number as a function of time is 565 I00 80 70 60 I I I I 50 40 30 I I t - - - - 20 - - - - 16- - X - - 5 12- 24 N E \ 20 W 0 CASE 12 - V CASE 15 0 C A S E 19 8- A C A S E 23 I I .I0 because we have replaced zb I I .I2 I I .I4 I I .I6 I I - - I .I8 with 2h(~t)’”.Expanding (18) around to in terms oft-’” gives Nu (t) = Nu (to) [ 1 t ( f ’ 1 2 - t;’”) c 4 t 0 ] . (19) As with the variation of Nusselt number with temperature, the variations due to changes in are approximately linear with t-‘/2. It will now be shown that when Nusselt number varies linearly as f ’ ” , as it approximately does for this problem, then the cooling of the conductive lid can be described by a similarity solution. The treatment of this solution follows Carslaw & Jaeger (1959), and the problem is similar to a moving boundary problem where there is a change of phase, which was first treated by Stefan (1891). The difference between the base of the boundary layer and the base of the lid is taken to be neghgible. The heat flux there is the maximum convective heat flux zb where K is the thermal conductivity. The position of zL is assumed to move with time at a rate given by zb = 2h ( K t ) ” 2 , where (21) X is a constant which depends on the Nusselt number of the system at a given time. 566 W.R. Buck The temperatures in the lid must satisfy the conductive heat transport equation. Such a solution is ~ ( z t, ) = B erf [ z / 2 ( ~ t ) "+~T,,.] (22) At depth z b the temperature is assumed to be T,, + To(see Fig. 1) so the constant B is given by T,,/erf(X), where T,, is the average temperature of the convecting region. Since T,, varied slowly in the numerical cases considered here, T,, is taken t o be constant when considering the temperature structure of the lid. If there were no convection, just conduction at all depths, then X would be infinite. Combining (20), (21) and (22), we find that the convective heat flux must vary linearly with t-'I2 for (21) to hold. This gives Next, we can show that, for this problem, the average temperature of the lid is constant in time and depends on A. The lid similarity temperature (TL)is defined as the average temperature down to the bottom of the lid divided by the temperature there (T,., + To). Integrating (22) down to z b and dividing by T,, + Toyields Thus TL is a function only of X and does not vary with time. When Tp- To= 0.9(T,,- To) then TL is analogous to the average lid temperature calculated for the numerical results which was shown to be nearly constant in time for the numerical calculations. Using this equation, a value of X can be calculated for a given value of TL. For the numerical cases the integration was done only down to the point where the temperature equalled 0.9 Tcr.Thus, the value of X obtained using (24) must be corrected to reflect the value of X which determines the rate of motion of the point where the temperature is T,,. The results of doing this are shown as values of h for the numerical cases in Table 3 . Comparison between theory and numerical results The combined effect of variations in T,, and t b on the convective heat flux is simple to get by differentiating and adding (1 5) and (19). The Nusselt number (Nu) is then linearly related to the convective heat flux (4, via (7) to give Fig. 1 1 shows plots of measured values of qcmaxversus f ' 1 2 for the numerical cases, and it is clear that the trend is linear in the region of 'transient equilibrium'. The behaviour is similar to that predicted for the Nusselt number (see Fig. lo). To compare the predictions of this theory against the results of the numerical calculations, we must calculate the parameters cz, c3 and c4 and be able to relate them to features of results of the model. Table 3 gives the values of the slopes from the plots in Fig. 1 1 along with the predicted values of the slopes based on the calculated values of c2, c3 and c4. The value of X for each calculation is estimated by calculating TL and relating it to h by (24). In the estimation of these parameters, no effort was made to adjust for the differences in the depth extent of the efficient cooling by convection (Az), which would affect the value of Cooling of a Variable-viscosityfluid 567 cz, nor was the value of to allowed to vary from case to case. Even so only for run 17 and case 22 was the predicted slope more than 1 0 per cent different than the calculated values. To determine X in terms of the coefficients of viscosity A , E and V, (23) can be differentiated and combined with (25), (7) and (8) t o give where R$ef is the critical Rayleigh number defined by (4) with p = k e f , and the heat flux qcmaxis taken to be zero when p ( t o ) equals pref.Instead of considering variations in the pre-exponential in the viscosity relation, A o , this expression has been written using the average viscosity, ,ci, which is directly proportional to A o . This relation shows that X does not depend on time for this theoretical development, but only on the physical parameters of the system. The most important parameter which affects X is the average viscosity, p. This is particularly interesting because the average viscosity of the asthenosphere may vary from place to place, as will be discussed later. In Fig. 12, the variation of X as a function of the log of the average viscosity is plotted. This is shown to be a nearly linear relation with a change of about 0.2 in X for a change of a factor of 10 in the average viscosity, for b (the exponent in Ra-Nu relation) equal to 0.25. Since the average viscosity is taken to be an exponential function of temperature, then X is essentially linear with changes in temperature 1.8 1.6 1.4 A 1.2 1.o 0.8 0.6 0 1 2 3 4 Log(Pref/p) Figure 12. The variation of h as a function of log [/+ef/(P)] as given by (26) assuming b = 0.25 and that a = 1.0 when p = 0.09 M.This assumes that a l l the variation in A is due to differencesin the average initial viscosity. Taking the cause of the viscosity variations to be differences in the initial temperature of the convecting region (T,) gives the axis at the top of the figure. The difference in T, from some reference value, T, ref, which would cause the viscosity variations on the lower axis is shown. The activation energy is taken t o be 1 0 0 k cal m o l ~and ’ T, ref, is 1200°C. 568 W.R. Buck around a reference value ( T , ref). For E = 100 kcal mol-' values of (T, - T, ref) are shown in Fig. 12 which correspond to the variations in average viscosity. Between cases 12 and 14, there is a change in the average viscosity of a factor of five, which would correspond to a difference in temperature of about 80°C for the value of E used. Equation (26) or Fig. 12 predicts that the difference in h for the two cases should be about 0.15, in excellent agreement with the value derived from the numerical experiments of 0.13. The change of h due to a change of the temperature dependence of viscosity E, is as simple to estimate as that due to changes in the average viscosity. This is because changing E causes a change in AT, the temperature difference across the convecting region. It is found that AT is inversely proportional to E. T h s can be understood as a consequence of the viscosity difference across the convecting region being roughly constant. This constancy is found not only in these calculations but is also noted by Jaupart (1981), who studied the linear stabdity of fluid with depth-dependent viscosity. Since E and AT are inversely proportional, the term c2c3 does not change for a change in E. Therefore only the change in AT in (26) will affect A. Since AT also enters into the Rayleigh number [see (4)],the change in the right-hand side of (26) due to a change in AT should be equal to the ratio of the old value of E over the changed value of E, all raised to the (1 + b) power. Case 18 had the same viscosity parameters as case 14 except that E was 80kcalmol-' as in case 14. For b = 0.25 the change in the right-hand side of(26) due to thischange inEisaboutafactorof 1.5 [=(110/80)1.25]. This is about the same change as would be predicted for a decrease in the average viscosity of a factor of five, which is the change between case 12 and cases 14 and 18. Therefore, it is not surprising that the values of X calculated for cases 18 and 12 are almost equal. It should be noted that the linear relation between qcmax and f - l i 2 shown for the numerical cases generally is not valid in the early times of model calculations. However, for case 15, which was the only case where a wavelength of flow was not imposed (see Buck & Parmentier 1986) the relation is valid at the earliest times (Fig. lo). Since this less constrained system is more likely to reflect the behaviour of a natural system, then the assumption of a linear relation between heat flux out of the convecting region and f-'", which is required for (26) to hold, is valid to earIier cooling times. For time r = 0 the approximation clearly breaks down because the heat flux out of the convecting region cannot be infinite. A parameterization of the effects of non-Newtonian viscosity was not explicitly considered. The stresses depend on the wavelength and Rayleigh number of the convective flow. For the small wavelengths and moderate Rayleigh numbers used in my model calculation, the effect of stress should not be large if it is appropriate to use a cut-off stress of about 10 barsbelowwhichthe rheology is stress independent, as is done in Fleitout & Yuen (1984). Application to geophysical problems In the preceding sections a mathematical description of the temperatures in a cooling lid over a variable viscosity convecting region was developed. The dependence of the only parameter (A) needed to describe the system on average viscosity, on the temperature dependence of viscosity (E), and on the pressure dependence of viscosity ( V ) was shown. The variation of the geophysical observables of lithospheric subsidence, local isostatic geoid height variations, and heat flow will be discussed in terms of their variation with A. Also, the use of the relations described here to estimate the rate of cooling of convecting magmatic intrusions will be briefly discussed. A word should be said about the assumptions used to relate the simple 1-D mathematical model presented here to the case of the 3-D mantle of the Earth. As has been shown, the I-D model matches the horizontally averaged temperature of the 2-D numerical calculations Cooling of a variable-viscosityfluid 569 quite well. Thus, this theory should be good when there is no convection other than that driven by cooling from above. Even if that flow is 3-Din nature, this should be true since the same kind of Ra-Nu relations [see (8)] should hold for a simple 3-D flow driven by cooling from above, as laboratory experiments show (Booker 1976). However, there are special problems in applying these results to the cooling of the oceanic lithosphere. First, there is a large scale of flow associated with the motion of the lithospheric plates. This flow should be perpendicular to the flow calculated in the numerical runs considered here (Richter 1973; Richter & Parsons 1975), as illustrated in Fig. 13. The present results are applicable to the cooling of the oceanic lithosphere only to the extent that there are no vertical gradients of velocity in the direction of the large-scale flow. The depth range for which this is a good assumption depends on the form of the large-scale flow. (b) 0 -h E Y ;r v I ICONC t DlFF ta w n !(OK) I I 0. ASTHENOSPHERE I 1 I I I I 400 2 I I 2c Figure 13. (a) Geometric relation envisioned between smallscale convection and the largescale flow under a ridge crest. (b) Shows the average temperatures for case 20 at model time 40 Myr and approximately how the model temperature profile is taken to relate to the oceanic lithosphere and asthenosphere. S denotes water depth or subsidence. 570 W.R. Buck The second assumption made in applying these results to the cooling of the oceanic lithosphere and asthenosphere is that cooling at all depths leads to subsidence of the surface. Much work has been done on the relation between temperature variations in a convecting fluid and surface deformation and geoid variations (McKenzie 1977; Parsons & Daly 1983). However, one must be able to specify the aspect ratio of the large-scale mantle flow, its depth extent, and its viscosity structure in order to relate the temperature variations caused by small-scale convection to the surface deformation and geoid. Therefore, I make the simple assumption that cooling at all depths is weighted the same. The heat flux out of the surface of the lithosphere and the subsidence due to thermal contraction scale simply with h. The surface heat flux is derived by differentiating the temperature profile [see ( 2 2 ) ] at z = 0. This gives a surface heat flux which is the same as that which would be calculated for conduction alone, increased by the value l/erf A. Fig. 13(b) illustrates how model temperature changes are related to subsidence. Since subsidence (S)should reflect thermal contraction due to the integrated heat flux out of the surface, it too will be equal to the conductive value times l/erf A. Therefore the predicted subsidence will vary linearly with the square root of age for this model, but the slope of subsidence versus (time)”2 of the seafloor will depend on the vigour of convection: If h scales with initial temperature of the mantle under the ridge then we can predict the change in the rate of subsidence for a given change in temperature. This is plotted in Fig. 14 which shows that the change in the rate of subsidence is only about 30 per cent for a change in mantle temperature of 200°C. The rate of subsidence is known to vary along different sections of ridge crest and systematic studies of subsidence are presently underway (Hayes 1986; Cochran 1986). The variations in surface heat flux should be of the same magnitude, Tm-Tmref( O K ) 50 100 200 300 I 0’71 2 - 0.6 -2E=+ 0.5 0.1 O’* Derivative of Subsidence with (time)’” A Convection 1 1.8 1.6 1.4 1.2 1 .o 0.8 Figure 14. The derivative of subsidence (S) with respect to versus h as predicted by (27). Also shown on the top axis is the difference in initial mantle temperature (T,) which would give rise to the indicated variations in h as discussed in the text for E = 100 kcalmol-’. The solid line shows the effect of variations in h and in (27). The dashed tine is the variation in the rate of subsidence for a conductive model which depends only on Tm not on A. Here K = 7.0 ~ l O - ’ r n ~ s -T, ~ , - To = 1300 K is taken and other constants as defined in Table 1 . Cooling of a variable-viscosityfluid 571 but would be difficult to resolve in the data due to the uncertainty of the measurements. The isostatic geoid height is very sensitive to the distribution of temperatures with depth and therefore to the effects of small-scale convection. I will show that convection under a cooling lithosphere will lead to isostatic geoid anomalies which decrease approximately linearly with age but that the predicted slope of the geoid height-age relation will depend strongly on the convective vigour. Before a relation between geoid height and h can be derived, it must be shown how the depth range of convection changes with time. In the numerical experiments described in Buck & Parmentier (1986), the depth range of convection is seen to increase with time. The linear gravity features observed over several areas of the oceanic plates may show little increase in wavelength with crustal age (Haxby & Weissel 1986), however, this may reflect the interaction of convective stresses with the growing elastic lithosphere. If the aspect ratio of the convection cells remains close to unity as they grow, then the wavelength will be about twice the depth range for convection. Since the dependence of viscosity on temperature limits the temperature drop across the convecting region, it is reasonable that the pressure dependence of viscosity controls the depth range of convection (2). The simplest assumption that can be used to derive an expression for 1 is that a constant ratio between the viscosities at the top and bottom of the convecting region must be maintained. Using equation (1) a given drop in temperature at the top of the region (T, - Tc,) will be related to 1 by where P is an average value of the pressure in the region. Next, the total amount of heat available for cooling of the convecting region must be estimated. The total amount of heat removed from the convecting region at any time (Qcr) is given by the integral of (21) minus the heat of cooling material added to the lid Qc, ( t ) can be related to the change in temperature of the convecting region and the thickness of the convecting region ( I ) through (21) Since Q,, depends on t'" then I will depend on ? I 4 . The rate of change of I is a strong function of the pressure dependence of viscosity, as it is proportional to VG'/'. Fig. 15 shows 1 as a function of time for two values of V', along with the convective wavelength assuming the convection cells have a unit aspect ratio. The size of the convection cells varies slowly, after an initial period of rapid growth. Thus, the assumption made earlier that the depth range of convective cooling (I) was constant in time is consistent with (30). The numerical calculations done so far do not allow comparison with the prediction of cell size versus time, since the box size and boundary conditions limited the growth of the cells, Appropriate calculations are now being carried out to check these predictions. The isostatic geoid height depends on the dipole moment of the density distribution (Haxby & Turcotte 1978) and thus on the shape of the temperature profile in the lid and in the convecting region. The contribution to the isostatic geoid height from the conductive lid is achieved by integration of the dipole moment of theetemperature distribution given by ( 2 3 ) down to depth ZL. The contribution due to the assumed drop of temperature, the 572 I 1 I I Figure 15. The variation in length scale of active convection ( I ) versus time given by equation 30 for two values of the activation volume V assuming A = 0.9 in both cases. Also shown is the wavelength for the convection assuming a unit aspect ratio for the cells. Constants as defined in Fig. (14). convecting region is then added to this t o give Time (rn.y.), 0.200 km -1.0 II 0 -15.0E v I 20.0- 0 50 75 100 Tirne(rn.y.), 0.400 km Figure 16. Plots of geoid height (If) versus time are shown for conductive cooling of a halfspace, the 25 results of numerical model case 15 and the analytic model given by (31) for A = 1.0. Two length scales (or box depths, D ) are used to dimensionalize the model results as discussed in Fig. 4. The analytic prediction is nearly linear with a slope almost twice that for conduction and also nearly matches the results of the numerical model which had the same value of A . Values for K , of and T, as in Fig. (14). Cooling of a Variable-viscosityfluid 573 where second order terms in the integration have been neglected. The predicted geoid height is very close to being linear with time, as shown in Fig. 16. Fig. 17 shows how strongly the slope of geoid height-age (dH/dt) depends on X and therefore on the initial temperature of the convecting region. A very different application of the parameterization derived here is to the cooling of magma bodies from above, which has been described by Carrigan (1984). He considers magma bodies that are insulated on their sides and that, because of temperature dependent viscosity of the magma, have a conductive lid over a convecting region. In his numerical calculation, he considers Rayleigh numbers in the same range as the numerical cases already described from Buck & Parmentier (1985), although the length scales involved are much smaller. The lower viscosity of magma offsets this. Carrigan (1984) develops a parameterization in terms of a set of ordinary differential equations which must be solved numerically. He is mainly concerned with the average temperature of the convecting region. In his treatment, he does not have to make the simplifying approximation that the depth extent of the convecting region is constant in time. That approximation must be made to derive the analytic parameterization described here. The advantage of this treatment is that someone who wishes to use the parameterization does not have to program the differential equations on a computer, but merely has to evaluate the expressions derived here. When only the temperature dependence of viscosity is considered in ( l ) , the temperature of the convecting region is approximated by: c3 where T, is the initial temperature, Nu(t,) is the initial Nusselt number for the system, and c2 and cg are as defined before. This was derived by combining (1 2) and (1 4) and integrating taking To= 0. Equation (32) predicts the exponential decrease in the interior temperatures w h c h were reported by Carrigan (1984). A direct comparison with h s results is difficult because, in the two cases he presents, he changes the initial conditions of temperature, the 03- Derivative of Geoid Height wlth Time x E -5. 02- E - - I U . 01 0.0 _ - / - - - - - - Conduction 1.8 1.6 1.4 1.2 1.0 0.8 A Figure 17. The derivative of isostatic geoid height ( H ) with respect to time as a function of h given by (31). As in Fig. (14)the solid line represents changes in the slope due only to changes in A and T,. The dashed line shows the effect of changes in Tm on the rate of change of H for a partly conductive model. Constants same as for Fig. (14). 574 W.R.Buck initial average viscosity, and the temperature dependence of viscosity. However, if the difference in initial conditions is a small effect compared to the temperature dependence of viscosity E, then (32) can be used to compare hls two cases. As noted before, the term c2c3 does not change for a change in E. Only the term Nu(t,)/c3 is important, and that term only depends on 1/E. The difference between his cases is that he takes E to be twice as large in the second case as in the first. He observes about one half the rate of change of temperature for the second case relative to the first. This is consistent with the prediction of (32). Discussion The linear gravity features observed over regions of the oceans (Haxby & Weissel 1986) may be a product of small-scale convection. Small-scale convection would produce change in the horizontally averaged temperature structure of the lithosphere and asthenosphere. The equations developed here can be used to estimate the observables that depend on the average temperature structure of the oceanic lithosphere and asthenosphere. The key question is whether there are any data which are better fit by the present model than by the standard half-space cooling model for the oceanic lithosphere. The average values of the surface heat flux, the subsidence rate, and the rate of change of the geoid height with age for the ocean basins (e.g. Parsons & Sclater 1977) can be fit equally well by either model. This is due to the uncertainty in the physical parameters which control the model predictions. However, the variations in the data from region to region, especially the data on rate of change of geoid height with age, appear to be more consistent with the present model. Andrews, Haxby & Buck (1986) find variations in the geoid height-age slope of about a factor of two. They consider areas where the lithosphere is younger than about 50Myr to avoid long wavelength contamination of the signal. These variations in slope may be a result of variations in the regional temperature of the asthenosphere. Seismic velocity models of the upper mantle (Woodhouse & Dziewonski 1984; Nakanishi & Anderson 1982) are consistent with variations of the average temperature of the upper mantle of about 200 K, assuming a value of the derivative of shear wave velocity with temperature of 3.4 x 10-4km (sK)-' taken from Simmons & Wang (1971). For the half-space cooling model the slope of the geoid height-age relation is directly proportional to the temperature difference between the surface and the asthenosphere (Haxby & Turcotte 1978). Thus, a 200K variation in temperature difference should give rise to less than 20 per cent of variation in the geoid height-age slope. On the other hand, a model which includes convective cooling of the asthenosphere gives much larger variations in geoid height-age slopes for the same temperature variations (see Fig. 17). This is primarily a result of the strong temperature dependence assumed for the viscosity of the Earth's mantle. Fig. 12 shows how a relatively small change in initial temperature of the convecting region causes a large change in the parameter h when the activation energy is taken to be 100 kcalmol-I. The effect of this change in h on the predicted geoid heigh-age relation is shown in Fig. 17. This figure shows that a change in initial asthenospheric temperature of about ISOK would produce an increase in the geoid height-age relation of about a factor of two. The subsidence and the surface heat flux should not be a good way to distinguish between the models. The slightly larger predicted variations in these parameters as a function of initial temperature for this model, compared with the half-space model, is probably not sufficient to be clearly resolved by the data. For example, the observed variations in the rate of subsidence are of order 30 per cent (Hayes 1986); this could be explained by either model. The reason the geoid height may be used to discriminate between the models is that Cooling of a variable-viscosity jluid 575 it is very sensitive to the density (temperature) distribution with depth. Convection causes much more cooling at large depth (in the asthenosphere) than does conductive cooling. The effect of assuming different values for the temperature and pressure dependence of viscosity on model predictions can be estimated using the derived equations. A very different value of the activation energy, E , would have a large effect on the model predictions for a given change in the initial temperature. If E is taken to be much larger than 100 kcalmol-', the temperature change required to produce a given change in h will be reduced. The reverse is true if E is smaller. If the pressure dependence of viscosity, V , is changed it will have a fairly small effect on the variation of h for a given change in initial temperature. However, when the ratio E/V is changed greatly there will be a large effect on the predicted depth extent of the flow [see (30)] and therefore also on the geoid height. A much larger value of V would lead to a thinner convecting region and to smaller changes in the predicted geoid height-age relation for a given change in the initial asthenosphere temperature. Conclusions Expressions are derived which give some theoretical understanding of the process of the cooling of a variable-viscosity fluid. Such a cooling process, where the lower viscosity portion of the fluid is driven to convect by the cooling from above, may have geophysical applications. The cooling fluid is separated into two regions which behave differently: the lid where heat is transported completely by conduction and a region where the dominant mode of heat transfer is by convection. The two regions are coupled because there must be a balance between the heat flux out of the convecting region and that which is conductively transported at the base of the lid. The rate of convective transport of heat controls the rate of thickening of the lid as the system cools. Because the rate of convective heat transfer behaves in a simple way, the conductive lid thickness increases linearly with and its temperature structure can be described by a similar solution. A simple mathematical description is given for the regular features seen in numerical experiments on cooling fluids with different viscosity parameters. These parameters include the temperature and pressure dependence of viscosity. Several approximations are used to show that simple equations can describe the lid temperature with time for this problem. Only one parameter (A) is needed to describe the cooling of the lid, and it is explicitly shown how to relate the viscosity and other parameters of the convecting system to that parameter. The theory developed here is shown to be consistent with the numerical results. With certain assumptions, it is claimed that this simple system could describe the cooling of the oceanic lithosphere and asthenosphere. The average subsidence and surface heat flow are related t o h, while the rate of change of the convective wavelength and the isostatic geoid height depend on the viscosity parameters as well as h. The parameterization derived here allows one t o estimate the effect of changes in model parameters (mainly the initial temperature of the convecting region) on the geophysically relevant results of the models without doing costly computer models. Temperatures in different areas of the asthenosphere may be different, and this is commonly assumed to be the cause of seismic velocity variations between different areas. Temperature variations will cause variations in the asthenospheric viscosity and in the rate of change of both subsidence and geoid height. The prediction of this model is that for changes in the average asthenospheric temperature of about 200 K there will be a relatively small change in the rate of subsidence (- 30 per cent) while the geoid height-age relationship could change by 100 per cent. This is about the magnitude of variations observed (Andrews et al. 1986). 576 W.R.Buck Acknowledgments Discussion with Marc Parmentier, Barry Parsons, Jean Andrews and Bill Haxby greatly improved this work. Thanks also for the suggestions of two anonymous reviewers. Support provided by an Exxon Graduate Teaching Fellowship while starting this work at MIT and a Larnont Post-Doctoral Fellowship at Columbia. Lamont contribution no. 4100. References Andrews, J . A., Haxby, W. F. & Buck, W. R., 1986. Regional variations along ridge crests: geoid height, depth to ridge crest and geoid height-age relationship (abstract), EOS, Trans. Am. geophys. Un., 67,356. Booker, J . R., 1976. Thermal convection with strongly temperature dependent viscosity,J. RuidMech., 76.741-754. Buck, W. R., 1983. Smallscale convection and the evolution of the lithosphere, EOS, Truns. Am. Geophys. Un., 64,309. Buck, W. R., 1984. Small-scale convection and the evolution of the lithosphere, PhD Thesis, Massachusetts Institute of Technology, Cambridge. Buck, W. R., 1985. When does small-scale convection begin beneath oceanic lithosphere?, Nature, 313, I 1 5 -117. Buck, W. R. & Parmentier, E. M., 1986. Convection beneath young oceanic lithosphere: implications for thermal structure and gravity,J. geophys. Res., 91,1961-1974. Carrigan, C. R., 1984. Time and temperature dependent models of cooling reservoirs: application to volcanic sills, Geophys. Res. Lett., 11,693-696. Carlslaw, H. S. & Jaeger, J. C., 1959. Conduction of Heat in Solids, Oxford University Press. Christensen, U., 1984. Convection with pressure- and temperaturedependent non-Newtonian rheology, Geophys. J. R. astr. Soc., 77,343-384. Cochran, J . R., 1986. Variations in subsidence rates along intermediate and fast spreading mid-oceanic ridges, Geophys. J. R. astr. Soc., 87,421-454. Curlet, N. W. E., 1976. Experimental and numerical modelling of natural convection in an enclosure, PhD thesis, Massachusetts Institute of Technology. Fleitout, L. & Yuen, D., 1984. Secondary convection and the growth of the oceanic lithosphere, Phys. Earth planet. Int., 36, 181-212. Goetze, C., 1978. The mechanisms of creep in olivine, Phil. Trans. R. SOC.A, 288,99-119. Haxby, w. F. & Turcotte, D. L., 1978. On isostatic geoid anomalies,J. geophys. Res., 83,5473-5418. Haxby, W. F. & Weissel, J. K., 1986. Evidence for smallscale convection from SEASAT altimeter data, J. geophys. Res., 91,3507-3520. Hayes, D. E., 1987. Age depth relationships and depth anomalies in the southeast Indian Ocean and south Atlantic Ocean, J. geophys. Rex, in press. Houseman, G . A. & McKenzie, D. P., 1982. Numerical experiments on the onset of convective instability in the Earth’s mantle,Geophys. J. R. ustr. Soc., 68, 133-164. Ingersoll, A., 1966. Convection instabilities in plane Couette flow, Physics Fluids, 9 , 6 8 2 4 8 9 . Jaupart, C., 1981. On the mechanisms of heat loss beneath continents and oceans, PhD thesis, Massachussets Institute of Technology. Kohlstedt, D. L., Nichols, H. P. K. & Hornack, P., 1980. The effect of pressure on the rate of dislocation recovery in olivine, J. geophys. Rex, 85,3125-3130. McKenzie, D. P., 1967. Some remarks o n heat flow and gravity anomalies, 1. geophys. Res., 72, 626 1-6273. McKenzie, D. P., 1977. Surface deformation, gravity anomalies and convection, Geophys. J. R . ustr. Soc., 48,211-238. McKenzie, D. P., Roberts, J . M. & Weiss, N. O., 1974. Convection in the Earth’s mantle: towards a numerical simulation, J. Fluid Mech., 62,465-538. Nakanishi, I. & Anderson, 0. L., 1982. World-wide distribution of group velocity of mantle Rayleigh waves as determined by spherical harmonic analysis,Bull. seism. Soc. Am., 72, 1185-1 194. Parker, R. L. & Oldenburg, D. W., 1973. Thermal model of ocean ridges,Nuture, 242, 137-139. Parmentier, E. M., 1975. Studies of thermal convection with application to convection in the Earth’s mantle, PhD thesis, Cornell University. Cooling of a variable-viscosityfluid 577 Parmentier, E. M., 1978. A study of thermal convection in non-Newtonian fluids, J. Fluid Mech., 84, 1-11. Parsons, B. & Daly, S., 1983. The relationship between surface topography, gravity anomalies and temperature structure of convection,J. geophys. Res., 88, 1129-1144. Parsons, B. & McKenzie, D., 1978. Mantle convection and the thermal structure of plates, J. geophys. Res., 8 3 , 4 4 8 5 - 4 4 9 6 . Parsons, B. & Sclater, J. G., 1977. An analysis o,f the variation of ocean floor bathymetry and heat flow with age, J. geophys. Res., 82,803-827. Reddy, K. P. R., Oh, S. M., Major. L. D. & Cooper, A. R., 1980. Oxygen diffusion in fosterite,J. geophys. Res., 85,322-326. Richter, F. M., 1973. Convection and the large scale circulation of the mantle, J. geophys. Res., 78, 8735-8745. Richter, F. M., Nataf, H.C. & Daly, S. F., 1983. Heat transfer and horizontally averaged temperature of convection with large viscosity variations,J. Fluid Mech., 129,173-192. Richter, F. M. & Parsons, B., 1975. On the interaction of two scales of convection in the mantle,J. geophys. Res., 80,2529-2541. Roberts, G. O., 1979. Fast viscous Bernard convection, Ceophys. Astrophys. Fluid Dyn., 12,235-272. Sammis, C. G., Smith, J. C. & Schubert, G., 1981. A critical assessment of estimation methods for activation volume,J. geophys. Res., 86,10707-10718. Schubert, G., Stevenson, D. & Cassen, P., 1980. Whole planet cooling and the radiogenic heat source contents of the Earth and Moon, J. geophys. Res., 8 5 , 2 5 3 1 -2538. Sharpe, H. N. & Peltier, W. R., 1978. Parameterized mantle convection and the Earth’s thermal history, Geophys. Res. Lett., 5 , 7 3 7 -740. Simmons, G. & Wang, H., 1971. Single Crustal Elastic Constants and Calculated Aggregate Properties; A Handbook, 2nd edn, Massachussets Institute of Technology Press, Cambridge, MA. Stefan, J., 1891. b e r die Theorie der Eisbildung, insbesondere uber die Eisbildung im Polar meere,Ann. Phys. Chem., 42,269-286. Turcotte, D. L. & Oxburgh, E. R., 1967. Finite amplitude convective cells and continental drift,J. Fluid Mech., 2 8 , 2 9 4 2 . Turcotte, D. L., Torrance, K. E. & Hsui, A. T., 1973. Convection in the Earth’s mantle, Meth. comp. Phys., 1 3 , 4 3 1 4 5 4 . Turcotte, D. L., Cook, F. & Willeman, R. J., 1979. Parameterized convection within the moon and terrestrial planets, Proc. 10th Lunar Planet. Sci. Con$, pp. 2375 -2392. Weertman, J. & Weertman, J. R., 1975. High temperature creep of rock and mantle viscosity, A. Rev. Earth planet. Sci., 3, 293-315. Woodhouse, J. H. & Dziewonski, A., 1984. Mapping the upper mantle: three-dimensional modelling of earth structure by inversion of seismic waveforms,J. geophys. Res., 89,5953-5986.
© Copyright 2026 Paperzz