Toward improving parameterization of cloud droplet activation in cloud and climate models, the integro–differential equation for supersaturation is solved analytically for the algebraic size spectrum of the cloud condensation nuclei (CCN) that is equivalent to the lognormal spectrum. The analytical solutions are obtained for four limiting cases that are combinations of two different values of the updraft vertical velocity (small and large) and two different values of the condensation coefficient that correspond to pure and polluted cloud drops. The characteristics of the CCN can vary within each limit. Thus, these four limits and interpolation among them cover the vast majority of cloudy conditions. Analytical expressions are obtained for the time of CCN activation, maximum supersaturation, and the concentration of activated droplets.
For small updraft vertical velocities, these quantities are the products of the power laws by six variables: CCN concentration, mean radius, soluble fraction, vertical velocities, surface tension, and condensation coefficient. At large updraft vertical velocities, the activation time and maximum supersaturation are the products of the power laws of only two variables—CCN concentration and vertical velocity—and are independent of the CCN physicochemical properties. The first limit is a generalization of the Twomey power laws, with Twomey’s coefficient CT and index k expressed via CCN physicochemical properties; the other three limits are new. The accuracy and regions of validity of these limits are determined by comparison with the exact numerical solution to the supersaturation equation. These solutions can be used for parameterization of drop activation in cloud and climate models and for control of numerical solutions. An advantage of this method is that it does not require running parcel models, and the drop concentrations can be obtained from lookup tables or as simple interpolation among the limiting solutions for the instantaneous model parameters.
The indirect aerosol effects on radiative forcing (e.g., Twomey 1977; Albrecht 1989) remain an outstanding uncertainty in simulations of global climate change (Foster et al. 2007). The importance of these effects has motivated development of improved schemes of drop activation for climate models. Drop activation schemes used prior to the 1990s were based mostly on the Twomey (1959) power law for concentration of activated cloud condensation nuclei (CCN), Ndr = CTsmk, where CT and k are the empirical coefficient and index and sm is the maximum supersaturation in a cooling air parcel. The drawbacks of Twomey’s parameterization are that Ndr is unlimited and can exceed CCN concentration Na and parameters CT and k are not related to CCN physicochemical properties.
A substantial revision and intensive development of new cloud drop activation parameterizations began in the 1990s (see reviews in Charlson et al. 2001; Lohmann and Feichter 2005; McFiggans et al. 2006; Svenningsson et al. 2006; Foster et al. 2007). In particular, Kulmala et al. (1993), Shulman et al. (1996), and Laaksonen et al. (1998) generalized Köhler’s theory for the presence of soluble trace gases and slightly soluble substances. Khvorostyanov and Curry (2007, hereafter KC07) generalized Köhler’s theory for the case of small soluble fraction and found analytical solutions to the Köhler equation and for the CCN wet size spectra at aerosol hygroscopic growth, as well as for activity spectra at drop activation without assumption of high dilution. New methods of calculating CCN activation were developed based on lognormal (e.g., Ghan et al. 1993, 1995; Feingold et al. 1994; Fountoukis and Nenes 2005) and algebraic (Cohard et al. 1998, 2000) CCN size and activity spectra, which limited the number of activated drops so that Ndr ≤ Na. Khvorostyanov and Curry (2006, hereafter KC06) and KC07 suggested another algebraic size and activity spectra of dry and wet CCN equivalent to the corresponding lognormal spectra, providing a generalization of the power laws that reconciled the lognormal and algebraic approaches; this approach was used in both bin and bulk cloud models (Morrison and Grabowski 2007, 2008). Parameterizations of drop nucleation have been developed using results from parcel models that are fit to various empirical functions of a few nondimensional parameters (Abdul-Razzak et al. 1998, hereafter AGR98; Abdul-Razzak and Ghan 2000, 2004; Fountoukis and Nenes 2005; Ming et al. 2006). Saleeby and Cotton (2004) developed another approach consisting of lookup tables compiled from many runs of a parcel model.
The necessity of running parcel models many times for compiling lookup tables or tuning the parameterizations makes the procedure time consuming and does not easily reveal the relevant physical relationships. An alternative approach was suggested in Khvorostyanov and Curry (2008, hereafter KC08), who derived the integro–differential equation for supersaturation using the algebraic size and activity spectra from KC06 and KC07 as a starting point. KC08 provided a numerical solution of this equation for the time of CCN activation tm, maximum supersaturation sm, and concentration of activated droplets Ndr as functions of the physicochemical properties and size spectrum of the CCN and the vertical velocity w.
A deficiency of all numerical solutions is that they do not demonstrate the general analytical dependencies of the drop activation on various parameters. The search for analytical solutions for tm, sm, and Ndr is desirable because they would allow the following applications: (i) control for the accuracy of various numerical solutions, especially in the cases where the general dependencies are unclear and spread of the parcel model runs can be significant (e.g., at strong variations of the condensation coefficient or surface tension); (ii) development of newer parameterizations based on the characteristic parameters occurring in these solutions; (iii) construction of the general solutions by interpolation among the particular analytical solutions; and (iv) design of laboratory and field experiments.
It has been known since the pioneering work by Twomey (1959) that the supersaturation equation does not allow general analytical solutions, even with the simplest power-law activity spectrum, unless certain simplifications are made. The situation becomes even more complicated when more complex activity spectra are used (e.g., lognormal or algebraic) because of the impossibility of exact analytical evaluation of the characteristic integrals of this problem. Twomey considered droplet growth with the condensation coefficient αc ∼ 1, that is, in the diffusion regime of droplet growth. Subsequent laboratory and field experiments and theoretical works showed that αc can be as low as 10−5; that is, droplets grow in the kinetic regime (Pruppacher and Klett 1997, hereafter PK97; Seinfeld and Pandis 1998; Buikov and Khvorostyanov 1979; Feingold and Chuang 2002; Chuang 2003). However, parameterizations of drop activation considered mostly the cases with αc ∼ 1 with some numerical estimates of the effects of αc (Nenes et al. 2002; Ming et al. 2006), but, to our knowledge, the general analytical dependencies of tm, sm, and Ndr on αc have not been derived.
Here, the analytical solutions to the general integral supersaturation equation from KC08 are obtained for four limiting cases: small and large updraft vertical velocities and two values of the condensation coefficient, corresponding to the diffusion and kinetic regimes of droplet growth. These solutions express the activation time, maximum supersaturation, and concentration of activated droplets as functions of the vertical velocity and physicochemical characteristics of the CCN including explicit dependencies on the condensation coefficient and surface tension.
2. Basic equations
The basic equations include the size spectrum of dry aerosol, the Köhler equation for dilute solutions, an equation for droplet radius growth rate accounting for kinetic effects, and the integro–differential equation for supersaturation. The size spectrum of dry aerosol fad(rad) by the dry radii rad is described by the lognormal distribution with the aerosol number concentration Na, dispersion σd, and mean geometric radius rd0 related to the modal radius rm = rd0 exp(−ln2σd). It was found in KC06 and KC07 that this lognormal size spectrum is equivalent to the algebraic size spectrum. CCN hygroscopic growth is described by the Köhler equation for supersaturation s = (ρυ − ρυs)/ρυs in dilute solutions (PK97):
where r is the radius of an aerosol particle, Ak is the Kelvin curvature parameter, and B describes the Raoult effect of soluble fraction of CCN:
Here ρυ, ρυs, and ρw are the densities of vapor, saturated vapor, and water; a list of notations is given in appendix A. The parameterization of B as the power law by the dry radius rad [second equation for B in (2.2)] is described in Khvorostyanov and Curry (1999), KC06, and KC07. The parameters b and β in (2.2) depend on the physicochemical properties of the soluble part of an aerosol particle and are given in appendix A. The values β = 1/2 and β = 0 correspond to the soluble fraction proportional to the volume (internally mixed aerosol) and to the surface area (e.g., a dust particle covered by a soluble film).
Here s0 is the mean geometric supersaturation and σs is the supersaturation dispersion:
The index ks0 is an analog of the Twomey (1959) power index kT and is a generalization for s space that accounts for the soluble fraction of the index introduced in Ghan et al. (1993) for the size spectrum, and parameters C0, and η0 are expressed via CCN microphysical properties:
The equation for droplet radius growth rate ṙd accounts for the kinetic correction ξ:
in which Dυ is the water vapor diffusion coefficient, Γ1 is the psychrometric correction to the growth rate due to latent heat release, Vw is the thermal speed of water vapor molecules (Γ1 and Vw are defined in appendix A), and αc is the condensation coefficient. The estimates for the typical conditions and detailed calculations in KC08 show that, if αc ∼ 1, then the kinetic correction ξ ≪ rd, can be neglected in (2.7), and the growth of newly activated droplet proceeds in a diffusion regime. If αc ≤ 0.01, then ξ ≫ rd in (2.7), and droplet growth is kinetic (free molecular).
If an air parcel is rising at vertical velocity w and crosses the condensation level, drops begin to activate, supersaturation increases and reaches its maximum value sm at the time tm when activation ceases, and the concentration of activated drops is Ndr(sm). Using the above assumptions for dry aerosol size spectra, particle activation, and droplet growth, KC08 derived the nonlinear integral–differential equation for the integral supersaturation that governs drop activation kinetics:
The parameter c1 is given in appendix A, τp,ac(t) = [4πDυNaract(t)]−1 is introduced in KC08 as the effective supersaturation relaxation time at the initial stage of condensation, ract(t) = J0(t) is an activation radius, and J0(t) is the integral:
where ref(t, t0) describes droplet growth after activation time t0 to the time tm
A solution to (2.8) can yield the values of sm, time tm, and Ndr(sm). As noted by Twomey, an analytical solution of the complicated nonlinear integral–differential equation (2.8) is a formidable task. Numerical solutions of the supersaturation equation in the parcel models may lead to substantial differences among various models due to different approximations in the description of physical processes (e.g., various choices of αc), finite difference schemes, and numerical techniques (e.g., Lin et al. 2002).
Therefore, instead of numerical calculation of sm and Ndr(sm), we apply the approach developed by Twomey (1959) and Sedunov (1967, 1974). It is based on approximation of the rhs of (2.8) to obtain analytically lower and upper bounds for solutions to (2.8). It will be shown that these bounds appear to be sufficiently close (differing by less than 10%–20%), and any solution between these bounds gives tm, sm, Ndr with sufficiently small errors (∼5%–10%) that do not exceed the typical errors in other parameterizations. The advantage of the analytical approach is that it provides the general analytical dependences of solutions on the input parameters that are often hidden in numerical solutions.
The simplest solution to (2.8) can be obtained by substituting the first term c1w on the rhs into the second term—that is, to search for a solution of the form y′(t) = s(t) = c1wt. Linear functions of this form were used by Twomey (1959) and Sedunov (1967, 1974) to obtain the analytical bounds. We transform the second term on the rhs in (2.8) using the linear approximation for y′(t) = s(t); then
with c5 defined in appendix A and J2 and J3 the dimensionless integrals:
where x = t/t0, z = x1/2, λs is the nondimensional parameter
The time tm when supersaturation in the parcel reaches a maximum sm is determined from the condition s′(tm) = 0 or y″(tm) = 0; then the lhs of (2.12) is zero and we obtain an algebraic expression
After tm is determined, it can be substituted into Λ(tm) in (2.12), which can be integrated neglecting y′ ≪ 1 on the left-hand side, and we obtain a time-dependent solution for s(t):
Equations (2.16) and (2.18) form a system of equations that allows for evaluation of tm and sm in various particular cases when the integrals J2 and J3 are specified. The asymptotic expressions for the integrals J2 and J3 will be obtained below and will allow us to derive four limits of the activation process, depending on λs and αc.
The lower and upper bounds for tm, sm, and Ndr can be evaluated using (2.11), y′(t) = s(t) = amt, and (2.18) to relate sm and tm. This is illustrated in Fig. 1, showing upper and lower bounds for s(t). It is usually presented as a schematic diagram [e.g., Fig. 1 in Twomey (1959) or Fig. 13–3 in PK97], but it is plotted here based on calculations of all curves using (2.17), (2.20), and (2.21). It was shown by Twomey (1959) and Sedunov (1967, 1974) that a choice
overestimates the integral J0 [of the type (2.9) but simpler] on the right-hand side of the supersaturation equation of the type (2.8) because the straight line sov(t) (DE in Fig. 1) lies above the exact curve s(t) and the area of the triangle DEB in Fig. 1 is greater than the area under the curve DCB equal to J0. Therefore, (2.20) underestimates sm and yields Twomey’s lower bound for sm. Twomey showed that another approximation sun,T(t) = c1w(t − tm) + sm underestimates the J0 because the straight line sun,T(t) (AC in Fig. 1) lies below s(t) and the area of the triangle ABC is less than the area below s(t) equal to J0. Therefore, using sun,T(t) overestimates sm(tm) and gives the upper bound for sm. Sedunov (1967, 1974) proved that another choice shown in Fig. 1,
also underestimates the rhs of (2.8) and gives an upper bound for tm and sm; this bound is closer to the exact solution because the line sun,S(t) (DC) lies above sun,T(t) (AC) and better approximates J0. Owing to the coefficient αk1 < 1, sm calculated with (2.18) is lower than point E and much closer to point C; thus (2.18) will be used for evaluation of the bounds.
The integral J0 in (2.9) here and J2 and J3 are much more complicated than J0 estimated by Twomey and Sedunov, which corresponds to the first limit considered in section 3, but the same is valid for these integrals in the other limits: (2.20) overestimates s(t) and J0 and yields a lower bound for sm, whereas (2.21) underestimates s(t) and J0 and yields an upper bound for sm. This is illustrated in Fig. 1 and will be proven for all four limits in subsequent sections.
The drop concentration can be calculated from (2.3) with s = sm and can be written as
The first form (2.22) is similar to Twomey’s power law with correction at large sm, as derived by Ghan et al. (1993, 1995) and modified in KC06 and KC08, accounting for the drop growth after activation and various soluble fractions. The second form (2.23) is the generalized power law; the coefficient C(sm) and index k(sm) are continuous functions of the maximum supersaturation or of the vertical velocity. The equations for C(sm) and k(sm) are given in KC06 and KC08:
Equation (2.16) for tm depends on Λ and contains the integrals J2(tm) or J3(tm), which themselves depend on tm and, in general, cannot be evaluated analytically without some simplifications. However, these equations are easily solved numerically and lookup tables are easily calculated, as illustrated in KC08. Then Ndr(sm) can be calculated in a cloud or climate model using the modified power law described in KC08 and these lookup tables.
Another way of studying the kinetics of drop nucleation is based on simplifying J2 or J3, finding the asymptotic limits of the solutions, and then constructing interpolations among them. There are four particular cases, or four limits, when the general solution described above can be simplified and approximate analytical solutions for tm, sm, and Ndr can be found. These cases are naturally separated by the values of the vertical velocities (or maximum supersaturations sm), determined by the values of λs in (2.15a), and of the condensation coefficients αc that determine diffusion growth of the activated droplets (sufficiently pure water drops) or kinetic growth regime (polluted water drops). Thus, these four limits are
λs ≪ 1 (small vertical velocities w), αc ∼ 1 (diffusion droplet growth limit);
λs ≪ 1 (small vertical velocities w), αc ≪ 1 (kinetic droplet growth limit);
λs ≫ 1 (large vertical velocities w), αc ∼ 1 (diffusion droplet growth limit); and
λs ≫ 1 (large vertical velocities w), αc ≪ 1 (kinetic droplet growth limit).
Equations (2.22) and (2.23) for Ndr(sm) and solutions for sm represent a generalization of the corresponding expressions from Twomey (1959) and Sedunov (1967, 1974), which were based on the power law for ϕs(s) and correspond to the first limit here, as described in section 3. These four limits, along with the situations intermediate among them, cover the vast majority of the situations that can be met in atmospheric clouds. Thus, if we find these limits and construct interpolations among them, the problem of parameterization can be solved. However, these limits have physical meaning without interpolation since each of them corresponds to some particular type of cloud with a certain range of vertical velocities. For example, limits 1 and 2 may correspond to stratiform clouds or fog with low vertical velocities w, and limits 3 and 4 correspond to convective clouds with higher w. Thus, for simulation of these cloud types, it can be sufficient to use these limits without interpolation over the range of vertical velocities.
The applicability of these limits is determined by the scaling parameters sm/s0 and λs = (sm/s0)ks0. Figure 2 shows that the values αc = 1010 (the physical value of αc ≤ 1, but αc = 1010 means absence of kinetic correction, ξ = 0) and αc = 1 yields close results; that is, the value αc = 1, which may be characteristic of pure water drops (e.g., AGR98), is close to Twomey’s case with ξ = 0. The regions of the four limits are depicted in Fig. 2b on the λs–w diagram. One can see that the first limit, λs ≪ 1, is valid with αc ∼ 1 and small w. The second limit with αc ≪ 1 is marked in Fig. 1 at w ≤ 1 around the ordinate axis. The third and fourth limits are reached at w > 1–2 m s−1. This representation is somewhat schematic and these regions may extend to larger areas with varying parameters. The applicability of these limits will be illustrated in more detail in the next sections by comparison of these limits with the exact solutions.
3. Limit 1: Small vertical velocity, diffusional growth regime
This limit corresponds to λs ≪ 1 (low w or low s) and αc ∼ 1 (diffusion growth regime of activated drops); thus, the denominator of the second fraction in the integrand of J3 [(2.14b)] tends to unity. This is equivalent to the absence of denominator in the differential CCN activity spectrum (2.3), which is then similar to the Twomey (1959) power law
When αc ∼ 1, we can neglect small ξ and also assume small rd(t0). With these assumptions and (3.1), limit 1 is equivalent to the model of activation considered by Twomey, Sedunov, and all subsequent similar models, but their empirical coefficient C0 and index ks0 were expressed in (2.5) and (2.6) via aerosol parameters. As shown in appendix B, J3 in this case is reduced to Twomey’s integral with the Euler beta function ℬ(x,z):
a. Lower bound
(The superscript i hereafter denotes the ith limit with i = 1–4, and the lower subscript l means hereafter lower bound). The maximum supersaturation is from (2.18) in this limit
and αk1 was defined in (2.19). Note that and are related as , that is, almost as in Twomey’s upper approximation (2.20) sov(t) but with the coefficient αk1. Table 1 shows that is only 11%–24% lower than sov(t). The droplet concentration from (2.22) is
The powers of w in the expressions for sm and Ndr are the same as derived by Twomey (1959), but his empirical indices and coefficients are expressed now via aerosol microphysical parameters. The coefficient αk1 is similar to the first major term in the Twomey power series expansion for the lower bound of sm [Twomey’s Eq. (7)]. The more detailed expressions can be obtained if s0 is expressed via rd0 and b from (2.4) and Ak is expressed via ζsa and T using (2.2):
Qw is given in appendix A. Equation (3.8) shows that tm decreases (i.e., activation ceases faster) with the increase of Na, rd0, solubility b (i.e., ɛυ), and updraft w and with the decrease of surface tension in the presence of surfactants (e.g., organics). In particular, with ks0 ∼ 2 (σd = 1.7, β = 0.5), (3.8) yields
Thus, sm increases with increasing w and ζsa and decreases with an increasing supersaturation absorption rate, that is, increasing Na, rd0, and b (or ɛυ). In particular, with ks0 ∼ 2, we have
The index of ζsa is negative—that is, the maximum drop concentration grows with decreasing surface tension [as with increasing concentration of organics; e.g., Facchini et al. (1999, 2000) and Abdul-Razzak and Ghan (2004)]—and the other five power indices are positive; thus, Ndr increases with the increase of any of the four factors: Na, rd0, b (or ɛυ), and w. The power by T is overwhelmed by decreasing , and Ndr decreases with increasing T, as obtained in KC08 in agreement with Saleeby and Cotton (2004). In particular, with ks0 ∼ 2, we have
The variations δNdr predicted by (3.11) at small variations of the surface tension δζsa are as follows:
In particular, with ks0 ∼ 2, then δNdr,l/Ndr,l ∼ (−3/4)δζsa/ζsa, which exactly coincides with the expression derived in Facchini et al. (1999). For a broader size spectrum with σd = 2.5 and β = 0.5, we have ks0 ∼ 1.16 (Table 1) and CNζ ≈ 0.55; that is, the dependence of Ndr,l on surface tension becomes weaker for broader CCN spectra, but the general qualitative dependence of Ndr,l on ζsa in the first limit is similar to that in Facchini et al. (1999) and Abdul-Razzak and Ghan (2004). Note that these conclusions are drawn since we here consider ζsa as a variable and do not consider effects of its variations due to redistribution of organics between the surface and volume in CCN, as in Li et al. (1998) and Abdul-Razzak and Ghan (2004). According to Facchini et al., a 30% decrease in ζsa should cause a ∼20% increase in Ndr, which should lead to an increase in top of the atmosphere albedo locally by ∼1% and could cause a global mean forcing with an upper limit of −1 W m−2. The opposite effects of decreasing Ndr with increasing organic coating of CCN and decrease of ζsa were predicted by Feingold and Chuang (2002) and other studies reviewed in Foster et al. (2007). The equations here allow estimations of the effects of surface tension with various other CCN properties. These effects can be noticeable for small w, but it will be shown in sections 5 and 6 that Ndr is highly insensitive to surface tension at high w (>1–2 m s−1).
b. Upper bound
Evaluation of the upper bound in this and other limits is a little more complicated and is done using the system of (2.16) and (2.18) to determine the two unknown, tm and sm. Substitution of J3(1) from (3.2) and am = sm/tm from (2.21) into (2.13a) yields Λ(tm). Substituting it into (2.16) and (2.18) and solving for tm, we obtain for , , and in the upper bound (subscript u) of the first limit almost the same expressions as (3.8), (3.10), and (3.11) but multiplied by the coefficient for in (3.8), by instead of αk1 for in (3.10), and instead of for in (3.11). The relation [similar to Twomey’s upper approximation (2.20)] arises again, but with the coefficient . The relation of the lower and upper bounds in the first limit is determined by the coefficient :
The coefficients αk1, , and along with the index ks0 are given in Table 1 for various size dispersions σd. One can see that ks0 decreases from 5.84 at σd = 1.2 (narrow spectra) to 1.16 at σd = 2.5 (wide spectra). All the coefficients are close to 1, and their variations are much smaller: αk1 decreases from 0.89 to 0.76 and decreases from 0.993 to 0.957. Thus, the lower bound for sm obtained with (2.18) is 11%–24% lower than would be obtained if applied the linear relation (2.20) for the maximum values as sm = c1wtm, and the upper bound is only 1%–4% lower. Calculated variations of the coefficients with β = 0 (not shown here) are even smaller. For all σd, ; that is, the condition is satisfied that the upper bound is higher than the lower bound. However, the difference between the bounds is rather small. The coefficient , which characterizes the ratio of the upper to the lower bounds, increases from 1.116 to 1.260 in this range; that is, the difference does not exceed 12% and 26% for the narrow and wide spectra, respectively. If one assumes that the true value is close to the mean of the upper and lower bounds, the error would be halved: ∼6% and 13%.
A comparison was performed of these approximate analytical limits for tm, sm, and Ndr with a more precise numerical solution of the complete supersaturation Eq. (2.8) with (2.22)–(2.26) as described in KC08, referred to hereafter as the “exact solution.” Figure 3 shows that the agreement between the first limit and exact solution for tm and sm is good for both bounds at w ≤ 30–50 cm s−1. A comparison of sm with the parameterization from AGR98 also shows satisfactory agreement; the curve calculated using AGR98 algorithm is closer to our lower bound for sm at small w and coincides with the exact solution at w > 0.5 m s−1.
Figure 3c shows a comparison of Ndr calculated using three methods. The first limit for Ndr (3.11), that is, the traditional power law (solid squares), is close to the exact solution only at w = 1–4 cm s−1 (diamonds). At greater w using the power law may significantly overestimate droplet concentrations. This curve and conclusion are in close agreement with those from Ghan et al. (1993, Fig. 7): This was the major reason for revisions of the Twomey theory. The upper bound of the “modified power law” (2.23) is close to the exact solution to w ∼ 0.5 m s−1: the error does not exceed 10%–15%. Thus, the modified power law for Ndr allows one to decrease the errors. This comparison justifies the validity of the first limit and shows the region of its applicability.
c. Comparison with the Twomey power law
A more detailed comparison with the Twomey power law allows expression of Twomey’s empirical coefficients CT and k via aerosol microphysical parameters. The expression for maximum supersaturation, derived in more general form in PK97, their Eq. (13-40), can be rewritten as
where CT is Twomey’s empirical constant of the activation spectrum and we introduced a new coefficient CST for convenience of comparison (a misprint in PK97 in the sign of the power index of CT is corrected). The coefficients A1 and A2 describe supersaturation generation and absorption, respectively, in the supersaturation equation [PK97, Eq. (13–30)]
where ρL is the liquid water content and A3 in (3.14b) is the coefficient in drop radius growth rate [Eq. (13–32) in PK97]. In our notation, A1 = c1, A3 = c3. The expression for A2 in PK97, their Eq. (13–31),
with esw being the saturated vapor pressure, can be simplified using the equations of state for water vapor and air as A2 = Γ1/ρvs, where Γ1 is the psychrometric correction defined in appendix A. Substitution of A1, A2, and A3 into (3.14b) yields
It follows from (2.3) or (3.1) that, for k = ks0, in this limit (sm/s0)ks0 ≪ 1. Then, writing c33/2 = c31/2c3 and using (2.7) for c3, we see that CST exactly coincides with our in (3.6). Equating our expression for the upper bound of sm (3.10) with correction (3.12) and Twomey’s (3.14a), dividing by , and using the definitions of b and Qw, we obtain
Equation (3.17) is valid for both β = 1/2 and β = 0 and sm measured in percent. Thus, Twomey’s empirical coefficient CT is expressed via Na, rd0, σd (in ks0), T, ζsa, β, b (or ɛv), ρs, and other properties of CCN. In addition, these dependencies are nonlinear—governed by the powers of ks0. It is not surprising that measurements made over a period of decades have found such different values for CT, varying by several orders of magnitude in various air masses and clouds (Hegg and Hobbs 1992; Table 9.1 in PK97; Seinfeld and Pandis 1998). Twomey’s empirical index k, equal to ks0 at s ≪ 1%, is also expressed with (2.5) via σd and β. Equations (3.17) and (2.5) explain many features of the observed values of CT and k. In particular, since CT ∼ Na, CT has higher values in continental than in maritime air masses, as well as greater values in polluted areas. Further, CT also increases with increasing soluble fraction ɛv, mean radius of CCN rd0, and with decreasing surface tension in the presence of surfactants.
Substituting any values of CCN parameters into (3.17) and (3.18) yields concrete values of CT and Ndr for a given air mass. It should be noted that the typical calculated values of CT and k in the limit s ≪ 1% are usually much greater than often measured, k ∼ 0.2–0.6 at s = 0.2%–1%, and then used in parameterizations. This reflects a substantial difference between Twomey’s model using typical k ∼ 0.2–0.6 (PK97) corresponding to some fixed and sufficiently high s and our model with calculated k(s), which is a generalization of Twomey’s model for arbitrary continuous s. This is illustrated in Fig. 4, which depicts C(sm) and k(sm) calculated with (2.24)–(2.26) and presented as functions of sm and w simultaneously: C(sm) ≈ 8000 cm−3 at w = 1 cm s−1 (sm = 0.043%) and decreases to the limit ∼500 cm−3 at w ≈ 1–2 m s−1 (sm = 0.5%–1%). Experimental data on C(sm) are not readily available. However, the calculated index k(sm), which also decreases from 1.67 to below 0.2 in this range of sm or w (Fig. 4b), is in a good agreement with data both from cloud chamber and field experiments (e.g., Jiusto and Lala 1981; Yum and Hudson 2001). This indicates that this model generalizes the Twomey model and allows correct prediction of decreasing C(sm) and k(sm) with increasing s. If these parameters are measured at some s, they can be easily recalculated for any value of s using (2.24)–(2.26), but this requires additional measurements or hypotheses on CCN size spectrum and chemical composition. If measurements of all variables are available, then (2.24)–(2.26) for C(sm) and k(sm) can be verified.
4. Limit 2: Small vertical velocity, kinetic growth regime
For small w, the activation spectrum is also approximated by the power law (3.1), but the first fraction and integrand in J3 are substantially different than that in the Twomey model owing to the kinetic drop growth rate. The integral J3 for this case is evaluated in appendix B:
In contrast to the first limit (3.2) for J3(1) in the diffusion regime that contained only the beta function, now J3(2) contains also several coefficients in dimensionless combination.
a. Lower bound
The lower bound for sm is calculated using (2.18):
The lower bound for Ndr is calculated from (2.22) taking into account that sm ≪ s0 (lower s limit):
For ks0 = 2, it follows from (4.8) in the kinetic regime:
For ks0 = 2, (4.9) gives in this kinetic regime
Comparing with the diffusion limit (3.10a), we see that the dependencies on Na, rd0, ζsa, T, and b are substantially weaker in the kinetic regime, and the dependence on w is slightly stronger. The droplet concentration from (4.6) and (4.7) converts into
In particular, for ks0 = 2, (4.10) gives in the kinetic regime
Comparing this to the diffusional growth limit (3.11a) from section 3, we see that the dependencies on the six parameters (Na, rd0, b, ζsa, T, w) become somewhat stronger in the kinetic limit. In addition, this limit includes explicit dependence on the condensation coefficient αc. An advantage of this method of asymptotic limits is that it allows for simple estimates of the effects of variations of the condensation coefficients on tm, sm, and Ndr. It follows from (4.8)–(4.10) that
b. Upper bound
Derivation of the upper bound, as for the first limit, is based on the use of the system of equations, (2.16) and (2.18). Substituting (4.1) for J3(2) with am = sm/tm from (2.21) into this system and solving for tm and sm, we obtain for , , and in the upper bound (subscript u) of the second limit (superscript 2) almost the same expressions as (4.8)–(4.10) but multiplied by the coefficient for in (4.8), by instead of αk1 for in (4.9), and by instead of for in (4.10). That is, the relation of the limits is
The coefficients αk1, , and are given in Table 1. As in the first limit, all the coefficients are close to 1, and is 2%–8% lower than 1. In this range of σd, increases from 1.099 to 1.20 and characterizes the ratio of tm and sm in the upper to lower bounds. Shown in Fig. 5 is a comparison of the upper bound of the second limit with the exact solution to the supersaturation Eq. (2.8). The agreement becomes sufficiently good at αc ≤ 0.04, and w ≤ 2–3 cm s−1. A surprising finding is that the value αc = 0.04, used in many cloud models, is close to the boundary of the kinetic regime, and the values of tm, sm, and Ndr are substantially greater in this regime than in the first limit with αc = 1 and a diffusion regime (cf. Figs. 3 and 5).
5. Limit 3: Large vertical velocity, diffusional growth regime
At large w—that is, in the limit sm ≫ s0 (λs ≫ 1)—the integral J3 in (2.14b) in general cannot be reduced to the beta function and it is more convenient to use (2.16) for evaluation of tm, with (2.13b) for Λ and (2.14a) for J2. The integral J2(3)(t) in this limit is evaluated in appendix B as
a. Lower bound
Substituting here am = c1w from (2.20) we obtain a simple asymptotic for the lower bound (subscript l) of the third limit (superscript 3) at large w:
This gives for sm from (2.18),
b. Upper bound
The upper bound, as in the previous sections, is obtained by substituting am = sm/tm from (2.21) into (5.2), then substituting the resulting relation tm(sm) into the system of equations, (2.16) and (2.18), and solving for tm, sm. Then we obtain tmu similar to (5.3) but multiplied by the coefficient and smu similar to (5.5) but with coefficient instead of αk1 so that
The coefficients αk1, , and are given in Table 1. They are also close to 1. The coefficient αk1 for the lower bound is the same for all four limits as the upper bound is lower than 1 by 3–7% and the ratio of the upper to lower limits, , is 1.09–1.23. That is, the error of any solution average between these bounds does not exceed 5%–11%.
Equations (5.3)–(5.9) show that tm and sm in the limit of high w are also related by the linear relation similar to (2.18) with some coefficients ≤ 1 and also have the form of power laws but only by Na and w. The dependencies of tm and sm on rd0, σd, b, β, and on curvature parameter Ak or surface tension ζsa, which were very substantial in the first and second limits at small w, disappear from the coefficients and power indices in the limit of high w. Thus, in the third limit with large w and αc ∼ 1 (sufficiently pure drops) we come to an unexpected result: the maximum supersaturation sm and activation time tm are independent of the CCN size spectrum and chemical composition. Thus, tm and sm depend only on CCN concentration Na and vertical velocities w, and the indices of these dependencies are constant and universal for various CCN; that is, they are the same for different values of ks0. This somewhat paradoxical result can be explained by the fact that the third limit deals with the region λs ≫ 1 where sm is much greater than the mean geometric s0. Thus, the vast majority of CCN will be activated, no matter what their individual properties are. This independence of the physical and chemical properties of CCN indicates that the effects of surface tension and chemical composition on the global albedo and radiative balance should be evident in stratiform clouds and become much weaker in convective clouds.
Figure 6 shows that the tm and sm in the upper bound of the third limit approach the corresponding exact solutions at w > 1–2 m s−1, but this conclusion is valid only for αc ∼ 1 and becomes invalid at αc = 0.1. At w < 0.7–1 m s−1, the third limit is invalid. This independence of rd, σd, and b (or ɛυ) at high w was illustrated in Figs. 3, 5, and 7 in KC08, where tm(w) and sm(w) in double-log coordinates represented almost straight lines with slopes slightly increasing toward large w. The corresponding curves tm(w) and sm(w) converged at w ≥ 0.5–2 m s−1 toward the limits with fixed slopes. The simple equations of the third limit here provide an explanation for this as well as exact values of the slopes. The expression (2.22) for Ndr shows that Ndr tends to Na at λs ≫ 1 (sm ≫ s0); we can approximate it by expanding the denominator in (2.22) into a power series by s0/sm ∼ 1/λs and retaining two terms—that is, Ndr or fraction activated Ndr/Na is a parabolic function of (sm/s0)−ks0:
and sm is defined in (5.5) and (5.7). Figure 6 shows that Ndr in the third limit at αc ∼ 1 also approaches the exact solution at w > 1–2 m s−1 but, at w ≤ 0.5 m s−1, Ndr increases with decreasing w; that is, the number of terms in expansion (5.10) becomes insufficient and (5.10) becomes invalid. In contrast to tm and sm, the value of Ndr depends of the CCN size spectrum and chemical composition via s0, but this dependence is substantially weaker than in the first and second limits.
6. Limit 4: Large vertical velocity, kinetic growth regime
When αc ≪ 1, then ξ ≫ rd(t0), and rd(t0) can be neglected in Us, so Us = Vs2; furthermore, (1 − x2) ≪ Us1/2 in J2, and (1 − x2) ≪ Vs. The integral J2 in the fourth limit is evaluated in appendix B:
a. Lower bound
b. Upper bound
The upper bound for tmu and smu is obtained as in the previous sections using (2.21) for am, substituting J2 from (6.1) into (2.16), (2.18), and solving the system of two equations. Then we obtain tmu similar to (6.2) but multiplied by the coefficient and smu similar to (6.4) but with coefficient instead of αk1, so that
The coefficients αk1, , and are given in Table 1. They also are close to 1. The coefficient αk1 for the lower limit is the same, for the upper limit is lower than 1 by 5%–10%, and , the ratio of the upper to lower limits, is 1.07–1.18. That is, the error of any solution mean between these limits does not exceed 3.5%–9%.
Thus, the activation time tm and maximum supersaturation sm in both bounds of the fourth limit do not depend on the CCN size spectrum (rd, σd), surface tension, soluble fraction β, b, or ɛυ, or chemical composition; they depend only on CCN concentration Na, vertical velocities w, and condensation coefficient αc. These dependencies have the form of the power laws, their indices are constant and universal for various CCN similar to the third limit, and—in contrast to the first and second limits at small w, where the indices depend on the size dispersions—the drop concentration Ndr can be calculated from (5.10) as a parabolic function of λs below Na. It depends on physicochemical properties of CCN as in the third limit but much weaker than in the first and second limits.
The characteristics of the fourth limit are compared with the exact solution in Fig. 7 for various αc. The fourth limit calculated with αc = 1 is far from the exact solution at all w, which is quite natural since αc = 1 is characteristic of diffusion regime. With αc = 0.04, the tm and sm in the fourth limit are already close to the exact solution. This means that αc = 0.04, often used in the models, is closer to the kinetic than to diffusion regime. For αc = 0.01, characteristic of polluted clouds, the parameters tm, sm, and Ndr almost merge with exact solutions at w ≥ 10 cm s−1.
7. Interpolation equations and comparison with exact solutions
General solutions for tm and sm can be constructed by interpolating between the limiting solutions with some interpolation function Ωw(w, wsc) or Ωα(α, αsc). For example, interpolations between the first and third limits can be constructed as
where wsc is a scaling parameter. The function Ωw(w, wsc) should satisfy the conditions Ωw(w = 0) → 1, Ωw(w ≫ wsc) → 0, which ensure corresponding limits at small and large w. The third term on the rhs of this equation is added to make these fitting functions closer to the exact solutions at the intermediate w, and the product Ωw(1 − Ωw) vanishes at w → 0 and w ≫ wsc.
Figure 8a shows sm calculated with Eq. (7.1) and Ωw(w, wsc) = exp(−w/wsc) with wsc = 1.8 m s−1 and Δsm13 = sm(3)(5 m s−1) − sm(3)(1 cm s−1). The fitted interpolated curve (solid circles) tends to the corresponding limits at w → 0 and w ≫ wsc, lies above both limits owing to addition of the term with Δsm13, and is very close to the exact curve. The Ndr(w) calculated with this interpolated sm(w) and the modified power laws (2.22) or (2.23) is also close to the exact curve. This example indicates that more precise interpolation expressions for Ndr can be obtained using (2.22) or (2.23) and interpolation among various analytical limits for sm. A detailed elaboration of this interpolation procedure among all four limits requires a special careful choice of several functions Ωw, which is beyond the scope of this paper and will be described elsewhere.
8. Summary and conclusions
The integro–differential equation for the integral water supersaturation in clouds, derived and solved numerically previously by the authors (KC08), is solved here analytically for some limiting cases. The size spectrum of dry CCN is assumed to be algebraic equivalent to lognormal with the concentration Na, mean geometric radius rd0, dispersion σd, and solubility parameter b proportional to the soluble fraction ɛυ. CCN may have variable chemical composition and surface tension ζsa. The analytical solutions are obtained for four limiting cases that are determined by the values of a nondimensional parameter λs (2.15a) and of the condensation coefficient αc. The parameter λs is proportional to w, rd0, and ɛυ, and for the typical for CCN values the λs criterion is equivalent to the two limits of small and large updrafts w. Thus, there are four combinations of two limits of w (small and large) and two limits of the condensation coefficient: αc ∼ 1 (pure cloud drops; diffusional regime of droplet growth) and αc ≪ 1 (polluted cloud drops; kinetic growth regime). The other characteristics of CCN (Na, rd0, and σd, composition, soluble fraction ɛυ and its distribution in the CCN, temperature, and pressure) can vary within each of these limits. Thus, these four limits and interpolation among them cover the vast majority of cloudy conditions.
Analytical expressions are obtained for the activation time tm, maximum supersaturation sm, and the concentration of activated droplets Ndr. Since the exact analytical solution of the integro–differential supersaturation equation is not feasible, the solutions are found using and generalizing the methods from Twomey (1959) and Sedunov (1967, 1974) as the lower and upper bounds for each quantity. The solutions for the bounds are almost identical and differ only by the coefficients. These coefficients are close to 1 and to each other and differ not more than 10%–20%; that is, any superposition between these bounds has a smaller error.
There is significant difference between the regimes of small updrafts w and large w. In the limit of small w in both the diffusion and kinetic regimes (first and second limits), the expressions for tm, sm, and Ndr are the products of six power laws of Na, rd0, b, w, ζsa, and αc with coefficients depending on pressure and indices depending on the size dispersion and soluble fraction. The modified power laws (2.22) and (2.23) substantially better approximate Ndr. The temperature dependence is negative (Ndr decreases with increasing T). The first limit of low w and diffusional growth is a generalization of the Twomey (1959) power laws, allowing Twomey’s empirical coefficient CT and power index k to be expressed directly as functions of the CCN microphysical parameters and chemical composition and extended from small supersaturations to any values. The limit of low w and kinetic growth provides for the first time an analytical dependence on the condensation coefficient and surface tension that can be verified both experimentally and using other theoretical methods, for example, parcel modeling. The often-used value αc = 0.04 is close to the boundary of kinetic regime. A comparison of these two limits with the exact numerical solutions from KC08 shows that the regime of “small w” can be valid up to w ≤ 0.2–0.5 m s−1, that is, for synoptic-scale updrafts and turbulent fluctuations of vertical velocity.
The solutions for large vertical velocities (w > 1–2 m s−1) are substantially different from the small vertical velocity solutions. The quantities tm and sm are also the products of power laws, but only by w and Na, and are independent of CCN properties. This fairly surprising finding can be explained by the fact that sm at large w is much greater than the mean geometric supersaturation s0, the majority of CCN have been activated, and therefore the dependence on CCN properties weakens and vanishes. The limits of w for onset of this regime increase with decreasing rd0 and soluble fraction. These expressions allow significant simplification of calculating drop nucleation in sufficiently strong updrafts in convective clouds.
In contrast to some previous parameterizations predicting power laws by sm and unlimited Ndr, the expression for Ndr found here for large w is described by the modified power laws (2.22) and (2.23) and is bounded by Na, as in Ghan et al. 1993, 1995; AGR98; Cohard et al. 1998, 2000; Fountoukis and Nenes 2005; KC06; KC08). A comparison of these two “large w” limits with exact numerical solutions from KC08 showed good agreement with accuracy of a few percent, indicating that the regime of large w can be valid for Ndr at w ≥ 1–2 m s−1.
These asymptotic analytical solutions can be used as controls for the accuracy and optimization of numerical solutions, estimating cloud drop spectral dispersions and the indirect dispersion effect on climate (e.g., Liu et al. 2006), design of field and laboratory experiments, parameterization of droplet activation by interpolation among the four limits for cloud and climate models, physically interpreting and understanding variations of the activation process with changing atmospheric composition, and investigations of aerosol indirect radiative forcing.
This research has been supported by NASA grant NNX07AQ26G and the NASA Modeling and Parameterization Program. The authors are grateful to the two anonymous reviewers for useful remarks that allowed us to clarify and improve this paper.
List of Notations
Kelvin curvature parameter
am slope in approximation s(t) = amt, (2.11)
parameter of Raoult effect of CCN soluble fraction
for β = 1/2
for β = 0
C(sm) coefficient of activity spectrum, (2.23)
C0 parameter of activity spectrum, (2.6)
CT Twomey’s empirical constant, (3.17)
c3 parameter of droplet growth rate, (2.7)
cp specific heat capacity of air
Dυ water vapor diffusion coefficient
g gravitational constant
k(sm) index of activity spectrum, (2.25)
kd0 = 4/(2πσd) power index of the size spectrum
ks0 index of activity spectrum at s ≪ 1, (2.5)
Le latent heat of condensation
Ma, Mw molecular weights of air and water
Ms molecular weight of the soluble fraction
ms mass of the soluble fraction
Na, Ndr aerosol and drop number concentration
Qw = (25/3Mw/3Rρw)3, (3.8)
R universal gas constant
Ra, Rυ gas constants for air and water vapor
rad, rd radius of a dry CCN and a droplet
rad1 scaling radius in b with β = 0
rd0 and rm the mean geometric and the modal radii of CCN size spectrum
s = (ρυ–ρυs)/ρυs supersaturation
s0 mean geometric supersaturation, (2.4)
sm maximum supersaturation in a parcel
sml, smu lower and upper bounds of sm
T temperature (K)
tm time when maximum sm is reached
tml, tmu lower and upper bounds of tm
Us, Vs, nondimensional parameters (2.15b)
thermal speed of water vapor molecules, (2.7)
w vertical velocity
the integral supersaturation
αc condensation coefficient
αk1 = (ks0 + 2)/(ks0 + 3), (2.19)
ℬ(x, z) Euler’s beta function
β describes composition of CCN in (2.2)
psychrometric correction to the drop growth rate, (2.7)
ɛm mass soluble fraction of CCN
ɛυ volume soluble fraction of CCN
ɛv0 reference soluble fraction for β = 0
ζsa surface tension at the CCN surface
η0 parameter of activity spectrum, (2.6)
λs = (amt/s0)ks0 parameter, (2.15a)
ν the number of ions in solution
ξ kinetic correction to radius growth rate, (2.7)
ρυ, ρυs, ρw densities of vapor, saturated vapor, and water
σd the CCN dry size spectrum dispersion
σs supersaturation spectrum dispersion, (2.4)
Φs osmotic potential, (2.2)
ϕs(s) CCN activity spectrum, (2.3)
χ(sm) parameter, (2.26)
Evaluation of the Integrals J2 and J3 for Four Limiting Cases
The J2 and J3 are the dimensionless integrals introduced in section 2:
Here x = t/t0, z = x1/2, and the three nondimensional parameters λs, Us, and Vs are
Limit 1: Small w, diffusional growth regime
In this case, λs ≪ 1; thus, the denominator of the second fraction in the integrand of J3 tends to unity, which corresponds to the power-law activity spectrum (3.1). We make additional approximations in (B.2), consider the case αc ∼ 1, neglect small ξ, and assume small rd(t0). Then Us and Vs vanish, and the integral J3 (B.2) is simplified and reduced to the Euler beta function ℬ(x,z), as in Twomey’s model:
Limit 2: Small w, αc ≪ 1
When αc ≪ 1, then ξ ≫ rd(t0) and rd(t0) can be neglected in (2.15b) or (B.4) for Us; then Us ≈ ξ2/(amc3t2) = Vs2. We also have in (B.2) for J3 the inequalities (1 − z) ≪ Us and (1 − z) ≪ Vs2. Expansion of the first fraction in the integrand of J3 in (B.2) by the small parameter (1 − z)/Vs2 yields
Substituting this expression for the first fraction into J3, we obtain
Limit 3: Large w, diffusional growth regime
At large w—that is, in the limit sm ≫ s0 (λs ≫ 1)—the integral J3 in (2.14b) or (B.2) in general cannot be reduced to the beta function, and it is more convenient to use J2(3)(t). Let us consider two functions in the subintegral expression for J2 in (B.1):
Then J2 can be written as
The function ϕnd(x) represents the algebraic CCN activity spectrum in nondimensional form. Its maximum evaluated from the condition d[ϕnd(x)]/dx = 0 is located at
For example, if ks0 = 2, then xm ≈ 0.1 for λs ∼ 30 and xm ≈ 0.18 for λs ∼ 10 (i.e., xm ≪ 1 at λs ≫ 1). It was shown in KC06 that this differential CCN activity spectrum has the shape of a “smoothed” Dirac delta function, and the nondimensional function ϕnd(x) also has this property. This is illustrated in Fig. B1, which shows that ϕnd(x) has a sharp maximum at x = xm, while ψ(x) is a smooth function. Thus, the major contribution to J2 comes from the region around xm, and we can use for evaluation of J2 the “quasi δ function” property of ϕnd(x), or the “mean value theorem” from the mathematical analysis:
The last integral is not equal to 1 and still should be evaluated since ϕnd(x) is proportional to the smoothed delta function but not equal. To find this integral, we use the property found in KC06 that the CCN differential spectrum is a smoothed δ function and is a derivative of the droplet concentration defined in (2.22), which is a smoothed Heaviside step function, found in a similar form by Ghan et al. (1993, 1995). The corresponding fraction activated Fnd(x) = Ndr(sm)/Na can be written from (2.22) in nondimensional form as
One can see that
The last approximate equation uses the fact that λs ≫ 1. With αc ∼ 1 and rd(t0) = 0, the values Us ≪ 1 and Vs ≪ 1, and ψ(xm) ≈ (1 − xm2). Figure B1 shows that at λs ≫ 1 and xm ≪ 1, the value of ψ(xm) is very close to 1—for instance, ψ(xm) = 0.997 in this example, and thus it can be assumed that ψ(xm) = 1. Substituting (B.15) into (B.12) with ψ(xm) = 1 and using (2.15a) for λs yields
The exact numerical calculation of J2 with the functions ϕnd (x) and ψ(x) shown in Fig. B1 yields J2(3) = 1.726 × 10−2, and calculation with approximation (B.16) and λs = 29.7, corresponding to Fig. B1, yields J2(3) = 1.795 × 10−2; that is, the error is less than 4%, so (B.16) can be used with good accuracy in this limit.
Limit 4: Large w (λs ≫ 1), kinetic growth regime (αc ≪ 1)
Now we consider the analytical evaluation of the integral J2 in (2.14a) or (B.1) in the fourth limit, sm ≫ s0 (λs ≫ 1), and αc ≪ 1. When αc ≪ 1, then ξ ≫ rd(t0), and rd(t0) can be neglected in Us; then Us = Vs2; besides, (1 − x2) ≪ Us, and (1 − x2) ≪ Vs2. Similar to the third limit in the previous section, we present the integrand of J2 (B1) as ϕnd (x)ψ(x) with functions defined in (B.9). Similar to the second limit, expansion of ψ(x) by the small parameter (1 − x2)/Vs2 yields
As for the third limit, the function ϕnd(x) has a sharp maximum at x = xm ≪ 1 when λs ≫ 1, and ψ(x) is a smooth function. Using again the quasi δ-function property of ϕnd (x) and the mean value theorem and substituting (B.17) into (B.12), accounting for (B.15), yields
Corresponding author address: Dr. J. A. Curry, School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, GA 30332–0340. Email: firstname.lastname@example.org