2 Integrated view of present state of knowledge re. aerosols
5 Properties and Distributions of the atmospheric aerosol
7 Measurement techniques and strategies
Tropospheric aerosols are suspensions of small particles, of diameter ranging from nanometers to micrometers, in the lower kilometers of the atmosphere. The particles derive from "primary" sources involving direct emissions of particles, and by "secondary" processes, reactions of gaseous precursor emissions in the atmosphere to form particulate matter.
Aerosols are important to global atmospheric chemistry for a number of reasons. Aerosol particles present surface area to reactive gaseous species that can provide new pathways for atmospheric chemical reactions and/or serves as sinks for reactive species such as free radicals. Aerosol particles serve as cloud condensation nuclei thereby altering the composition of cloud water and influencing the course of aqueous reactions in clouds. Through scattering and absorption of radiation in the solar spectrum aerosol particles influence the rate of primary photochemical reactions.
Aerosol particles in the lower troposphere exhibit an average residence time of roughly one week which is sufficient to allow transport on continental to intercontinental scales, with resultant influence on global atmospheric processes. Their global distribution has been demonstrated by satellite remote sensing for some time over the oceans. An example from the Polder/ADEOS satellite is given for the month of June 1997 in Figure Intro.1. The top part of the figure clearly shows the dust plumes from subtropical desert regions. In northern mid-latitude regions of the bottom part of Fig. Intro.1 small, mainly combustion-derived particles in the continental plumes of the northern hemisphere are indicated by the spectral dependence of the upwelling radiation (see also Higurashi and Nakajima, 1999).
| Species | Natural processes | Anthropogenic processes or effects | Present day aerosol compared to pre-industrial time |
| Primary Particles | |||
| Mineral dust Sea salt Biogenic particles Black carbon |
Wind Wind Wind Vegetation fires |
Land use change, Climate change? Climate change? Land use change, Climate change? Fossil fuel and biomass burning |
Increased dust Change in sea salt Decreased biogenic particles? Increased black carbon |
| Precursors of Secondary Particles | |||
| Dimethylsulfide SO2 NH3 NOx Volatile organic componds |
Phytoplankton degradation Volcanic emissions Microbial activity Lightning Emission from vegetation |
Increased oxidation capacity Fossil fuel combustion Agriculture Fossil fuel combustion Increased oxidation capacity Industrial processes |
Increased sulfate Increased sulfate Increased ammonium nitrate Increased nitrate Increased secondary organic particulate matter |
2.2. Aerosol processes (Raes)
Figure Intro 3 depicts schematically the microphysical processes that influence the size distribution and chemical composition of the atmospheric aerosol. It illustrates the wide range of sizes that is involved in the formation and evolution of aerosol particles, the existence of primary particles and secondary particles. Figure Intro 3 also highlights how aerosols participate in atmospheric chemical processes through homogeneous, heterogeneous and in- cloud reactions. Traditionally, atmospheric aerosol particles have been divided into two size classes: coarse (D p > 1 µm) and fine (Dp < 1 µm), reflecting the two major particle source processes: mechanical ones for coarse and combustion plus secondary for fine material. Both populations strongly overlap, however, in the 0.1-1 µm diameter range. 2.3. Role of atmospheric general circulation (Raes)
Studies of aerosol formation in specific atmospheric compartments, such as the marine boundary layer, the free troposphere etc. have shown that aerosol characteristics observed within a certain compartment cannot always be explained by processes occurring within that compartment. The characteristic time of many of the microphysical aerosol processes is days up to several weeks, hence longer than the residence time of the aerosol within a typical atmospheric compartment. Hence, to understand aerosol properties, one cannot confine the discussion to such compartments, but one needs to view aerosol microphysical phenomena within the context of atmospheric dynamics that connects those compartments (Raes et al., 2000)
3.1. Primary emissions
3.1.1. Mineral soil derived dust (Marticorena)
3.1.2. Sea spray (O'Dowd)
3.1.3. Primary biogenic particles (Artaxo and Jaenicke)
3.1.4. Combustion Aerosol (Bond, Liousse and Cachier)
3.1.4.1. Particulate Organic Matter (POM)
3.2. Precursors to secondary aerosols
Figure Intro 3: Schematic picture of the microphysical processes that influence the size distribution and chemical composition of the atmospheric aerosol
Primary particles that are derived from division of bulk material and subsequent suspension by the wind, such as sea salt, soil dust, and biological material, have most of their mass associated in the coarse particle range. However their highest numbers occur in the 0.1-1 µm range. Because of their low concentrations and large sizes, primary particles derived from division of bulk material generally do not coagulate with one another, but can become mixed with other species through uptake of condensable material from the gas phase.
Another important type of primary particles is so-called soot, emitted from combustion, consisting of black carbonaceous material which has not been fully oxidized in the combustion process, often mixed with refractory metal oxides. Soot is formed in the combustion of carbonaceous fuels as particles with a diameter 5 to 20 nm. However, such particles rapidly coagulate to form fractal- like aggregates, which in turn collapse to more compact structures having a diameter of several tens of nanometers under the influence of capillary forces of condensing vapors.
Gas to particle conversion in the atmosphere has traditionally been thought to occur when a volatile species reaches a concentration that exceeds its equilibrium vapor pressure, resulting in a thermodynamic driving force for condensation. In the atmosphere, this situation has generally been thought to be driven by chemical reactions leading to a products with very low vapor pressure. Alternatively this situation can be reached simply by a reduction in temperature, which has the effect of reducing the species' equilibrium vapor pressure. A recently suggested alternative possible mechanism is the co- condensation of several species, (discussed in detail in the section on aerosol transformation).
Formation of new particles (nucleation) has been thought to occur only when the vapor pressure of the condensing species exceeds the equilibrium vapor pressure of the condensed phase material by an amount sufficient to overcome the energy barrier to new particle formation due to the strong influence of surface tension associated with such nuclei. Because the equilibrium vapor pressure increases with increasing curvature of the surface, the equilibrium vapor pressure over molecular clusters formed by nucleation is much larger than that over a flat surface or a film on a pre-existing particle. Consequently, nucleation is energetically less favorable than condensation, and will be favored by the absence of pre-existing particulate surface or by very low temperatures. Recent work suggests the possibility of barrierless ternary nucleation of sulfuric acid, water vapor, and ammonia (discussed in detail in the section on particle formation) Once nucleation occurs, the new particles grow by condensation and self-coagulation. As particles reach a diameter of the order of the mean free path length of the condensing molecule, condensation becomes diffusion-limited and slows down. Also, self-coagulation, which is a second- order process, eventually slows down as number concentrations fall. Under background tropospheric conditions, particles formed initially by nucleation require days to weeks to grow larger than about 0.1 µm solely by condensation and coagulation (Walter, 1973). Under polluted, urban type conditions, this growth can occur within a day because of the strong supply of condensable material from homogeneous gas phase reactions (Raes et al., 1995).
Another more complicated but important growth process is by chemical processing in non-precipitating clouds (Friedlander, 1978; Hoppel et al., 1990; Hoppel et al., 1994a; Mason, 1971). This process begins with the uptake of water vapor. According to traditional Köhler theory, a critical supersaturation for condensational growth can be defined which depends on particle size and on the amount of soluble material in the particle. More generally, particle size, wettability of its surface and the combined solubilities of the chemical mixture comprising the particle control the equilibrium size of a particle as a function of water vapor saturation ratio and that of other condensable gases. Once a drop of sufficient volume is formed, gaseous species like SO2 can dissolve and be oxidized in the aqueous phase. On average, nine out of ten clouds evaporate rather than precipitate. When the droplets evaporate, larger particles reform as a result of the additional oxidized material, e.g. sulfate (Birmili et al., 1999b; Yuskiewicz et al., 1998). BR
Reactions that occur in cloud water also occur in solution droplets in subsaturated (non-cloudy) atmospheres, however with different efficiencies because of the larger ionic strength in such droplets. Additionally, adsorbed gases react on the particle surfaces yielding products that might either remain on the particle or return to the gas phase.
Aerosols are removed from the atmosphere by dry and wet processes. For particles below Dp » 0.1 µm the dominant dry removal mechanism involves diffusion to the surface, a process which becomes less efficient as particle size increases. Coarse particles (Dp > 1 µm) settle gravitationally, a process which becomes less efficient as particle size decreases. In the range 0.1 < Dp < 1 µm, dry removal is very slow, and the formation and growth processes discussed above tend to accumulate condensed material in this size range in the sense of extreme atmospheric residence times. Provided their condensational growth is not hindered by any of the above factors, these particles are removed mainly by growth to cloud drops during cloud formation and subsequent removal from the atmosphere in precipitation.
Figure Intro 3 gives a rough indication of particle size categories. The category "fine particles" is further divided into "accumulation" (mode) particles (0.1 < Dp < 1 µm), "Aitken" (mode) particles" (0.01 < Dp < 0.1 µm) and "nucleation (mode) particles" (Dp < 0.01 µm). Reference to "modes" is made because measurements of the complete particle size distribution indeed show the existence of such modes (Covert et al., 1996b; Hoppel, 1988; Whitby, 1978). These modes result from specific processes. The nucleation mode is the result of recent nucleation events involving formation of new particles from gases, in part associated with combustion processes, and the Aitken mode results from condensation and self coagulation and a tail of high temperature primary emissions.
The sections of this chapter follow the schematic process picture given in Figure Intro 3. After a discussion of natural and anthropogenic aerosol sources and major aerosol transformation processes in the atmosphere the experimental and modeling tools of atmospheric aerosol research are reviewed. Our state of knowledge on distribution and properties of the atmospheric aerosols are given.
3. Emissions
Mineral soil-derived particles are produced by aeolian erosion mainly in the arid and semi-arid regions of the world. During their transport in the atmosphere, dust particles are known to influence the radiation budget, and thus climate, mainly by back scattering and absorbing incoming and outgoing radiation (Li et al., 1996; Sokolik and Toon, 1996; Sokolik et al., 1998). Natural or human-induced climatic changes influence the emission strength and locations, which are sensitive to wind velocity and precipitation.
It is also recognised now that mineral dust plays a significant role in tropospheric chemical reactions. In particular, mineral particles interact, via heterogeneous chemistry reactions, with the nitrogen (Manabe and Gotlieb, 1992; Wu and Okada, 1994) and the sulfur cycle (Dentener et al., 1996; Parungo et al., 1995) and thus influences particle acidity (Varma, 1989; Zhao et al., 1988), the gas phase composition and the number, the size distribution and the chemical composition of aerosol particles.
Since the seventies, estimates of the global mineral dust emissions have been published by various authors (see the summary by Duce, 1995). These estimations range between 100 and 3000 Tg yr-1 , most of them being comprised between 500 and 1500 Tg yr-1 . This large range of uncertainty results for a part from the different approaches used by the authors, but also from the size-range considered as describing the "dust" size distribution. The size spectrum of the mineral particles close to the source regions is wide, including particles from the submicrometer range up to several tens of micrometers (d'Almeida et al., 1991; Gomes and Bergametti, 1990; Hoyningen-Huene von et al., 1999) As a result, the complete size distribution is difficult to sample and measure with good precision. Because very few experiments covered the complete size range, the dust size distribution in source regions remains poorly documented.
Assessments of the global dust emissions show that mineral aerosol is presently a major source of tropospheric aerosol since it represents about 50% of the total particulate mass injected every year in the atmosphere.
Mineral dust emission by wind erosion involves physical processes controlled both by meteorological parameters and the soil surface features. The erosion threshold wind velocity is one of the key parameters to describe these emissions: the dust event frequency is determined by the number of times with wind velocity exceeding this threshold while their intensity depends on the magnitude of this excess. The combination of these non-linear processes with the heterogeneity of the surface characteristics makes the dust emissions highly heterogeneous in space and time. As a result, the estimation of the strength and location of the mineral dust emissions at regional or global scale requires the knowledge of (i) the related physical processes and (ii) the surface features of the source regions. Thus, physical models of mineral dust emissions have been recently developed (Marticorena and Bergametti, 1995; Shao et al., 1996). They are based on an explicit description of the main physical processes involved during dust production and they have been validated by comparison with micro-scale experimental data. When applied to large scales (like the Saharan desert) these models simulate with a good accuracy the dust events frequencies and their intensity (Marticorena et al., 1997). Moreover, their capability to reproduce specific dust events is quite satisfying showing that the major factors and processes controlling dust emissions have been captured in such models.
Mineral dust is mainly produced by sand-blasting processes (Gillette, 1978; Gomes and Bergametti, 1990; Shao et al., 1993) that are not fully understood. Because the efficiency of the sand-blasting processes is not physically parametrized, these models predict quite precisely the mass of emitted dust but they can not simulate the associated size distribution. This is a severe limitation because most of the atmospheric effects of dust are strongly size- dependent. Recently, wind-tunnel studies of sandblasting processes have documented the size distribution of the mineral dust (Alfaro et al., 1997). They show the dependence of the size distribution on wind conditions and soil characteristics (Alfaro et al., 1998). Thus, it is reasonable to expect that this kind of studies could soon provide an operational parameterization of the processes controlling the dust size distribution.
Another strong limitation in the use of physical dust emission schemes for global atmospheric transport modelling is due to the required input parameters, especially those characterizing the soil type and the surface roughness. Various approaches have been developed to retrieve from satellite observations information on soil mineralogy (Escafadal, 1994) or surface roughness (Greeley et al., 1997) in arid regions. These works concern high resolution data and thus can hardly be extended to retrieve surface parameters on global scale. However, they can be viewed as a first step towards the establishment of similar treatment for satellite products with coarser resolution but on a global scale.
For semi-arid regions, the assessment of the surface features is more complex since the seasonal precipitation pattern induces the growth of the vegetation and thus a change in surface roughness, soil moisture and the protection against wind erosion. Moreover, increasing human land use in these regions (for example by overgrazing and changes in agricultural practices) also affects the response of the soil surface to aeolian erosion. In a recent study, Tegen et al. (1995) have estimated that 30 to 50% of the present dust emissions result from disturbed soils. Additional developments on both the natural and anthropogenic factors are still necessary to model the mineral dust emissions in semi-arid regions.
Sea-salt particles are produced at the ocean surface by the bursting of air bubbles resulting from entrainment of air induced by wind stress. On bursting, these bubbles produce film and jet drops (Andreas, 1998; Blanchard, 1963; Blanchard and Woodcock, 1957; Fitzgerald, 1991). These bubbles are most concentrated in whitecaps associated with the breaking of waves which commences at wind speed approximately 3-4 m s-1 . Depending on its size, each bubble can generate as many as 10 jet drops with a typical size of 0.5 µm diameter (although extending to sizes greater than 20 µm), and up to several hundred film drops in the sub-micrometer range (Cipriano and Blanchard, 1981; Woodcock, 1972; Woolf et al., 1987). At wind speeds in excess of 7-11 m s-1, the tearing of wave crests results in the injection into the marine boundary layer of ultra large spume sea-salt particles (Monahan, 1986; O'Dowd et al., 1997b; Smith et al., 1989; Wang and Street, 1978; Wu et al., 1984).
Sea-salt size distributions with three modes between 100 nm and 300 µm were found over the North East Atlantic (O'Dowd et al., 1997b), thought to correspond to the three different production mechanisms (cf. Figure O'Dowd 5 1). A mode, with diameter of 200 nm, thought to result from film drop production, was found to dominate the number concentration. A second mode at around 2 µm was ascribed to jet drop production. A third mode, ascribed to spume drop production, was observed at 12 - 20 µm and extending to diameters greater than 300 µm under the highest wind speed conditions. 1
Although number concentration is dominated by the smallest, sub- micrometer salt particles, surface area is dominated by the jet drops and volume is dominated by spume drops when present. Number, surface and volume size distributions, constructed from log-normal curves derived from measurements (O'Dowd et al., 1997b), are shown in Figure O'Dowd 1 to highlight the relative importance of each mode. In this figure, distributions are shown for a relative humidity of 80%, leading to particles sizes » twice their dry size.
Figure O'Dowd 1: Sea-salt number, surface and volume size distributions (for relative humidity 80%) for conditions typical of 17 m s-1 wind speeds (N » 70 cm-3 ). Also shown is a typical accumulation mode sulphate number distribution for N = 100 cm-3 O'Dowd, 1997 #9702.
Number and mass concentrations are strongly dependent on wind speed, typically adhering to an exponential relationship in the form of log C = aU10 + b, where C is the concentration, U10 is the wind speed at 10 meter height, and a and b are parameters which depend on particle size (de Leeuw, 1986; Exton et al., 1985; Gras and Ayers, 1983; Lovett, 1978; O'Dowd and Smith, 1993; O'Dowd et al., 1997b; Smith et al., 1989; Woodcock, 1953). For moderately high wind speeds of 17-18 m s-1 , these relationships predict number concentrations of 80-100 cm-3 in the film drop mode, 3-5 cm-3 in the jet drop mode, and » 0.005 cm-3 for the spume drop mode.
Investigations of the relationship between sea-salt mass, measured using impactor techniques (e.g., review by Fitzgerald, 1991) exhibit good agreement with the optical particle counter techniques for relevant size overlap. Other experimental techniques such as growth factor analysis (Berg et al., 1998) and mass spectroscopy (Murphy et al., 1998a) also confirmed the presence of significant concentrations (up to 100 cm-3 ) of sea-salt particles at sub- micrometer sizes over the southern oceans.
Vegetation is long recognized as an important source of both primary and secondary aerosol particles. Two thirds of the Earth's land mass is covered by plants. This surface is much larger than that of deserts. Forest vegetation is the principal global source of atmospheric organic particles (Cachier et al., 1985), and in a tropical forest, natural vegetation plays a major role in airborne particle concentrations (Andreae and Crutzen, 1997). Only a few studies of natural biogenic aerosols from vegetation in tropical rain forests have been undertaken (Artaxo et al., 1994; Artaxo et al., 1990; Artaxo et al., 1988; Echalar et al., 1998). The natural biogenic aerosol comprises many different types of particles, including pollen, spores, bacteria, algae, protozoa, fungi, fragments of leaves, excrement and fragments of insects. This aerosol component is mainly in the coarse size fraction. The mechanisms of particle emission are still not well understood, but probably include mechanical abrasion by wind, biological activity of micro-organisms on plant surfaces and forest litter, and plants physiological processes such as transpiration and guttation. These processes may generate particles containing biogenic related elements such as Na, Mg, P, S, K, Ca, Zn, and Rb (Beauford et al., 1975; Beauford et al., 1977; Nemeruyk, 1970). Also, a significant fraction of the aerosol is comprised of secondary biogenic particles, formed by gas-to-particle conversion of organic and sulfur-related natural biogenic gases. Bacteria in forested areas were found in the size range of 0.5 to 2.5 µm (Jaenicke and Matthias-Maser, 1992). Biological activity of micro-organisms on leaf surfaces and forest litter results in airborne particles. Windblown pollens certainly contribute to coarse particles in forested areas. Particulate material containing Zn, Pb and Cu are produced by higher plants (Beauford et al., 1975; Beauford et al., 1977). The transpiration of plants can lead to migration of Ca2+ , SO42- , Cl- , K+ , Mg2+ and Na+ to the atmosphere. The elements (K, P, S, Zn, Rb and others) are essential to superior plants. They are present in the fluids circulating in the plant and are released from the leaves to the atmosphere (Nemeruyk, 1970).
Natural biogenic aerosol particles emitted by plants play an important role in nutrient cycling in tropical ecosystems. The tropical ecosystems are in a delicate nutrient balance characterized by intense internal recycling, and depend on atmospheric input of certain mineral nutrients to achieve this balance (Vitousek and Sanford, 1986). For the past twenty years most of the tropical forest areas have been under strong pressure by rapid change of land cover, with forest and adjacent savannahs being cleared, most of the time with the use of fire, and converted into pasture and agricultural fields at a substantial rate (Kaufman et al., 1998). Pyrogenic and natural biogenic emissions to the atmosphere in the Amazon Basin may have an impact on the global tropospheric chemistry, because this region exhibits intense convective activity that may inject gases and particles to high altitudes where they can be transported over long distances (Echalar et al., 1998).
Most of the analysed elements are dominant in the coarse mode, with the only exception of sulfur. The particulate mass for particles less than 10 µm is 5-20 µg m-3 . Phosphorus is strongly present in the coarse mode with more than 90% of P-concentrations in particles larger than 2 µm.
A different approach to the problem of biogenic particles has been taken by classifying them according to a certain method that is sensitive to the ubiquitous protein in primary biological material. Mathias-Maser (1998) gives a summary of the methodology and of the corresponding results.
The total content (by number or mass) for particles > 0.4 µm diameter is rather constant throughout the year (»25 % of the particles > 0.4 µm diameter). The biological matrix, however, is changing. In spring pollen prevail. Other seasons are characterized by abrasion products, fragments, micro-organisms, and other material. Unpublished measurements from Ireland complemented the picture for maritime air. For all size ranges the protein-containing number fraction is greater in marine (34 - 76%) than in continental (29 - 46%) air masses.
There are two main reasons for studying carbonaceous particles (TC): due to their noticeable optical properties, these particles may have a global radiative impact (Penner et al., 1992) and their exceptional surface reactivity generates important chemical interaction with gases (O3 , PAN, NOx ). Furthermore, as other atmospheric particles they affect photolysis rate of key species, and the role of BC particles could counteract that of other particles (Dickerson et al., 1997). More data exist for biomass burning where TC is more dominant particulate matter than for fossil fuel sources.
Primary particles emitted directly from combustion processes consist mainly of carbonaceous and inorganic compounds, which are found in both the fine and coarse modes. The mineral matter, termed fly ash, comes directly from the fuel; (see review of mechanisms of particle formation by Flagan and Friedlander, 1978). Submicrometer carbonaceous particles (black carbon or BC) are formed within the flame by a complicated process (e.g., Haynes and Wagner, 1981), and larger particles result from the incomplete burnout of solid fuel. Organic products of incomplete combustion, as well as sulfur, may also condense on the particles.
The formation of the particulate matter (PM) depends strongly on properties of the burning process, especially the temperature, the amount of oxidizer, and the fuel. Production of both submicrometer fly ash and BC requires high temperatures, (Glassman et al., 1994), although light-absorbing material may also be produced by the oxidation of existing carbonaceous material. Emissions also depend on post-combustion, size-dependent emission controls.
Source strength estimates for primary particles
Global source strength estimates for primary particulate matter are available only for black carbon (BC: Cooke and Wilson, 1996; Liousse et al., 1996). These inventories may count some fly ash as BC, overestimating the source strength of BC (Bond et al., 1998).
Mineral matter is a significant component of primary PM when a region has a large component of industrial coal combustion and few emission controls. As a region becomes industrialized, its combustion practices becomes more efficient and large-scale, and the primary PM emissions contain more ash and have a higher single-scatter albedo. As further development occurs, the optical properties of the emissions may not change, but their total mass will decrease as controls are installed.
Research needs concerning primary combustion particles
Characterization of primary particulate matter should include divisions by mass into the fractions that are optically relevant: ash, organic carbon, black carbon, and possibly sulfates. Particular attention is needed for the combustion practices thought to contribute the largest fractions to the global burden of primary particles, such as domestic coal and wood burning and industrial combustion in less-developed countries. The hygroscopicity of these particles should be investigated as a first step in estimating their lifetimes and contribution to cloud droplet nuclei. These studies are most important for long- lived submicrometer particles.
A tabulation of combustion practice by region is needed for more accurate development of emission inventories. Information on practice in non-OECD countries is particularly lacking. There are no quantitative estimates of global fly ash source strength, in particular concerning fly ash from biomass burning. There is also missing information on primary mineral matter from sources other than combustion for heat and power.
For either biomass burning or fossil fuel sources, POM is the predominant fraction in the carbonaceous particulate material. BC/POM ratio is, however, highly dependent on the type of sources. The BC/POM ratio is higher for fossil fuel sources than for biomass burning. Also, BC/POM is linked to the type of combustion: the most efficient combustion (at high temperatures) produces relatively more BC. The primary or secondary origin of anthropogenic POM emissions is difficult to assess as both fast and slow chemical reactions produce anthropogenic organic particles. Emission inventories of organic particles may be compiled based on observations in environmental conditions close to source areas (Liousse et al., 1996). POM budget is underestimated by a factor of two when « real » primary POM is only considered (Cooke et al., 1999). Another type of source for secondary organic particles (SOA) is the conversion of natural organic gases such as terpenes (cf. section 4.2). Figure Liousse 1 presents the state of art for existing global inventories. These global budgets are in agreement with previous rough estimates (Cachier et al., 1986; Seiler and Crutzen, 1980; Turco et al., 1983).
Figure Liousse 1 Comparison of global inventories for BC and POM particles
Many steps are required for such an exercise: global fuel maps, type of combustion, emission factors (EF) with information on particle size distribution, source seasonal variability and height of emission. The first global emission inventory (column 1 in Figure Liousse.1) was constructed by Penner et al. (1993) for BC and POM emitted by tropical savannah and forest fires assuming same emission factors for both sources. With an indirect approach using SO2 global distribution and scaling with regional BC/SO2 ratios the inventory was largely overestimated. In 1996, Cooke and Wilson presented a global BC inventory including biomass burning sources (BB) and an exhaustive description of fossil fuel sources (FF). Using satellite data the fire seasonality parametrization has been improved (Cooke et al., 1996).
The Cooke et al. (1999) inventory is devoted to the production of BC particles by fossil fuel only. This inventory (column 5 in Figure Liousse 1) has new EF which are of the order of recent measurements (Bond et al., 1998). Also, improvement is due to the consideration of particulate size distribution and country development level. In 1996, Liousse et al. have developed an inventory including both BC and POM with a simple description for fossil fuel and an exhaustive consideration for biomass burning sources. Tropical savannah and forest, agricultural and domestic fires were taken into account. Recent experiments (Andreae et al., 1998; Reid et al., 1998a) have confirmed EF and BC/POM ratio choices of this study. Main uncertainties are linked to the estimations of tropical burnt areas. Improvements are ongoing using satellite data. The most recent inventory is complemented by the addition of boreal and temperate fires available for the years 1965-1995 but numerous assumptions are still necessary for Russian sources. In Liousse et al. (1996) natural SOA sources have been also considered with a simple scaling method where they represent 7.8 Tg POM per year. More recently, Kanakidou used a chemical modeling to derive SOA, and obtained a global budget of 6-15 Tg SOA per year (Kanakidou, 1998).
The best guess for anthropogenic carbonaceous particle sources may be found in inventories 4 and 6 for BB and 5 for FF(BC) and numbers 4 and 5 for FF(POM), (cf. Figure Liousse 1). The total annual amount of combustion aerosol emissions is comparable to that of anthropogenic sulphates. These inventories are available for a mean year representative of the 1980-1990 decade which means that there is an urgent need for updates. A slight decrease in Europe and an important increase in Asia would be expected in the budget.
Research needs concerning particulate organic matter
Important uncertainties still remain on crucial parameters and measurements are needed to understand better the emissions and characteristics of carbonaceous particles, to determine optical properties of BC and POM including information on size distribution, light back scattering properties or presence of external/internal mixture. Investigations are also necessary to describe better the hygroscopic properties of these particles. Finally, information on the vertical distribution of carbonaceous particles and their properties is crucial.
| Phase of activity | Emission S Tg/yr |
| Pre-eruption and intra-eruption | 5 ±2 |
| Post-eruption and extra-eruption | 5 ±2 |
| Explosive | 4 ±2 |
| Total | 14 ±2 |
| Volcano / Year | Emission S, Tg |
| Katmai | 5 |
| Agung | 3.0 |
| El Chichón (1982) | 3.5 |
| Pinatubo (1991) | 8-10 |
3.2.3. SO2 from fossil fuel combustion (Benkovitz)
For the oxides of sulfur (SOx = SO2 + primary sulfate) efforts to estimate total global emissions from anthropogenic sources were begun in the 1970s using global fossil fuel combustion statistics and globally averaged emission factors (emissions per unit activity, e.g., kg of sulfur per barrel of oil burned). Some past estimates of annual anthropogenic emissions of SOx for the same base year differed by as much as 50% and no estimates of the uncertainty associated with these emission values are available (Benkovitz et al., 1996).
Total global emissions estimates are of limited utility because a) the environmental issues associated with SOx are highly non uniform globally because of the short atmospheric residence times of the sulfur species, and b) the relationships of emissions to environmental effects are non-linear (Liu et al., 1987; Misra et al., 1989). Clearly, therefore, gridded global inventories are essential from both scientific and management perspectives. Allocation of emissions onto global grids has been constrained to use of a variety of country- based statistics despite recognition of complexities such as the comparability of statistics among different countries. These exercises have generally relied on population, location of industrial clusters, or other surrogates to distribute geographically the emissions. A number of countries (United States, Canada, Australia, Japan, South Africa) and regions (Western Europe, Eastern Europe, parts of Asia) have developed their own detailed emissions inventories using the same basic methodology (emissions = activity rate x emission factor x control technology x other parameters) by developing more accurate regional statistics and emission factors, spatial identification of large sources, and improved surrogate statistics for spatial apportionment of diffuse sources (Graedel et al., 1993); where available, these inventories are much more reliable than those generated using globally averaged emissions parameters.
In 1990 the Global Emissions Inventory Activity (GEIA) of the International Global Atmospheric Chemistry Program (IGAC) was initiated. The GEIA Working Group on Anthropogenic Emissions of SOx and NOx prepared a global inventory of SOx emissions for circa 1985 (Benkovitz et al., 1996). The compilation of this inventory started with a default global inventory which was based on national fuel statistics and global emission factors and gridded based on population statistics. These default emissions were replaced by emissions from regional inventories which, when evaluated, were considered to be more accurate. Thus emissions estimates in the resulting inventory were not of uniform quality, and in certain regions emissions from certain source categories, such as international shipping, biomass burning etc. were incomplete or not included. Still, the GEIA inventory incorporated the most up-to-date knowledge of anthropogenic sulfur emissions. Resulting emissions are presented in Plate Benkovitz 1; the data are available from http://www.info.ortech.on.ca/cgeic/.
3.2.4. Ammonia emissions and interactions with particles (Dentener and Metzger)
3.2.5. Oxides of nitrogen
3.2.6. Volatile organic compounds (Hoffmann and Seinfeld)
3.4 Aviation-produced particles (Kärcher) 3.5 Particle formation, particle characteristics 4.1. Nucleation (McMurry and Kulmala)
4.1.1. Laboratory studies
4.1.2. Field studies
4.2. Condensational secondary organic aerosol formation (Hoffmann and Seinfeld)
4.2.1. Theory
4.2.2. Summary of observed particulate SOA products
4.3. Dilute liquid phase reactions (Fuzzi and Herrmann)
Clouds constitute an efficient reaction medium for chemical transformations. Chemical constituents in the liquid phase of clouds derive from the incorporation of soluble species contained in the aerosol particles on which cloud droplet grow and the dissolution of trace gases in cloud water. The different species introduced in cloud droplets can eventually react in the liquid phase to form other products. 4.3.1. Liquid phase chemical reactions and cloud properties
4.4. Concentrated liquid phase reactions (Hermann)
4.4.1. Tropospheric physical Matrices
4.4.2. Thermodynamic background and implications of concentrated liquid phase chemical reactions
4.4.3. Kinetic Concepts and chemical conversions
4.5. Processes leading to internal aerosol mixtures (Raes)
In treating multicomponent aerosols, the issue arises whether the components are mixed within a single particle (internal mixing), or whether the various components are present as pure particles (external mixing). The manner in which different aerosol species are mixed in individual particles affects both optical and hygroscopic properties (Heintzenberg and Covert, 1990). The only processes depicted in Figure Intro 3 that lead to externally mixed aerosols are nucleation and primary emissions. The remainder of the processes lead to internal aerosol mixtures. The degree of "internal mixing", that is the degree to which the chemical composition of each individual particles resembles that of the average particle composition, will increase with the time available for interaction, hence with the residence time of the aerosol in the atmosphere. In urban and polluted continental conditions, the characteristic times of many of these interactions (cf. Table Raes 1) are short, and internal mixing will occur on a short time scale. However, these regions provide generally also sources for primary and nucleated secondary aerosol, enhancing external mixing. Field observations have indeed shown that in continental areas, two distinct aerosol types with different hygroscopic properties are often present, a direct indication of externally mixed aerosol (Svenningsson et al., 1994; Zhang et al., 1993). In an aged polluted air masses, such as found in continental plumes over the ocean, this feature is not observed, suggesting a chemically more homogeneous (sub-micrometer) aerosol (Putaud et al., 2000), in agreement with the long residence, hence interaction times.
4.5.1. Sulphur oxidation in sea salt (O'Dowd and Sievering)
5.1. Size distribution
The general features of the size distribution of the atmospheric aerosol are well-established since the pioneering work of K. Whitby and his group (Whitby, 1978). During the past ten years improved electrical and aerodynamic techniques have provided more detailed and accurate data down to three nanometer and up to 10 micrometer. Consequently, now 4 - 5 submicrometer modes are frequently being measured (Covert et al., 1996b; Yuskiewicz et al., 1998) and one or two supermicrometer modes (Heintzenberg et al., 1998). Instead of giving a comprehensive review of size distribution data this section focuses on a few issues that are relevant to atmospheric chemistry.
Despite its importance for particle formation and climate, relatively little effort has been spent on the understanding of sources and removal processes of NH3 . Most work on atmospheric ammonia has been performed with respect to eutrophication and acidification close to these sources; large scale transport and chemistry of NH3 and NH4+ received much less attention. Emissions of NH3 are associated with animal waste, fertilizers, biomass combustion, soils and vegetation and some other minor sources (Bouwman et al., 1997). The microbial breakdown of urea and uric acid present in animal waste produces ammonium, which subsequently partly volatilizes as NH3 . The overall emission of NH3 from waste is dependent on the specific N-excretion per animal, and the NH3 loss during housing, storage of waste outside the stable, grazing and application of manure on grassland or arable land. Further important properties influencing NH3 volatilisation involve soil pH and moisture, and temperature.
Some Northern European countries have measured and calculated country- and animal-specific emission factors. The applications of such emission factors to calculate animal related emissions elsewhere, may be quite problematic, since agricultural practice and climatic factors may differ substantially from those in Northern Europe. In addition, the number of animals per country may fluctuate strongly from year-to-year. The most recent estimate for global NH3 emissions (Bouwman et al., 1997) from animals relied on constant emission factors and amounted to 21.7 Tg N y-1 , which is of similar magnitude as fossil fuel related global NOx -N emissions. The second most important emission category is N- containing synthetic fertiliser. Again, huge differences in agricultural practice and environmental conditions, cause a large variation of emissions factors. Overall global emission of ammonia derived from nitrogen fertiliser was estimated to 9 Tg y-1 , which is 10 % of the amount applied. Interestingly, ammonia losses from application of urea fertiliser to rice paddies seem to contribute strongly to this. Other anthropogenic sources, such as biomass burning, cropland and humans additionally emit about Tg. Natural sources such as soils, vegetation and oceans, emit about 10-20 Tg (Bouwman et al., 1997; Schlesinger and Hartley, 1992) and are highly uncertain. The global source strength of ammonia is about 55 Tg N y-1 , which is of similar magnitude as global NOx -N emissions.
Dry deposition and reaction with acidic particles and particle precursor gases are the main removal mechanisms for NH3 . Oxidation chemistry of NH3 was evaluated by Dentener and Crutzen to play a relatively minor role (Dentener and Crutzen, 1994). Because ammonia emissions occur almost exclusively close to the earth's surface, and plants utilise nitrogen in their metabolism, dry deposition is a very efficient process which may remove 40-60 % of all emissions. Ammonia which has reacted with sulfuric or nitric acid to form ammonium is much less efficiently removed by dry deposition and is removed mainly by wet deposition with a average residence time in the atmosphere of up to one week, compared to the much shorter residence time of gas phase ammonia, which is less than one day.
The understanding of the atmospheric ammonia cycle is still limited, because:
Despite all uncertainties involved, several studies (e.g., Galloway et al., 1995) have indicated the significance of ammonia for the global nitrogen cycle. Recent modeling studies (Adams et al., 1999; Metzger et al., 1999) indicated the potential significance of ammonium and nitrate for aerosol burden and composition. These studies were performed using thermodynamic equilibrium models, developed originally for urban smog conditions. Substantial effort has been spent on extending these schemes to global modeling (e.g., Nenes et al., 1998).
The global modeling results indicate that substantial amounts of nitrate may contribute to the particulate mass in polluted regions such as central Europe and the east-coast of North America, especially if one accounts for recent concentration changes in the particle precursor gases as mentioned above. In contrast to sulfate, nitrate particles are semi-volatile and their formation stingily depends, besides the presence of other particles and particle precursor gases, on temperature and relative humidity. Therefore ammonium nitrate exhibits a pronounced diurnal and seasonal cycle with maxima at night and winter, in contrast to sulfate which is preferentially produced at high solar radiation conditions. Nitrate-containing particles have a considerably lower deliquescence relative humidity than ammonium sulfate particles resulting in an enhanced uptake of water by the aerosol particles at lower relative humidities. Model results indicate an significant enhancement of the particle volume and scattering cross section, due to the particle-associated water. Moreover, aqueous particles may allow multi-phase chemical processes that do not occur or occur more slowly on their dry counterparts.
A major challenge is presented by increasing the resolution of the models and developing sub-grid parameterisations that represent the variability of ammonia emissions and the resulting effect on particle composition. Long term, representative and reliable measurements of ammonium, sulfate and nitrate are needed in conjunction with deposition measurements to close further the ammonium budget.
These important natural and anthropogenic aerosol precursors are discussed in chapter X.
Reactive organic gases (ROGs) are emitted into the troposphere from anthropogenic and biogenic sources. Anthropogenic sources comprise organics such as alkanes, alkenes, aromatics, and carbonyls, whereas biogenic sources include organics such as isoprene, mono- and sesquiterpenes as well as a series of oxygen-containing compounds. Because the volatility of the oxidation products is one of the most important parameters that determines the significance of precursor gases in terms of their particle forming potential, only hydrocarbons with more than six carbon atoms are considered to contribute to secondary organic particles under atmospheric conditions (Seinfeld and Pandis, 1998). Thus, natural SOA formation is believed to result mainly from the oxidation of C10 - and C15 -terpenoids. Based on laboratory data of biogenic ROG particle yields and emission inventories, (Andreae and Crutzen, 1997) estimated 30 to 270 Tg year-1 for the production of secondary organic particles from natural ROGs. Because the estimated global emissions of anthropogenic ROGs are significantly lower, their contribution to organic particles on a global scale would appear to be small when compared with natural terrestrial sources. However in the urban atmosphere during severe smog episodes, anthropogenic SOA formation can make a major contribution to fine particulate mass. Among the anthropogenic precursor gases, aromatics have been identified to dominate the process of SOA formation; this suggests that anthropogenic SOA formation in an urban air shed can be modeled based on the aromatic content of the complex hydrocarbon mixture (Odum et al., 1997a).
Two major particle types are produced by jet aircraft engines: (1) nanoparticles from nucleation during cooling and dilution of the exhaust, mainly composed of water and sulfuric acid, but also organics and possibly and other chemical species (Curtius et al., 1998; Kärcher et al., 1998a; Schröder et al., 1998; Yu et al., 1999a); (2) accumulation mode soot particles (number mean diameter range 20-60 nm), composed of carbonaceous agglomerates formed during fuel combustion containing volatile sulfur and organic compounds (Kärcher et al., 1996; Petzold et al., 1999; Petzold and Schröder, 1998).
As demonstrated by observations and models, charged molecular clusters produced via high temperature chemical reactions in the combustion chamber (chemi-ions) play a central role in the formation and evolution of aviation- produced liquid particles (Curtius et al., 1998; Kärcher et al., 1998a; Kärcher et al., 1998b; Yu and Turco, 1997). The particles grow in size by coagulation and condensation processes, forming internal soot-H2SO4 /H2O mixtures.
Observations and models demonstrate that aircraft produce of the order 1017 sulfate particles per kg of fuel consumed (Penner et al., 1999). In contrast, contemporary engines emit 1014 - 1015 soot particles per kg fuel (Penner et al., 1999). Recent in situ observations support earlier conjectures that modern engines emit less (but also smaller) soot particles than old-technology engines, yielding a fleet-averaged emission index of about 0.04 g soot/kg fuel (Petzold et al., 1999).
Sulfur injections of 0.06 Tg/yr (1992 figure) into the upper troposphere and lower stratosphere by the current aircraft fleet are about 3-5 times less than natural S-sources (SO2 , DMS, OCS) in non-volcanic periods (Penner et al., 1999). However, the contribution of aircraft-produced sulfate particles to the aerosol in the tropopause region is larger than suggested by a comparison of the sulfur mass because the impact on cloud formation may partly be controlled by an increase in the number of aerosol particles acting as condensation or ice nuclei.
Aircraft are estimated to emit 0.006 Tg C (1992 figure) as soot into the atmosphere (Penner et al., 1999). Observed soot masses range between 0.1-2 ng m-3 at 10 km altitude (Blake and Kato, 1995; Pueschel et al., 1992; Pueschel et al., 1997), Measurements of absorbing material (presumably soot) in cirrus clouds at 8-12 km showed higher values (10 ng m-3 ), correlating with local aircraft fuel consumption (Ström and Ohlsson, 1998b).
The aircraft fleet may increase cirrus cloudiness (Boucher, 1999). Observations from space (Mannstein et al., 1999; Minnis et al., 1998) and from in situ measurements (Schröder et al., 1999) demonstrate that persistent contrails may develop into cirrus clouds. Aircraft-produced particles may also trigger cirrus indirectly, that is, without contrails or after the disappearance of short-lived contrails. The potential for heterogeneous nuclei to cause ice formation at relative humidities that are lower than those needed to freeze ice homogeneously in sulfate particles raised concerns about the role of aircraft soot in modifying existing or nucleating new cirrus clouds (Jensen and Toon, 1997; Kärcher et al., 1998a).
Sulfate particles formed in aircraft exhaust are likely too small to contribute to cirrus formation in the atmosphere. Larger particles, such as soot, could be more efficient ice nuclei (IN). For aircraft-generated soot to influence cirrus properties, the soot particles would need to be more efficient or more abundant than other atmospheric IN. Concentrations of upper tropospheric IN are thought to be small (a few per liter), but recent observations show that this is not always the case (Rogers et al., 1998). Observational evidence exists for soot influencing cirrus in terms of enhanced ice crystal number densities in cloud regions perturbed by aircraft exhaust (Ström and Ohlsson, 1998a), but the measurements do not reveal the related physical mechanisms. Changes in the radiative forcing due to aviation-induced changes in cirrus cover, ice crystal sizes, and number densities could be significant (Meerkötter et al., 1999; Ponater et al., 1996; Wyser and Ström, 1998).
Whereas the direct radiative forcings associated with aircraft-produced sulfate and soot particles are estimated to be small compared with those originating from contrails or greenhouse gases, the indirect impact of these particles on cloud formation and modification - and its associated radiative forcing - could be substantial (Penner et al., 1999). A better understanding of the physico-chemical surface properties and the ice forming ability of soot particles is required before the impact of aircraft soot on cloud formation and heterogeneous chemistry in the tropopause region can be fully addressed.
4. Transformations
Homogeneous nucleation has been investigated in the laboratory for several inorganic and organic chemical systems pertinent to the atmosphere. The inorganic system that has received most attention is the binary H2SO4 -H2O system. A limited work amount of work on ion-induced nucleation has been reported. Nucleation from a variety of gas phase organic precursors has been studied in smog chambers.
Species that nucleate in the atmosphere often form new particles at exceedingly low (sub-ppt) concentrations of the precursor gases. Because these nucleating species are present at such low concentrations they are difficult to measure. Furthermore, many species that nucleate are "sticky" and are lost rapidly to surfaces. For example, undetectable NH3 may affect measurements of binary H2SO4 -H2O nucleation rates in laboratory studies.
Inorganic systems:
The prediction of binary nucleation theory that small quantities of H2SO4 would significantly reduce the water vapor supersaturation required for homogeneous nucleation was first confirmed in expansion cloud chamber experiments of Reiss et al. (Reiss et al., 1976). Shortly thereafter, based on measurements in flow reactors with 1010 < [H2SO4 ] < 1011 cm-3 , reasonable agreement was found, (Boulaud et al., 1977), between the number of particles produced in mixtures of H2SO4 and H2O and theoretical predictions (Mirabel and Katz, 1974). Good agreement was reported between diffusion cloud chamber theory and measurements done at 10-3 < H2SO4 Relative Acidity < 10-8 and relative humidities exceeding 100% if the formation of sulfuric acid hydrates in the vapor phase was taken into account (Mirabel and Clavelin, 1978). (Relative acidity of sulfuric acid is the ratio of the actual partial pressure of the acid to the equilibrium vapor pressure above a flat surface of neat acid at the same temperature.) Based on measurements in a continuous flow reactor it was concluded that discrepancies between measured and theoretical nucleation rates were correlated with the number of acid molecules in the critical cluster (Wyslouzil et al., 1991). Their results were found to be roughly an order of magnitude higher than the results of Mirabel and Clavelin upon extrapolation to the high RHs used in the latter study. In experiments at 25 ŗC in a steady flow reactor with 38% < RH < 52% and with 1x1010 < [H2SO4 ] < 3x1010 molecules 25 cm-3 reasonable agreement with previous work and with the binary theory was reported when hydrates were included (Viisanen and Kulmala, 1997). These laboratory studies have all been done at sulfuric acid vapor concentrations that are much higher than values measured in the atmosphere which have seldom exceed 10 8 molecules cm-3 , even when nucleation is occurring (Clarke et al., 1998a; Weber et al., 1996). These low values may be explained in part by the lower temperatures encountered during many of the atmospheric observations and by the possible participation of other species in atmospheric nucleation processes.
Several other laboratory studies of multicomponent nucleation significant to the atmosphere have been reported e.g., on measurements of binary nucleation in the methanesulfonic acid (MSA)/water system. There it was found that nucleation occurred at partial pressures below those required for either pure species (Kreidenweis et al., 1989). It is unlikely, however, that atmospheric concentrations of MSA are high enough for binary MSA/H2O nucleation to play an important role although ternary nucleation involving H2SO4 /MSA/H2O might be important (Kreidenweis and Seinfeld, 1988). Recent experiments have shown that multicomponent nucleation rates are even enhanced for substances that are immiscible in bulk solutions (Strey and Viisanen, 1995).
Particle formation from dimethylsulfide and dimethyldisulfide precursors was investigated (Kreidenweis et al., 1991) and (Raes et al., 1992) reported similar work on particle formation from SO2 . In both cases measured size distributions were compared with results obtained by solving a detailed model for chemical transformations and particle nucleation and growth. Uncertainties in rate constants and other parameters precluded definitive comparisons between binary nucleation theory and experiment. Such studies offer the advantage that nucleation is driven by chemical transformations, much as occurs in the atmosphere.
Studies of particle production by ionizing radiation have also been reported (Bricard et al., 1972; He and Hopke, 1995; Kim et al., 1998; Mäkelä, 1992; Raes and Janssens, 1985; Raes et al., 1985; Yu and Turco, 1997; Yu and Turco, 1999). These studies have shown that small ions enhance nucleation rates in systems that are chemically similar to the atmosphere and that NH3 enhances nucleation rates when SO2 is irradiated in air containing H2O. Understanding of clustering on small ions is not yet adequate, however, to permit quantitative estimates of the importance of ion-induced nucleation in the atmosphere.
Nucleation of atmospheric organic systems:
Data from smog chamber studies of the formation of organic aerosols have been analyzed to make inferences about homogeneous nucleation (Pandis et al., 1991; Stern et al., 1987; Wang et al., 1992a). However, organic precursors typically produce multiple particulate products, and the molecular identity and physical and chemical properties of these products is often unknown. Therefore, although these studies may provide insights into nucleation tendencies for various organic gas phase precursors, extrapolation to the atmosphere would involve large uncertainties.
The formation of new particles by nucleation often occurs in marine and continental atmospheres. These nucleation events sometimes follow regular diurnal patterns whereas at other times they occur in response to an atmospheric perturbation such as cloud processing or source emissions.
Discoveries about the formation of new atmospheric particles by homogeneous nucleation have accelerated in the past decade with the development of instruments for detecting freshly nucleated particles and their gas phase precursors (cf. section 7).
Nucleation is often observed during daylight hours in the vicinity of convective clouds in continental (Radke and Hobbs, 1991) and marine environments (Clarke et al., 1998b; Hegg et al., 1990; Martinsson et al., 1999; Perry and Hobbs, 1994). A definitive explanation for this phenomenon is not yet established. It is known that nucleation tends to occur in air masses where the sulfuric acid relative acidity is elevated (Clarke et al., 1998b; Weber et al., 1999). It has been postulated that nucleation occurs because clouds deplete the particle surface area by coalescence or precipitation while they simultaneously pump precursor gases from the ocean's surface to outflow regions with high actinic flux. Nucleation then occurs following the rapid photochemical production of condensable products that may include H2SO4 produced by photo oxidation of dimethylsulfide (Clarke et al., 1998b; Perry and Hobbs, 1994). Atmospheric experiments (Weber et al., 1999) yielded that nucleation in cloud outflows often occurs at H2SO4 relative acidities consistent with those predicted by the classical binary theory for H2SO4 -H2O nucleation (Jaecker-Voirol and Mirabel, 1989; Kulmala et al., 1998a; Wilemski, 1984). Observations suggest that nucleation in the upper tropical troposphere is a significant global source of atmospheric particles (Brock et al., 1995; Clarke, 1993; Clarke et al., 1998b). Evidence has also been reported for nucleation in the midlatitude upper free troposphere (Clarke, 1993; Schröder and Ström, 1997).
Evidence of nucleation has also been observed in the marine boundary layer (Covert et al., 1992; Hoppel et al., 1994a). Clarke et al. (1998a) reported on a well documented MBL nucleation event that was linked to dimethylsulfide emissions. However, other field measurement campaigns have provided little or no evidence for significant nucleation in the marine boundary layer (Bates et al., 1989; Clarke et al., 1998b; Covert et al., 1996b; Raes et al., 1997). Several groups (Covert et al., 1996a; Wiedensohler et al., 1996) have argued that nucleation mode particles detected in the MBL are probably produced aloft in FT cloud outflows and transported to the surface.
Nucleation has also been observed at ground level continental sites (Birmili, 1998; Bradbury and Meuron, 1938; Hogan, 1968; Hörrak et al., 1998; Koutsenogii and Jaenicke, 1994; Kulmala et al., 1998b; Mäkelä et al., 1997; Pirjola et al., 1998; Went, 1966). These events typically follow regular diurnal patterns, with concentrations of nucleation mode particles increasing several hours after sunrise and reaching peak concentrations in the range 104 cm-3 or higher after several hours. The chemical mechanisms of nucleation in these regions have not yet been established. It has been postulated that organics or small ions may play a role, and that nucleation may occur above the boundary layer and be detected at ground level after atmospheric mixing occurs.
Several groups have observed nucleation in regions impacted by local sources. Nucleation events have been observed at the coast in Mace Head, Ireland, during on-shore flow (Grenfell et al., 1999; McGovern et al., 1996; O'Dowd et al., 1999a; O'Dowd et al., 1998). These events only occur in sunlight during low tide. Particles are detected within seconds to minutes after air flows over the beach, and particle concentrations as high as 40,000 cm-3 and estimated particle production rates as high as 10,000 cm-3 s-1 have been reported. Particle growth rates at this site must be several orders of magnitude higher than values reported for aerosols in the remote marine or continental environments, which are typically of order 1-2 nm h-1 (Kulmala et al., 1998b; Weber et al., 1997; Weber et al., 1998a). Nucleation events were detected downwind of penguin colonies on Macquarie Island (Weber et al., 1998a). They postulated that as clean marine air containing H2SO4 flowed over the island it entrained NH3 or other biogenic compounds that participated in particle production. Nucleation at this site occurred at H2SO4 activities much lower than can be explained by the binary H2SO4 -H2O theory.
Nucleation has also been observed on mountains (Marti, 1990; Raes et al., 1997; Shaw, 1989; Weber et al., 1996; Weber et al., 1997; Wiedensohler et al., 1997). Weber et al. found that nucleation at Mauna Loa, HI, and Idaho Hill, CO, follows a regular diurnal pattern. Rates of new particle formation ( nucleation rates) were found to be orders of magnitude higher than can be explained by the H2SO4 -H2O theory and varied as ~[H2SO4 ]2 rather than ~[H2SO4 ]10 , as predicted by the binary theory. Raes et al. observed freshly nucleated particles during daytime at high elevation but not at sea level on Tenerife, and attributed this to photo induced nucleation in upslope air that was impacted by local biogenic and anthropogenic emissions (Raes et al., 1997). The observations of Wiedensohler et al. are intriguing as they document the occurrence of freshly nucleated particles downwind of an orographic cloud on two occasions at night (Wiedensohler et al., 1997). Nucleation models including several chemical systems (H2SO4 -H2O, HCl-H2O, HCl-H2O-NH3 ) were unable to explain their observations.
Research needs
The binary H2SO4 -H2O theory appears to be consistent with models and observations for nucleation in some locations and inconsistent in others. Nucleation that occurs near the surface typically occurs at rates greatly exceeding those predicted by the binary theory. There is a need for laboratory and field studies to identify nucleation precursors (including small ions) and to establish validated models of atmospheric nucleation.
Secondary organic particulate matter is formed in the atmosphere by the mass transfer of low vapor pressure substances to the atmospheric condensed phase. The condensable species are formed in the gas phase by the reaction of organic gases with the principal atmospheric oxidizing agents - O3 , OH and NO3 . This fraction is termed secondary organic aerosol (SOA).
The basic picture of the chemical processes leading to secondary organic particles is that a series of condensable products (P '1 , P '2 , ...) with different stoichiometric yields (a '1 , a '2 , ...) are formed in the gas-phase reaction of a parent ROG with an atmospheric oxidant
where kOH , kO3 and kNO3 are the rate constants for the individual oxidation reactions. The products formed in this first oxidation step may subsequently undergo gas-phase reaction themselves, creating a second generation of condensable species (P ''1 , P ''2 , ...).
Over the last decade several groups have investigated the formation of secondary organic particles from individual hydrocarbons by carrying out smog chamber experiments (Forstner et al., 1997a; Forstner et al., 1997b; Griffin et al., 1999; Hoffmann et al., 1997; Izumi and Fukuyama, 1990; Odum et al., 1996; Odum et al., 1997a; Pandis et al., 1991; Wang et al., 1992a; Wang et al., 1992b; Zhang et al., 1992). Typically, the hydrocarbon of interest is injected into a large volume reaction chambers and the evolution of the particle size distribution is measured simultaneously with the degradation of the parent ROG. The SOA yield (Y) can then be defined as the fraction of a hydrocarbon that is converted to particulate matter. Initially it was believed that each ROG should possess a unique value of its SOA yield (Grosjean and Seinfeld, 1989; Pandis et al., 1992; Pandis et al., 1993), but measured yields for an individual ROG exhibited a degree of variation that could not be reconciled in terms of a single, unique SOA yield for each parent ROG. Following Pankow (1994a; 1994b), Odum et al. (1996) formulated a framework for explaining observed SOA yield data. A comprehensive derivation of the gas/particle partitioning model can be found in Seinfeld and Pandis (1998).
In principle, the relative particle-forming potential of a group of organics could be determined based on their oxidation products and the thermodynamic properties of these products. This ab initio approach represents a goal that is not yet attainable because of incomplete knowledge of the semi-volatile oxidation products of the important particle-forming compounds. Thus, it is necessary to rely on experimentally measured particle yields which over the last several years have been measured for over 30 aromatic and biogenic organics in the California Institute of Technology outdoor smog chamber.
The stoichiometric and partitioning parameters allow direct evaluation of the particle-forming potential of the parent organics. SOA potentials are given by Odum et al. (Odum et al., 1996; Odum et al., 1997b) for 17 aromatic precursors and by Griffin et al. (Griffin et al., 1999) for 14 biogenic precursors. Odum et al. (Odum et al., 1996; Odum et al., 1997b) showed, moreover, that particle formation from the photo oxidation of a mixture of parent hydrocarbons can be predicted simply from the SOA yields for the individual parent compounds. This suggests that, at least for the case of a pure organic absorbing phase, oxidation products of different parent hydrocarbons are as soluble in a mixed organic product phase as in an organic phase consisting exclusively of their own oxidation products.
As a result of a series of field studies carried out in the last few years, it became evident that secondary organic particle formation, as described above, is believed to account for a significant fraction of the atmospheric aerosol (Andreae and Crutzen, 1997; Rodhe et al., 1998). However, since little is known about the chemical composition of the organic fraction of atmospheric particles, the necessary determination of the relative contributions of natural and anthropogenic sources remains uncertain. This task is especially difficult since primary as well as secondary organic material can contribute to the organic fraction of tropospheric particles. Only recently have a number of laboratories systematically identified components of secondary organic particles.
In the case of biogenic ROGs, gas-phase ozonolysis has been especially investigated in detail, and various new products have been identified, such as multifunctional carboxylic acids (e.g., C8 and C9 -diacids or hydroxy-oxo- carboxylic acids) (Christoffersen et al., 1998; Glasius et al., 1999; Hoffmann et al., 1998; Jang and Kamens, 1999; Yu et al., 1999b). The low vapor pressure of these products dictate that a substantial fraction will be found in the particle phase.
Closure of the carbon mass balance is not yet possible, however, for all relevant biogenic ROGs. This is especially true since less is known about product formation of the two other relevant oxidation pathways under tropospheric conditions, namely OH and NO3 initiated ROG degradation. The significance of the two oxidants in terms of particle formation in the ambient atmosphere is not sufficiently known yet.
In general, the available information indicates that the reaction of biogenics with NO3 radicals represent efficient routes to the formation of condensable products (Griffin et al., 1999; Hoffmann et al., 1997). Although some degree of grouping of biogenic hydrocarbons based on structural characteristics in terms of their particle forming potentials is possible, it seems still necessary to account individually for most of the biogenic ROGs when modeling secondary organic particle formation. While smog chamber studies on the a-pinene oxidation indicate that the OH reaction is a less effective route to the formation of SOA (Hoffmann et al., 1997; Wirtz et al., 1998), other biogenic ROGs appear to possess a very high particle formation potential after the attack of OH to a double bond (Griffin et al., 1999). Clearly, more data are needed especially on the particle forming potential of the OH initiated oxidation of biogenic ROGs under low NOx conditions, in particular to assess the production of secondary organic particles in remote areas.
Anthropogenic SOA precursors include organics such as alkanes, alkenes, aromatics and carbonyls. Atmospheric photo oxidation of aromatics has been identified to be particularly important in the formation of secondary organic particles in the urban atmosphere. Currently only about 15-30% of the organic particulate mass can be assigned to specific compounds, mainly saturated and unsaturated anhydrides (Forstner et al., 1997a; Neusüß et al., 2000; Odum et al., 1997b).
Research needs
The processing of chemical compounds by liquid water droplets influences the chemical composition of the troposphere (e.g. Lelieveld and Crutzen, 1991). A proper description of liquid phase chemical processes is therefore necessary to assess the role of clouds in a changing atmosphere. Understanding the chemistry of clouds requires a detailed knowledge of the factors controlling the chemical composition and the concentration of chemical species in droplets.
Two families of chemical species are key participants in the cloud liquid phase chemical reactions: sulphur species and organic compounds. The chemistry of oxidised (NOx) and reduced (NH3) nitrogen species is of lesser importance, although gas phase-derived HNO3 and NH3 are key compounds in determining pH of cloud droplets and thus affecting the overall liquid water chemistry.
S(IV) oxidation reactions occur in clouds at a much faster rate than in the clear air: model calculations Langner and Rodhe (1991) have shown that, on a global scale, tropospheric in-cloud SO2 oxidation is from two to five times more important than out-of-cloud oxidation. Laboratory studies have identified the reactions responsible for S(IV) oxidation in the atmospheric liquid phase and in determining the associated rate coefficients (Warneck, 1996).
Aqueous chemical conversions are driven not only by classical oxidants such as hydrogen peroxide, organic peroxides or ozone but also by free radicals. Radicals may originate from the gas phase as for OH, HO2 , NO3 , CH3O2 and others (Davidovits et al., 1995). These species are referred to as primary radicals, as opposed to secondary radicals which are formed within aqueous atmospheric particles (SOx - (x = 3,4,5), Cl/Cl2 - , Br/ Br2- , CO3- ).
Radicals exist in solution in very small concentration which can currently be assessed only from models. However, they exhibit a very high reactivity towards inorganic and organic constituents of cloud droplets. Two more recent treatments summarise the state of knowledge as of 1995 (Huie, 1995; Zellner and Herrmann, 1995).
Very little is known on aqueous phase chemistry of organic species. In current box models the description of organic chemistry in cloud water is limited to gas phase-derived C1 and C2 compound (Herrmann et al., 1999). Recent work (Facchini et al., 1999a; Saxena and Hildemann, 1996; Zappoli et al., 1999) has, however, shown that CCN contain a high percentage of soluble organic species. Very little however is still known of the nature, physico- chemical properties and liquid phase chemical reactions of these species.
A further complication of the picture concerning cloud water chemical reactions derives from the size-dependence of cloud water chemical composition which was stressed by Ogren and Charlson (1992). Model results (e.g., Hegg and Larson, 1990; Pandis et al., 1990) have also shown the importance of an investigation of size-dependent cloud droplet chemistry demonstrating that bulk cloud water parameters are not applicable to processes taking place within individual cloud droplets. There is a general lack of experimental data on size- segregated cloud chemical composition to validate cloud model outputs (Collett Jr. et al., 1994; Laj et al., 1997; Ogren et al., 1992).
Interaction between the different phases must be considered when discussing chemical processes in clouds. Modelling results, for example, have indicated strong effects of the presence of clouds on ozone formation/destruction processes (Lelieveld and Crutzen, 1991; Mölders et al., 1994).
Upon cloud evaporation, gas and particles are released back to the atmosphere. The particles resulting from the evaporation of cloud droplets is likely to be quite different (in physical and chemical properties) from that which entered the cloud, because of in-cloud processes. Both modelling and experimental results have shown that sulfate particle concentrations before and after passage through a cloud differs significantly, with much larger concentrations and modified size-distribution in the outflow of the cloud system (Bower et al., 1997; Laj et al., 1997; Wiedensohler et al., 1997). The effects of multiple cloud passages on CCN have important implications for the direct aerosol radiative forcing by increasing the efficiency of light scattering due to in- cloud particle growth (Yuskiewicz et al., 1999).
In addition, physico-chemical properties of cloud droplets may also be altered because of the in situ chemical generation of products altering water surface tension and vapour pressure which corresponds to water activity (Shulman et al., 1996). Also, this effect has important implications for indirect aerosol radiative forcing by affecting the drop size distribution (Facchini et al., 28 1999b).
Research needs
Current research is centered on the interaction of radicals with organics which may also heavily influence the rate of S(IV) oxidation in droplets (Warneck, 1996), alter the pattern of precipitation composition, and may also lead to the production of harmful substances (Lüttke et al., 1997). The radical- driven oxidation of S(IV) to S(VI), after a great deal of research, constitutes an extensive research topic of its own. Attention needs to be given to understand better occurrence and liquid phase chemistry of organic species via laboratory studies, focused field campaigns and modelling efforts. Current efforts to measure and model size-dependent cloud chemical composition should also be improved.
Tropospheric aerosol particles are exposed to ubiquitous water vapor. The water content of particles significantly influences the physico-chemical properties of these particles. Particles containing water-soluble substances are subject to growth under conditions of high humidity as a consequence of uptake of water vapor that is driven by the chemical affinity between water and the soluble materials. Even at low humidity particles generally contain liquid water, and probably even a particle considered 'dry' is covered by a water film consisting of several layers of water molecules. This picture directly raises the question whether chemical processes at water coated particles should be treated as `heterogeneous' or 'multiphase' processes (see discussion by Ravishankara, 1997).
The physical matrix of highly concentrated or even supersaturated aqueous electrolyte particles suspended in air may host a variety of chemical species which may be involved in solution-phase reactions. The sum of (i) chemical reactions currently either already implemented in special (mostly marine cases) aerosol models (see for mechanisms e.g., Sander and Crutzen, 1996; Vogt et al., 1996) plus (ii) the more complex variety of reactions implemented in cloud chemistry models (see Herrmann et al., 2000 and references therein) and, finally, (iii) such reactions which may be important in aqueous tropospheric particles but have not been implemented in any model up to now, have to be considered to occur also as concentrated liquid phase chemical aerosol reactions. Of class (iii) reactions of organic particle constituents may be of special importance but up to now have been implemented only for C1 - and C2 - organics (Herrmann et al., 2000) in multiphase mechanisms.
In subsaturated air in the troposphere the following systems may be differentiated, in part on the basis of relative humidity or water activity. Continental aerosol and haze. Principal anion species are sulfate and nitrate; principal cation species are ammonium and hydrogen ion. Ionic strengths of particles in urban polluted areas of 8 to 19 M have been identified (Stelson and Seinfeld, 1981). Chemical conversions in liquid sulfuric acid particles have been treated with special attention to stratospheric heterogeneous chemistry (Hanson et al., 1994). Typical electrolyte concentrations in haze particles are about 1 M, corresponding to a relative humidity of ~95% or higher. Marine aerosol. Electrolyte concentrations are approximately 5 M, corresponding to relative humidity of about 80% (Sander and Crutzen, 1996). Chemical conversions within concentrated liquid phase particles of marine (sea salt) aerosols have been treated recently with regard to (i) halogen atom activation and (ii) the destruction of Arctic ozone resulting from release of bromine to the atmosphere as a consequence of aqueous phase aerosol chemical reactions
Dust particles. Chemical conversions on or in dust particles are largely unassessed. Electrolyte contents may reach 15 to 20 M.
The measure of the deviation from ideal behaviour for a solute-containing aqueous solution is expressed as the solutes activity coefficient. Data on activity coefficients are available from extensive tabulations (e.g., Goldberg, 1981; Hamer and Wu, 1972; Pethybridge and Prue, 1972 and references therein) or may be calculated according to the theoretical apparatus developed by Pitzer and his co-workers (Pitzer, 1991; Pitzer, 1995). For some basic inorganic aerosol systems a thermodynamic analysis (Carslaw et al., 1995; Clegg et al., 1998a; Clegg et al., 1998b) can be performed via the world-wide-web making use of a program maintained by S. Clegg ( http://www.uea.ac.uk/~e770/aim.html) .
A large variety of aerosol studies exist where particle thermodynamics are treated by the concept of deriving activity coefficients for the aqueous particle inorganic solutes. These studies are successful in describing particle thermodynamics. The chemistry involved here, however, is limited to the physical presence of inorganic solutes and their dissociation equilibria as the only chemical reactions involved and their influence on water activity, which, at a more basic physico-chemical interpretation level corresponds to the disturbance of the original water structure by dissolved ions.
Many more chemical reactions are expected to occur in concentrated electrolyte tropospheric aqueous particles. Further development of the activity approach demands the determination of an applicable activity coefficient to be carried with each reactant's concentration in the implemented chemical reaction scheme. Complications arise from the fact that, within tropospheric particles, activity coefficients are not restricted to a purely inorganic system, but organics will also dissolve into the high electrolyte concentration aqueous solution. Such organics may either dissolve in their original form, or alter their molecular form by dissociation (organic acids, diacids, hydroxy-acids) or hydration (aldehydes). The dissolved organics will also influence water activity so that activity coefficients derived for a pure inorganic system cannot necessarily be regarded valid in a solution containing organic species.
A chemical reaction for an electrolyte solution with given composition, acidity, and temperature could be described correctly when all activity coefficients of educts and products as well as the rate constant for the reaction under the give conditions are known. As outlined in section 2 of this contribution, problems will arise in the determination of the needed activity coefficients. This section will outline possible treatments for deriving the needed rate coefficients for a reaction under high electrolyte concentrations as found in tropospheric aerosol particles.
The usual Debye-Hückel description of the so-called primary kinetic salt effect and its most wide-spread extensions are only applicable up to ionic strengths of about 0.5 M. It should be noted that an extension of the Debye- Hückel theory up to ionic strengths of several mol/l has been suggested by Glueckauf (1969). Extended Debye-Hückel treatments for the rate constants of ion-ion-reactions have successfully been applied for accounting for primary kinetic strength effects in cloud chemistry models (see, e.g., Jacob, 1986). For clarity, primary kinetic strength effects in ion-ion reactions may be addressed as 'type 1'.
In reactions involving neutrals only, or one ion and a neutral, a primary kinetic salt effect is clearly identified (Debye and McAulay, 1925; Herrmann and Zellner, 1998) but this can currently not be treated other than by examining each reaction in each matrix separately. It is suggested to address such kinetic salt effects involving neutral molecules as primary kinetic salt effects (type 2). No cloud- or aerosol chemistry model is known where these effects are explicitly considered.
Equilibrium constants are affected by ionic strength (secondary kinetic salt effect) as they represent the ratio of two rate constants which are both subject to the above primary kinetic salt effects (Harned and Owen, 1958) and Robinson and Stokes (1959). These effects have been considered in cloud models in the same manner as the primary kinetic salt effect for ion-ion-reactions (Jacob, 1986).
The most drastic approach used in simple tropospheric aerosol models is to adjust only the ionic concentration to the matrix studied and to a first approximation just neglect the changing activities of all species in a high concentration electrolyte. This approach may be refined in that for stable, ionic species for which the necessary activity coefficients are available (e.g., after Pitzer) these will be used in the mechanism, whereas, for other species like neutrals and radicals no activity corrections are performed. This approach is much used in current marine aerosol models (e.g., Sander and Crutzen, 1996).
As even the last-mentioned approach cannot be found satisfactory, the need arises to find ways to describe the rate constant for reactions (i) between two ions, (ii) a neutral species and one ion and (iii) between two neutral species under high ionic strength conditions. As current treatments will not allow the prediction of rate constants from extrakinetic parameters in the range of ionic strengths as encountered in the marine or continental aerosol, i.e. at least 5 M, another approach may have to be followed in future treatments. This concept may be addressed as the kinetic ion pairing approach. Ion pairing in solution is a phenomenon which is verified for different systems by laboratory measurements involving electrochemical (conductivity) and spectroscopic techniques. Ionic reactants existing as anions may, at higher electrolyte contents in a given matrix, form ion pairs with the abundant counterion. Stability constants for such ion pairs are available from literature (Högfeldt, 1982; Perrin, 1979) or may be calculated according to the Fuoss-Eigen equation (see Davies, 1962 for an overview). The observed rate constant as a function of ion concentration then results from two parallel elementary reactions coupled by the ion pair formation equilibrium.
Non-radical reactions
The effect of added electrolyte on the reaction rate of ozone with S(IV) in aqueous solution was analyzed by Maahs (1983). The intepretations, however, are critically discussed by Hoffmann (1986) in his review, who assigns the observed effect to the effect of ionic strength on the S(IV) equilibria, i.e. to the occurrence of a secondary ionic strength effect. The reaction has later been studied by Lagrange et al (1993). Electrolyte effects in the reaction of H2O2 with HSO3- have been investigated by Lagrange et al. (1993). Reaction rates were found to be significantly increased in the presence of chloride and ammonium ions. Results on the influence of electrolytes on the oxidation of sulfite by ClO - are described by Lagrange et al. (1995). Very recently, salt effects in the formation and decomposition of hydroxy-methanesulfonic acid (HMSA) are discussed by Lagrange et al. (1999b).
Whereas the above studies directly relate to tropospheric multiphase S(IV) oxidation, kinetic salt effects of NaCl and Na2SO4 have recently been described in the reaction of nitrite by H2O2 (Lagrange et al., 1999a).
Radical reactions
No studies on kinetic ionic strength effects in any OH reactions in aqueous solution are known at this time.
Several NO3 reactions in aqueous solution with anions and neutral organics have been investigated. A large primary kinetic salt effect for the reaction of NO3 with Cl- is described by Exner et al. (1992) applying NaClO4 as an electrolyte and is suggested to account for a large difference in the directly measured rate constant when compared to older literature data (see Herrmann and Zellner, 1998 for a discussion). The effect has not been found in a recent study by Imanura et al. (1997) using another experimental approach and varying added NaNO3 up to 1 M. The reaction requires further study.
Further primary kinetic ionic strength effects in the reactions of NO3 with dissolved acetaldehyde (Zellner et al., 1995, (see also Herrmann and Zellner, 1998) and oxalic acid (Zellner et al., 1995) are described. The effects can be treated according to Olson and Simonson (1949) allowing a kinetic description and extrapolation of results.
A kinetic salt effect slowing down the observed rate constant is described for the reactions of the dichloride radical anion (Cl2- ) with methanol as well as hydrated formaldehyde (Jacobi et al., 1999).
Research needs
The experimental data set for a more complete description of aerosol chemistry is very sparse. Further systematic laboratory studies are required. Apart from experimental data for single reactions such studies may also result in correlations which, in the future, may allow the estimation of rate constants at elevated ionic strength in different cases chosen according to electrolyte compositions.
The current status of understanding chemical reaction in matrices at elevated ionic strengths is unsatisfactory and obviously requires further extensive study. Here the approach of fitting experimental data to an ion-pairing mechanism secures the use of experimentally determined data only, but, at least at present, leads to the need of an explicit investigation for every single reaction in each given electrolyte system. The ion-pairing approach for chemical conversion reactions of highly reactive species implies a clear physico-chemical mechanism to describe kinetically at least ion-neutral reactions. It should be combined with the treatment of stable and moderately reactive species by their activities obtained from the Pitzer approach or by extended Debye-Hückel treatments as far as applicable. More experimental studies on kinetic primary ionic strength effects, resulting correlations and the development of sound theories are required in order to better understand tropospheric aerosol chemistry.
Sea-salt aerosol in the marine boundary layer, is about 80% water. This particle-associated water is at a pH of 8. The alkalinity of this sea-salt water is derived largely from the carbonate, bicarbonate, and borate present in bulk surface sea water. The magnitude of this alkalinity is an important factor in determining the role of sea-salt particles as chemical reaction sites for SO2 among other trace gases. Measurements in various marine boundary layer environments indicate that 50% to as much as 90% of nss-sulfate mass is internally mixed with sea-salt (O'Dowd et al., 1997a; Posfai et al., 1995; Sievering, 1984). The sea-salt and sulfate cycles are, thus, inextricably linked in the marine boundary layer. A large fraction of sulfate production in the marine boundary layer, even a substantial majority in remote areas removed from sulfur pollution sources (Sievering, 1984), may occur in sea-salt-bound condensed water. It has been shown that ozone oxidation of SO2 in sea-salt-bound condensed water contributes to the observed NSS-sulfate (Sievering et al., 1991),. Modeling of the marine boundary layer sulfur cycle (Chameides and Stelson, 1992) generally confirmed this. Further, Chameides and Stelson (1992) showed that, as O3 oxidation of SO2 proceeds in the sea-salt water, its pH will drop until, at a pH of 6 or less, O3 oxidation is quenched. The magnitude of NSS-sulfate produced by this mechanism should equal the alkalinity present in the sea-salt-bound condensed water. This implies that no more than about 1 nmol m-3 of the total NSS-sulfate observed at any location within the marine boundary layer can be due to this mechanism. However, further oxidation in sea-salt-bound condensed water may be triggered by O3 oxidation. Hydrogen peroxide oxidation of SO2 contributes in a minor way once pH of 6 or less is obtained (Chameides and Stelson, 1992).
The halogen release mechanism proposed by Vogt et al. (1996) and discussed further by Keene et al. (1998) may also contribute to the large amount of nss-sulfate found internally mixed with sea-salt particles. However, it is important to note that no experimental evidence is available to support the Vogt-et-al.-mechanism. Other approaches are needed to complement the known high pH ozone pathway in order to fully explain the NSS-sulfate found in supermicrometer particles during ACE 1.
Supermicrometer particulate calcium excess (i.e., non-soil-derived Ca++ ) was found to be nearly equal to the observed alkalinity during ACE 1. On the assumption that this calcium excess was present as CaCO3 when sea salt particles were emitted from the sea surface, the buffering capacity of ACE 1 sea salt particles is roughly doubled. This sum of alkalinity and calcium excess buffering capacity is sufficient to attribute 70-90% of the supermicrometer NSS- sulfate production to the high-pH O3 oxidation of SO2 pathway (Sievering et al., 1999), (see also contribution on calcium enrichment in sea salt particles).
An alternative explanation for this high fraction of nss-sulfate found internally-mixed in sea-salt particles may be due to the enhanced buffering capacity of cloud droplets activated upon sea-salt nuclei as opposed to calcium enhancement of the aerosol. In effect, the cloud behaves as a virtual buffer, maintaining pH high, and allowing further rapid sulfate production, above the carbonate buffer equivalent, to proceed (O'Dowd et al., 2000). Simulations of sulfate production, using a Pitzer non-ideal solution effect aerosol-cloud- chemistry model illustrate that pH in sea-salt particles > 1 µm is typically of the order of seven, and during cloud sulfate production during cloud formation, the increasing water volume maintains a pH of the order of seven (O'Dowd et al., 2000) .
From the possible sulfate production pathways outlined above, there are possible sulfate production mechanisms, both in cloud and out of cloud, that have the potential to explain the high fraction of nss-sulfate mass associated with sea-salt particles; however, it is not clear at this stage which mechanism dominates, and consequently, the dominant heterogeneous sink for SO2 , and the dominant production mechanism for sulfate particles, in the marine boundary layer remains unclear.
5. Properties and distributions of the atmospheric aerosol
| BC | POM | References | |
| Size range (µm; s) Range Best guess |
(0.0236-0.1); (1.6-2.1) 0.06; 2 |
(0.06-0.12); (1.7-2) 0.08;2 |
BC = 1, 2, 3, 4, 5, 6,7 POM = 8, 9, 10, 11, 12, 13, 14 |
| Density (g cm-3 ) Range Best guess |
0.6-2 1.5 |
1-2 1.5 |
5, 8, 15 |
| Refractive Index (n;m) Range Best guess |
(1.5-2);(0.5-1) 1.95;0.66 |
(1.4-1.6);(0-0.005) 1.55;0.005 |
BC = 5, 16, 17, 18, 19 POM = 8, 10, 12, 20 |
| Mass scattering efficiency (m2 g-1 ) Best guess |
2 |
4 |
9, 14, 21, 22 |
| Mass absorption efficiency (m2 g-1 ) Range Best guess |
(7-13) 7 |
- - |
5.3. Cloud Condensation Nuclei (Snider, Jennings)
The cloud condensation nuclei (CCN) are a subset of the total particle population of the atmospheric aerosol that forms cloud drops characterized by decreasing equilibrium vapor pressure of water with increasing water content and resultant spontaneous growth at supersaturation greater than a specified value. The concentration of CCN as a function of water vapor supersaturation is referred to as the CCN activation spectrum. This spectrum, together with the updraft speed at cloud base, and the degree of mixing of cloudy parcels with dryer air from the surrounding environment, strongly influence the microphysical properties of warm clouds, i.e. clouds that do not contain the ice phase. 5.4. The utility of the CCN concept (Heintzenberg and O“Dowd)
The concept of CCN has been accepted by the cloud physics community and most of the aerosol research community for many decades. Its attractiveness stems partly from the pioneering work of Twomey (1959) who introduced measured supersaturation (CCN) spectra into cloud modelling. Visualising a distinct "subset" of the very complex atmospheric aerosol as a well-defined and measurable quantity which is sufficient for a description of the interaction of aerosols and clouds is another factor contributing to the attractiveness and the wide application of the CCN concept. 5.5. Ice forming nuclei (Bigg)
Homogenous nucleation of liquid water to form ice becomes effective at temperatures below about 36C; particles which will induce water to freeze at warmer temperatures are known as ice forming nuclei, or IFN. Although those active at temperatures warmer than 10C constitute less than one-millionth of the total number concentration, they have a potential for modification of cloud processes an climate modification out of proportion to their numbers. This arises through their direct role in the production of precipitation, its intensity and duration and the effects of that precipitation on atmospheric dynamics and surface albedo. Clouds which are transformed completely to ice crystals can continue to survive in humidities where water clouds would evaporate, so that cloud lifetime and therefore radiation can be affected. Cloud albedo can also be affected directly (but not greatly, except in unusual circumstances) through the removal of water drops and the fallout of the ice which removed them. A further possibility is that concentrations of IFN capable of forming ice crystals at humidities below water saturation can modify the formation of visible cloud. Modeling studies, though unrealistic from the point of view of the ice phase, showed the effects of precipitation on atmospheric circulation and dynamics to be substantial (Fowler et al., 1996). There is an excellent review of the effects of ice formation in clouds (Baker, 1997). 5.6. The 3-dimensional distribution of the atmospheric aerosol (Heintzenberg)
Knowledge of the spatial distribution of aerosol concentrations and chemical and physical properties is necessary for quantitative assessment of aerosol effects. A basic prerequisite for any quantitative regional or global assessment of aerosol effects on the Earth's system is the knowledge of their distribution in time and space. The numerical models with which such assessments are made might calculate the particle distributions and properties from first principles governing the atmospheric aerosol life cycle or by means of climatologies that are meant to be representative of typical aerosol properties, but to date the latter approach has predominated (cf. the section on aerosol modelling), albeit with a slowly growing range of details. 6.1
A surrogate of the CCN activation spectrum is measured by exposing particles to a controlled supersaturation while monitoring cloud droplet nucleation and growth. The measurement represents an assessment of the evolution of aerosol particles to cloud drops in an environment which is not perturbed by the additional processes that influence droplet size and concentration within natural clouds (i.e., entrainment and drizzle scavenging).
Relatively few measurements of CCN exist and there is no co-ordinated measurement program for the measurement of CCN properties over a wide range of air masses and meteorological conditions. The World Meteorological Organisation (WMO) Scientific Advisory Group on Aerosols (and Aerosol Optical Depth) have prepared a draft document on CCN Measurement Procedures) for the WMO Global Atmosphere Watch (GAW) Aerosol Programme. The draft report includes methods of measurement, recommended operational mode and supersaturation, and calibration and standardization of equipment.
Three recent advances have occurred in the area of CCN instrumentation. Ji et al. (1998) produce a range of applied supersaturations (0.1 to 2%), in a steady-flow chamber to perimt discrimination between activated and unactivated particles based on their fall velocity. The size distribution of the unactivated fraction is characterized by a differential mobility analyzer. The CCN activation spectrum is derived as the difference between the total and the non-activating number concentrations. This technique requires approximately one hour of quasi stationary conditions to obtain the complete CCN activation spectrum.
Two improvements have been made to static chamber CCN counters. Twomey and co-workers have deployed a video-based detection scheme which counts growing drops in real time, obviating the task of manually counting drop images (Ramsey-Bell and Covert, 1992). Also working with a static chamber design, Vali and co-workers (Delene et al., 1998; Oliviera and Vali, 1995) have developed a device which permits inference of the CCN concentration from a continuous measurement of the light scattering produced by the activating drops; the device is capable of measuring the CCN concentration in about 20 seconds.
Questions concerning the absolute accuracy of many CCN measurement systems have arisen from studies that have concurrently measured CCN activation spectra and particle size distributions (Andreae et al., 1995; Covert et al., 1998; Gras, 1995; Hegg et al., 1996; Quinn et al., 1993). These workers show that the number concentration of CCN measured at or extrapolated to a specified supersaturation can be substantially smaller than predictions based on the accompanying particle size distribution measurement. These comparisons are based on Köhler theory which relates measurements of physical size, and the water soluble fraction (often assumed), to the critical supersaturation. This apparent disparity may be due to nucleation inhibition by some component of the aerosol, presumably organic, or may be a consequence of systematic error in the measurements. Bower and co-workers observed a similar disparity in their analysis of measurements made within adiabatic clouds at the Great Dun Fell in northern England (Bower et al., 1997).
Research needs
The need to predict particle size distributions from a consideration of aerosol source and sink processes (Fitzgerald et al., 1998; Kerminen and Wexler, 1997; O'Dowd et al., 1998; Raes, 1995), and to apply this understanding to the prediction of CCN activation spectra in climate models (Ghan et al., 1993) provides a strong motivation for identifying the source of the apparent CCN disparities noted above. Future investigations will need to examine the influence of surface active materials on water uptake by aerosol particles and drops. There is need for a co-ordinated effort to investigate CCN properties. There has been a general lack of activity in this area for nearly two decades.
However, there are over-simplifications inherent to this concept that become increasingly difficult to justify with the rapidly increasing experimental and modelling capabilities of atmospheric aerosol research. The concept or definition of CCN is at best operational and any quantification of CCN with existing instrumentation is strongly dependent on the thermodynamic evolution forced upon an aerosol sample in any particular CCN counter. An intrinsic problem is that the thermodynamic evolution occurring in any experimental CCN device bears rather little resemblance to the condensational growth as it happens in any real cloud. There are many physical and chemical factors contributing the actual subset of the total particle population which is incorporated into a particular growing cloud. Even certain gaseous species can strongly affect the growth of cloud elements (Kulmala et al., 1995; Kulmala et al., 1993; Kulmala and Mattila, 1996; Laaksonen et al., 1997).
Additionally, the concentration of CCN at a particular supersaturation is not necessarily applicable to the prediction of cloud droplet concentration. Even the generally accepted hypothesis that an increased in concentration of CCN leads to an increase in cloud droplet concentration is not necessarily correct, particularly in a mixed population of CCN species. For example, an enhancement in CCN concentration measured at 0.2% supersaturation, typical of moderate stratocumulus clouds, would suggest a similar increase in cloud droplet concentration within these cloud types; however, if the additional CCN are activated at supersaturations considerably lower than 0.2% the resulting depletion of water vapour and a suppression of peak supersaturation in a real cloud can inhibit activation of a significant fraction of the pre-existing CCN population resulting in a reduction in cloud drops (O'Dowd et al., 1999b).
Avenues of research that may be expected to resolve the CCN-problem include more detailed experimental studies of the condensational growth properties of size-resolved atmospheric particles and ensuing parameterizations for cloud models, and cloud simulation experiments with more realistic flow- through cloud simulators.
The measurement of concentrations of IFN that would be active in natural clouds is complicated by the fact that the probability of a given nucleus causing freezing at a particular temperature depends upon whether it was immersed in water before being cooled, contacted a supercooled drop, or was subjected only to ice-supersaturated water vapour (Vali, 1981). The only type of instrument that could reproduce these three conditions together with a realistic thermal and hygrometric history of nuclei participating in cloud formation is a very large expansion chamber, the walls of which are cooled in synchronism with the expansion. Such an instrument has been built and operated satisfactorily but its size, the time taken for a single expansion and the difficulty of counting the ice crystals in a sufficiently large volume make it unsuitable for counting the relatively rare natural ice nuclei (Hoppel et al., 1994b). More practical methods, useful for finding sources of IFN but of doubtful value in predicting the number of IFN active in a supercooled cloud include: freezing of water drops containing the aerosol, collection of particles on a membrane filter, then detecting the IFN by developing ice crystals on the filter in a thermal gradient diffusion chamber, rapid expansion chambers or continuous flow thermal diffusion counters.
Supercooled clouds themselves are of course the final arbiters of the effectiveness of IFN. Ice crystal replicators and, later, imaging probes for use on aircraft have allowed extensive observations of ice crystals within clouds. Interpretation of what is found in terms of IFN concentrations however is a very difficult problem. There can be recycling of ice crystals from evaporated clouds, ice crystals falling from overlying clouds and a number of mechanisms that create secondary ice crystals. Concentrations may also vary very widely from one part of a cloud to another. It became clear at least thirty years ago that the best estimates of IFN concentration were usually lower than actual ice crystal concentrations and in many clouds several orders of magnitude lower. The early work has been reviewed by Mossop (1985). A more recent study by Bower et al. (1996) has again found ice crystal concentrations to be several orders of magnitude higher than could be attributed to primary nucleation by IFN at cloud top temperatures. Obviously there must be processes which lead to the generation of new ice crystals from those formed on IFN for there to be such an enormous variability relative to IFN concentrations.
Probably the most common secondary ice crystal generation process was identified by Hallett and Mossop (1974), who showed that capture of relatively large (>25 µm diameter) cloud drops by ice crystals led to the ejection of ice splinters as the drops froze. The process has been calculated to take only about 30 min to change cloud element concentrations from 0.01 L-1 to 100 L-1 for common cloud conditions. However, Rangno and Hobbs (1994) believe that an even faster process, which depends on the concentration of large drops in the tail of the cloud drop spectrum, (for a debate on the issue, see Blyth and Latham, 1998).
The multiplication processes that depend on cloud droplet spectrum therefore also depend on other factors such as liquid water content and the spectrum of cloud condensation nuclei. Realistic modeling of ice formation requires the simulation of the evolution of the cloud drop spectrum as well as the IFN concentration at -5C must be calculated. Latent heat release and consequent effects on cloud properties, lifetime, development of precipitation and atmospheric dynamics must be accounted for. The rapid and complete glaciation of clouds through multiplication effects must far outweigh the effects of IFN in clouds in which there is no secondary generation. However, the time taken for the process to reach full glaciation depends considerably on the initial concentration of very active IFN.
The nature of IFN that could produce ice crystals at 5C, the very specific temperature at which the Hallet-Mossop process operates efficiently, was a complete mystery until the discovery by Schnell and Vali (1972) that a proportion of common vegetation-living and marine bacteria possessed this property. It has since been found that some fungi and other organisms also have significant IFN activity.
The numerous experiments over the past 40 years involving increase of IFN have not demonstrated any large sensitivity to their concentrations. Is the same likely to be true of an experiment which decreased their numbers? Assuming bacteria or other micro-organisms to provide the most effective IFN, removal of vegetation by over-grazing or tree-felling is one such experiment and desertification has often resulted. However, changes in surface moisture and albedo and concentrations of cloud condensation nuclei also accompanied the "experiments" so the extent to which IFN may have contributed is unknown.
The remaining property of IFN that could be important is the ability of a proportion of them to create ice crystals at ice supersaturation but water subsaturation. There have been very few reliable experiments on this property.
Perhaps the best practical IFN counter currently available with sufficient humidity control to make such measurements was described by Rogers (1993). He found that the concentrations of natural nuclei did not decrease dramatically as humidities were lowered below water saturation. This implies that in the case of clouds produced by slow uplift at temperatures well below 0C, a few ice crystals should grow and fall through the rising air before a visible (water saturation) cloud occurs. Reduction of the humidity by these growing crystals might then reduce the albedo of the cloud that formed, or in critical cases even prevent its formation. Unambiguous detection of such ice crystals will be difficult because the humidity above cloud top is rarely uniform and it takes time for them to grow to a detectable size, by which time they are not in the air parcel in which they were formed. Heymsfield et al. (1998) reduced this problem by making measurements in the more uniform vapor fields in the vicinity of orographic wave clouds but still had the problem of growth time to contend with. Their conclusion was that there was no detectable ice crystal formation below water saturation at temperatures above -39C but that at -55C ice crystal formation was observed at 75% humidity relative to water.
A different approach has been to study the particles found within ice crystals in cirrus. Mineral particles were found to dominate both interstitial aerosol and ice crystal residues but having a substantial difference in elemental composition (Heintzenberg et al., 1996) . This could be interpreted to mean that a subset of the mineral particles formed ice crystals by heterogeneous nucleation. A similar conclusion was reached by Chen et al. (1998) although they found that particles containing soluble material were also involved .
Research needs
Should IFN be regarded as insignificant contributors to climate control (as the general aerosol was until the beginning of the decade) and no further work on them be done? Their potential for dynamic and precipitation effects through the ice crystal multiplication processes suggest that this could be a short-sighted attitude. As an alternative we would have to develop a climatology of IFN capable of acting as freezing nuclei (the mode of ice formation that appears to be appropriate for material from micro-organisms) at a temperature of -5C, together with the associated CCN, so that ice crystal multiplication processes could be modeled. The first step would be to develop an easy method of counting such IFN, the second to determine the exact nature of the Rangno and Hobbs (1994) ice crystal multiplication processes. If that could be achieved a modeling approach incorporating ice crystal multiplication might give a quantitative guide as to the role of IFN (Fowler et al., 1996). A determination of whether IFN acting at water subsaturations anywhere in the atmosphere are ever sufficient to influence subsequent cloud formation would also seem to be a worthwhile endeavour.
Several reasons can be given for the rather slow progress in the acquisition of aerosol climatologies. By far the great majority of measurements have been made at the surface and mainly over continents. The use of still rather disparate methodologies makes it difficult to harmonise the results. Indeed, no comprehensive parametrization of continental aerosols has been accomplished since the pioneering work of K. T. Whitby (1978), despite the availability of commercial and specialised methodology which can reveal many more details than was possible 25 years ago (e.g., Heintzenberg et al., 1998).
Over marine areas the distribution of the atmospheric aerosol is known with much less certainty than over the continents. Only for about a quarter of the oceanic surfaces well-calibrated and parametrized submicrometer size distributions are available (Heintzenberg et al., 2000). For bulk chemical composition the corresponding coverage is somewhat better (» 60%, cf. Heintzenberg et al., 2000) but still far from satisfactory. Much of the progress in marine aerosol characterisation came through the large IGAC field experiments of the mid-nineties (ACE 1, Bates et al., 1998), (ACE 2, Raes et al., 2000)
Knowledge about the vertical aerosol distribution is even more limited than the surface distribution. Over the continents there has been only one long- term effort in vertical aerosol profiling by in situ measurements, (Hofmann, 1993; Hofmann et al., 1998), yielding valuable data on the effects of volcanoes, of the increase of air traffic and of decreasing surface emissions on the upper troposphere and lower stratosphere. A series of research flights by A. Clarke and others complemented to some extent the continental data over the Pacific and north Atlantic region (Clarke, 1993; Clarke et al., 1999; Clarke et al., 1996; Clarke et al., 1997; Heintzenberg et al., 1991; Weber et al., 1999).
After about five years of technical development a first aerosol payload is now flying frequently on a commercial aircraft (Brenningkmeijer et al., 1999). On flights between Europe and the Indian Ocean the first aerosol climatology for the tropopause region as been accumulated by these flights (Hermann, 1999), showing clear influences of surface aerosol sources in the tropics and in the mid-latitude over Europe.
A few aerosol lidars have monitored the vertical aerosol distribution over extended time scales (Ansmann et al., 1997; Jäger and Carnuth, 1994; Reiter and Jäger, 1986).
A major factor limiting the extent and quality of present aerosol climatologies is the lack of well-defined and calibrated aerosol monitoring programs. For lack of suitable methodology and local competence the initial aerosol monitoring program designed by WMO (1993) has been of somewhat limited value so far. This program had adopted a design philosophy of monitoring stable long-lived trace gases, which adds little of value to interpretation of aerosol measurements. Only a few stations operated under US and Australian scientific auspices, have produced valuable background data in marine of free-tropospheric air masses (e.g., Ayers and Gras, 1991; Bodhaine, 1983; Heintzenberg and Bigg, 1990), Major findings at these stations were:
Rather recently, a few continental stations with high methodological standards have been added in the US, (Koloutsou-Vakakis et al., 1998), and Europe (Mészįros et al., 1998) supplying valuable information on particle properties relevant to their direct radiative forcing and chemical apportionments of these properties to natural and anthropogenic particle sources. Renewed efforts of extending the Global Atmospheric Watch (GAW) program to detailed aerosol measurements, (GAW, 1994), have not yet resulted in sufficient coverage in terms of aerosol parameters and regionally representative stations.
6. Aerosols and climate
| Location | Mass Scattering Efficiency, m2 g-1 | ||
| nss SO42-d D < 10 µm | Sea salt D <1.0 µm | Sea salt 1.0 | |
| 70S-40Sa | 5.1 ±0.43 | 5.5 ±0.22 | 0.68 ±0.08 |
| 20S-20Na | 3.0 ±1.5 | 4.1 ±2.1 | 0.79 ±0.13 |
| 20S-20Na | (4.0 ±1.2) | (8.7 ±4.4) | (1.9 ±0.28) |
| 30N-54Na | 7.4 ±2.1 | 3.5 ±0.62 | 1.1 ±0.14 |
| 48N,127Wb | 4.2 ±0.48 | 3.7 ±0.38 | 0.9 ±0.07 |
| 46S,145Ec | 1.3 ±0.55 | 4.5 ±1.1 | 1.2 ±0.18 |
6.2. Contribution of boundary layer clouds to the indirect aerosol effect on climate (Brenguier)
The indirect aerosol effect (IAE) is presently the most uncertain forcing mechanism in the prediction of climate change. In particular, it is likely to counteract part of the warming due to greenhouse gases (Slingo, 1990). The main contribution to the IAE comes from marine stratocumulus clouds (Randall et al., 1984). The IAE refers to potential changes of cloud radiative properties at the global scale due to changes in physical and chemical properties of those particles which form cloud drops. Modifications of these particles are reflected in a cloud by changes in the cloud droplet number concentration (CDNC). Variations of CDNC are then likely to affect cloud radiative properties via numerous mechanisms. The basis of the IAE is a direct link that has been established by Twomey, (1977), between aerosol properties, CDNC and cloud albedo. The additional feedback mechanisms potentially responsible for the uncertainty in the prediction of the IAE are: the extension of cloud cover with increasing global temperature (Arking, 1991), the reduction of cloud precipitation efficiency in clouds (Albrecht, 1989), the coupling between diabatic processes and cloud dynamics (Martin et al., 1997; Pincus and Baker, 1994), and the radiative effect of in-cloud absorption on short wave radiation (Boers and Mitchell, 1994). 6.3. Aerosol-climate relationships (Feichter)
The direct effect of aerosol particles on radiative fluxes in almost all model studies is taken into account by prescribing optical particle properties which have been calculated off-line by Mie-calculations or are based on observations, assuming a uniform particle size, density and particle composition for each of the aerosol components. Despite the fact that particles of different chemical composition are internally mixed, most of the studies considered the different types as independent or assume a uniform internally mixed aerosol. The first studies of the indirect effect considered anthropogenic sulfate as calculated by Langner and Rodhe (1991) as a surrogate for all particles produced by anthropogenic emissions and used an empirical relationship to relate the particle mass to the number of cloud drops (Boucher and Lohmann, 1995; Hegg, 1994; Jones et al., 1994). These studies calculated only the possible effect of anthropogenic sulfate particles on the cloud albedo. In a next step the sulfur chemistry has been coupled with the cloud microphysics to estimate the effect of particles on the cloud life-time and the cloud water content (Jones et al., 1999; Lohmann and Feichter, 1997; Rotstayn, 1999a). Mechanistic approaches which relate the particle number to the cloud droplet number concentration has been proposed by several groups (Chuang et al., 1997b; Ghan et al., 1997; Lohmann et al., 1999a).
6.3.1. Radiative forcing
6.3.2. Climate response
7.1. Aerosol sampling (Huebert)
It is difficult to measure accurately concentrations and properties of ambient aerosols (Vincent, 1989; Willeke and Baron, 1993). Numerous potential artefacts can modify the ambient population as it is being sampled, so that the analytical system generates data that is a poor representation of reality. Since many of these problems are sufficiently subtle (or simply not widely understood) that users of analytical equipment are either unaware of or unable to quantify them, it is possible that significant amounts of published aerosol data is misleading in one way or another. The potential for experiencing sampling artefacts depends strongly on the size and chemical properties of the aerosols being sampled. 7.2. On-line chemical analysis of individual particles (Wexler)
A class of new instruments are now becoming more available for atmospheric sampling. These instruments analyze chemically individual particles on-line and in real time. The technique is referred to as single particle analysis (SPA). As with all instruments, SPAs have strengths and weakness. Some of the weaknesses will be overcome by continued research and development. Others are inherent. The paragraphs that follow, outline some SPA strengths and weaknesses, the different incarnations of current SPA instruments, and some improvements on the horizon:
7.3. Analysis of aerosol samples with physical techniques (Swietlicki)
Numerous analytical techniques based on physical principles have been employed to study the composition of atmospheric aerosol particles. Techniques are also available that can offer isotopic information and surface analysis. This section will mainly focus on electron and ion beam analytical techniques, and mass spectrometric methods.
7.3.1. Ion and electron beam analytical techniques
7.3.2. Mass spectrometric methods
7.3.3. Other methods
7.4. In situ physical measurement techniques (Wiedensohler and Wendisch)
In situ measurements (ground- or aircraft-based) provide microphysical properties of aerosol particles which can not be retrieved by remote sensing techniques. However, the in-situ measurement techniques described below modify the ambient aerosol mainly due to sampling artefacts (cf. section on aerosol sampling). Inlet or transport pipes cause diffusion and impaction losses for very small and large particle, respectively. Due to changes in relative humidity, the aerosol is often measured under dry or undefined conditions which are not representative for the undisturbed ambient aerosol. The spatial and temporal resolution of in-situ aircraft measurements is limited by the speed of the moving measurement platform. Fast on-line measurements can create Poisson errors whereas bag sampling systems limit the spatial resolution. Ground-based measurements can provide a more complete set of aerosol parameters with a high time and/or size resolution, however, the aerosol characterisation in the vertical column is missing.
7.4.1. Particle number concentration
7.4.2. Number size distribution
7.4.3. Particle volume extinction coefficients
7.4.4. Particle volume scattering coefficients
7.4.5. Hygroscopicity
7.5. Remote sensing of aerosols: (Kaufman and Kent)
7.5.1. Past Satellite observations of aerosol effects
7.5.2. Expected improvement in the next decade
7.5.3. Ground based columnar remote sensing
7.5.4. Remote sensing of upper tropospheric and stratospheric aerosols
7.5.5. Remote sensing of particles by Lidar
7.6. Closure in atmospheric aerosol research (Heintzenberg)
The concept of closure experiments or studies was introduced formally into atmospheric aerosol science at the Dahlem workshop on Aerosol Forcing of Climate (Ogren, 1995), the main motivation for this kind of exercise being the need to reduce the uncertainties in assessments of aerosol forcing of climate. 8.1. Modeling aerosol-dynamic processes (Stratmann and Wilck)
Modeling aerosol dynamics requires describing the flow, heat and mass transfer, gas and particle phase chemistry and particle dynamics in the system of interest. For atmospheric applications this means describing the meteorological, gas phase chemical and particle dynamic processes including particle phase chemistry. 8.2. Mesoscale aerosol modelling (Ackermann, Hass and Schell)
8.3. Global aerosol models (Feichter)
For a few years now global chemical transport models (CTM) and general circulation models (GCM) of the atmosphere have been explored to evaluate the atmospheric transport and the interactions of gaseous and particulate constituents. Whereas observations of atmospheric constituents are limited in space and time and contain only a subset of parameters controlling the climate system, numerical models permit exploration of the full range of parameters in space and time. Ultimately the ability of these mathematical algorithms to represent the real world has to be validated by comparison to observational data.
8.3.1. Global chemistry and aerosol models
8.3.2. Static climatologies derived from observations and model- calculations
8.3.3. Dynamically calculated aerosol mass distributions
8.3.4. Dynamically calculated aerosol mass and number concentration
8.3.5. Modelled aerosol components
8.4. Upscaling of parameterizations in large-scale models (Lohmann)
Processes which act on spatial scales smaller than the grid box or temporal scales shorter than the time step of any model cannot be resolved and therefore have to be parametrized i.e., these processes have to be expressed in terms of large scale variables. The question is how to develop a parameterization for, for instance, a general circulation model, which typically has a time step of 20 minutes and a spatial resolution of 3-5 degrees times 3-5 degrees. There are four possibilities to derive parameterizations:
The decade of the nineties has been one of major progress in the recognition of the role of aerosols in global atmospheric chemistry, in describing concentrations and properties of tropospheric aerosols and their geographical distribution, and understanding the controlling processes.
Experimental evidence of the IAE is a challenge because cloud radiative properties are primarily determined by the cloud geometrical thickness (H), or by its liquid water path (LWP). These parameters are highly variable. Observed differences in cloud albedo can be attributed to changes in CDNC only if LWP or H are precisely measured. Boers et al. (1998) report an example of IAE related to changes in the distributions of natural CCN over the ocean. The IAE is illustrated by a variation of the relationship between optical thickness and LWP for summer versus winter clouds. Indications of IAE are also presented in Rosenfeld, and Lensky (1998). These authors analyze AVHRR images of cumulus fields, to derive the cloud top temperature and effective droplet diameter. Variations of the relationship between those two parameters can be also interpreted as a sign of IAE due to continental pollution. More recently, Brenguier et al. (2000) have reported observations in marine stratocumulus during the second Aerosol Characterization Experiment (ACE-2). Simultaneous in situ measurements of cloud microphysical properties and remote sensing of the cloud radiances in the visible and near-infrared provide evidence of changes in cloud radiative properties related to changes in the aerosol background. It is remarkable, however, that all three studies demonstrate that the droplet effective diameter is primarily dependent on cloud depth and that the aerosols do not directly affect the effective diameter but rather its dependence on the cloud depth.
Parameterization of the IAE is also a challenge. Up to now, the radiative forcing by boundary layer clouds in GCM has been parametrized with LWP and the effective diameter. Both parameters and the cloud extent have been adjusted to fit climatological observations. The procedure is efficient as long as the relationship between these parameters is unchanged. On the opposite, for a prediction of the IAE it is necessary to develop « physical » parameterizations, i.e. to capture the essence of the potential feedback mechanisms involved in the IAE. The preliminary step is to validate process parameterizations with closure experiments (cf. section on closure). The IAE involves interactions between aerosol particles, cloud microstructure and radiative properties.
For the interaction between particle properties and droplet concentration, closure experiments reveal serious discrepancies, with overprediction of droplet concentrations with regards to the values measured in situ. There are still no firm conclusions about the origin of the overestimation factor of two to ten between these two quantities (Chuang et al., 2000; Wood et al., 2000), though closure experiments between CCN activation spectra directly measured at cloud base and droplet concentration measured slightly above show a good agreement (Snider and Brenguier, 2000). Therefore, the discrepancy is more likely related to the prediction of the aerosol nucleating properties from their measured physical and chemical properties.
Following the diagnostic of CDNC changes in response to a modification of the aerosol background, it is then necessary to establish a relationship between CDNC and cloud radiative properties. Most of the radiative transfer calculations in boundary layer clouds have been performed using plane-parallel models, with horizontally and vertically uniform microphysical fields . In these models, microphysics is characterized by the droplet effective diameter. Simulations of the IAE have then been conducted by reducing the value of this parameter, in response to an increase in CDNC (Jones et al., 1994).
However, the vertical profile of cloud microphysics in cloud is not uniform because liquid water content (LWC) increases with altitude above cloud base (Martin et al., 1994), and so does the effective diameter. The reduction in droplet effective diameter thus depends on the cloud depth. The novel flight procedure (series of ascent and descent through the cloud layer), applied during the Cloudy-Column experiment within ACE-2, was specifically designed for the statistical description of the vertical profiles of cloud microphysics (Pawlowska and Brenguier, 2000). Various situations with very different aerosol backgrounds have been documented, thus validating experimentally the adiabatic cloud model. More realistic parameterizations are obtained by using adiabatic profiles of cloud microphysics (Boers and Mitchell, 1994). In particular, they establish direct relationships between CDNC and the cloud parameters required for radiative transfer calculations.
Another source of discrepancy between plane parallel models and actual clouds is the horizontal variability of the microphysical fields. Their statistics have been characterized with satellite and in situ observations and it has been demonstrated that the albedo of actual clouds is smaller than the albedo of a horizontally uniform model with the same LWP (Davis et al., 1997; Davis et al., 1996; Hignett and Taylor, 1996; Pincus et al., 1999).
Finally, the formation of drizzle precipitation can have a significant impact on parameterizations of the IAE. Numerical simulations (Feingold et al., 1997) suggest that precipitation formation leads to droplet spectra broadening and depletion in CDNC, so that cloud optical thickness does not scale with CDNC1/3 , as predicted by the adiabatic model, and that the susceptibility of cloud radiative properties to changes in CDNC is significantly increased. In addition, the precipitation efficiency in boundary layer clouds with low values of LWC is particularly sensitive to CDNC. Doubling CDNC, as it happens currently during pollution outbreaks, can prevent the formation of drizzle particles, thus increasing cloud extent and life-time. For all these reasons, it is likely that the second indirect effect, via precipitation efficiency, could be more significant than the Twomey effect.
Prospective for GCM simulation of the IAE and research needs
Two key processes still resist to the most recent column closure experiments: the CCN activation at cloud base and the radiative properties of a cloudy column. Additional efforts are therefore needed for the analysis of the most recent closure experiments. The experimental community is still missing an instrument for in situ measurements of the activation process in the range 0.2 to 8 micrometers. This range is crucial for understanding particle growth and their activation as cloud drops. The development of such an airborne instrument is highly desirable for the future field campaigns.
The challenge for the prediction of the IAE is to develop physically based GCM parameterizations from process parameterizations. The predictable parameters in a GCM are much less numerous than the measured parameters in a closure experiment and they characterize properties averaged over a scale much larger than the typical process scale. The basic information for simulation of the IAE are the characterization of aerosol nucleation properties and the dynamic and thermodynamic properties of the cloud field. Numerous questions have to be answered between experimentalists and modelers such as:
An efficient solution for extrapolating experimental results from the process scale to the GCM scale is to use 3-D simulations of cloud fields (Bechtold et al., 1996) for deriving 1-D parameterizations (Feingold and Heymsfield, 1992), that can be tested further with forcing techniques at the GCM scale (Lohmann et al., 1999c).
The radiative forcing, defined as the change in the radiative fluxes at the top of the troposphere or top of model domain due to a given aerosol component, has been calculated by off-line and by on-line models for a wide variety of different particle distributions and emission scenarios (e.g., Houghton et al., 1996; Shine and de F. Forster, 1999; Tegen and Miller, 1998). Estimates of the global mean direct forcing range for sulfate between -0.20 (Hansen et al., 1997) and -0.82 W m-2 (Haywood and Ramaswamy, 1998). Based on the very few studies estimating effects of other materials than sulfate, the radiative forcing of carbonaceous particles is about -0.4 W m-2 (Cooke et al., 1999; Hansen et al., 1997; Penner et al., 1998), that of anthropogenically disturbed mineral dust +0.09 W m-2 (with -0.25 in the solar and +0.35 in the infrared part of the spectrum) (Miller and Tegen, 1998).
Haywood et al. (1999) calculated short-wave forcing by major natural and anthropogenic constituents . Reported top-of-atmosphere global-annual forcings (in W m-2 ) were natural sulfate, -0.93; anthropogenic sulfate, -0.72; organic carbon, -1.02; black carbon + 0.17; natural dust -0.58; anthropogenic dust -0.54. Two estimates were given for sea salt based on low and high emission values, - 1.51 and -5.03 W m-2 , respectively. Current estimates of the indirect forcing range quite widely; differences arise mainly from the parametrization of cloud droplet number concentration on particle mass. For indirect forcing by sulfate estimates range from -0.4 to -2.1 W m-2 (Boucher and Lohmann, 1995; Chuang et al., 1997a; Feichter et al., 1997; Jones et al., 1999; Jones and Slingo, 1996; Kiehl et al., 2000; Lohmann and Feichter, 1997; Lohmann et al., 1999b; Rotstayn, 1999b). Again in view of the many uncertainties associated with understanding and description of aerosol sources and processes and with parametrization of these processes, such forcings must still be considered rather tentative.
Global distributions of direct radiative forcing by black carbon (BC) and particulate organic matter (POM) have been calculated using global concentration maps (obtained by a TM2z transport model), and the optical properties given in Table Liousse.1. The back scatter fraction was taken from Hobbs et al. (1997) for biomass burning particles. Results are presented in Plate Liousse 1 for July.
Plate Liousse 1: Direct radiative forcing (DRF) due to Black Carbon (BC) and Particulate Organic Matter (POM) for July.
It can be seen in Plate Liousse.1 that global patterns are linked to source regions. A global cooling is calculated for the Northern hemisphere with negative values over Europe of the order of -1.0 W m-2 . This result is strongly dependent on the importance of BC in the aerosol mix. However, it may be underlined that even in these areas of strongest forcing, the cooling is dominant if POM is taken into account (and not only sulphate). The cooling effect is higher in the southern hemisphere where mean values may reach -8 W m-2 over Amazonia. As expected, a positive forcing is found for areas with high surface albedo.
The relationship between radiative forcing and climate response was examined for a wide range of different artificial perturbations by Hansen et al. (1997). The direct and indirect climate effects of anthropogenic sulfate were examined for equilibrium (constant) or transient changes in sulfate distributions, and with prescribed or interactively calculated forcing. Tegen and Miller (1998) calculated the climate effect of mineral dust interactively in an equilibrium simulation. Besides the uncertainties associated with the distribution and the optical properties of aerosol particles, climate models exhibit a wide scatter in the climate sensitivity (defined as the ratio of the change in the global mean surface air temperature to the global mean radiative forcing) arising from differences in approaches to treating cloud physical processes and in the assumed optical properties of clouds (e.g. Cess et al., 1996).
7. Measurement techniques and strategies
One of the most thoroughly documented types of sampling artefacts involves gas/aerosol interactions. Appel et al. (1978) were among the first to note that gaseous nitric acid and particulate nitrate can interconvert on filter media. The collection of nitric acid vapor by a filter intended to collect particulate nitrate generates a positive artefact, since the analyst will find extra nitrate and interpret it as having been ambient nitrate aerosol. Conversely, aerosol species that have significant vapor pressures such as ammonium nitrate, can evaporate when conditions change, causing a negative particulate nitrate artefact (Zhang and McMurry, 1987).
There are methods for minimizing these artefacts, however. Many widely used filter media contain sites that can retain acidic gases such as nitric acid and sulfur dioxide and therefore cause positive artefacts. The filter medium Whatman 41, for instance, collects total nitrate very efficiently. The use of Teflon or acid-washed quartz filters can minimize such problems. Long sampling times contribute to evaporative artefacts, since it is more likely that ambient conditions will change during a long sample, so the shortest viable sampling times also helps to minimize negative artefacts. One of the most rigorous methods for avoiding gas/aerosol artefacts is the use of diffusion denuders (Gormley and Kennedy, 1949; Sickles II et al., 1990). These devices first denude the air stream of condensable gases and then sample the aerosols with a filter pack that can collect both the aerosol and any volatilized products from it.
Organic aerosols pose significant problems, which have yet to be resolved (Appel et al., 1989). Many organic species have significant vapor pressures, so that exchange between condensed and vapor phases is continuous. Any change in pressure or temperature during sampling will displace this equilibrium, distorting the ambient distribution between the phases. Furthermore, many organic vapors adsorb strongly on filter media, generating positive artefacts. The most common approach to correcting for this artefact involves the use of a second filter in series with the first (Novakov et al., 1997). However, this relies on the often-questionable assumption that vapors are only weakly adsorbed, so that the same artefact will be generated on the second filter as on the first. Much remains to be done before organic aerosol observations can be used with confidence.
Inertial effects make it difficult to bring even non-volatile particles into sampling devices (Fuchs, 1975; Okazaki et al., 1987). The tendency of particles to continue moving with the ambient flow rather than entering the flow of sampling inlets tends to generate artefacts with larger particles. Every inlet system has some size above which it discriminates against particles, making it particularly difficult to get representative samples of sea salt and mineral dust, for both of which most of the mass is larger than one micrometer. In order to standardize sampling for specific purposes, considerable effort has gone into designing inlets that are well-characterized for cut-offs at 2.5 and 10 µm diameters (Lundgren et al., 1996; Wedding and Weigand, 1982). It is important that users of published data look for evidence that the generators of the data thought carefully about what size ranges their inlets admitted.
Inertial problems are magnified when aerosols are sampled from aircraft, which often fly at 100 m s-1 or faster. The deceleration of air within inlet systems generates intense turbulence that propels many of the particles to the walls to the inlet diffuser cone. This results in transmission efficiencies for supermicrometer species of only 5 to 20%, whereas submicrometer particles may be collected with 50 to 100% efficiency (Huebert and Lee, 1990; Sheridan and Norton, 1998). Recent measurements during the ACE 2 experiment (Schmid et al., 2000) demonstrated the impact of this on measurements of the extinction of radiation by dust and sea salt: a nephelometer (sampling from an inlet) underestimated the former by about 85% and the latter by about half relative to inlet-less approaches. Similarly, a comparison of a total aerosol sampler with a conventional curved-tube inlet during the PEM-Tropics experiment found that MSA and non-sea salt sulfate were underestimated by a factor of 20-30 by filters behind the conventional inlet (Huebert et al., 1998). The result is that many of the published concentrations of aerosols above the surface are likely to be underestimates of the true ambient concentrations. Users of this data must be aware of this pervasive artefact.
There are new technologies being developed, however, which will improve the situation. A new total aerosol sampler (TAS) has been demonstrated to collect all aerosols that enter its tip, without artefact (Huebert et al., 1998). The disadvantage of TAS is that all size information is lost in the extraction process. To enable size-dependent sampling, a new laminar-flow inlet (Seebaugh and Lafleur, 1996) being developed at Denver University will be test-flown in 2000.
Aerosol sampling problems have no doubt compromized much of the data now in the refereed literature. However, as awareness of these issues grows and technological improvements are realised, we should begin to see much more representative values in the future. A significant fraction of the recent progress has been motivated by IGAC-sponsored aerosol research programs.
Primary Strengths
Primary Weaknesses
The Near Future
All known single particle references are available at http://www.me.udel.edu/wexler/spa_refs.html
In PIXE (Particle Induced X-ray Emission) analysis, X-ray quanta are emitted from the sample when irradiated with a beam of ions (normally protons) having energies of a few MeV/amu (Johansson et al., 1995). During this ion bombardment, each element emits several characteristic X-rays, which are used to identify and quantify the elements present in the sample. PIXE is nowadays a routine analytical technique for the study of atmospheric aerosols (Maenhaut, 1992). The reasons for this are the many advantages it can offer: multi-elemental (Na - U), low detection limits, quantitative with good accuracy and precision, non-destructive, fast, inexpensive, and minor sample preparation.
For an aerosol sample collected in continental background air masses, 10- 20 elements can normally be detected and quantified with PIXE within an error of 10-15%. An additional advantage is that the detection limits vary smoothly with Z. Absolute and relative detection limits down to 10 -12 g and 0.1 µg/g respectively can be achieved. Owing to the high sensitivity of the method PIXE can be used to analyse aerosol samples collected in air masses with low particle concentrations or with a high time or size resolution. The analysis itself can often be automated. The qualities listed above make PIXE well suited for long- term aerosol monitoring programs, such as the US IMPROVE (Sisler and Malm, 1994) and European EMEP networks (Kemp, 1993), (see Maenhaut, 1990 for a review of PIXE and other competing trace element analytical techniques).
The major detection limiting factor for PIXE and many other analytical techniques is the backing substrate on which the particles are deposited during collection and analysis (Cahill and Wakabayashi, 1993). There is a need for backing substrates that contain only a few elements of the periodic table (e.g., plastics or metals that are free from other contaminants), and have very low area mass density yet are mechanically robust. Recently, there has been some promising progress regarding new backings (Papaspiropoulos et al., 1999), showing that more can be done to push detection limits even further.
During irradiation with MeV/amu ions, not only X-rays are emitted, but also other types of radiation that can be used to obtain additional information regarding the sample composition. Over the last decade, a range of complementary ion beam analytical (IBA) techniques have been developed which can provide fully quantitative analysis of low-Z elements (H-Mg) simultaneously with PIXE for high-Z. These complementary IBA techniques include PIGE (Particle Induced Gamma-ray Emission), PESA (Particle Elastic Scattering Analysis) also denoted RBS (Rutherford Back scattering), cPESA (Coincidence-enhanced Particle Elastic Scattering Analysis), pNRA (Photon- tagged Nuclear Reaction Analysis) and others (Cahill, 1990; Cohen, 1998; Kristiansson and Martinsson, 1997). In principle, the entire periodic table (H- U) can therefore be analysed concurrently. The detection limit for H using cPESA is as low as 10 pg cm-2 (Martinsson and Kristiansson, 1993). For aerosol samples, pNRA can be used for Li, Be, B, F, Na and Mg, and PESA (RBS) for C, N and O. The sensitivity for the lighter elements vary considerably with Z and typically range between 10 up to a few hundred ng cm-2 (Mentes et al., 1999).
In nuclear microprobe (NMP) setups (Watt and Grime, 1987), ion beams of MeV/amu energies can be focused down to 1 µm2 or less, and be used for fully quantitative studies of single aerosol particles with PIXE detection limits (1-10 ppm) that are at least an order of magnitude better than for electron microprobes (Artaxo et al., 1993). In practice, NMP analysis is limited to super-micrometer particles and is very time-consuming. It would be possible to use various types of beam-induced radiation to locate and then analyse the individual particles automatically, similar to automated EPMA (Electron Probe Micro Analysis), but this has still not been demonstrated. NMP is best suited for trace element analysis of particles in the 1-10 µm size range in samples of e.g. soil dust, sea-salt and fly ash. The potential for low-Z analysis using complementary IBA techniques in parallel with PIXE has not yet been fully 1 exploited in NMP aerosol work.
EPMA or SEM (Scanning Electron Microscopy) if imaging is the main objective is the most widely used technique for analysis of the composition of sub-micrometer single particles. EPMA is based on the emission of characteristic X-rays during electron bombardment and offers multi-element analysis of elements above Z=13 with reasonable accuracy and detection limits down to 0.1 % in particles > 50 nm. Automated-EPMA work only on particles > 150-200 nm. Liquid nitrogen cooling of the sample during electron irradiation can ameliorate the losses of volatile compounds in vacuum. Important low-Z elements such as C, N and O can be analysed in particles > 50 nm if a windowless X-ray detector is used that does not attenuate the soft X-rays excessively. The aerosol sampling is then done on very thin aluminium or beryllium foils or on substrates on which a layer of LiBH4 has been evaporated.
The matrix corrections on which the C, N, O quantification rely are however rather uncertain, thus affecting the accuracy. Recent results nevertheless show promise for aerosol studies (Ro et al., 1999). The potential aerosol applications include analysis of organic compounds of both anthropogenic and biogenic origin (even purely organic particles) and nitrates. This combination of both high- and low-Z analysis capability in all but the very smallest particles would greatly facilitate studies of the degree of external or internal aerosol mixture.
Energy-dispersive X-ray (EDX) analysis of single particles down to 5 nm and imaging down to 1 nm can be performed with Scanning Transmission Electron Microscopy (STEM), but the analysis is very time-consuming and only limited number of aerosol particles can be studied this way. Very few aerosol studies have been performed so far.
STEM can be combined with EELS (Electron Energy Loss Spectroscopy) for analysis of sub-micrometer particles, in practice only for lighter elements (Z < ca. 30). Reasonable data acquisition times and volatility losses can be obtained when the full electron energy loss spectrum is acquired simultaneously (Parallel-EELS, PEELS). The method has been demonstrated to work for 20 nm atmospheric aerosol particles (Maynard and Brown, 1992) with detection limits of 4% for carbon, equalling 30 carbon atoms! EELS yields information regarding the chemical bonds and can for instance be used to distinguish between graphitic and amorphous carbon. Although very time-consuming, both STEM-EDX and PEELS might be worthwhile to pursue in special cases, e.g. to study particle nucleation events.
The combination of electron microscopy and microchemistry is not new (see e.g. Bigg et al., 1974) but could most likely be developed further. Here, the substrate and the collected aerosol particles are exposed to various vapours that react with a specific compound in the aerosol particles. Organic compounds can be dissolved in organic solvents such as decane or xylene, or be forced to react with RuO4 leading to an oxidation of the organic particles. The resulting Ru-deposit can be detected down to 30 nm.
Methods based on atomic or molecular mass spectrometry (MS) are destructive, and the evaluation of the MS spectra obtained are often complicated by interferences, i.e. superpositions of different atomic or molecular species with indistinguishable mass-to-charge ratios. MS techniques used for aerosol studies include ICP-MS (Inductively Coupled Plasma Mass Spectrometry), SIMS (Secondary Ion Mass Spectrometry) and LAMMS (Laser Microprobe Mass Spectrometry).
Of these, ICP-MS is rapidly growing in popularity. It is a fully quantitative multi-elemental technique capable of analysing nearly all elements in the periodic table with excellent detection limits (0.1 ng ml-1 or better). Isotopic information can also be obtained, since the ICP ionisation is almost complete. ICP-MS suffers from matrix effects that can be severe and difficult to predict or estimate, and the aerosol samples have to be dissolved in liquid (Sah, 1995).
In TOF-SIMS, secondary ions are sputtered from the target particle during irradiation with an ion beam of low energy (keV) and analysed with a time-of- flight (TOF) mass spectrometer. Secondary ions are only generated down to a depth of a few molecular layers, but repeated sputtering allows (destructive) depth profiling to be made with a lateral spatial resolution of a few micrometers.
In conventional LAMMS, a high-power laser is used to ionise the sample collected on a backing substrate, and the ions produced are brought into a TOF- MS. The small laser focus (0.5-1 µm) allows analysis of single aerosol particles. The method is time-consuming and semi-quantitative at the very best, but can nevertheless provide useful information on single-particle inorganic and organic composition and state of mixture (e.g., Gieray et al., 1997).
For more information on single-particle analysis using various physical analytical techniques, (see the extensive review of Jambers et al., 1995).
Some additional analytical techniques that have been used for aerosol studies but were not described here are INAA (Instrumental Neutron Activation Analysis), EDXRF (Energy Dispersive X-ray Fluorescence Analysis), TXRF (Total Reflection X-ray Analysis), SXRF (Synchrotron X-ray Fluorescence Analysis), SAED (Selected Area Electron Diffraction), XPS (X-ray Photoelectron Spectroscopy or ESCA), and Micro-Raman Spectroscopy (Maenhaut, 1992; Spurny, 1986; Van Grieken and Markowicz, 1992).
Research needs
Physical techniques will continue to play a vital role in atmospheric science. The high sensitivity that can be achieved means that extremely small sample masses are required. Bulk aerosol compositional data provided by these techniques can be used for studies of environmental cycling of various elements, aerosol mass closure studies and source apportionment based on receptor modelling.
Single-particle analysis can certainly tell very interesting tales on the origin and evolution of the atmospheric aerosol, for instance regarding the degree of external and internal mixture and new particle formation. Single-particle compositional data still remain to be reconciled with measurements of other properties performed in-situ on individual particles, such as hygroscopic and cloud-nucleating behaviour.
The total number concentration of the aerosol particles as the integral of the number size distribution can be measured with a high time resolution. State of the art instruments to determine the particle number concentration are continuous flow condensation particle counters (CPC's) (Agarwal and Sem, 1980). This type of instrument counts each individual aerosol particle and therefore has an upper detection limit in particle concentration. Common CPCs have lower detection efficiency diameters of 10-15 nm, and an upper number concentration limit of about 10,000 cm-3 without significant coincidence. The development of the Ultrafine-CPC (Stolzenburg and McMurry, 1991) enabled number concentration measurements down to 3 nm in diameter with an upper detection limit of 100,000 cm-3 . Measurements of particle number concentration and size distribution in the range below 8 nm became possible by the Pulse- Height-Analysis-UCPC (Weber et al., 1998b).
DMA-based size spectrometers
The development of DMA-based size spectrometers (DMA = Differential Mobility Analyzer), in the last years, enabled number size distribution measurements almost over the whole submicrometer size range. Two different sizing systems are presently used for atmospheric measurements. DMPS- systems (Differential Mobility Particle Sizer) measure stepwise while SMPS (Scanning Mobility Particle Sizer) systems continuously scan through the size range. DMPS-systems need a scan time of 10-15 min and provide relatively good counting statistics in each size bin. In contrast, SMPS-systems can scan much faster (1-5 min), possibly leading to higher statistical uncertainties and artificial distortions at the ends of the size distributions. Those artefacts may lead to large errors in calculations of higher moments of the size distribution such as surface area, scattering coefficient, volume, or mass.
DMPS systems are commercially available since the eighties, however, were limited only to the size range above 10 nm. The development of DMAs for nanometer particles (<10 nm) and the availability of the UCPC (UCPC = Ultrafine Condensation Particle Counter) led to the design of "ultrafine"-DMA- based size spectrometers (Birmili et al., 1999a; Russell et al., 1996; Winklmayr et al., 1991). However, the statistical uncertainty in the range below 10 nm is still quite large for atmospheric measurements in each DMA- based sizing system due to decreasing DMA penetration efficiency, low charging probability and the low aerosol flow rate of the UCPC (Wiedensohler et al., 1994).
Optical Particle Counters (OPC)
These are instruments which size each individual aerosol particle according to their optical properties. The particles are aspirated into a scattering chamber and then directed into the centre of a beam of laser or a white light. The light is scattered by the particle; its intensity is related to the size of the particle via Mie theory for homogeneous spheres. In order to suppress sizing ambiguities due to oscillations of the Mie curves, the scattered light is collected over a wide angular region or a white light source is used. During sampling and within the scattering chamber the particles are dried and therefore the particles are not sized in ambient, undisturbed conditions (Strapp et al., 1992). Further problems are caused by the refractive index and shape-dependence of the measurements. With intra-cavity laser illumination, lower size detection limits of such instruments of 70 nm can be reached. As other single particle counters OPCs have upper concentration limits which are given by the probability of particle coincidence in the sensing volume.
In combination with a DMA the OPC can be used to quantify external mixtures of light absorbing and light scattering particles in the size range about 0.3 µm (Covert et al., 1990).
The classical method for the measurement of light extinction by aerosols utilises the Lambert-Beer law. Attenuation of white light or spectral channels is recorded after passage through a distance in the atmosphere (e.g., Goes, 1963; Goes, 1964). By combining the classical attenuation method with Differential Optical Absorption Spectroscopy (DOAS) a promising new development of this technique has been made which, in principle allows to determine in the same ambient aerosol beam spectral aerosol extinction, humidity and trace gas concentration (Flentje et al., 1997). To date there is no technique available to measure in situ the absorption component of particle extinction. Because of the low levels of absorption in the atmospheric aerosol filter-based quasi-real time techniques have been developed since the early 80ties (Hansen et al., 1982; Reid et al., 1998b). For climate studies the need remains, however, to extend these sampling techniques to the ambient airborne aerosol.
The particle scattering coefficient can directly be measured using so-called nephelometers (Heintzenberg and Charlson, 1996). In these instruments a white light source illuminates an air volume. The scattered light is collect within a certain scattering angle region (»7 - »170 degrees) and certain spectral intervals. With some reasonable assumptions these data yield the volume scattering coefficient of the particles. Nephelometers, however, influence the relative humidity and thus the scattering coefficients by heating-up the aerosol sample. Usually good agreement (<20%) is obtained for dry conditions between integrated (using the particle size distribution) and directly measured (nephelometer) particle scattering coefficients by using a newly designed nephelometer (Anderson et al., 1996b). Several approaches have been followed to extend the dry scattering measurements to ambient humidity conditions reaching from completely enclosed sensing chambers (Heintzenberg and Erfurt, 2000) via partly enclosed chambers (Garland and Rae, 1970; Malm et al., 1996), to completely open nephelometers with reduced sensitivity and high systematic errors (Ruppersberg, 1964).
In the eighties and nineties, measurements of hygroscopic properties of aerosol particles were significantly improved by the development the Hygroscopic-TMDAs (Tandem Differential Mobility Analyser, Rader and McMurry, 1986). For the first time, the size-resolved external mixture of aerosol particles in terms of hygroscopicity could be determined. However, these measurements are limited to the size range below 300 nm in diameter, and do thus not give the entire climate-relevant information for cloud or fog processes, and scattering of solar radiation of aerosol particles under ambient conditions. No technique exist that measures the hygroscopicity directly and independently of filter or impactor samples for the size range larger than 300 nm in diameter.
In the last decade, satellites have been used for semi-quantitative global monitoring of aerosol distribution in one (Herman et al., 1997) or two (Higurashi and Nakajima, 1999; Husar et al., 1997) spectral channels. The results are described for AVHRR as an effective aerosol optical thickness. TOMS data are interpreted as an index that may be related to aerosol optical thickness provided either external information is available about the type of aerosol or assumptions are made about the height of the aerosol and the nature of the underlying surface (Torres et al., 1998). Correlation between aerosol loading and cloud properties as determined from satellites was used to estimate the aerosol indirect forcing: the aerosol effect on cloud droplet size and reflectance of sunlight (Charlson et al., 1992; Kaufman and Fraser, 1997). But uncertainty in these estimates is about factor of four, due mainly to the difficulty to estimate the concentration of the background "natural" aerosol.
The POLDER instrument flown on the ADEOS-I mission in 1996 (Deuzé et al., 1999) was the first designed for quantitative estimate of the particle properties, column burden and radiative forcing. It combines spectral measurements in four spectral channels (0.4 - 0.9 µm) with polarization measurements and angular measurements in a geometry similar to a "fish eye camera" observing the earth with a wide range of view angles. The results are global maps of aerosol optical thickness, estimates of refractive index and wavelength exponent of particulate extinction over the ocean and an aerosol index over the land. Sampling of most of the view directions permits comparison with the results with the spectral flux reflected by the aerosol to space. But the large 6 km size of the foot print restricts the cloud free regions that can be used for the aerosol evaluation.
The geostationary satellites, GOES and METEOSAT do not have many spectral bands and their calibration is an issue but they can measure regionally the diurnal cycle and yield estimates of the aerosol column burden and properties (Prins et al., 1998).
European, Japanese and US space research organizations plan to launch in the next several years satellite systems with instrumentation designed for qualitative aerosol measurements and for evaluation of the aerosol effect on cloud microphysics and solar radiation. Several satellites will be launched by NASA as part of the Earth Observing system (EOS). Together with France the first active aerosol and cloud directed satellite PICASSO-CENA mission is planned for 2003. A new POLDER system for polarized aerosol measurements is due in 2001 with NASDA“s ADEOS-II.
A large array of ground based federated AErosol RObotic NETwork (AERONET) of autonomous radiometers will be used to evaluate and complement the space borne information (Holben et al., 1998). These instruments (100 of them world wide) measure the aerosol spectral optical thickness (0.38-1.02 µm) and the sky spectral radiance. Columnar averages of particle size distribution (Nakajima et al., 1983), estimated refractive index (Yamasoe et al., 1998) and single scattering albedo (Dubovik et al., 1998) are being derived from the sky data. This information is used to derive aerosol dynamic climatology, that can be used in models and inversion schemes for the satellite data (Remer et al., 1998a; Remer et al., 1998b). The information is also used to derive a climatology of the diurnal cycle of the aerosol.
Upper tropospheric aerosols are not easy to study remotely. One set of instruments that has been used with some success is the solar occultation series SAM II, SAGE I and SAGE II (Kent et al., 1998; Kent et al., 1995). Although intended for the study of stratospheric aerosols these instruments can, in the absence of high cloud, make measurements well down into the troposphere. One disadvantage of the solar occultation technique is the low sampling rate. Approximately 30 events are obtained per day separated by about 240 in longitude and at two specific latitudes normally in opposite hemispheres. For SAGE II a period of about 3 weeks is required to obtained reasonable global coverage. Long term climatologies, extending over years, can be readily built up but the instrument is not able to follow shorter duration events such as the movement of an aerosol plume. The instruments are self calibrated at each event and are thus very suitable for studying long-term trends in aerosol concentrations (Kent et al., 1998).
The (AVHRR, Husar et al., 1997) measures the total aerosol optical thickness which, except in cases where there are intense elevated aerosol layers, is dependent mainly upon the boundary layer aerosol with its higher aerosol concentrations. The Total Ozone Mapping Spectrometer (TOMS, Hsu et al., 1996). cannot readily detect aerosols near to the earth's surface but has a good sensitivity above about 1 km altitude. Both altitude ranges are in contrast to those of SAGE II and its fellow instruments where most of the data reported relates to altitudes above 6 km. AVHRR and TOMS are also nadir looking and have a good horizontal resolution. SAGE II in contrast, with its tangential measurement geometry, has a relatively poor horizontal resolution. Nevertheless, on a global scale, all these instruments can supply important information about aerosol distributions and movements and, by virtue of their sensitivity to different altitude ranges, collectively create a more complete picture than is available from a single instrument. Data from SAGE II and the other solar occultation instruments are inverted to produce a vertical profile of aerosol extinction, while that from the AVHRR is in terms of the total aerosol optical depth.
Lidar has been used to make remote observations of the atmospheric aerosol for many years. The basic lidar measurement consists of a profile of back scattered signal intensity as a function of range. This may be inverted to obtain separate profiles of back scatter and extinction provided assumptions can be made, or information is available, about the relationship between these two optical quantities. Additional inversion difficulties may arise on occasion due to secondary scattering of the laser radiation. Most commonly, and particularly where the extinction is low, the derived parameter is either the aerosol back scattering cross-section or the ratio of the aerosol back scattering cross-section to that of the molecular atmosphere. The range of useful lidar measurements is determined by the size and power of the lidar system and by the optical density of the scattering medium. A further problem that occurs in lidar measurements is the difficulty of obtaining an absolute calibration of the lidar system and data is often normalised by measuring the strength of the returned from a standard target, often chosen to be the molecular atmosphere. Measurements made from a single ground site are usually directed towards a local target, such as an industrial plume, or aircraft contrails, or into the stratosphere, particularly after volcanic aerosol injection. Eye-safe autonomous systems have been developed, which may also have scanning capabilities that can be used to map the spatial distribution of the aerosol.
Although the caveats noted above can make absolute lidar measurements of aerosol optical properties subject to error, the technique has shown itself unique in its ability to map aerosol with a vertical resolution that is unmatched by any passive technique. The range resolution of a lidar is typically 10-30 m and the accuracy of an altitude determination is normally only limited by the knowledge of the altitude of the lidar platform. Thus, the first lidar in earth orbit, LITE was able to study with great precision aerosol phenomena such as pollution outflows from industrial centres, smoke from biomass burning, the marine boundary layer and desert dust outbreaks. The data is most valuable in its ability to be used to verify regional and global transport models.
Similar studies have been made on a more regional scale by airborne lidar. Typical measurements include studies of Arctic aerosols showing Arctic haze (Radke et al., 1989), stratospheric intrusions and smoke plumes (Browell et al., 1992), remote oceanic measurements (Cutten et al., 1996), and particles from biomass burning (Anderson et al., 1996a). Most profitably, lidar data is obtained in conjunction with in situ measurements of particle size distributions and composition as well as meteorological data - information not directly available from the lidar measurement itself.
Raman aerosol lidars, (Althausen et al., 2000; Ansmann et al., 1990), are used with increasing success in the lower atmospheric region (Ansmann et al., 2000), yielding profiles of absolute optical properties of particles and derived physical properties. An important step for the acquisition of 3- dimensional aerosol data is the installation of the European Lidar Network (EARLINET) in 2000.
From the beginning closure studies were seen as tools to summarise critically and quantitatively the state of knowledge about characteristics, processes and effects of the atmospheric aerosol. Thus any closure exercise should necessarily lead to conclusions concerning the improvement of experimental and/or modelling methodology.
Redundant information or an over-determined system is a core component of a closure study, another one being a model of the investigated system which connects the different experimentally determined characteristics of the system. As a theoretically straightforward example, measurements of aerosol extinction in parallel with scattering and absorption measurements yield a closure test that does not need more than the principle of energy conservation. A third essential component and prerequisite for any useful conclusion to be drawn from the exercise is the quantification of the uncertainties of all experimental data. In the classic case of a purely experimental closure study, aerosol characteristics determined with different independent methods are compared and the system is considered to be closed if the results agree with the experimental uncertainties. Obviously, large uncertainties facilitate the achievement of closure. However, not many useful conclusions about the improvement of experimental and/or modelling methodology can be expected in such a case.
Another essential requirement in closure studies is a high cross sensitivity of the different measured aerosol parameters. Measured dependent aerosol parameters (e.g., spectral optical extinction coefficients) should have a sufficiently close connection with the measured independent aerosol parameters (e.g., size dependent particle concentrations and chemical composition) so that any significant change (i.e. a change greater than the experimental uncertainty) of an independent aerosol variable causes a significant change in the dependent aerosol property. As an inappropriate example of cross sensitivity, it is very easy to close total number and scattering coefficients of an aerosol because light scattering and total number are only loosely connected in most atmospheric aerosols. Consequently, not much can be learnt from this closure test.
In principle the closure concept can be applied to aerosol systems ranging from micrometer, as for the characteristics of an individual particle suspended in an electromagnetic trap, to the planetary scale, as in the case of aerosol retrievals from satellite measurements for full atmospheric column. However, the degree of closure attainable in the atmospheric aerosol systematically diminishes as we increase the size of the physical domain because the number of processes that need to be taken into account and the number of atmospheric variables that need to be known or modelled increases. Thus, expectations on closure and ensuing conclusions must be reduced when going from zero to higher dimensions. Nevertheless, even on a three-dimensional regional scale useful information about atmospheric processes can be derived from closure studies. A good example can be found in Alpert et al. (1998) who could estimate atmospheric heating by dust from the "errors" in a model/data assimilation system (Alpert et al., 1998).
After rather few closure studies in the seventies and eighties the concept is being exploited in a rapidly increasing number of experiments in the nineties (cf. a first review of the subject including specific recommendations for future exercise in Quinn et al., 1996a). Since then three large scale atmospheric aerosol experiments have been conducted which comprised several closure exercises. The results of the first one (TARFOX) have been published recently (Russell et al., 1999). A clear sky column (Russell and Heintzenberg, 2000) and a cloudy column closure experiment (Brenguier et al., 2000) were part of the Second Aerosol Characterisation experiment over the polluted eastern Atlantic ocean and adjoining coast. The Indian Ocean Experiment and the planned Asian aerosol characterisation experiment also comprise closure tests from the local volumetric to the three dimensional column scale.
8. Modeling of aerosols
In order to simulate particle dynamics, changes of the particle size distribution due to nucleation, condensation, absorption, chemical reactions, coagulation and external forces such as the gravitational force must be taken into account.
Here two different aspects have to be considered:
a) the description of the actual microphysical and chemical processes (taken up in the section on transformations)
b) the technique used to describe the changes of the size distribution
The most widely used method for modeling the changes of particle size distribution today is the so-called sectional technique. In this technique, the particle size axis is divided in intervals and certain moments, e.g. the number or mass, in size each interval are solved for by the model (Gelbard et al., 1980). This technique has been extended to account for particle composition at an early stage (Gelbard and Seinfeld, 1980) and is used in several contemporary atmospheric models (Dabdub et al., 1998; Kleeman and Cass, 1998; Lurmann et al., 1997; Meng et al., 1998; Sun and Wexler, 1998). Recent improvement concerning the sectional technique are the consideration of moving sections (Jacobson, 1997; Kim and Seinfeld, 1990) and the introduction of a 2-moment sectional approach (Harrington and Kreidenweis, 1998a; Harrington and Kreidenweis, 1998b)
Another class of aerosol models uses the so-called method of moments in which the time evolution of some moments of the size distribution is solved for. The moments of the particle size distribution are related to significant characteristic properties (total number, surface and volume of particles) of the particle population. Using a moment technique, these properties can be calculated efficiently without detailed information about the size distribution. However, a problem concerning this technique is to obtain a closed set of equations for the moments of the particle size distribution.
One possibility to solve this problem is to assume an analytical function for the size distribution. This approach is used e.g. in the modal aerosol dynamics technique, assuming that the aerosol may be divided into several distinct populations, so-called modes, which evolve in time almost independently. The size distributions of the individual populations are described by e.g., the log normal distribution. For the description of the time evolution of the modes, including multi-component systems, the method of moments is used (Wilck, 1998).
Another promising technique to obtain a closed set of equations for particle size distribution moments is the quadrature technique of McGraw (1997), in which a set of moments is used to derive points and weights for a Gaussian quadrature rule. These points and weights are then used to calculate the integrals over the size distribution appearing in the equations of change for the moments. The quadrature technique has already been applied in large scale models (Wright et al., 2000).
Research needs
There is still a strong need for improvement concerning the time efficiency and accuracy of these techniques, especially for applications in atmospheric transport models.
One of the most important questions today is how to describe properly gas to particle mass transfer especially for organic material, i.e. how to account for particle nucleation and growth processes when organic species are involved. Models to calculate heteromolecular nucleation rates for organic species in atmospheric systems are still either completely lacking or have large uncertainties, and probably will continue to do so in the near future (cf. section on transformations).
Tropospheric aerosol modelling on regional scales has been done on an episodic as well as long-term basis with Eulerian models utilizing different techniques (e.g. sectional and modal; see previous section). In this context mesoscale models are defined as models that can be applied to urban and regional scales up to »1000 km and that have a vertical scale reaching into in the free troposphere.
Most of the urban (Lurmann et al., 1997; Meng et al., 1998; Sun and Wexler, 1998; Wexler et al., 1994) as well as mesoscale episodic models (Dabdub et al., 1998; Jacobson, 1997) are sectional models that have been applied to Southern California. A modal model has been developed by Binkowski and Shankar (1995) and applied to Eastern North America (Binkowski, 1999). A modified version of this model has also been applied to Central Europe (Ackermann et al., 1998). Middleton (1997) presented an application of a regional aerosol model to the Denver region . A comparative review of the thermodynamic algorithms used in these models can be found in Zhang et al. (1999), and a general comparison regarding the applicability of these models to address PM standards was given by Seigneur et al. (1999). All of these models include primary as well as secondary particles and are linked to a gas-phase photochemical model in order to represent the interactions between the phases.
Other model applications on the same scale are aiming at describing long- term concentrations rather than episodes and consequently use a simplified description of the size distribution and/or aerosol chemistry using a parametrised gas-phase chemistry or assume a mono-disperse inorganic aerosol system (Tarrasón and Tsyro, 1998).
Secondary organic aerosols (SOA, cf. section 4.2) are difficult to determine from measurements directly. Therefore a number of direct modelling methods have been applied (e.g., Bowman et al., 1997; Pandis et al., 1992; Schell et al., 1999; Strader et al., 1999). Models have the advantage that they describe the formation of SOA based on chemical and physical principles.
Emissions of precursor gases, photochemical degradation, dispersion, formation of low-vapour-pressure products and gas/particle partitioning of these products are implemented in the models to predict SOA concentrations. It was proposed that secondary organic particle production can be described with a two product model (cf. section 4.2), an approach that is widely used in regional aerosol modelling nowadays.
Three-dimensional global atmospheric chemistry models simulate the emissions and the transport of chemical constituents due to large-scale winds and due to subgrid-scale processes like turbulent vertical exchange and cloud processes, the deposition at ground, the dynamic evolution of particles, their removal by precipitation and chemical transformations. Models which calculate the temporal and spatial variability of concentrations of atmospheric constituents may be classified as climatological or episodic, and as on-line or off-line: Chemical transport models (CTM) which calculate the tracer distributions based on a prescribed meteorology are called off-line and general circulation models (GCM) which add extra dependent variables are called on-line models. CTMs are driven by climatological mean (Langner and Rodhe, 1991; Pham et al., 1995; Zimmermann et al., 1988) or by instantaneous wind fields (Heimann and Keeling, 1989; Mahlman and Moxim, 1978; Prather, 1987). Wind fields are provided by GCMs or by data assimilation of observed winds. GCMs treat the transport of atmospheric constituents similar to that of water vapor by introducing additional prognostic variables on-line with the model's meteorology (Feichter et al., 1991; Hunt and Manabe, 1968; Rasch et al., 2000; Tegen and Miller, 1998). The advantage of such an approach is the high temporal resolution and that parameters needed to simulate the evolution of particles, like three-dimensional cloud water distribution, cloud cover and the generation of precipitation, are provided in a consistent way by the GCM. Furthermore, the use of a GCM allows feedbacks between the atmospheric constituents and the transporting winds. The disadvantage is that the computational expense is very high and that the simulated meteorology is by no doubts not as good as observed data. Recently Newtonian relaxation was introduced to force a GCM to simulate a specific weather episode (Dentener et al., 1999; Feichter and Lohmann, 1999; Jeuken et al., 1996). This technique, also called nudging, relaxes the model state toward observational data by adding at each time-step an additional term to the model equations. Aerosols have been represented in global models by a variety of approaches of increasing complexity, as briefly outlined below.
Concentrations and geographical distributions of different aerosol types (d'Almeida et al., 1991; Tanré et al., 1984) are prescribed. Optical properties of different aerosol types (e.g., urban, rural, remote etc.) (Krekov, 1992; Shettle and Fenn, 1979) are derived from measurements and Mie-calculations. Most current climate models use such climatologies to account for the direct effect of aerosol particles on radiative fluxes. Indirect aerosol effects are neglected or taken into account in a very simple way by assuming different effective cloud droplet radii over land and over sea which affect the optical properties of clouds and the autoconversion of cloud drops to precipitation. This approach suffers from obvious limitations, such as the inability to account for correlations between particle concentrations and dynamic variables in the GCM or feedbacks between the aerosol and the rest of the climate system.
Some GCMs and sub-global models treat the particle precursor chemistry and the evolution of the particle mass interactively with meteorology taking into account the complex interactions between cloud processes, heterogeneous chemistry and wet removal (e.g., Benkovitz and Schwartz, 1997; Feichter et al., 1997; Koch et al., 1999; Rasch et al., 2000; Roelofs et al., 1998). Such approaches give a consistent distribution of clouds and particles and allow to represent the high temporal and spatial variability of aerosol particle mass distributions. These models require considerable parameterization of aerosol processes because of lack of explicit treatment of aerosol dynamics.
Additionally they are limited by present understanding of aerosol sources and removal processes. They have been most successful for simple, relatively well understood chemical systems such as dust and sulfur.
Recently attempts have been undertaken to calculate not just the particulate mass but also the particle number concentration by parameterizing particle formation and dynamic processes. Two kinds of aerosol dynamics models have been developed: spectral schemes and bin schemes. In the spectral scheme one or more aerosol modes of the particle distribution are described by log normal distributions (cf. section 8.1). Assumes a constant width of a mode, the distribution is described by two prognostic variables, the particle mass and number concentration. The mass median diameter is then calculated for a given particle density. Such a scheme has been applied by Schulz et al. (1998) to mineral dust and by Wilson et al. (1999) to sulfuric acid. Both considered three aerosol modes and used the off-line transport model TM3.
Bin schemes distribute the particle mass on different size classes. Each bin is characterized by its geometric mean diameter. With a large number of size classes this scheme allows to reproduce the change in size distribution accurately. Schulz et al. (1998) did sensitivity tests varying the number of size classes between five and 100 and found based on mineral dust simulations that 20 bins are sufficient to avoid numerical inaccuracies . Such schemes have been applied for mineral dust studies (Schulz et al., 1998; Tegen and Lacis, 1996) and for sea-salt particles (Gong et al., 1997a).
Global modeling studies so far explored mainly the role of anthropogenic sulfur emissions considering sulfate as a surrogate for anthropogenic pollution but encompass recently a greater number of aerosol species. Emission inventories and global distributions of primary carbonaceous particles have been derived by three groups: LLNL (Liousse et al., 1996; Penner et al., 1993), JRC (Cooke and Wilson, 1996) and CNRS (Cooke et al., 1999). Mineral dust distributions were calculated with global models where dust emissions are parametrized in terms of soil moisture, surface wind speed, soil texture and vegetation (Dentener et al., 1996; Genthon, 1992; Marticorena and Bergametti, 1995; Tegen and Fung, 1994). Tegen et al. (1996) estimated that about 50% of the dust load originates from disturbed soils affected by cultivation, deforestation, overgrazing and erosion . Sea-salt particle distributions have been calculated by several groups (Genthon, 1992; Gong et al., 1997b, for details see section 3). Whereas most of the model studies consider an externally mixed aerosol Adams et al. (1999) simulated an internal mixture of nitrate, sulfate and ammonium applying a thermodynamic model to determine the partitioning of ammonium and nitrate between gas and particle phase. In view of the limited understanding of many of the pertinent processes such models may perhaps best be considered exploratory and requiring much evaluation by comparison with observation.
Research needs
The spatial and temporal variability of aerosol properties is determined by a complex combination of the processes of generation, transformation and evolution of an ensemble of particles, their transport from the sources, their removal from the atmosphere and the ambient temperature and humidity (besides other trace gases which are in phase exchange with the particulate phase). An isolated consideration of these processes or of specific aerosol compounds is rather arbitrary, as all of them are interconnected. Moreover, all global climate model studies so far have only considered the bulk particulate mass of some specific components rather than the size spectra of an externally and internally mixed aerosol and their size-dependent chemical composition, despite recognition that the size distribution and the chemical composition of the particles controls the activation of particles to cloud elements. The cycles of aerosol particles are closely connected with hydrological processes. Therefore, models have to consider the complexity of a multi-phase system. Certainly a weak point of the climate models is the parametrization of cloud processes and properties and the cloud-radiation feedback. Uncertainties in the calculated cloud properties are hard to quantify because satellite retrievals are not very accurate and in-situ measurements cannot be extrapolated.
Future needs include representation of size resolved aerosol distributions; inclusion of heterogeneous chemistry on particle surfaces and interactions between different aerosol species; and perhaps treatment of particles as ice nuclei (cf. section on ice forming particles). Necessary model evaluation includes comparison of calculated 3-dimensional cloud and aerosol distribution, particle size distribution, and microphysical properties to in-situ measurements and ground-based and satellite remote sensing data. More detailed comparison of processes such as the rates of key processes and concentrations of gas-phase precursors and intermediates will also lend improved confidence in models.
Four examples for the different possibilities including their advantages and caveats are discussed below:
For some processes analytical solutions can be obtained by some simplifications. An example is the cloud droplet activation process, where in equilibrium, the particle size as a function of relative humidity is described by the Köhler curve. Cloud droplet formation depends on supersaturation and the size spectrum and hygroscopicity of the particles. Supersaturation itself is created by cooling due to e.g., adiabatic expansion and depleted by condensation onto the existing cloud drops. Ghan et al. (1993) used the Köhler curve as the starting point for a parameterization of cloud droplet nucleation and assumed that the condensation rate can be related to dry particle diameter, such that droplet formation can be written as a function of total number of particles, an activation parameter and the CGM-variable vertical velocity. However, the number of cloud drops activated is determined by a small scale updraft which is much larger than the mean vertical velocity of a GCM grid. To take subgrid scale variations into account Lohmann et al. (1999a) modified the grid-box mean vertical velocity with the square root of the turbulent kinetic energy. This is justified if only one cloud type exists in the grid box. If cloud types are varying in one grid box than integration over a probability distribution function of vertical velocities is to be preferred (Ghan et al., 1997).
Dry deposition velocities for aerosol particles are sometimes derived from wind tunnel studies. However, these velocities are often lower than those obtained from field measurements because of the absence of surface roughness and the increased surface area of vegetation (Seinfeld and Pandis, 1998). Another example is the parameterization of the gas exchange between the ocean surface water and the atmosphere, which Liss and Merlivat derived from wind tunnel experiments (Liss and Merlivat, 1996). In reality organic films on the water surface or vertical mixing within the mixing layer may affect the exchange rate and is not taken into account.
An example for the use of field data in deriving parameterizations is given by Quinn et al. (1993) who derived a relationship between sulfate mass and cloud condensation nuclei (CCN) concentrations from four long term averaged data pairs at one site . These data are representative of the temporal scale of a GCM, but cannot represent the spatial variability. Boucher and Lohmann (1995) compiled measurements of sulfate particulate mass, CCN and number of cloud drops (CDNC) from various continental and marine sites in clean and polluted air, for a variety of weather situations and experiments including long term data and data from measurement campaigns. From this compilation they derived a relationship between sulfate mass and CDNC. This data set should be able to represent the spatial and temporal variability of CDNC in a GCM but is limited as it takes sulfate particulate mass as a surrogate for total particulate mass.
In-cloud scavenging of sulfate particles was measured to be between 30% and 90% by different authors (cf. a review in Boucher and Lohmann, 1995). This large uncertainty can be reduced by making use of simulations with a detailed aerosol-cloud microphysical model as for instance done by Flossmann (1996). She found that nucleation scavenging removes 80-90% of the soluble particulate mass. These findings were then used in a GCM study to justify that all sulfate mass undergoes nucleation scavenging (Feichter et al., 1996).
In summary, parameterizations based on field studies which encompass either the spatial scale of a few grid boxes or are conducted over at least one month are probably the best suited for use in GCMs as they represent real conditions. Laboratory studies can give valuable insight in specific controlling processes.
9. Conclusions
A major driving force for the upsurge in research that has led to this progress was the seminal paper by Charlson and others (1987) that proposed that sulfate aerosols from marine phytoplankton not only influence the microphysical properties of marine stratus clouds but actually participate in a feedback loop that is capable of stabilizing climate. Although this suggestion was criticized on a number of grounds, it nonetheless stimulated much research that has attempted to identify the sources of marine aerosols and to characterize their life cycle. More broadly, recognition of a strong sensitivity of climate to radiative forcing of aerosol particles by both the direct and indirect mechanisms and recognition also that the uncertainty in aerosol forcing of climate must be greatly reduced in order to properly assess climate sensitivity to increased concentrations of greenhouse gases intensified the effort to determine the global distribution of aerosols and the absolute and relative contributions of natural and anthropogenic aerosols. This research focus built on and became superimposed on research on atmospheric aerosols that had been ongoing for some years, motivated by concerns over aerosols as air pollutants influencing health, visibility, and, through deposition, natural and managed ecosystems and structures of economic or cultural importance. The enhanced understanding resulting from this research is making its way into numerical models describing the emissions, transport, transformation, and deposition of aerosols and their precursors on a variety of scales, from urban air sheds to global. The results of this revolution in research on atmospheric aerosols are reflected in the material that has been presented in this Chapter.
Much of this revolution has been fuelled by new measurement capabilities. Extension of the range of DMAs and condensation particle counters down to particle diameters of a few nanometers has revealed the episodic nature of nucleation events, though the substances responsible for these events still remain a mystery. Single-particle mass spectrometers have revealed that the internal and external state of mixing of particles is much richer than had previously been imagined. This richness of course is a challenge to explain but also provides a means of identifying the processes responsible for formation of particulate matter in the atmosphere. Combinations of older techniques such as controlled-temperature volatilization coupled to mobility analysis also have provided a wealth of insight, such as the frequent observation in the upper marine troposphere that particles consist of volatile material plus a core of refractory material that comprises a small fraction of the mass of the particles but whose number is conserved on heating. Likewise coupling of relative-humidity control to mobility and light scattering measurements is providing much insight to the hygroscopic properties of ambient aerosols.
To great extent the revolution in atmospheric aerosol research can be attributed also to a change in the way aerosol measurements and atmospheric chemistry measurements generally are being conducted. Investigators and funding agencies are recognizing that meaningful progress in understanding atmospheric chemistry processes requires a large number of measurement capabilities to be brought to bear at the same time. The recognition that aerosol particles are closely coupled chemically to the soup of reactive gases in which they are embedded has not only led aerosol scientists to recognize the need to describe the gas phase chemistry, but has also forced the gas-phase community to recognize that aerosol particles can influence gas-phase chemistry not just as sinks of reactive species, or by diminishing photo-actinic flux, but also as reaction vessels that can contribute reactive species back to the gas phase. The latter recognition of course has gained much support from the recognition of the crucial role of aerosol processes in the catalytic cycle responsible for springtime destruction of polar stratospheric ozone. A related advance in the past decade is the conduct of aerosol investigations over rather large arenas, thousands of kilometers or more, in order to capture significant aerosol processes that are taking place over such geographical scales. These studies require resources - ships, aircraft, surface-based stations as well as large numbers of investigators and specialized instruments - resources that are often beyond those available to single agencies or even single countries. This new model for carrying out aerosol research represents an important sociological change. Another positive sociological change is the open sharing of data among project participants and with the broader research community, thereby greatly enhancing the utilization and value of the data. The organizers of such large multinational projects are especially to be commended for coordinating activities and resources in such endeavors, as are funding agencies for making such activities possible. Quality control of the host of particle instruments remains a crucial problem that limits the gains of these large experiments as compared to focused small scale process studies.
Large-scale projects also require the active participation of meteorologists and of modelers who can provide a description of the governing transport fields that is necessary for the interpretation of the observations and recently even in forecast mode to aid in targeting aircraft missions. Much of the ability of the meteorologists to contribute to these studies relies on the availability of synoptic-scale meteorological data from the operational weather forecasting community. The availability of these data for these purposes relies in the first instance on the cooperation of this community; the open sharing of meteorological data and derived data products is a major international achievement. Timely utilization of meteorological data has also been greatly enhanced by the availability of high speed computers and data communications.
Although much aerosol research has traditionally focused on in-situ measurements, a major exception has been remote sensing by sun photometry. The past decade has seen the fielding of networks of well-calibrated instruments capable of measuring aerosol optical thickness with high accuracy. The availability of such data on the web greatly enhances their utility in developing aerosol climatologies and as observations against which models can be evaluated. Surface-based Lidars are beginning to become routinely employed in networks, and limited experience shows that these can provide vertical resolution that is otherwise not available, and if multiple wavelengths are employed, useful microphysical information. Another extremely valuable source of data to the aerosol research community is satellite-borne instruments. Most data so far have come from passive instruments such as radiometers; often the aerosol signal is obtained as a residual after subtraction of other known or assumed contributions, and products such as aerosol optical thickness are somewhat model dependent. Still the availability of such data has been a great stimulus to aerosol researchers. Active devices such as satellite-borne Lidars are in their infancy, but they hold great promise for the future.
An important new paradigm of atmospheric aerosol research that has taken hold in the past decade is that of so-called "closure" experiments in which a nominally over-determined set of measurements is made and then examined for self-consistency. Alternatively a closure experiment may be viewed as the testing of a hypothesis, where a subset of the measurements serves as input to a calculation, and the calculated quantity is examined for agreement with the observation within the uncertainties of the observations, the hypothesis under examination being the algorithm of the calculation. This approach is adding new rigor to aerosol field measurements.
The need for improved aerosol models to address questions ranging from local and regional air pollution to the direct and indirect radiative influence of aerosols is placing increasing demands on the aerosol modeling community. Most work to date has focused on particulate mass, particulate matter being treated as one or more chemical species analogous to a gaseous compounds, without treatment of the dynamics governing the size distribution. Substantial progress has been made at these various scales, as measured by the extent of agreement between modeled and measured concentrations of aerosol constituents. Because of the dominant influence of meteorological variability on the concentrations and geographical distributions of atmospheric substances such as aerosols having lifetimes of several days, it is increasingly becoming recognized that the meteorological driver that is used in such models must employ observationally derived data in order to permit meaningful evaluation of model performance by comparison with observations. This approach has met considerable success in urban regions for which there is an extensive emissions data base and detailed characterization of the three-dimensional wind field. At the subhemispheric scale measured and modeled daily-average concentrations of sulfate agree typically within a factor of two. Several groups are representing in global scale models mass loadings of key aerosol constituents--sulfate, nitrate, organics, elemental carbon, sea salt, dust--but the accuracy of these models is not yet well established. Major questions remain regarding the chemical species contributing to particle mass, the chemical reactions responsible for producing the condensable material contributing to gas-to-particle conversion, mechanisms responsible for new particle formation, and the role of clouds in evolution of particles. These questions can be resolved only by further research into these processes.
Considerable effort is being directed to developing models of aerosol microphysical processes to gain predictive capability for the aerosol size distribution and size distributed composition. Representation of size-resolved aerosols is necessary for accurate evaluation of radiative forcing, especially the indirect forcing, in which cloud and aerosol microphysics play the central role. Likewise, from the perspective of developing strategies to control local and regional particulate air pollution, it is necessary to develop model-based representations of the processes responsible for aerosol loading that can confidently relate this loading to sources of particulate matter and to other controlling variables. However in view of the still limited understanding of many of the processes that must be represented in the models it is clear that much effort must be directed to gaining enhanced understanding before these processes can confidently be represented in models.
A major objective of global-scale aerosol modeling is evaluation of aerosol influences on climate at present, over the industrial period, and for the future, for assumed emissions scenarios. Most such assessments to date have used temporal-average aerosol fields calculated in transport-transformation models that are run separately from the climate model that is used to assess the climate influence; that is, the aerosol field is calculated "off line", or have represented aerosol influences as changes in surface albedo, again based on aerosol fields calculated off-line. It is now recognized that such an approach may miss- represent key features of aerosol-climate interactions, for example correlations between relative humidity and particulate mass that might enhance direct forcing, or more intrinsically, the influence of aerosols on precipitation development and resultant changes in the hydrological cycle. For these reasons it is considered necessary ultimately to represent aerosols "on-line" in climate models. This may be expected to add considerable computational burden to climate models, and it will be a challenge to the aerosol modeling community to develop accurate and efficient methods of representing size- and composition- dependent aerosol processes suitable for incorporation in climate models.
10. References
Last modified: Thu Apr 27 10:21:40 CEST 2000