A one-dimensional sectional aerosol model integrated with mesoscale meteorological data to study marine boundary layer aerosol dynamics



[1] The dynamics of aerosols in the marine boundary layer are simulated with a one-dimensional, multicomponent, sectional aerosol model using vertical profiles of turbulence, relative humidity, temperature, vertical velocity, cloud cover, and precipitation provided by 3-D mesoscale meteorological model output. The Naval Research Laboratory's (NRL) sectional aerosol model MARBLES (Fitzgerald et al., 1998a) was adapted to use hourly meteorological input taken from NRL's Coupled Ocean-Atmosphere Prediction System (COAMPS). COAMPS-generated turbulent mixing coefficients and large-scale vertical velocities determine vertical exchange within the marine boundary layer and exchange with the free troposphere. Air mass back trajectories were used to define the air column history along which the meteorology was retrieved for use with the aerosol model. Details on the integration of these models are described here, as well as a description of improvements made to the aerosol model, including transport by large-scale vertical motions (such as subsidence and lifting), a revised sea-salt aerosol source function, and separate tracking of sulfate mass from each of the five sources (free tropospheric, nucleated, condensed from gas phase oxidation products, cloud-processed, and produced from heterogeneous oxidation of S(IV) on sea-salt aerosol). Results from modeling air masses arriving at Oahu, Hawaii, are presented, and the relative contribution of free-tropospheric sulfate particles versus sea-salt aerosol from the surface to CCN concentrations is discussed. Limitations and benefits of the method are presented, as are sensitivity analyses of the effect of large-scale vertical motions versus turbulent mixing.

1. Introduction

[2] Understanding and predicting aerosol chemistry and dynamics in the natural environment is a research priority in the atmospheric sciences community. While aerosol effects on climate continue to drive much of this interest, aerosols also play an important role in a number of other disciplines including (1) aerosol corrections to remote sensing measurements, (2) understanding deteriorating regional visibility, (3) evaluating military electro-optical systems performance, (4) health effects, (5) understanding the role aerosols play in determining the microstructure and colloidal stability of clouds, and (6) understanding the transfer of mass from gas-phase pollutants to particulate matter (atmospheric heterogeneous chemistry). The last decade has seen a number of large multi-institutional field campaigns aimed at understanding aerosol in different geographical locations (for example, Aerosol Characterization Experiments (ACE-1, ACE-2, ACE-Asia), INDOEX, etc.). The interplay of the large number of physical and chemical processes in the atmosphere which determine the evolution of the aerosol size distribution is so complicated that numerical models are required as diagnostic tools to adequately interpret the results of these experiments and to validate whether all relevant process are included, and included correctly.

[3] Thus there are a large number of aerosol models to simulate aerosol loading in the atmosphere and the effects of aerosols on various physical and chemical processes. At one extreme lie box (zero-dimensional/parcel) models which include extensive and detailed treatment of all relevant microphysical and chemical processes (i.e., coagulation, nucleation, condensational growth, gas-phase chemistry, etc.) [see, e.g., Raes, 1995; Katoshevski et al., 1999; Jacobson, 2002]. These box models do not contain any atmospheric transport processes such as horizontal advection and turbulent dispersion. At the other extreme are global circulation and chemical transport models to which aerosol mass is added as a prognostic variable; an example would be production of sulfate aerosol mass by the oxidation of gaseous sulfur compounds (eleven such models are compared by Barrie et al. [2001] and Lohmann et al. [2001]). If the actual size distribution is required, the aerosol mass is partitioned according to some specified size distribution (as opposed to a model determination of the size distribution). However, with increasing computational power, detailed microphysical processes can be added explicitly to large-scale meteorological models at each grid point [e.g., Toon et al., 1988; Binkowski and Shankar, 1995; Gong and Barrie, 2003]. This can be done in real time, but is usually done off-line using the output of the large-scale models to drive the aerosol transport processes. If the aerosol model is run off-line, any aerosol effects on radiative transport and cloud microphysics will not be reflected in the atmospheric dynamics. Adding aerosol dynamics to large-scale models make it possible to evaluate the effects which aerosol have on radiative transport in regional, global, and climate change models. However the large grid spacing of 10s to 100s of kilometers in large-scale models does not allow capture of the smaller and local-scale features (determined by such processes as fallout of larger particles, nucleation events, precipitation scavenging, etc.)

[4] Study of marine boundary layer (MBL) aerosols has often been accomplished using a box (0-D) model where the box can be viewed as a well-mixed column extending from the ocean surface to the top of the MBL, with exchange occurring at the surface and with the free troposphere (hereafter, “FT”) at the top of the MBL. The treatment of chemical and microphysical processes within the column can be quite rigorous and very effective in studying the dynamics of the aerosol size and chemical distributions under various generic scenarios. As notable examples we reference here only two of a number of studies: Raes [1995] and Katoshevski et al. [1999], both of which looked at a number of physical processes occurring within the MBL as well as exchange at the surface and with the FT. In a variation of the box model Garrett et al. [2003] stacked three box models, two within the MBL and one above, to study the temporal evolution of the effect of an elevated dust layer on radiative transfer. Exchange between the boxes provide vertical exchange/removal.

[5] One-dimensional (Lagrangian) models of the MBL have been developed by Fitzgerald et al. [1998a, 1998b] and Gelbard et al. [1998] and by Suhre et al. [2000]. The Fitzgerald et al. [1998a, 1998b] model contains detailed aerosol microphysics and requires meteorological inputs, such as vertical profiles of temperature, humidity, turbulent mixing coefficients, clouds and precipitation, to be supplied externally, i.e., via input tables. The model, using meteorological data climatologically representative of that which occurs off the east coast of the United States, was used to study the evolution of the aerosol size distribution in an air mass advected from the United States out over the Atlantic [Fitzgerald et al., 1998b]. Suhre et al. [2000] added limited aerosol microphysics to a 1-D (Lagrangian) boundary layer model [Suhre et al., 1998] to simulate the evolution of aerosols in the ACE-2 Lagrangian experiment. Suhre et al. [2000] discuss the advantages of a 1-D model over a box model.

[6] The Fitzgerald et al. [1998a, 1998b] model is the Naval Research Laboratory's (NRL) high-resolution, one-dimensional (Lagrangian) sectional aerosol model MARBLES (Marine Boundary Layer Aerosol Model), and it aims to predict accurately the vertical structure of aerosol in the MBL. It contains all aerosol microphysical processes thought to be important in the MBL and has the novel feature of accurately treating transport through humidity gradients without introducing numerical dispersion. There are no restrictions on the number of size sections, chemical components (gaseous or particulate), or vertical resolution, but it is typically run with 37 size sections, four chemical components (sea salt, sulfate, dust, and water), and a highly resolved vertical column (cell depths from 2 m near the surface to 50 m at higher altitudes). The required meteorological variables are supplied externally by the user via an input table, with inputs at any spatial or temporal resolution desired.

[7] Since many of the aerosol processes depend critically on meteorological variables (wind, temperature and humidity fields and vertical profile of turbulent mixing coefficient), assimilation of high-resolution meteorological data occurring concurrently with the experimental aerosol data is critical. Here an important extension of MARBLES which interfaces the 1-D aerosol model with the output of a nested grid mesoscale meteorological model is described. The mesoscale model employed is the Navy's 3-D Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS) model. From any point of interest in the MBL, back trajectories are calculated from COAMPS and the required meteorological data for the aerosol model is extracted along that trajectory.

[8] In this paper we (1) describe improvements which have been made in MARBLES since the earlier publication; (2) describe the method of generating back trajectories and extracting the meteorological data from the mesoscale meteorological model for use in MARBLES; (3) show model results generated for two days with significantly different back trajectories and discuss implications with regard to sulfur speciation resulting from the various chemical processes and the relative contribution of sea-salt aerosol and FT aerosol to the CCN population; and (4) discuss the sensitivity of the results to selected parameters such as the FT aerosol, large-scale vertical velocities, and the inadequacies of current mesoscale models as drivers of MARBLES.

2. Aerosol Model Description

[9] The original version of MARBLES as described by Fitzgerald et al. [1998a, 1998b] and Gelbard et al. [1998] included the following microphysical processes: (1) coagulation, (2) oxidation of trace gases, (3) growth of core material by condensation of nonvolatile gas-phase reaction products, (4) particle growth resulting from in-cloud oxidation of SO2 to sulfate, (5) nucleation of new sulfate particles, (6) precipitation scavenging, (7) gravitational settling, and (8) sea-salt aerosol production at the ocean surface; the following transport processes are also included: (9) vertical turbulent mixing and (10) exchange with the free troposphere via an entrainment velocity applied at the top of the MBL (top of original model domain). The model assumes that the aerosol chemical components are internally mixed within a given size section. The numerical implementation is such that mass is conserved during coagulation and both mass and particle number are conserved during growth (in radial space) and during vertical mixing. Numerical diffusion due to changing wet size during transport through humidity gradients is eliminated by expressing size-dependent processes in terms of dry radius with the humidity dependencies included in the (rate) coefficients. For example, the fall velocity and coagulation equations are expressed in terms of dry radius with humidity dependence of the radius included in spatially and temporal dependent rate coefficients, so that fall velocity and coagulation rates proceed at a rate dictated by their local wet size. This is particularly important for vertical mixing in the MBL where there are significant humidity gradients coupled with rapid turbulent mixing. Furthermore it was shown that using the wet radius (nonconserved property during transport) as an independent variable is fundamentally incorrect [Fitzgerald et al., 1998a].

[10] MARBLES is here considered as a vertical column and is advected along a trajectory determined by the meteorological model. The meteorological data required by the aerosol model is extrapolated using the results at meteorological model grid points that lie closest to the points on the back trajectory. Because of the well-mixed character of the MBL, the concept of a 1-D column moving with the mean wind speed in the MBL is a reasonable approximation. The extension of the column into the lower free troposphere to simulate MBL-FT exchange is more questionable, because of possible velocity gradients across the boundary and lack of mixing. In some large-scale experiments, a FT exchange velocity is derived from the measured exchange of ozone or water vapor, in which case we would terminate the column at the top of the MBL and use the measured exchange velocity to simulate exchange between the MBL and FT. For the present case where we have the numerically generated turbulent mixing coefficients, the exchange coefficients themselves define the extent of the MBL and the exchange with the FT is determined by the magnitude of these coefficients, plus any exchange that results from synoptic-scale vertical velocities generated by the mesoscale model.

[11] The following subsections describe improvements made to the aerosol model since 1998 [Fitzgerald et al., 1998a, 1998b]. The first of these changes, the addition of vertical velocity transport, was necessary for successfully coupling the aerosol model with the output of the meteorological model. Also described are changes to the sea-salt aerosol source function and the inclusion of heterogeneous oxidation of sulfur dioxide in sea-salt aerosol to sulfate aerosol mass.

2.1. Addition of Vertical Velocity to the 1-D Aerosol Model

[12] Vertical transport caused by smaller-scale motions, such as turbulent eddies whose size is on the order of the height of the MBL and smaller, is accounted for by the turbulent mixing coefficients, which describe their statistical effect on mixing in the boundary layer. The mesoscale meteorological model calculates these (subgrid) turbulent mixing coefficients by its boundary layer parameterization. Large-scale subsidence/lifting motions (large compared to the grid scale) are resolved by the numerical solutions to the fluid mechanics equations. The characteristic length which separates the largest eddies from the large-scale vertical velocities generated by the model is, of course, not well defined. However, the same division of the two transport mechanisms, which results from the mesoscale model treatment, is also then necessarily used in the aerosol calculations.

[13] Locally, the effects of turbulent mixing in the MBL are much stronger than effects of large-scale vertical velocities, which can often be neglected over short periods of time. Over longer periods of time and for exchange with the free troposphere, the large-scale vertical motions become important. Use of the output of the mesoscale meteorological model gives a set of vertical velocities and mixing coefficients that should be consistent and realistic for use in the aerosol model.

[14] The vertical velocities generated by the mesoscale model are the result of the 3-D mass conservation equation, and are generated by horizontal divergence of the wind field. In the 1-D aerosol model the change in the vertical velocity between vertical cells must be compensated for by horizontal flow through the walls of the column, which is done in a manner which is consistent with the fluid continuity equation and is described in detail by Hoppel et al. [2005a]. This (1-D) limitation arises because the air that enters or leaves the column laterally, at any designated height, must have the same aerosol load both inside and outside the column. For regions with large-scale subsidence, the 1-D limitation does not represent a significant limitation because the column is affected by inflow from the FT, whereas lateral flow is out of the column. For regions with large-scale lifting, assuming 1 km (horizontal) grid spacing and uniform lifting of 1 cm s−1, it would take approximately 5 days for a nonuniformity at 10 km distance to reach the central column moving with the mean wind speed, a time which is comparable to the life time of MBL aerosol. This suggests that if the MBL is nearly uniform over a distance of 5 to 10 km, the horizontally uniform assumption should be quite good (for more detail see Hoppel et al. [2005a]).

[15] Since the vertical velocity for each size particle is the sum of the fluid plus the gravitational fall velocity, the vertical movement in each size section must be treated individually in the aerosol model. The implementation of the numerical scheme in the model is quite tedious because we have kept the same density function representation of aerosol mass across radial sections and across the vertical cells as that in the original model. Aerosol mass and number concentration are still conserved. The details of the procedure together with many tests of the algorithm are given by Hoppel et al. [2005a].

2.2. Sea-Salt Aerosol Source Function

[16] For numerical aerosol modeling, the size-resolved sea-salt aerosol flux emanating from the surface is required. We will refer to this flux as the sea-salt aerosol source function (SSASF), which is a function of size, wind speed, and any other state variable necessary to specify the SSASF. It is clear that aerosol generation will be a complex function of wind speed, wind history, atmospheric stability, sea-state and possibly other variables such as ocean temperature, relative humidity, and surface-active material at the ocean surface. Intuition would suggest that the most important variables might be wind speed and sea-state (especially the amount of white water). Since wind speed is the primary determinant of sea-state, one might expect that, for a single-parameter formulation, wind speed would be the most appropriate parameter. Past formulations of the SSASF use only the current wind speed and in one case the current and average wind speed over a previous time period [Gathman, 1983].

[17] Originally MARBLES allowed for the use of either the sea-salt source function proposed by Monahan et al. [1986] (default source function) or that suggested by Smith et al. [1993]. Later, the hybrid source function [Hoppel et al., 2002] based on Monahan's size distribution in the 0.3 to 5 μm radius range and the modified Smith et al. [1993] source function in the 5 to 30 μm radius range to account for the spume-generated aerosol was added to the user menu. While acknowledging the existence of sea-salt aerosol at radii smaller than 0.3 μm, the sea-salt aerosol source function was cut off at 0.3 μm since there were no measurements of the source function below about 0.5 μm. However, two recent papers [Martensson et al., 2003; Clarke et al., 2006] have proposed source functions which extend the range down to about 0.01 μm, and Martensson et al. [2003] also propose a SSASF which includes a temperature dependence. Here we use the results of these two papers to extend a source function given by Hoppel et al. [2002] down to 0.01 μm as prescribed below. Since particles with an equivalent dry size greater than about 0.03 μm act as CCN it is important to include particles at least as small as about 0.02 μm if the effect of aerosols on cloud properties and the change in the size distribution due to cloud processing are to be included.

[18] The SSASFs proposed by Monahan et al. [1986], Smith et al. [1993], Fairall et al. [1983], and Reid et al. [2001] were recently reviewed by Hoppel et al. [2002] and shown in Figure 1 for wind speeds of 9 ms−1. The source functions for Fairall et al. [1983] and Reid et al. [2001] are given for fixed wind speeds, as indicated in Figure 1, and therefore do not exactly match the wind speeds of 9 and 20 m s−1. In deference to the latter two papers [Martensson et al., 2003; Clarke et al., 2006], the SSASF are plotted in terms of the logarithmic derivative, i.e., dF/d[log(r)] as opposed to dF/dr.

equation image

(Note that log(r), as it often appears in the literature, is really short for log (r/ro) where ro is a reference size taken to be 1 μm.)

Figure 1.

Summary of sea-salt aerosol source functions used in deriving the current MARBLES formulation (red line, see text for details), for a 10 m wind speed of 9 m s−1.

[19] The SSASFs plotted in Figures 1 and 2are for dry size sea-salt particles. Often the SSASFs are given for a RH of 80%, a typical value for the marine environment. To convert from the wet size to dry size we have used the activity given by Tang [1997]. The factor α which relates the dry size to the size at 80% RH (Rdry = α R80) is about 0.5. In Figure 1 the solid black line labeled “corrected Smith,” is the Smith et al. [1993] function as corrected by Hoppel et al. [2002], which uses the deposition velocity for a uniform surface source of particles rather than a source from above. It is also truncated at 2 μm, since the equilibrium assumption used to obtain the Smith et al. [1993] SSASF becomes increasingly inaccurate at radii below ∼5 to 10 μm wet (ambient) radius [see Hoppel et al., 2002]. Monahan et al. [1986] covered a radius range of about 0.3 to 15 μm at 80% RH (or 0.15 to 7.5 μm dry radius). The dashed blue line which extends the range to smaller radius is arbitrary and for comparison purposes only.

Figure 2.

Current MARBLES SSASF, as in Figure 1, for 10 m wind speeds of 5, 10, 20, and 30 m s−1. Solid line is for a source introduced at a height of 10 m, and the dashed line is for a source flux introduced at height of 1 m above the sea surface [i.e., see Hoppel et al., 2005b].

[20] Also shown in Figure 1 are the SSASFs recently proposed by Martensson et al. [2003] and Clarke et al. [2006] for fine particles. Martensson et al. [2003] measured the aerosol flux generated by forcing air through a sintered glass filter into salt water of different salinities and temperatures. The size-resolved aerosol flux together with the fraction of the surface area covered by the bubbles determines the particle flux per unit of area and is used as a surrogate for the flux generated by the open ocean. To extend these measured fluxes to obtain the open ocean SSASF, the particle flux per unit area of bubble coverage was multiplied by the expression given by Monahan and O'Muircheartaigh [1980], for the white cap coverage as a function of wind speed. The SSASF thus obtained is shown in Figure 1 (black curves with open and closed diamonds) for temperatures of 5 and 25°C. The SSASFs for intervening temperatures lie roughly within the envelope defined by the SSASFs of these two temperatures.

[21] The SSASF presented by Clarke et al. [2006] (light blue curve with square symbols) was obtained by observing the flux of sea-salt particles produced by breaking waves on a beach and advected onshore by an onshore breeze. Observations of the amount of white water during the coastal flux measurements were used to determine the flux per unit area of white water produced by the breaking waves. To get the open ocean SSASF as a function of wind speed the flux per unit area of white water was multiplied by the fraction of open ocean that is covered by white caps as given by Monahan and O'Muircheartaigh [1980] (also as given by Monahan et al. [1986]).

[22] The SSASFs shown in Figure 1 were obtained by a number of different methods, all of which require one or more questionable assumptions to arrive at an open ocean SSASF. The Monahan, Clarke, and Martensson SSASFs all assume that the shape of the SSASF does not change with wind speed; that is, the SSASF can be written as the product of a size-dependent function and a wind-dependent amplitude. All three also use the same wind-dependent function. They differ in the manner in which the size-resolved flux is obtained. Monahan obtained the flux from measurements of aerosols and fractional whitewater coverage in a large wave tank. Martensson et al. [2003] obtained the flux from bubbling air through salt water in a small temperature controlled bubble chamber, and Clarke et al. [2006] obtained the flux from measurements of sea-salt aerosol produced by waves breaking at the shoreline and the fractional whitewater in the surf zone. The Smith et al. [1993] method requires the equilibrium assumption; that is, the upward source flux is balanced by the downward gravitational flux. This assumption is probably valid for particles greater than about 5 μm to 10 μm ambient (wet) radius [Hoppel et al., 2002]. While the equilibrium assumption severely limits the size range to large particles, from a fundamental point of view, it should give the best determination of the flux in the 10 to 30 μm ambient radius range. Since the Smith et al. [1993] SSASF is obtained from real atmospheric data it also includes the wind blown spume component of the aerosol flux. The SSASF of Reid et al. [2001] (light gray lines) and Fairall et al. [1983] (green lines) are based on atmospheric measurements and given mainly for comparison and an indication of the limits of uncertainty encountered in flux determinations.

[23] In our prior paper [Hoppel et al., 2002], we concluded that a viable interim solution was to take the larger of the Monahan and Smith formulations throughout the 0.3 to 30 μm range (at 80% RH) for wind speeds above the onset of spume, about 9 m s−1 and the Monahan formulation by itself at wind speeds less than 9 m s−1. Given the large uncertainties resulting from the measurements and the underlying assumptions regarding how representative data from wave tanks, bubble chambers, and shore-breaking waves are of the open ocean we see no reason to change the SSASF which was proposed by Hoppel et al. [2002] for the size range from 0.15 to 15 μm dry size. However the new data of Martensson et al. [2003] and Clarke et al. [2006] at small radii gives us an opportunity to extend the size range to lower radii. At small particle sizes both the Martensson and Clarke SSASFs show a maximum below 0.1 μm with a declining SSASF at 0.01 μm. Considering the assumption and measurement uncertainties involved in the determinations of the SSASF, we find the better than order of magnitude agreement of the Monahan and the newly proposed SSASFs by Martensson et al. [2003] and Clarke et al. [2006] in the 0.1 to 1 μm region quite acceptable. We have chosen to extend the SSASF down to 0.01 μm by applying a factor which modifies the Monahan et al. [1986] SSASF at small radii as shown in Figure 1 by the red line with plus symbols. This is obtained by multiplying the Monahan et al. [1986] SSASF by the following multiplicative factor

equation image

where r is the radius in microns.

[24] Figure 2 shows this SSASF as used in this study throughout the entire range 0.01–30 μm for wind speeds of 5, 10, 20, and 30 m s−1. The black line is for a source at 10 m and the dashed line is for a source introduced at 1 m. At large particle radii the source at 1 m is larger than at 10 m because fewer large particles reach the height of 10 m. For this study, the lowest MARBLES cell is 2 m in height. Sea-salt particles are therefore injected at 1 m, the midpoint of the lowest cell. Also shown in Figure 2 is the typical critical radius below which particles are too small to serve as CCN. In the aerosol model particles larger than 0.033 μm (dry radius) are assumed to act as CCN. This lower limit is based upon the position of the cloud-processing minimum found over the open oceans as discussed in Hoppel et al. [1986, 1996] and is the same value as given by Fitzgerald et al. [1998a]. Uncertainties in the SSASF below the CCN limit will have little effect on the results discussed in this paper.

2.3. Heterogeneous Oxidation of Sulfur Dioxide in Sea-Salt Aerosol

[25] In the current version of MARBLES sulfate from the various sources is carried as different species so that it is easier to compare the relative contributions due to the various mechanisms. In the earlier version sulfate was generated from SO2 by (1) condensation of gas-phase H2SO4 generated by OH oxidation of SO2, (2) liquid-phase oxidation of SO2 during cloud processing, and (3) nucleation of H2SO4 aerosol directly from H2SO4(g). In addition to keeping each contribution as a separate species, the new version also includes heterogeneous oxidation of SO2 by ozone in sea-salt aerosol. This section describes the formation of non-sea-salt (nss) sulfate by oxidation of SO2 in sea-salt aerosol, and how it is calculated in the model.

[26] During the oxidation of S(IV) to S(VI) by ozone in a sea-salt droplet, the gas-phase flux of both ozone and SO2 to the droplet must equal the aqueous-phase oxidation rate within the droplet. In units of moles per liter per second (M s−1) this balance can be written as [Hoppel and Caffrey, 2005]

equation image
equation image

where kmt is the gas-phase mass transport coefficient as defined by Schwartz [1986] and given by

equation image

where a is the radius of the particle, equation image is the mean molecular speed, α is the accommodation coefficient, and Dg is the gas-phase diffusion coefficient. The latter three parameters are specific to either ozone or SO2, depending on whether kmt is used in equation (3) or (4). The mass transport coefficient contains all the effects of gradient and interfacial gas-phase transport. The effective second order rate constant, keff, can be written in terms of the usual O3-S(IV) reaction rate constants as

equation image

where k0, k1, and k2 are the rate constants for ozone oxidation of dissolved SO2, HSO3, and SO32−, respectively, and K1 and K2 are the equilibrium dissociation constants for HSO3 and SO32−. The effective Henry's law constant for S(IV) is

equation image

where HSO2 is Henry's constant for dissolved SO2.

[27] The right side of equations (3) and (4) is the aqueous-phase reaction rate, where the overbar indicates the average reactant concentration over the volume of the entire droplet and the subscript s on the reactant concentration indicates the value at the surface. The expressions for the average concentrations of the reactants in the droplets and the numerical solution to the coupled set of equations (3) and (4) are given in Hoppel and Caffrey [2005] and the method of solution will not be repeated here. The results of the numerical solution to the coupled equations show that for typical atmospheric concentrations of the reactants (1) the dissolved ozone concentration at the surface is (very nearly) in equilibrium with the gas-phase concentration, but greatly depleted as the ozone moves inward toward the center of the droplet because of reaction with much more abundant [S(IV)] and (2) the opposite is true of S(IV). Because of the extreme solubility of S(IV), the fractional depletion of S(IV) caused by its reaction with O3 is negligible, but the low gas-phase concentration of SO2 relative to O3 causes the gas-phase transport of SO2 to be important at high pH. This allows us to write equation (4) as

equation image

where the equations are decoupled as a result of taking the dissolved O3 at the droplet surface to be in equilibrium with gas-phase O3. In keeping with the (numerical) result that the gas-phase transport limitation is that due to SO2, equation (4), instead of equation (3), must be used. Q is the factor that relates the value of [O3] at the surface to the average value and is given by Schwartz and Freiberg [1981] as

equation image


equation image

Daq is the aqueous-phase diffusion coefficient and image is the first-order rate constant for ozone, and in this case is just keff[S(IV)]s. The value of [S(IV)]s could be obtained numerically from equation (8) using an implicit equation solver but here we use the following approximation.

[28] [S(IV)]s is approximated in the expression for Q by image the equilibrium value, and equation (8) can then be solved for [S(IV)]s. The transport-limited rate (right side) of equation (8) is then

equation image

where the first-order rate constants are given by

equation image


equation image

The factor Q in the numerator of equation (11) is the fractional decrease in the unconstrained rate due to the liquid-phase depletion of ozone and the inverse of the denominator is the fractional decrease due to the gas-phase transport limitation of SO2.

[29] Figure 3 compares the approximate transport-limited rate given by equation (11) (red curves) with the exact solution (black curves) by Hoppel and Caffrey [2005] for ambient O3 and SO2 concentrations of 25 ppb and 0.05 ppb, respectively. The agreement is better than we expected given the increasingly divergent value of Q obtained by assuming that (Henry's Law) equilibrium for [S(IV)] rather than using the transport limited [S(IV)]s when calculating Q. The divergence of the two values of Q is quite large as the pH increases above about 7.5. The good agreement of the transport limited rates shown in Figure 3 can be traced to the fact that at high pH the second term of the denominator is much greater than one, and the Q in the denominator cancels that in the numerator. Physically, this occurs because the gas-phase transport limitation becomes dominant at high pH (seen in Figure 3), as the rate approaches the maximum gas-phase transport limits (horizontal dotted lines), obtained by setting the SO2 concentration at the surface to zero. The approximate solution given by equation (11) was compared with the exact solution over a range of O3 and SO2 concentration typical of that which might reasonably be encountered in the atmosphere and the agreement found to be surprisingly good over the entire range.

Figure 3.

Reaction rates for the O3-S(IV) reaction as a function of pH for droplets of 1, 5, 10, and 20 μm radius, respectively, in descending order in the figure. The dashed line is the unconstrained (i.e., no mass transport limitations) rates. Dotted lines indicate the absolute maximum rate limits allowed by gas-phase transport, calculated by assuming an aqueous phase concentration of zero for the appropriate reactant. For image = 0.05 ppbv; image = 25 ppbv.

[30] To calculate the transport-limited rate given by equation (11), a value of pH (expressed as [H+]) must be specified. As shown in Hoppel and Caffrey [2005], the pH of the sea-salt aerosol remains high until the alkalinity is expended, at which time the pH will drop precipitously with any further addition of sulfate. The reaction is essentially extinguished when the nss-sulfate is equal to half the alkalinity (the sulfate ion is doubly charged). Any additional nss-sulfate formation will then occur at a much slower rate and involve other reactions, such as the oxidation of S(IV) by H2O2. The changing pH of the droplet can be calculated from the changing alkalinity using the ionic charge balance equation as done by Hoppel and Caffrey [2005], but that involves solving an implicit equation for [H+] numerically. In the model we use equation (11) for the transport-limited rate evaluated at a constant pH of 8, as long as there is alkalinity in the droplet. Justification for using a constant value is found in the flatness of the rate curve shown in Figure 3 at high pH. The alkalinity and pH of seawater are about 0.0023 M and 8.1, respectively. In a seawater droplet, in water equilibrium with an atmosphere at 80% RH, the alkalinity and pH are roughly, 0.018 M and 8.6. For a high value of pH of 8, Q can be approximated by the first term in equation (9) as 3/q. Also at high pH only the last terms of equations (12) and (13) are important (O3 reaction with sulfite), thus simplifying the calculation in the model.

[31] The alkalinity in size section i (moles m−3) is calculated at a time t as

equation image

where Miss is the mass concentration (kg m−3) of sea-salt in section i and f is the potential alkalinity in moles of alkalinity per kg of dry sea salt (7.0 × 10−2) and {SO42−} is the mass concentration of sulfate in moles m−3. New sea-salt aerosol mass is introduced into the lowest cell in accordance with the SSASF at its potential alkalinity and added to existing sea-salt mass at the current alkalinity. Sea-salt aerosol is transported, mixed, and removed at the current value of alkalinity of the section.

3. Meteorological Data

[32] As MARBLES requires meteorology as an input, it is run as essentially an “off-line” model; that is, it is not run concurrently with the meteorological model. This has disadvantages that (1) there is no feedback between the aerosol processes and the meteorological processes and (2) MARBLES time resolution is limited to the frequency of archived met data. For MBL aerosol dynamics, issue 1 is not a significant concern, and as we were able to run COAMPS in house, all necessary meteorological data could be saved as often as desired. In this study, data was saved in 1 hour intervals.

[33] What follows is a description of the COAMPS reanalysis modeling that generated the gridded meteorological data, and the details and issues of extracting and using such data with MARBLES via air mass back trajectories. This work was prompted by involvement in the Roughness Evaporation Duct (RED) experiment conducted in August–September 2001 off the east coast of Oahu, Hawaii [Anderson et al., 2004], and therefore all modeling described here is focused in the area of the Pacific ocean upwind of Hawaii.

3.1. COAMPS Reanalysis

[34] The U.S. Navy's mesoscale analysis and prediction systems COAMPS (Coupled Ocean-Atmosphere Mesoscale Prediction System) was used with previously archived global analyses produced by the Navy's global analysis and prediction system NOGAPS (Navy Operational Global Atmospheric Prediction System) as the initial and boundary conditions (detailed description of NOGAPS is given by Rosmond [1992], Hogan and Rosmond [1991], and Hogan and Brody [1993]).

[35] The COAMPS reanalysis was conducted using the atmospheric component of COAMPS [Hodur, 1997] with nonhydrostatic dynamics conducted with a 12-hour cycle, at which archived observations available for the NOGAPS analysis within the COAMPS domain were assimilated. The reanalysis approach, similar to the one used by Shi et al. [2004], employs analysis and prediction systems to perform data assimilation of historical observations, thus obtaining the most realistic atmospheric conditions for some periods in the past. A reanalysis typically starts from climatology or an archived analysis of coarse spatial resolution, and proceeds forward one data assimilation cycle at a time. These cycles consist of a short-range 12-hour forecast by the prediction model, which produces the first guess fields for the data assimilation at the next update cycle.

[36] The reanalysis run here was started at 0000 UTC 13 August 2001 with a 12-hour data assimilation cycle until 0000 UTC 20 September 2001, and COAMPS reanalysis fields were archived hourly. Quadruple-nested grids of COAMPS were centered over eastern Pacific Ocean with horizontal resolutions of 81 and 27 km for the two outer grids, and centered near the Hawaiian Islands with horizontal resolution of 9 and 3 km for the two inner grids (Figure 4). There were 30 vertical levels with higher vertical resolution in the planetary boundary layer, with the lower cell midpoints at 10, 30, 55, 90, 140, 215, 330, 500, 750, and 1100 m. For the three outer grids, the Kain-Fritsch cumulus parameterization is employed [Kain and Fritsch, 1993], while explicit moist physics [Rutledge and Hobbs, 1983] is used in the innermost grid and for stratiform precipitation in the outer grids. The planetary boundary layer is parameterized on the basis of the turbulent kinetic energy equation and the mixing length. The Monin-Obukhov similarity is used for the surface layer [Louis et al., 1982]. Detailed documentation of COAMPS can also be found at the COAMPS model website (http://www.nrlmry.navy.mil/coamps-web/web/home).

Figure 4.

COAMPS quadruple nested grid setup.

[37] Observations used in the reanalysis were (1) surface observations consisting of unclassified and declassified surface observations, fixed and drifting buoys, civilian ship reports, declassified Navy ship reports, and Australian sea level pressure; (2) upper air observations, including pilot balloon observations, unclassified and declassified rawinsonde observations, and aircraft reports; and (3) satellite retrievals of temperature soundings from National Oceanic and Atmospheric Administration (NOAA) and Defense Meteorological Satellite Program (DMSP) satellites, and cloud track winds from geostationary satellites. More details about the data sources are given by Westphal et al. [1999] and Shi et al. [2004].

[38] Operationally, COAMPS uses the NOGAPS forecast fields for its lateral boundary conditions on the outer mesh utilizing the Davies [1976] method. The Fleet Numerical Meteorology and Oceanography Center (FNMOC) global data sets of terrain height, surface albedo, climatological deep soil temperature, climatological ground wetness and surface roughness were used by NOGAPS and COAMPS, after interpolation to their respective model grids. The spatial resolution of these parameters was approximately 1° longitude and latitude, except for terrain, which originates from a master data set with a 100-m resolution. The global sea surface temperature (SST) was from the daily archived fields from the Navy's ocean analysis system OTIS [Cummings et al., 1997], for which the primary observations were Multi-Channel Sea Surface Temperatures (MCSST), buoy, ship, and bathymetry data. The MCSST retrievals were produced at the Naval Oceanographic Office from high-resolution satellite infrared imagery. The mesoscale SST was analyzed on each COAMPS grid with the standard COAMPS Ocean Data Assimilation system using the same data sources.

[39] The data assimilation for the COAMPS is similar to that in the NOGAPS. In data sparse regions of the outer grid, selected points from NOGAPS analyzed fields are automatically inserted as synthetic observations. The data sets are also more complete since the data cutoff constraint imposed for operational runs during the same period time no longer existed. Some declassified data (e.g., reports from Navy ships) had also become available and merged into the archived observations for NOGAPS.

3.2. Trajectories and Air Mass Meteorology

[40] The HySplit (HYbrid Single-Particle Lagrangian Integrated Trajectory) program was used directly with COAMPS reanalysis data to derive air mass back trajectories along which the aerosol model input data would be extracted. Details on HySplit are available from Draxler and Hess [1997] at http://www.arl.noaa.gov/ready/hysplit4.html, and from Draxler and Hess [1998]. The COAMPS reanalysis data set was processed into the “ARL packed” format required by HySplit via modification of the available NOAA FORTRAN code to accommodate the COAMPS-3 data files used here (these modifications are now included in the HySplit public distribution), and included all four of the available COAMPS nested grids.

[41] The top of the HySplit model domain was manually set to 34800 m to coincide with the COAMPS model domain, and air mass back trajectories were run starting at the location of the Scripps Institution of Oceanography Research Floating Instrument Platform (FLIP) during the RED experiment (21.649°N, 157.841°W), moored about 10 km off the northeast coast of Oahu [Anderson et al., 2004]. As run in trajectory mode, HySplit only computes advection from the u,v,w wind components and does not consider any thermodynamic or turbulent effects. Back trajectories run with vertical motion calculated directly from the w component of the 3-D wind field show descending motion from levels far above the MBL. However, as the MBL is well mixed, we propose that trajectories calculated along isobaric surfaces best represent boundary layer air mass histories. As wind speeds in the MBL above the lower 100 m surface layer are generally constant with height, we chose a back trajectory starting level of 330 m. Figure 5 shows a sample back trajectory result and illustrates the similarity of six day back trajectories starting at 10, 90, 215, and 330 m.

Figure 5.

(top) Air mass back trajectory results for case 1, 11 September 2001. (bottom) Vertical position of each trajectory, started on four different heights of 10, 90, 215, and 330 m. Vertical motion calculated along isobaric surfaces.

[42] Vertical profiles of pressure, temperature, eddy diffusivity (heat, Kh), relative humidity, vertical cloud location, rain rate and location, 10-m wind speed, and vertical velocity were then extracted from the COAMPS data set via bilinear interpolation on each sigma (height) level at each hourly trajectory location. The resulting series of vertical profiles along each air mass trajectory are supplied to MARBLES via the input file as a function of time and the midpoint of each sigma level. With the exception of the location of clouds, MARBLES internally interpolates in the vertical to remap to its own vertical cell spacing, which is 2 m at the surface, increasing to 50 m intervals at the top of the MBL.

[43] Figure 6 contains a time series of meteorological data retrieved from the 330 m trajectory shown in Figure 5. The boundary layer is clearly defined from both the eddy diffusivity values and the humidity profiles, extending from 500 to 1000 m along this and most other trajectories studied from this reanalysis. Several issues arose in the application of these meteorological profiles to MARBLES, which are described below.

Figure 6.

COAMPS data extracted along air mass trajectory for case 1 (11 September). Each separate line represents a 12 hour average, with time running from 0 to 141 hours from black to red.

3.2.1. Description of Rain and Clouds

[44] COAMPS contains a general bulk microphysics scheme for the prediction of clouds and precipitation as well as cumulus scheme for convective processes not resolved on COAMPS grids greater than 10 km [Chen et al., 2003]. As the parameterization cannot resolve individual clouds, it instead parameterizes the impact of the cloud, predicting precipitation, etc., in the output, without placing a cloud at the corresponding grid point and vertical level. Thus, for the purposes of generating input fields for MARBLES, clouds were placed (1) where predicted by the bulk microphysics scheme and (2) at the COAMPS σ level corresponding to the top of the boundary layer (defined as the lowest level where the eddy diffusivity, Kh, is <1.0 m2s−1) at grid points where convective precipitation is predicted to have accumulated at the surface during the previous hour. Rain rates are input as the sum of both stable and convective rainfall amounts per hour.

3.2.2. Grid Spacing

[45] Only the 2 coarsest COAMPS grids (grids 1 and 2, 81 and 27 km grid spacing, respectively) were used to both calculate the back trajectories and retrieve the meteorological variables for use with MARBLES. Initial use of the 2 smaller grids near the final trajectory location (9 and 3 km grid spacing) showed considerably larger vertical velocity data than seen on the larger grids, often diluting the boundary layer aerosol concentrations by a factor of 2 or more within a 12 hour time span at the end of the trajectory. While not verified, it is suspected that this is a result of explicit moist physics calculations used on the more highly resolved grids 3 and 4 [Chen et al., 2003].

[46] Given that these high-resolution grid data are only available for approximately the last 12 hours of each run, we chose not to use the data generated from these grid spacings and suspect it is statistically unrepresentative of the conditions affecting the Lagrangian column during each run.

4. Aerosol Model Runs and Results

[47] A group of trajectories were computed to use as test cases from the COAMPS data set described in the previous section. From these, two trajectories, shown in Figure 7, have been chosen as test cases to use with the aerosol model, contrasting both location and the wind speed for this study area. One trajectory, starting on 10 September 2001 at 1000 UTC, is seen to curl back from the west coast of the United States before entering the trade wind zone and extends back for 10 days with generally low winds (4–9 m/s), whereas the other back trajectory (beginning on 11 September 2001, 0000 UTC) extends directly up the coast and travels off the COAMPS grid after 144 hours (6 days) and has higher winds (6–11 m/s).

Figure 7.

Back trajectories ending at Oahu, Hawaii, on 10 (case 2) and 11 (case 1) September 2001. Color scale indicates COAMPS predicted 10-m wind speed.

[48] MARBLES results from model runs made along these two trajectories are presented here, focusing first and most extensively on the results from the 11 September run, with additional discussion of some of the unique features of the results from the second trajectory.

4.1. Initial and Boundary Conditions

[49] The initial and boundary conditions are inputs to the aerosol model, and thus are especially important in MARBLES modeling.

4.1.1. Self-Initialization and Aerosol Components

[50] MARBLES requires initialization of the aerosol mass in each section (size) at each height (cell). Since these initial values are unknown, values believed to be reasonable can be assigned at t = 0. We have seen in the modeling presented here that within a couple of days, the model results are insensitive to the initial values and are determined by the model processes and the boundary conditions. This process of self-initialization was examined by Fitzgerald et al. [1998b] in modeling the evolution of the aerosol size distribution off the east coast of the United States, and there model results were found to be independent of the initial conditions after 5 to 10 days. The five day transition period required to erase the initial conditions is longer than we observed here because (1) Fitzgerald et al. [1998b] had a changing free tropospheric (FT) boundary to account for the decay of FT aerosol and SO2 concentrations from typical continental to remote Atlantic values and (2) the MBL over the Pacific is lower than that found off the east coast of the United States.

[51] Thus we have chosen to initialize the boundary layer with no aerosol mass at any size (i.e., a MBL devoid of aerosol) and allow the model to self-initialize. As described below, boundary conditions specified at the top (Free Troposphere) and bottom (sea surface) of the model domain include size-resolved aerosol concentrations and fluxes, respectively. As done in the previous version of the model, two chemical aerosol component besides water are included: sulfate and sea-salt, with the sulfate mass speciated between its sources as described earlier. An insoluble component (dust) is also carried in the calculations, but no source of dust was included and thus dust concentrations remained void at all times in all model runs. Also, it is recognized that organic matter contributes to the marine-derived aerosol flux [Blanchard, 1964; Hoffman and Duce, 1974], and that recent studies have measured a significant fraction of the submicron aerosol mass to be of organic composition and postulated to be from primary marine aerosol production processes [Cavalli et al., 2004; O'Dowd et al., 2004; Leck and Bigg, 2005; Tervahattu et al., 2002]. However, we have not included any organic fraction in the aerosol modeling presented here, focusing instead on the model improvements described earlier. The presence of organic material in marine aerosols has potentially significant implications in modeling everything from MBL aerosol and cloud dynamics to climate change, and is something we hope to include in future studies.

[52] Gas-phase concentrations were initialized with boundary layer concentrations of SO2(g) of 50 pptv, having been chosen as a midpoint of values listed by Warneck [2000] for the remote marine troposphere (9–90 ppt, 22–71 remote Pacific Ocean), while DMS(g) has been initialized at 300 pptv [Yvon et al., 1996; Fitzgerald et al., 1998b], and H2SO4(g) at 0.14 pptv [Weber et al., 1997].

4.1.2. Free Troposphere Boundary Conditions Aerosol and Sulfur Gas Concentrations

[53] Much more important to the final result than initialization are the FT values imposed as upper boundary conditions on the model and the surface fluxes assumed in the model.

[54] Field measurements made from aircraft during PEM-A and PEM-B in the equatorial Pacific south of Hawaii by Clarke and coworkers [Clarke and Kapustin, 2002; Clarke et al., 1999; Moore et al., 2003] have shown a gradient in the aerosol size distribution from a nucleation mode near a radius of 10 nm in the upper troposphere (<4 km) that increases to 25–30 nm at lower elevations just above the MBL (∼2 km) [i.e., see Moore et al., 2003, Figure 5; Clarke et al., 1999, Figure 4; Clarke and Kapustin, 2002, Figure 11]. Mode number concentrations above the inversion are 750 cm−3 [Clarke et al., 1999, Flight 10, Figure 4], 600–800 cm−3 [Clarke and Kapustin, 2002, Figure 11], and 600–700 cm−3 [Moore et al., 2003, Figure 5a].

[55] Measurements made at the Mauna Loa Observatory, Hawaii, located 3.4 km amsl and well within the free troposphere, show distributions centered near a radius of 20 nm [Clarke et al., 1996] and 10–15 nm [Weber and McMurry, 1996, Figure 1] during downslope conditions indicative of air mass transport from above. Clarke's measurements were made in July 1995 and lack the smaller nucleation mode, whereas Weber reports a frequency size distribution compiled from 432 DMA scans over a 72 hour period, also measured in July 1992, that, when averaged produces the curve shown in Figure 8.

Figure 8.

Free tropospheric size distributions used as input to the aerosol model (red line), as derived from measurements by Weber and McMurry [1996] (black line and symbols) and Clarke et al. [1996] (green).

[56] Therefore, owing to the consistency of 10–30 nm mode in both types of data sets, we select a distribution representative of the FT just above the MBL as the lognormal fit to the average of the Mauna Loa data of Weber and McMurry [1996] and Clarke et al. [1996], as seen in Figure 8, with a total concentration of 460 cm−3, geometric mean diameter, rg = 23 nm, and the geometric standard deviation, σg = 2.00.

[57] The FT sulfur gases SO2, DMS and H2SO4 concentrations are set to 46 ppt, 36 ppt [Blomquist et al., 1996], and 0.04 ppt [Weber et al., 1995], respectively, at the top of the domain. Free Troposphere Mixing

[58] In meteorological models, vertical velocities are calculated from the horizontal divergences of the wind field. Since even the highest-resolution mesoscale models have grid spacings greater than 1 km, vertical velocities caused by horizontal divergences on a scale less than about 10 km cannot be captured. If smaller-scale phenomena are important to the dynamics they must be parameterized; hence the parameterization of the effect of MBL turbulence on dynamics and effect of clouds and precipitation on energy balance. While the effects of vertical motions on a scale intermediate to MBL turbulence (less than the height of the MBL) and those resolved by the mesoscale model, may have little effect on the meteorological variables, they are known to be of importance in describing dispersion of a scalar contaminant in the FT.

[59] Chatfield and Crutzen [1984] documented the problems eddy diffusion models have in describing transport in the entire troposphere, beyond the boundary layer, arguing that inclusion of localized, rapid vertical mixing by convective systems is necessary to accurately model atmospheric compounds whose lifetimes vary rapidly in the vertical, such as sulfur compounds. Recently, Lenschow and Gurarie [2002] have shown that equivalent eddy diffusivities between 1 and 10 m2s−1 for the upper troposphere are effective in predicting mean vertical structure and fluctuations in trace gas concentrations in the troposphere, and state that “the concept of an eddy diffusivity in the free troposphere is not very robust because much of the vertical transport occurs by processes other than small-scale turbulent eddies…” [Lenschow and Gurarie, 2002, pp. 29-4-5]. Radioactive tracer studies have derived free tropospheric values of eddy diffusivity between 4 and 20 m2s−1 [Hunten, 1975]. Lee and Larsen [1997], taking Hanna's [Hanna, 1995] advice, arbitrarily set Kh above the boundary layer equal to 10% of the maximum BL value, and found better agreement with measured 222Rn values, as the value of 0.1 m2s−1 was “too small for Rn to diffuse continually to the layer aloft” from the boundary layer. These 10% values ranged from 0.2 to 20 m2s−1 [see Lee and Larsen, 1997, Figure 7].

[60] In running MARBLES using the turbulent mixing coefficients and vertical velocities generated by COAMPS (to be discussed shortly), it was found that the lack of any mixing in the FT resulted in significant depletion of aerosol concentrations which extended for several hundred meters into the FT and which increased in depth with time. To remedy this, we have (also following the suggestion by Hanna [1995]) arbitrarily increased the mixing coefficient in the FT from zero to a value of 5 m2 s−1, on the order of one tenth that of the maximum in the MBL. To capture the very stable region at the top of the MBL where there is an inversion in the potential temperature, we let K(z) go linearly to a value of 0.2 m2 s−1 (equivalent to an exchange velocity, ve, of 0.4 cm s−1).

4.1.3. Sea Surface Boundary Conditions

[61] The formulation for the flux of sea-salt particles from the ocean surface has previously been described in section 2.2. DMS Flux

[62] The flux of dimethyl sulfide from the ocean surface is highly variable, dependent on both the surface water concentration and boundary layer meteorology. Measurements typically report flux values ranging from 1 to 40 μmol m−2d−1 and generally increase with wind speed, although variability has been shown to be not fully explained by wind speed [Yvon et al., 1996; Zemmelink et al., 2004; Hintsa et al., 2004; Aranami et al., 2002; Huebert et al., 2004]. We have chosen a constant flux of 8 μmol m−2d−1, an intermediate value in this range and based on measurements made in the equatorial Pacific based on the work of Yvon et al. [1996] (see further discussion by Fitzgerald et al. [1998b]). The oxidation of DMS by OH and NO3 and the production of SO2 remains as given by Fitzgerald et al. [1998a]. Deposition Velocity

[63] The deposition velocities are those given by Hoppel et al. [2005b] and are different depending on whether or not the particles have a source at the surface or source “above.” As suggested by Hoppel et al. [2005b, section 7], for a given size section, the particle gradient between the lowest two cells is calculated at each time step. If the net flux is upward, the particles in that section are assumed to originate at the surface; whereas, if the net flux is downward the particles are treated in the more conventional manner as diffusing downward from above. Since the concentration in the lowest cell is used as the reference concentration in calculating the deposition velocity, and since the particle source is at the midpoint of the lowest cell, the deposition velocity for a surface source is just the gravitational settling velocity [see Hoppel et al., 2005b, equation (7)]. For a source above, equation (35) of Hoppel et al. [2005b] is used for the deposition velocity where the reference concentration and height is taken at the midpoint of the lowest cell (1 m).

4.2. Case 1 Results (11 September Simulation)

4.2.1. Evolution of MBL Aerosol

[64] Figure 9b shows wind speed along the trajectory (by the white line associated with the right-hand scale) and contours of the turbulent mixing strength (black lines) and vertical velocities (given by color contours) as a function of altitude. Also shown by symbols are periods where there is cloud cover. The heights of the symbols are located at the middle of the COAMPS cells that indicates the presence of clouds. MARBLES then places clouds in all MARBLES cells which populate that COAMPS cell; for example, the symbols at 750 m indicates clouds are present from 600 m to 850 m. Because of the low vertical resolution of COAMPS, the cloud can often extend 50 to 100 m into the FT. The top of the MBL is defined as the height to which the eddy diffusion contours extend. The weakest eddy diffusion contour (2 m2s−1) is typically at 750 m with periodic excursions up to approximately 1000 m. While the data assimilation provides a linear interpolation of the COAMPS generated turbulent mixing coefficients onto the more dense MARBLES cells, the rather discrete jumps in the height of the MBL at 500 m, 750 m, and 1100 m are related to the low vertical resolution of COAMPS. The high MBL prior to −130 hours occurs at the very beginning of the trajectory shown in Figure 7 and prior to the air mass encountering easterly trade wind flow. Subsidence (downward vertical motion indicated by orange) dominates the large-scale vertical motions, ranging up to 3 cm s−1 in both directions, but most frequently in the 0–1 cm s−1 range (lightest shades of orange/blue). Upward velocities are indicated by blue.

Figure 9.

Interstitial particle (rdry < 0.03 μm) number concentrations (N) for case 1 (11 September) as a function of time and height, illustrating the effects of mixing and vertical transport in the MARBLES domain. (a) N for particles originating in the free troposphere. (b) Meteorology used in the MARBLES run as derived from COAMPS along the air mass back trajectory, showing large-scale vertical velocities (downward motion in orange, upward motion in blue, both in increments of 1 cm s−1), and eddy diffusion coefficients Kh (black lines) as a function of height, and surface wind speed (white line, right side ordinate) in m s−1. Clouds are shown as white diamonds and are placed on the midpoint of the COAMPS vertical cells (three of which are shown as white-dashed lines). (c) N for particles originating as sea-salt particles and (d) total N (sum of Figures 9a and 9c).

[65] Figures 9a and 9c show the cumulative number of particles smaller than 0.03 μm radius, for sulfate and sea-salt particles, respectively. We have chosen particles smaller than the cloud-processing minimum to illustrate the effects of vertical transport on the aerosol. Number concentrations are calculated assuming that particles in a given size section are either free troposphere sulfate or sea salt. While this is not strictly true, it is a good approximation for particles smaller than 0.03 μm radius because internal mixing, resulting from both coagulation and condensational growth (by sulfate from DMS), is small during the lifetime of the particle. (A typical size distribution is shown in Figure 11a and will be discussed later.)

[66] The initial aerosol concentrations are seen at the very beginning of Figures 9a and 9c, where a uniform FT concentration extends down to about 600 m, below which there are no particles. There is then a period of “self initialization” over a period of a day or so. In this particular case, the initial mixing dilutes the FT aerosol to a height of 1200 m because of the downward mixing of FT aerosol into the empty MBL. After the height of the MBL drops from 1200 m to 700 m at about −130 hours (as indicated by contours of mixing coefficients), the height of the unperturbed FT aerosol concentration descends from about 1200 m to 800 m over the next 12 hours by subsidence and residual FT mixing as described in section 4.1.2. Once the FT aerosol reaches the MBL it is mixed uniformly throughout on a timescale less than a few hours, as evidenced by the nearly vertical uniform color in the MBL at any point in time. There is a loss of the small particles in the MBL primarily due to their growth to larger sizes by condensation of sulfate (from DMS). To a much lesser extent these particles are also lost by coagulation with larger particles and deposition at the ocean surface. The time constant for the loss of particles smaller than CCN by the above processes is estimated by Hoppel and Frick [1990] to be the order of 10 to 30 hours. The time constant for “filling” of the MBL with FT aerosol is found by dividing the height of the MBL by the subsidence velocity. If the MBL height is 700 m and the subsidence velocity is 1 cm s−1, the filling time is 20 hours, and comparable to the aerosol lifetime, indicating that there would be significant loss of particles in this size range during the filling process, giving rise to lower concentrations in the MBL. All of the above processes are calculated explicitly in the model, giving much better simulation of actual conditions than any such estimate.

[67] Events during which there is significant eddy mixing extending above the typical height of 700 m can greatly enhance the exchange with the FT, which for small particles is evidenced by a decrease in concentration in the lower FT and increase in the MBL. Such events occur at −108 hours and −70 hours (among others), as seen in Figures 9a and 9c with concurrent mixing shown in Figure 9b. The large influx of FT aerosol into the MBL after −36 hours is the result of both the upward mixing events at about −36 hours and strong subsidence.

[68] The evolution of the sea-salt aerosol particles with radii less than 0.03 μm is shown in Figure 9c. The source strength of sea-salt particles is a strong function of wind speed (see Figure 2), with the maximum sea-salt concentration limited only by removal processes. Since the lifetimes of these small particles are of the order of tens of hours, we see an initially slow build up of the sea-salt concentration during the first 20 hours. The time period from −120 hours to −96 hours sees an increase in wind speed to about 10 m s−1 which caused a further increase in sea-salt particles indicated by the green color after time −100 hours. Upward mixing events, such as that which occurred at −108 hours, and to a lesser extent, upward large-scale motions, disperse a small amount of sea salt into the lower FT as seen in Figure 9c by the purple color. The drop in wind speed around t = −36 hours coupled with strong subsidence causes the decrease in MBL sea-salt aerosol between about t = −24 to t = −5 hours. Subsidence dilutes the MBL with FT air devoid of sea-salt particles. Figure 9d gives the total concentration with r < 0.03 μm, i.e., the sum of Figures 9a and 9c.

[69] Figure 10a shows the concentration of particles with rdry > 0.5 μm. These particles are essentially all sea-salt particles originating at the surface with the overwhelming majority too small to exhibit a gravitationally induced gradient. The MBL concentrations are sensitive to wind speed (generation), subsidence (dilution), detrainment (upward mixing), and height of the MBL (confinement). For example, the sharp decrease in concentration after −36 hours results from a drop in wind speed and subsequent dilution by subsidence shown in Figure 10c.

Figure 10.

Supermicron particle number concentration results (rdry > 0.50 μm), (a) with and (b) without large-scale vertical motion, (c) COAMPS met data as in Figure 9, and (d) gravitationally induced near-surface gradients for large particles ((rdry > 5.0 μm).

[70] Figure 10d shows the concentration for particles larger than 5 μm dry radius. The large particles are very much dependent on wind speed (Figure 10c) and increase dramatically at the onset of spume generated particles above about 9 m s−1 as discussed in section 2.2. The vertical scale on Figure 10d extends to only 50 m so that the strong gravitationally induced vertical gradient can be captured in Figure 10d. This gravitationally induced gradient is enhanced by the vertical gradient of relative humidity. The corresponding wet particle radius increases dramatically with altitude causing more rapidly settling of a given dry-size particle with increasing height as will be discussed in section 4.2.6.

4.2.2. Sensitivity to Vertical Velocity

[71] Considerable effort was made to include the effects of large-scale vertical velocity fields in the new version of MARBLES (as discussed in section 2.1) and vertical velocity can be set to zero as an option in the input file. In the MBL the effect of mesoscale vertical velocities are much weaker than effects of turbulent mixing and contribute little to short-term aerosol dynamics. Over the long term, the main effect of vertical velocity is to mix FT air into the MBL. As stated earlier, upward gradients in velocity have little effect since the air entering the column at any point has the same aerosol concentration as that in the column (horizontally homogeneous assumption). Any gradients in concentrations are advected up or down by the vertical velocity. This vertical advection of aerosol gradients is important in the FT where there is little turbulent mixing, but is not important in the MBL where strong turbulence works to destroy vertical gradients. For subsidence, the effect of dilution of MBL air with FT air is significant over periods of time the order of a day. This is shown by the comparison of the total particle concentration of particles with r > 0.5 μm in Figures 10a and 10b, where the vertical velocity has been set to zero in Figure 10b. The concentrations in Figure 10a are sometimes 50% lower than that of Figure 10b because of the dilution of MBL aerosol by FT air which is essential free of sea-salt aerosol. If in the small particle case discussed in conjunction with Figure 9, we had shown concentrations from the run with no vertical velocity, very little difference would have been seen in the total concentration (≤±10%) because the dilution of small sea salt particles by downward vertical motion is largely made up for by the entrainment of FT particles. The chemical composition would change accordingly between sea salt and sulfate, and it is shown in section 4.2.5 that the fraction of sea-salt particles which are CCN when there is no vertical velocity is increased by more than 50%.

4.2.3. Size Distributions

[72] Dry size distributions at the end of the trajectory are shown in Figure 11. Figure 11a shows the number distribution at four altitudes. The solid line and the short-dashed black line show the size distribution within the MBL at 12 m and 500 m. Because of the strong mixing in the MBL these size distributions are almost identical except at the largest radii where there is a small difference (on exponential scale) due to gravitationally induced vertical gradients. It should be recognized that the actual (wet) size distributions are different because of the uptake of water at different relative humidities at the two altitudes. The blue line gives the size distribution in the FT used as the upper boundary value. The solid line shows the size distribution at 1150 m and the long-dashed line the size distribution at 1450 m. Some detrainment of particles from the MBL into the lower FT is indicated by the hump in the lower FT size distributions at larger sizes indicating the presence of some sea-salt particles and the hint of the cloud processing minimum at 0.04 μm radius. For the smallest particles the concentration is lower in the MBL than the FT because of the loss of interstitial aerosol to cloud droplets during cloud processing and condensation of sulfate (from oxidation of DMS and SO2) causing growth to larger diameters. No nucleation is observed in the domain of the model.

Figure 11.

Aerosol number and mass size distributions at the final output time (t = 0 on Figures 10 and 12) for case 1 (11 September): (a) total number at four different altitudes and that used to represent the free troposphere (blue), (b) total number (black) and number distributions of particles originating in the free troposphere (red) and as sea-salt aerosol (green), (c) same as Figure 11b but as mass concentrations, and (d) mass concentrations for total (black dashed) and individual sulfate species and for sea-salt aerosol (green).

[73] Figure 11b shows the size distribution in the MBL (at 475 m) and the relative contributions of FT aerosol (red) and sea-salt aerosol (green). Since the model only keeps track of the size resolved mass concentrations, the number concentrations of the two components is calculated under the reasonable assumption that the sulfate formed during cloud processing is distributed equally on all activated particles (sea salt and FT; condensed sulfate mass is negligible as seen in Figure 11d). For the conditions encountered during this run particles that originate in the FT are dominant at radii less than about 0.3 μm and sea-salt particles are dominant at larger radii. The total number concentrations of the FT and sea-salt aerosol as shown in Figures 11a and 11b are 405 and 157 particles cm−3, respectively. As will be discussed later the fraction of sea-salt particles which are CCN (i.e., large enough to be activated in clouds) is about 20% at the final time.

[74] Figure 11c shows the total mass distribution as the dashed line, as well as the mass distribution of sulfate (red) and sea salt (green). While the number of sulfate particles is much greater than the number of sea-salt particles, the bulk of the aerosol mass is sea salt; total mass concentrations are 5.7 × 10−10 and 1.8 × 10−8 kg m−3, respectively, for sulfate and sea-salt aerosol.

4.2.4. Sulfur Speciation

[75] This version of MARBLES keeps track of the various sources of sulfate as individual species. The mass distribution from each source, together with the total sulfur (black dashed line) is shown in Figure 11d. Also shown is the sea-salt mass in green. Over the entire size range, the mass of sulfate produced by gas-phase condensation of sulfate from DMS (light blue line) is negligible compared to the total. The sulfate produced by cloud processing (dark blue line) is a major contributor to the sulfate found in the cloud-processing maximum. However, we were somewhat surprised to find that the FT sulfate contributed as much to the total sulfate at the cloud processing maximum as did the new sulfate produced by cloud processing, at least under these remote conditions. The sulfate added by cloud processing is responsible for moving FT sulfate mass from the region of the cloud-processing minimum so that it accumulates in the region of the cloud processing maximum.

[76] As can be seen from Figures 11a and 11b, the concentration of particles in the FT is much larger than in the MBL right at the size of the cloud-processing minimum. This particle gradient across the boundary causes enhanced entrainment of particles into the MBL by mixing and by subsidence at the radius associated with the cloud-processing minimum. Once in the MBL, cloud processing then increases the radius of those entrained particles so that they grow to the size of the cloud processing maximum. This accumulation of FT mass at the cloud-processing maximum is evident in Figure 11d. At the cloud-processing maximum the gradient flux is upward so that any gradient flux is from the MBL into the FT. This process is discussed by Fitzgerald et al. [1998b, Figure 6] where the transport across the FT boundary is the result of mixing via an entrainment velocity. Experimental evidence of this inverted entrainment at the cloud processing maximum is given by Clarke et al. [1996]. This new version of the model includes both mixing and large-scale vertical motions, which can behave differently. For example, while both mixing and subsidence will entrain particles from the FT into the MBL at the cloud processing minimum, at the maximum, mixing will move particles from the MBL into the FT, but subsidence will move particles from the FT to the MBL, decreasing the concentration at the cloud processing maximum without affecting the FT.

[77] The sulfate mass formed heterogeneously by the process described in section 2.3 is most interesting. In an environment where the ozone and SO2 concentrations are 25 ppb and 0.05 ppb respectively, Figure 3 indicates the rate of S(VI) formation in a 10 μm radius droplet is about 5 × 10−7 M s−1 at a pH greater than 7.5. If the RH is 80% the alkalinity in the sea-salt droplet will be about 0.02 M (about 8 times greater than sea water) and the corresponding dry size will be about 5 μm in radius. Under these conditions the alkalinity will be used up in about 6 hours (a factor of one half enters because the sulfate ion is doubly charged). Under the same conditions a 1 μm radius particle will lose its alkalinity in about 5 min. In Figure 11d the heterogeneously formed sulfate (magenta curve) mirrors that of the sea-salt mass with the fraction of heterogeneously formed sulfate mass being 0.0034 that of sea salt for particles smaller than about 5 μm. This is the exact amount of sulfate required to neutralize the alkalinity of sea salt. For larger particles the fraction decreases and at the largest size the ratio is less than 0.0001. For the larger particles, not only is the transport-limited rate much less than for smaller particles (Figure 3), but their lifetime is shorter because of gravitational settling, such that the alkalinity in the largest particles is only partially neutralized before being lost to fallout. Since the flux of sea-salt aerosol increases dramatically with wind speed and since the reaction is so fast, we would anticipate that the scavenging of SO2 by sea salt would be increasingly effective at high wind speeds, decreasing the concentration of SO2. Less SO2 would then be available for conversion during cloud processing. This is discussed further in section 5.3.

[78] Figure 12 shows the sulfur speciation as a function of time. The sulfur mixed into the MBL is quite constant, which reflects the “self-regulating” influence of entrainment across the MBL-FT boundary by both mixing and subsidence. Both mechanisms tend to drive the FT sulfate in the MBL toward the concentration in the FT. Most of the variability in the total MBL sulfate is due to cloud processing, which is a major source of MBL sulfate. While sulfate formed by heterogeneous reactions in sea-salt aerosol is only about 10% or less in this run (with moderate wind speeds), it is shown in section 5.3 that at wind speeds above 10 m s−1 (owing to the increase in sea-salt aerosol source flux (see Figures 2 and 3)), the heterogeneous reaction becomes the dominant sink of MBL SO2.

Figure 12.

Time series of sulfate aerosol mass concentrations for individual sulfate species at the surface (12 m) for case 1 (11 September).

4.2.5. Sea-Salt Versus Free Troposphere CCN

[79] As mentioned above in connection with Figure 11b, the number of CCN attributed to entrainment of FT aerosol and sea salt aerosol can be calculated under the assumption that the sulfate produced by cloud processing is distributed equally on all particles in the section. All particles greater than the critical size for activation are CCN. While cloud processing can move particles from the cloud-processing minimum to larger sizes, growth processes, excluding coagulation, do not remove particles from this size range. Figure 13 shows the total number of CCN as a function of time by the black line, which is about 200 cm−3 after an initialization period of a couple of days. The green line shows the ratio of sea-salt CCN to the total number of CCN and ranges from 20 to 30%, depending primarily on wind speed. The increase in the sea-salt fraction from time −120 to −96 hours is largely due to the higher winds during this period. Since CCN have lifetimes of the order of a few days the effect of wind speed on the sea-salt particles is accumulative and lags the instantaneous wind. As seen earlier only the largest particles respond rapidly to wind speed (Figure 10d). The decrease in the sea salt fraction after about −40 hours is the result of dilution, by upward mixing of sea-salt particles into the FT occurring around −36 hours, and strong subsidence, as well as a drop in wind speed. The up tick in the fraction at the very end of the run is a result of the increase in wind speed and lowering of the mixed layer. When entrainment increases the flux of FT aerosol into the MBL, it dilutes the concentration of sea-salt particles, making the total concentration of CCN less variable than either component (during this run there was no removal of CCN by precipitation scavenging).

Figure 13.

CCN (rdry > 0.033 μm) and total number concentrations at 575 m, and the fraction of CCN originating as sea-salt aerosol, with (solid lines) and without (dashed lines) vertical velocity for case 1 (11 September).

[80] CCN concentration and speciation sensitivity to large-scale vertical motion is shown in Figure 13 by the dashed lines; model runs conducted without transport by vertical velocity motions differed little in CCN and total number concentrations. However, as discussed in section 4.2.2, the dilution of sea-salt particles is largely offset by the entrainment of FT particles, and thus the fraction of sea-salt CCN increases by 50–70%, as shown in the dashed green line on Figure 13.

4.2.6. Vertical Profiles of Aerosol Concentrations

[81] Since MARBLES includes vertical mixing, size-resolved gravitational settling, and high vertical resolution, it has the capability to resolve gravitationally induced gradients which are important for larger particles. In Figure 14 the lines without symbols show the vertical profiles of sea-salt particles in several size sections from about 2 μm to 23 μm in dry radius. Concentrations are normalized to the concentration in the lowest cell (2 m in height) where the sea-salt particles are introduced. Figure 14 is for time −16 hours during a period when there is subsidence.

Figure 14.

Vertical concentration profiles of individual aerosol size bins (dry sizes), with geometric average radii indicated on each curve, at the final model output time (t = 0) for case 1 (11 September). Black dashed line is the relative humidity profile.

[82] As shown in Figure 14b, many of the larger particles are removed by fall out before reaching about 30 m. As the particles rise in the surface layer they experience stronger mixing which decreases the gradient in the central region of the MBL where mixing is greatest. As the large particles near the top of the MBL the mixing decreases; this, coupled with increasing fall velocity due to water uptake at the higher RH, causes a rapid decrease in the concentration of large particles at the top of the MBL. The increasing vertical gradient with increasing size in the central region of the MBL is, of course, related to the increased gravitational force experienced by larger particles; a larger gradient is required to suspend larger particles. The large drop in particles in the surface layer (first few cells) is caused by the decrease in K(z) to zero at the surface, and, in this particular case, is enhanced by subsidence. It should be clear that the actual (wet) size of the particles will be considerably larger than the dry size. At any height, the settling velocity is determined by the RH at that height; for example, a 5 μm dry size particle will be about 10 μm at 80% RH, and 15 μm at 95% RH.

[83] The monotonically decreasing normalized concentrations for particles with increasing radius in the 1–5 μm dry radius range shown in Figure 14, while expected, turns out to be quite rare. The interplay between increasing/decreasing SSASF as a function of wind speed, the fall velocity as a function of height (RH), and the strength of mixing as a function of height can make for some very strange results with respect to concentration as a function of size. For example, on one occasion during decreasing wind speeds the normalized concentration of 4 μm particles was greater than that of 2 and 3 μm particles, largely because of the fact that the net flux of 4 μm particles was downward or zero (suspended), increasing the concentration in most of the MBL, whereas the 2 to 3 μm radius particles were still being dispersed by upward mixing. The implication of the complicated behavior just described raises questions about the interpretation of aerosol measurements in the MBL and their relation to the SSASF. The large particles in the FT seen in Figure 14 are primarily the result of upward mixing events extending above the 800 m level at times prior to −16 hours as seen in Figure 10c.

[84] Also shown in Figure 14 are profiles of particles with radii of 0.011 μm and 0.009 μm, where the symbols are at the midpoints of each MARBLES cell. At 0.011 μm (blue line), particles are being supplied to the MBL both by the downward mixing of sulfate particles from the FT and upward mixing of sea-salt particles from the surface. Since turbulent mixing in the model is done by chemical component, the profile of 0.011 μm particles can be viewed as the sum of the FT sulfate particle and sea-salt particle profiles. In the model the SSASF extends only down to 0.011 μm, so that the 0.009 μm particles originate only in the FT and the profile is normalized to the concentration at 1550 m. The positive gradient exhibited by the concentration of 0.009 μm particles (green line) indicates that these particles are moving down from the FT to the surface, where some are being removed by surface deposition.

[85] Figure 15 shows the vertical concentration profiles for a number of size sections between 1 μm and 18.7 μm where the size is the size at ambient RH. Concentrations are normalized to the concentration in the lowest cell. At each output time MARBLES converts the dry size distribution to a wet size distribution using the local RH and remaps the wet size distribution onto the same size sections as used for the dry size distribution. In regions where the slope of the size distribution is negative, the swelling of the particles with increasing RH cause an increase in particle concentration at a fixed size. If the negative slope of the size distribution is sufficiently large, then for a given increase in RH with height, the particle concentration at ambient RH can increase with height. For sizes larger than the cloud-processing maximum the slope of the distribution is nearly always negative and at large radii is very steep. The increase in the normalized vertical profile of a given wet size seen in Figure 15 is the result of the interplay of vertical gradient of the dry size profile (Figure 14) and the slope of the SD. For very large particles (dashed lines in Figure 15) the slope of the size distribution remains nearly constant so that the variations seen as the radius increases from 11.6 to 18.7 μm is due largely to the decreasing gravitationally induced vertical gradient, as discussed in connection with Figure 14. The increase with height for smaller particles shown in Figure 15 decreases at smaller sizes because of the decrease in slope of the size distribution near half a micron seen in Figure 11a. Very near the surface (Figure 15b) the vertical gradient in dry size is so great that the increase due to swelling cannot overcome the decrease due to the large vertical concentration gradient near the surface (caused by the rapidly decreasing K(z)). Also the model resolution plays a role near the surface. The RH in the lowest COAMPS cell (0–20 m) is also assigned at z = 0, so that there is no RH gradient in the lowest three MARBLES cells whose centers are at 1, 3, and 12 m (see RH curve in Figure 14b. While one might expect adiabatic mixing with decreasing RH right down to the interfacial layer, it is also true that right at the surface the saturation vapor pressure of sea water would dictate a higher RH.

Figure 15.

Vertical concentration profiles of individual aerosol size bins, as in Figure 14, for ambient (wet) sizes, for case 1 (11 September). Black dashed line on right side of plot is relative humidity profile.

[86] The behavior of the ambient concentration profile has important ramifications for interpreting aerosol measurements made at ambient RH. Many optical particle counters and impaction devices measure concentrations at ambient sizes and interpretation of profiles thus obtained is complicated as seen above. Also any comparison of modeled and measured results would require measurements to be averaged over periods of time long compared to the turbulent mixing timescale (approximately 20 min) to correspond to modeled results.

4.2.7. Sulfur Gas Concentrations

[87] Dimethyl sulfide (DMS) and sulfur dioxide (SO2) gas phase concentrations are initialized in the boundary layer at 300 pptv and 50 pptv, respectively. DMS undergoes oxidation by both the hydroxyl (OH) and nitrate radicals (NO3). As detailed by Fitzgerald et al. [1998a], the aerosol model does not predict the concentrations of either oxidant, but instead allows OH to vary diurnally between [OH]max = 2.5 × 106 molecules cm−3 at solar midday and a minimum of 10−3[OH]max as prescribed by a fit to the clean marine measurements of Toon et al. [1987]. Nitrate radical concentrations, also based on Toon et al. [1987], are set at 1.0 × 103 molecules cm−3 between 0700 and 1900 and minimum values of 10−3[NO3]max occurring between 1900 and 0700 local time.

[88] A time series of sulfur gas concentrations from near the middle of the boundary layer (475 m) are displayed in Figure 16. As modeled here, DMS is emitted from the ocean surface at a constant rate of 8.0 μmol m−2 s−1 (see section 4.1) and lost via oxidation to SO2 and detrainment to the free troposphere. Given these conditions, predicted boundary layer concentrations are a function of (1) the depth of the boundary layer into which the surface flux is mixed and (2) the rate of oxidation by OH and NO3. With the prescribed kinetics for OH and NO3 yielding a maximum solar midday oxidation rate of 1.5 × 10−5 s−1, the MBL DMS concentration at steady state should range between 130 and 285 ppt for MBL heights ranging from 1100 to 500 m, respectively, assuming the surface flux is well mixed throughout the boundary layer. Figure 16c shows model predicted DMS ranging between 180 and 270 ppt for this model run, a consequence of the boundary layer height primarily ranging between 500 and 750 m. The initial concentration of 300 ppt is initially diluted through the full 1100 m MBL height, dropping the concentration from 300 to 170, or by nearly the ratio of MBL heights. DMS gas concentrations for the second model run (10 September, reported in the next section, and not shown here) show DMS boundary layer concentrations ranging between 300 and 400 pptv that are a result of lower boundary layer heights for that model run.

Figure 16.

Sulfur gas concentrations for case 1 (11 September), at 475 m, in pptv at 298 K.

[89] The time series of SO2 concentrations produced via oxidation of DMS (Figure 16b) clearly shows the effect of the diurnal variation of the OH concentration on the production of SO2 in the MBL. An inverse pattern can be seen in the DMS concentrations, although not as pronounced, as the DMS source flux is of the same order of magnitude as the oxidation rate. Sulfur dioxide concentrations are seen to vary between 10 and 80 ppt, centered on our starting concentration of 50 ppt and typical of measured SO2 concentrations in the remote MBL. The peak SO2 concentration at −120 hours corresponds to a cloud free MBL period of time extending from the beginning of the run to just after −110 hours. The effect of MBL clouds on the SO2 concentration is seen in its sharp decrease at the onset of clouds throughout the boundary layer at −96 hours, even though occurring at solar midday in the presence of peak OH concentrations. Clouds present anywhere in the model domain, from the surface to 1550 m are indicated in Figure 16b by the symbols; however, their existence throughout the boundary layer (i.e., fog) is seen only at −96 hours (see Figure 9b for vertical and temporal location of clouds). Clouds present at other times are located at the top of the boundary layer, and do not have as pronounced an effect on the peak SO2 concentrations.

[90] Sulfuric acid gas concentrations follow a much sharper diurnal cycle (Figure 16a, note log scale on vertical axis), as in the absence of OH it is efficiently scavenged by condensation on existing aerosol surface area. Peak H2SO4(g) concentrations of approximately 0.2 pptv are predicted during peak SO2 concentrations of 60 ppt, decreasing by 3 orders of magnitude during the night (a result of the OH decrease of 3 orders of magnitude during night hours).

4.2.8. Aerosol Optical Extinction and Optical Depth

[91] MARBLES calculates the wet particle size assuming water vapor equilibrium at each spatial location using the water uptake algorithm developed in Fitzgerald et al. [1998a]. Within a given size section all particles are assumed to have the same mixture of components, in this case sulfate and sea salt. While the total number of particles may remain almost uniform with height in the MBL, the (wet) area distribution will increase dramatically with height (because of increase in RH), as already illustrated in Figure 15. To illustrate the effect of growth with RH in the boundary layer, Figure 17a is a contour plot of the total number of particles with wet radius greater than 0.1 μm. The increase in number in the regions of high RH is the result of growth of smaller particles into the size range rw > 0.1 μm (not an increase in the absolute number).

Figure 17.

(a) Relative number concentrations for rwet > 0.10 μm, as compared with the computed aerosol extinction coefficient, β (m−1), at 550 nm for (b) sulfate aerosol and (c) total aerosol. Contour lines indicate relative humidity at levels of 90, 95, and 98% in Figures 17b and 17c. (d) Total aerosol optical depth through the model domain (1550 m) at 550 nm for the total (black line) and the sea-salt (green) and sulfate (red) components separately.

[92] Also, incorporated into MARBLES is a program for calculating Mie extinction coefficients from the wet size distributions at output times. Figures 17b and 17c show the extinction resulting from sulfate particles that originate in the FT and the total extinction at a wavelength of 0.55 μm, respectively. The difference is due to extinction from sea-salt particles. For this case, even though the wind speeds are moderate, sea salt is responsible for most of the aerosol extinction. This is more clearly seen in Figure 17d, where the optical depth is plotted as a function of time. The total aerosol optical depth (AOD) is given by the black line, and the amounts due to sea salt and sulfate by the green and red lines, respectively. The extinction coefficients and optical depths shown here are for clear sky and do not include any clouds. They are calculated from the wet aerosol size distributions equilibrated to the RH profiles generated by COAMPS, and are therefore the AOD associated with the MBL (or more precisely, the 1500 m domain of MARBLES).

4.3. Case 2 Results (10 September Simulation)

4.3.1. Evolution of MBL Dynamics and Aerosol

[93] The back trajectory for 10 September starts 10 days earlier in westerly flow along the 36th parallel as shown in Figure 7. After about 5 days in weak westerly flow the trajectory turns south and encounters easterly flow, spending the last four days guided by the trade winds. Here we will present only enough data so that this case with lower wind speeds can be contrasted to the 11 September case that had higher winds and a trajectory which originated in a subsiding air near the west coast of the United States. Figure 18 shows the evolution of the MBL dynamics and sea-salt aerosol. Figures 18b and 18c reveal the low wind speed and poorly defined MBL with clouds and humidity extending to over 1200 m in height for the first four days. As the trajectory turns south the wind speed increases, the MBL height descends and by the sixth day, −96 hours, the air mass is well into the easterly flow. Figure 18a shows the evolution of the sea-salt particles with r > 0.5 μm, starting with very low concentrations prior to −132 hours and the build up in concentration as the wind speed increases from under 4 m s−1 to 8 to 9 m s−1 between −120 to −72 hours. These concentrations can be compared to those in Figure 10a for 11 September, for higher wind conditions (note change in color scale).

Figure 18.

Case 2 (10 September, 1000 UTC): (a) total N, for rdry > 0.50 μm; (b) MBL dynamics input to MARBLES, as derived from COAMPS (same as Figure 9b); (c) aerosol optical extinction coefficient β (m−1, for 0.55 μm) (d) aerosol optical depth at 0.55 μm for the MARBLES domain up to 1550 m; and (e) sea-salt aerosol CCN fraction (left axis) and total (black dashed), free tropospheric (red), and sea-salt aerosol (green) CCN concentrations. Contour lines in Figure 18c indicate relative humidity at levels of 90, 95, and 98%.

4.3.2. Optical Extinction and Optical Depth

[94] Figure 18c shows the evolution of the aerosol extinction coefficient for the 0.55 μm wavelength and its dependence on wind speed, humidity and height of the MBL. The integrated (vertical) optical depth between the surface and 1500 m is shown in Figure 18d. Also shown is the individual contribution due to sea-salt aerosol (green) and sulfate aerosol (red) originating in the FT. The FT contribution is quite constant with time and is the dominate component at very low wind speeds, whereas when the wind speed increases the sea-salt component of the aerosol is the biggest contributor to the optical depth (after −108 hours). Because of its wind speed dependence the sea-salt component is much more variable than the sulfate component. Figure 18d can be compared to Figure 17d where the optical depth resulting from sea-salt aerosol is always much greater than the contribution due to sulfate. It is also clear that if optical depth were calculated for longer wavelengths the relative contribution of sea salt to the total would be even greater because of the larger concentration of large sea-salt particles.

4.3.3. Sea-Salt Versus Free Troposphere CCN

[95] Figure 18e shows the total CCN (black dashed line) in the MBL and ratio of sea-salt CCN to the sum of both the sea-salt and FT CCN (black line). The FT component (red line) shows very little variability and since the FT is the dominant species the total CCN concentration exhibits little variability. The sea salt contribution increases from about 5% before −120 hours to almost 30% as the wind speed increases from under 4 m s−1 to over 8 m s−1 after −108 hours. Aerosol size distributions and sulfur speciation figures are not shown for the 10 September case, because little more can be learned than that which was illustrated earlier in connection with the 11 September simulation.

5. Discussion

[96] The predicted aerosol concentrations depend upon (1) atmospheric dynamics, (2) aerosol transformation processes, and (3) aerosol boundary conditions at the surface and the free troposphere. A basic assumption here is that COAMPS gives a reasonable, accurate description of average atmospheric dynamics along the back trajectory, despite the resolution limitations discussed in section 5.5 below. Studies of the sensitivity of the aerosol size distribution to different aerosol transformation processes were performed by Fitzgerald et al. [1998b] for nucleation, surface deposition, coagulation, condensation (of sulfate), turbulent mixing, cloud processing, and precipitation scavenging. However, results presented here show that the magnitude of change seen by modifying the aerosol boundary conditions indicates that the greatest aerosol uncertainty lies in these boundary conditions imposed on MARBLES, i.e., the SSASF and aerosol concentrations in the lower free troposphere together with the MBL-FT exchange mechanisms.

5.1. Sensitivity of MBL Aerosol to Free Troposphere Aerosol and Sea-Salt Aerosol Source Function

[97] Sensitivity of MBL aerosol concentrations to the FT aerosol concentrations and sea-salt source flux was evaluated by making separate model runs with FT concentrations and SSASF doubled and halved, separately and together, for the 11 September model run. In each case, the concentrations responded nearly linearly to the change in the concentration/flux; that is, doubling of the FT aerosol concentrations resulted in a doubling of the fraction of MBL CCN which originate in the FT, and so on.

5.2. Sea-Salt Versus FT Source of CCN

[98] Because of their hygroscopicity, sea-salt aerosol was the earliest candidate suspected of playing a role in cloud and precipitation formation [Woodcock and Gifford, 1949; Twomey, 1954]. To determine whether these CCN were sea-salt or sulfate particles, volatility measurements [Twomey, 1968] were used to show that volatile particles such as sulfate (and possibly organic) are globally much more abundant than the rather involatile sea-salt CCN, except possibly in the MBL. Following Twomey, Dinger et al. [1970] found in the Caribbean that sea-salt particles accounted for about 15% to 50% of the total CCN concentration active at about 0.7% SS and the fraction decreases to essentially zero above the marine temperature inversion.

[99] More recently there has been a renewed interest in the abundance of sea-salt nuclei relative to FT sulfate particles as the source of CCN for MBL clouds [e.g., O'Dowd et al., 1999; Gong and Barrie, 2003; Clarke et al., 2006]. This interest stems from the need of climate change studies to accurately assess the impact that increased anthropogenic SO2 will have on CCN concentrations and hence cloud albedo. O'Dowd et al. [1999], using an optical particle sizing system (OPC) together with volatility discrimination, measured aerosol with r > 0.1 μm and found that during 21 flights over the eastern Pacific, where wind speeds rarely exceeded 10 m s−1, 90% of the particles were deemed to be sulfate, leaving 10% to be sea-salt particles. However, during 6 north Atlantic flights where the wind speed was high (17 m s−1), they found that out of a total of about 250 cm−3 particles with r > 0.1 μm, about 75 cm−3 were sea salt. Actual cloud droplet concentrations were a modest 50 cm−3. Further analysis suggested that about 80% of the particles actually activated in clouds (CCN) would have been sea-salt particles. In the tropical Pacific, Clarke and Porter [1993], making volatility-discriminated size distribution measurements with an optical particle counter in the radius range 0.08 < r < 0.5 μm found that sea-salt particle concentrations averaged about 10 cm−3, which was about 12% of the total particle concentration of 80 cm−3.

[100] Hoppel and Frick [1990] made volatility-discriminated aerosol size distributions with a differential mobility chamber (DMA) over the size range of 0.006 < r < 0.5 μm during a cruise from Hawaii to Tahiti to 55°S latitude. Analysis of this data set together with an Atlantic data set is given by Hoppel et al. [1996] where the number of particles in the cloud-processing mode is used to deduce the number of CCN. The frequency of occurrence histogram of the CCN concentration, drawn from 373 size distributions taken between Hawaii and Tahiti, indicated that the maximum in the CCN frequency distribution lay in the 100–150 cm−3 range with a spread between 50 and 250 CCN cm−3 [Hoppel et al., 1996]. Wind speeds throughout the Hawaii to Tahiti leg were low (typically easterly winds between 5 and 10 m s−1) and volatility discrimination indicated that fewer than 5% of the particles were sea salt. During the leg south of Tahiti to 55° south latitude, much higher winds were encountered. During one stormy period with winds of 12 m s−1 and wave heights of 3–4 m and very clean air, sandwiched between periods of higher wind speeds and higher seas (too high to allow data to be taken), about a third of the particles (r < 0.5 μm) were sea salt (22 out of 67 cm−3) and nearly all particles in the 0.2 to 0.5 μm were sea salt [Hoppel and Frick, 1990, Figure 23].

[101] From the above discussion of measurements over the Pacific, especially in the tropics and trade wind zone with wind speeds less than about 10 m s−1, it would appear that typical concentrations of CCN would be in the 50 to 200 cm−3 range with less than 10% being sea salt. The total CN concentration is about double that amount, or 100–400 cm−3. The modeled values given in Figures 13 and 18e can be compared to the measured values just cited. In both cases the CCN concentrations are about 200 cm−3 and the total about 500–600 cm−3. The fraction contributed by sea salt is less than 5% for the period of very low wind speeds in Figure 18e, increasing to 20% to 30% when the wind speed increases to 9 m s−1, and more in line with the larger predicted fraction for the higher wind speed day shown in Figure 13.

[102] The increase in the predicted sea-salt CCN fraction from less than 5% at wind speeds of 4–5 m s−1 to 20–30% for wind speeds in the 8–9 m s−1 indicated by Figures 13 and 18e appear to be higher than the measured values cited above. While the modeled values are for a specific set of conditions generated by COAMPS, we would still expect that the modeled concentrations should be within the measured range for similar wind speeds. It is clear that any errors in specifying the FT aerosol and the SSASF will translate into errors of similar magnitude in each component within the MBL. Since the FT component is dominant in the submicron size range, except possibly at high wind speeds, errors in the FT size distribution will have a greater effect than will errors in the SSASF on the submicron aerosol.

5.3. Sulfur Speciation

[103] Oxidation of available SO2 to nss-sulfate occurs by three competing processes: (1) OH oxidation of SO2 in the gas phase with subsequent condensation of H2SO4 on existing particles, (2) aqueous-phase oxidation of dissolved SO2 during cloud processing, and (3) heterogeneous oxidation in sea-salt particles. A fourth process, nucleation, was not observed in any of our runs. Figures 11d and 12 show how available sulfur was partitioned for the 11 September run (note that the FT sulfate shown in Figures 11d and 12 is particulate sulfate from the FT and therefore not derived from one of the above conversion processes). While particulate sulfate produced from condensation of gas-phase reaction products can be important at very small radii and in the growth of very small particles to larger radii, it does not appear to be a significant source of sulfate mass in the remote MBL. Thus the primary competing mechanisms are cloud processing and heterogeneous conversion in sea-salt aerosol. Since both cloud formation and sea-salt aerosol generation are highly dependent on meteorological conditions, we would expect significant changes in the partitioning as meteorological conditions change. For the conditions which existed during the 11 September run, with significant cloud cover and moderate wind speeds, Figure 12 indicates that conversion during cloud processing is clearly the dominant process after an initial spin up time during which there were few clouds. The effects of cloud conversion on the gaseous SO2 concentration were also discussed in connection with Figure 16.

[104] The very rapid increase in the SSASF with wind speed shown in Figure 2 would indicate that the heterogeneous reaction would overtake cloud processing rapidly as the wind increased above that encountered on 11 September. To test the sensitivity of the sulfate speciation between these two processes to wind speed, we reran the 11 September data and arbitrarily doubled the wind speed so that the wind speed originally in the range of 5 to 10 m s−1 increased to 10 to 20 m s−1 (and, accordingly, an order of magnitude increase in sea-salt particle flux). As a result, the increased amount of sea-salt aerosol produced by the higher winds essentially scrubbed all the SO2(g) from the boundary layer as fast as it was produced by oxidation from DMS, as shown in Figure 19a, and reduced the amount of cloud-processed-produced sulfate to 1/5 the amount produced with the original wind speeds (see Figure 12). In fact, the diurnal variation in production of SO2(g), shown for the original run in Figure 16 is now seen in the time series of heterogeneous sulfate mass. Very small increases in SO2 are still discernible at 24 hour intervals before being removed by the sea-salt aerosol reaction. Additionally, sulfuric acid gas concentrations (not shown) dropped by approximately 2 orders of magnitude, resulting in even lower amounts of condensed sulfate.

Figure 19.

(a) SO2(g) and (b) total aerosol sulfate mass concentrations for case 1 (11 September run) run with a doubling of the surface wind speed, resulting in an order of magnitude increase in sea-salt particle flux and greatly increased rate of heterogeneous sulfate oxidation.

[105] Comparison with Figure 12 shows that while the partitioning of nss-sulfate shifted from cloud-processing to the heterogeneous reaction, the total mass concentration of sulfate did not significantly change, as it is simply added to the existing aerosol via a different process and shifted the nss-sulfate from the cloud processing peak at r = 0.1 μm to the peak in the sea-salt mass distribution at 1.0 μm. The sulfate mass distribution now contains 2 roughly equivalent modes at these 2 radii.

[106] A similar run with 2× the wind speed was made for the model run of 10 September described in section 4.3, as this run contained a longer trajectory and more varied wind speeds (i.e., lower). Examination of the results of both days would indicate that the results shown in Figure 19 would be typical of those obtained under sustained winds of approximately 16 m s−1.

[107] The heterogeneous SO2 reaction rate essentially shuts off once the droplet alkalinity is consumed (see section 2.3), and yields a mass ratio of 0.0034 (heterogeneous sulfate/sea salt). As the heterogeneous reaction is faster on small particles because of mass transport limitations (see Figure 3), examination of this mass ratio shows the alkalinity on the small sea-salt particles preferentially consumed, as seen earlier in Figure 11d. Figure 20 plots this ratio as a function of particle size for a 11 hour period during the daytime with peak SO2(g) production for both the normal (Figure 20, curves marked by A) and 2× wind speed run (Figure 20, curves marked by B). Curves marked by B in Figure 20 begin at hour −110 with the mass ratio of approximately 0.0025, indicating incomplete neutralization of the available alkalinity due to SO2(g) depletion during nighttime hours. As SO2(g) begins to be generated with the increase in OH during daylight hours, the curves, shown in 1 hour intervals, indicate that the smaller aerosol particles get neutralized first, increasing to larger radii over the course of the 12 hour period, until all particles with radii <2 μm are saturated. For the high wind case, this diurnal cycling was observed daily as new sea-salt aerosol was added at night with insufficient SO2 to neutralize even the smallest particles. This contrasts with the original run, shown in Figure 20 (curves marked by A), which have saturation in particles up to a radius of 10 μm; incomplete neutralization at larger sizes is due not to lack of available SO2(g), but to the short lifetime due to gravitational settling of these larger particles.

Figure 20.

Ratio of heterogeneous sulfate mass to sea salt mass as a function of size, for the 11 September run, where curves are for each hour during daylight hours from −110 to −100 hours. A ratio of 0.0034 (dashed line) indicates all sea-salt aerosol alkalinity is consumed. Curves marked by B are results from the simulation with wind speeds twice the COAMPS predicted values, and curves marked by A are those using base case values. Comparison with Figure 19 indicate that at t = −110 hours, SO2 (g) is completely depleted, and some sea-salt aerosol alkalinity remains at all radii. As the SO2 begins to increase, the alkalinity is preferentially consumed first on the smallest particles, as can be seen in the series of curves marked by B shown at 1 hour intervals, eventually being generated fast enough to consume the alkalinity in particles up to approximately 2 μm radius. Results marked by A are shown for comparison for the run with base case wind speeds and show typically that the reaction runs to completion up to r of 10 μm for this time interval, even during nighttime hours.

5.4. Role of Precipitation

[108] For both the 10 and 11 September runs discussed above there was no precipitation along the back trajectory generated from the COAMPS data. Again, the limited resolution of precipitation in COAMPS presents limitations on the aerosol model. There are two precipitation schemes in COAMPS. The first simulates convective precipitation events smaller than the grid spacing. If the model determines, at a grid point, that convective clouds should occur, COAMPS sets out to determine the energy and moisture tendencies supplied to the model grid and calculates a parameterized subgrid precipitation (averaged over the grid spacing). In the second scheme, for larger systems where there is resolved supersaturation at a grid point, then the precipitation is calculated as stable precipitation. Contour maps of the frequency of precipitation during the 45 day period for which the COAMPS reanalysis was executed show low frequency of precipitation in the region traversed by the trajectories ending at FLIP (such as the two shown in Figure 8). Higher frequencies of precipitation, both convective and stable, were seen to the south (equatorial) side of the trajectories, with higher frequencies of convective precipitation (no stable precipitation) near and to the west of Hawaii.

[109] Katoshevski et al. [1999] used a box model to study MBL aerosol processes and assumed that precipitation occurred once every nine times clouds were encountered, with attendant scavenging of a portion of MBL particles from the box. Doubling the rain rate from 5 mm h−1 (base case) to 10 mm h−1 decreased the average total concentration in the column by less than 20%. Decreasing the frequency of rain from once every nine cloud cycles (base case) to once every 18 cloud cycles increased the average concentration in the column by less than 20%. Thus Katoshevski et al. [1999] conclude that variations in precipitation frequency have only a minor effect on the average total concentration found in the MBL.

[110] Fitzgerald et al. [1998b] used the earlier version of MARBLES to simulate aerosol evolution over 10 days in air masses advecting off the east coast of the United States. In the base case precipitation occurred during two 12 hour precipitation events centered on day 4 and 8 with rate of 1 mm h−1 (to give climatologically 10 day average) and the size distribution at day 10 was used as the end point for comparisons. Increasing the second rain rate event (centered on day 8) from 1 mm h−1 to 10 mm h−1 decreased the concentration of particles under the cloud-processing maximum by a factor of about three (on day 10), but had no noticeable effect on the interstitial particles. Changing the FT exchange velocity from 0.1 to 1.0 cm s−1 had a much larger effect. These prior results, together with those found here, indicate that the aerosol size distribution is quite robust in the face of precipitation events associated with smaller scale disturbances; that is, recovery of the size distribution via exchange with the FT, cloud processing, and surface production are the dominant processes shaping the SD over a period of several days (even in the face of convective precipitation). The above argument is not meant to imply that wet deposition is not an important aerosol sink; to the contrary, wet deposition is undoubtedly the major removal mechanism for submicron particles. This occurs globally, to some extent, by small-scale convective processes, but more importantly, through large-scale precipitation events associated with frontal storm systems, which are not encountered here.

5.5. Adequacy of COAMPS as a Driver of MARBLES

[111] Aerosol concentrations at a given time are sensitive to processes occurring over a periods of several days prior to that time. This suggests that the meteorological model should be of a scale comparable to that of mesoscale models for the aerosol to be within the domain of the model for a number of days. However it is not at all clear that typical mesoscale models have the resolution required for high vertical resolution of MBL aerosol as described here. The horizontal resolution of COAMPS can be increased by the user and the nested grid capability can be used to increase resolution over a targeted area. The vertical resolution of COAMPS is typically fixed and cannot be altered without significant coding changes. The height and dynamics of the MBL are on a scale too small to be resolved by the equations of motion used in meteorological models so they are parameterized, as are other subgrid phenomena such as clouds and precipitation. The accuracy with which the model determines the height of the MBL and the magnitude of the vertical profile of mixing coefficients affect the aerosol dynamics. For small particles the MBL can be considered well mixed for nearly all times, so the accuracy of the mixing strength in the interior of the MBL is not as important as getting the height and mixing across the top of the MBL correct. For very large particles gravitational fall out prevents the particles from reaching the top of the MBL and the strength of the mixing, which suspends the particles against gravitational settling, is more important than height of the MBL. At intermediate sizes the accuracy of both the mixing strength and height of the MBL are important.

[112] COAMPS has eight levels between the surface and 1000 m, with expanding depth such that the cell centered at 750 m is about 300 m deep. While interesting and useful results were obtained using the output of COAMPS, the lack of vertical resolution at the top of the MBL was the most severe limitation imposed by the use of COAMPS data here. While interpolation of the COAMPS values onto the more dense MARBLES cells gave good numerical resolution for MARBLES, the basic limitations of the COAMPS resolution were not eliminated. The lack of resolution at the top of the MBL is seen in the rather discrete jumps in the height of the MBL as evidenced in the contours of turbulent mixing coefficients in Figure 9b. Also the lack of resolution is seen in the clouds extending from the MBL into the lower FT as discussed earlier. In a few instances, COAMPS places clouds in a cell where the mixing is zero; just above what we would interpret as the top of the MBL. An example can be found at −132 hours in Figure 9b. There are periods such as the one at −12 hours in Figure 9b were both clouds and mixing spike upward above the mean height of the MBL. We assume these represent periods of penetrative convection with cloud formation above the mean height of the MBL.

[113] Vertical motions on a scale intermediate to MBL turbulence and the large-scale vertical motions resolved by the numerical solution to the fluid mechanical equation are not accounted for in mesoscale models. While this intermediate scale may have little effect on meteorological predictions, they are known to be important in FT tracer studies as discussed in section 4.1.2. In the modeling reported here, adding a small amount of turbulent mixing to the FT simulated the effect of FT mixing. This mixing was necessary to replenish aerosol lost to the MBL through entrainment.

[114] The successful use of COAMPS as a driver of MARBLES as described in this paper, despite these limitations, argues that future efforts use mesoscale models with increased vertical resolution. We know of no fundamental reason why the vertical resolution of existing mesoscale models cannot be increased. Increased resolution, of course, will come at the expense of computational time. Future efforts should seek the mesoscale model that can best resolve quantities at the top of the MBL and that incorporate the most accurate parameterizations of MBL processes, cloud formation, and precipitation rates.

6. Conclusions

[115] The remote marine boundary layer is a unique setting to study aerosol microphysics, with a simple domain and relatively small number of, if poorly characterized, sources and sinks. A number of models have been developed to study the complex physical and chemical properties which govern the atmospheric size distribution there. These, for the most part, have used generic inputs, such as fixed exchange with the FT, and particle size ranges small enough to treat the MBL as a well-mixed box with exchange at the top (with the FT) and the bottom (with the ocean surface). In the work presented here we have linked a 1-D, highly resolved (column) aerosol model with mesoscale meteorological data so that the required meteorological and dynamical variables can provide a more realistic representation of the atmosphere at a desired real time. This linking, accomplished via the use of air mass trajectories and hourly vertical profiles of the necessary meteorological data, has allowed for the important and unique use of both turbulent mixing coefficients (describing both the strength of mixing and the height of the MBL) and large-scale vertical velocities.

[116] The high vertical resolution of the size-resolved aerosol components, as exemplified by the profiles of dry and wet sizes shown in Figures 14 and 15, are unique capabilities of this model and make it valuable in the prediction of aerosol optical extinction, as shown in Figures 17c and 18c. The particle profiles include not only the effect of particle swelling with changing relative humidity, but also the effect of gravitationally induced gradients resulting from the turbulent mixing and the large-scale vertical velocity given by the meteorological model.

[117] Results presented here have demonstrated the utility of the model using COAMPS-generated output during the RED experiment conducted off Oahu, Hawaii, in 2001. The modeled evolution of MBL aerosol over 2 separate open-ocean trajectories have shown that:

[118] 1. Interstitial particle number and CCN concentrations are controlled largely by a balance between (1) entrainment of FT aerosol into the MBL and (2) generation of sea salt at the surface (which is variable with wind speed). Newly introduced aerosol mixes uniformly within the MBL within a few hours, and the contribution of small particle FT aerosol remains rather steady at approximately 2/3 the FT value. Small particle sea salt aerosol concentrations are much more variable, dependent on wind speed.

[119] 2. CCN (rdry > 0.033 μm) concentrations were predicted to be stable near 200 cm−3 after an initialization period of approximately 24 hours, with sea-salt particles contributing 10–30% to the total during periods of typical moderate wind speed (5–10 m s−1). This result is largely determined by the imposed boundary condition and is somewhat larger than indicated by prior measurements in the tropics.

[120] 3. Aerosol transport by large-scale vertical motions (vertical velocities), while negligible over short timescales, was important in diluting the MBL with FT air over times on the order of 1 day and greater, and in predicting entrainment of FT aerosol. Sea-salt particle concentrations were diluted by approximately 50% as compared to model runs made without transport by large-scale vertical velocities (i.e., over those runs using only a FT exchange velocity).

[121] 4. Sulfate mass, under moderate winds of 5–10 m s−1, was split between influx from the free troposphere and that generated by cloud processing, with heterogeneous sulfate contributing another ∼10% to the total sulfate mass. However, under high winds, (10–20 m s−1), the larger sea-salt flux resulted in greatly enhanced heterogeneous oxidation, preferentially depleting available SO2 (g) and reducing the cloud processed sulfate mass by 80% of its value run with moderate winds. Sulfate aerosol mass from condensation of gas-phase oxidation products was negligible compared to these other processes at all sizes.

[122] 5. Under moderate wind speeds and greater, sea-salt aerosol is responsible for the bulk of the aerosol optical extinction, with MBL sulfate AOD ranging from 0.01 to 0.03 and sea-salt AOD ranging from 0.01 to 0.10 (0.05 to 0.10 under moderate winds).

[123] While the particular use of mesoscale meteorological model in this study imposed undesirable limitations on the aerosol model, we believe that future efforts following a similar methodology with higher-resolution meteorological models with improved cloud, precipitation, and MBL parameterizations will be of increasing importance and usefulness.


[124] This research was sponsored by the Office of Naval Research. The authors would like to thank the investigators in the RED experiment, especially Ken Anderson and Jeff Reid for their support of our participation in RED.