Volume 118, Issue 11 p. 5380-5552
Regular Article
Open Access

Bounding the role of black carbon in the climate system: A scientific assessment

T. C. Bond

Corresponding Author

T. C. Bond

University of Illinois at Urbana-Champaign, Urbana, Illinois, USA

Corresponding author: T. C. Bond, University of Illinois at Urbana-Champaign, Urbana, IL, USA. ([email protected])Search for more papers by this author
S. J. Doherty

S. J. Doherty

Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, Seattle, Washington, USA

Search for more papers by this author
D. W. Fahey

D. W. Fahey

NOAA Earth System Research Laboratory and Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado, USA

Search for more papers by this author
P. M. Forster

P. M. Forster

University of Leeds, Leeds, UK

Search for more papers by this author
T. Berntsen

T. Berntsen

Center for International Climate and Environmental Research-Oslo and Department of Geosciences, University of Oslo, Oslo, Norway

Search for more papers by this author
B. J. DeAngelo

B. J. DeAngelo

US Environmental Protection Agency, Washington, District of Columbia, USA

Search for more papers by this author
M. G. Flanner

M. G. Flanner

University of Michigan, Ann Arbor, Michigan, USA

Search for more papers by this author
S. Ghan

S. Ghan

Pacific Northwest National Laboratory, Richland, Washington, USA

Search for more papers by this author
B. Kärcher

B. Kärcher

Deutsches Zentrum für Luft- und Raumfahrt Oberpfaffenhofen, Wessling, Germany

Search for more papers by this author
D. Koch

D. Koch

US Department of Energy, Washington, District of Columbia, USA

Search for more papers by this author
S. Kinne

S. Kinne

Max Planck Institute, Hamburg, Germany

Search for more papers by this author
Y. Kondo

Y. Kondo

University of Tokyo, Tokyo, Japan

Search for more papers by this author
P. K. Quinn

P. K. Quinn

NOAA Pacific Marine Environment Laboratory, Seattle, Washington, USA

Search for more papers by this author
M. C. Sarofim

M. C. Sarofim

US Environmental Protection Agency, Washington, District of Columbia, USA

Search for more papers by this author
M. G. Schultz

M. G. Schultz

Forschungszentrum Jülich GmbH, Jülich, Germany

Search for more papers by this author
M. Schulz

M. Schulz

Norwegian Meteorological Institute, Oslo, Norway

Search for more papers by this author
C. Venkataraman

C. Venkataraman

Indian Institute of Technology, Bombay, India

Search for more papers by this author
H. Zhang

H. Zhang

China Meteorological Administration, Beijing, China

Search for more papers by this author
S. Zhang

S. Zhang

Peking University, Beijing, China

Search for more papers by this author
N. Bellouin

N. Bellouin

Met Office Hadley Centre, Exeter, UK

Search for more papers by this author
S. K. Guttikunda

S. K. Guttikunda

Division of Atmospheric Sciences, Desert Research Institute, Reno, Nevada, USA

Search for more papers by this author
P. K. Hopke

P. K. Hopke

Clarkson University, Potsdam, New York, USA

Search for more papers by this author
M. Z. Jacobson

M. Z. Jacobson

Stanford University, Stanford, California, USA

Search for more papers by this author
J. W. Kaiser

J. W. Kaiser

European Centre for Medium-range Weather Forecasts, Reading, UK; King's College London, London UK; Max Planck Institute for Chemistry, Mainz, Germany

Search for more papers by this author
Z. Klimont

Z. Klimont

International Institute for Applied System Analysis, Laxenburg, Austria

Search for more papers by this author
U. Lohmann

U. Lohmann

Eidgenössische Technische Hochschule Zürich, Zurich, Switzerland

Search for more papers by this author
J. P. Schwarz

J. P. Schwarz

NOAA Earth System Research Laboratory and Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado, USA

Search for more papers by this author
D. Shindell

D. Shindell

NASA Goddard Institute for Space Studies, New York, New York, USA

Search for more papers by this author
T. Storelvmo

T. Storelvmo

Yale University, New Haven, Connecticut, USA

Search for more papers by this author
S. G. Warren

S. G. Warren

University of Washington, Seattle, Washington, USA

Search for more papers by this author
C. S. Zender

C. S. Zender

University of California, Irvine, California, USA

Search for more papers by this author
First published: 15 January 2013
Citations: 243


[1] Black carbon aerosol plays a unique and important role in Earth's climate system. Black carbon is a type of carbonaceous material with a unique combination of physical properties. This assessment provides an evaluation of black-carbon climate forcing that is comprehensive in its inclusion of all known and relevant processes and that is quantitative in providing best estimates and uncertainties of the main forcing terms: direct solar absorption; influence on liquid, mixed phase, and ice clouds; and deposition on snow and ice. These effects are calculated with climate models, but when possible, they are evaluated with both microphysical measurements and field observations. Predominant sources are combustion related, namely, fossil fuels for transportation, solid fuels for industrial and residential uses, and open burning of biomass. Total global emissions of black carbon using bottom-up inventory methods are 7500 Gg yr−1 in the year 2000 with an uncertainty range of 2000 to 29000. However, global atmospheric absorption attributable to black carbon is too low in many models and should be increased by a factor of almost 3. After this scaling, the best estimate for the industrial-era (1750 to 2005) direct radiative forcing of atmospheric black carbon is +0.71 W m−2 with 90% uncertainty bounds of (+0.08, +1.27) W m−2. Total direct forcing by all black carbon sources, without subtracting the preindustrial background, is estimated as +0.88 (+0.17, +1.48) W m−2. Direct radiative forcing alone does not capture important rapid adjustment mechanisms. A framework is described and used for quantifying climate forcings, including rapid adjustments. The best estimate of industrial-era climate forcing of black carbon through all forcing mechanisms, including clouds and cryosphere forcing, is +1.1 W m−2 with 90% uncertainty bounds of +0.17 to +2.1 W m−2. Thus, there is a very high probability that black carbon emissions, independent of co-emitted species, have a positive forcing and warm the climate. We estimate that black carbon, with a total climate forcing of +1.1 W m−2, is the second most important human emission in terms of its climate forcing in the present-day atmosphere; only carbon dioxide is estimated to have a greater forcing. Sources that emit black carbon also emit other short-lived species that may either cool or warm climate. Climate forcings from co-emitted species are estimated and used in the framework described herein. When the principal effects of short-lived co-emissions, including cooling agents such as sulfur dioxide, are included in net forcing, energy-related sources (fossil fuel and biofuel) have an industrial-era climate forcing of +0.22 (−0.50 to +1.08) W m−2 during the first year after emission. For a few of these sources, such as diesel engines and possibly residential biofuels, warming is strong enough that eliminating all short-lived emissions from these sources would reduce net climate forcing (i.e., produce cooling). When open burning emissions, which emit high levels of organic matter, are included in the total, the best estimate of net industrial-era climate forcing by all short-lived species from black-carbon-rich sources becomes slightly negative (−0.06 W m−2 with 90% uncertainty bounds of −1.45 to +1.29 W m−2). The uncertainties in net climate forcing from black-carbon-rich sources are substantial, largely due to lack of knowledge about cloud interactions with both black carbon and co-emitted organic carbon. In prioritizing potential black-carbon mitigation actions, non-science factors, such as technical feasibility, costs, policy design, and implementation feasibility play important roles. The major sources of black carbon are presently in different stages with regard to the feasibility for near-term mitigation. This assessment, by evaluating the large number and complexity of the associated physical and radiative processes in black-carbon climate forcing, sets a baseline from which to improve future climate forcing estimates.

1 Executive Summary

1.1 Background and Motivation

[2] Black carbon is emitted in a variety of combustion processes and is found throughout the Earth system. Black carbon has a unique and important role in the Earth's climate system because it absorbs solar radiation, influences cloud processes, and alters the melting of snow and ice cover. A large fraction of atmospheric black carbon concentrations is due to anthropogenic activities. Concentrations respond quickly to reductions in emissions because black carbon is rapidly removed from the atmosphere by deposition. Thus, black carbon emission reductions represent a potential mitigation strategy that could reduce global climate forcing from anthropogenic activities in the short term and slow the associated rate of climate change.

[3] Previous studies have shown large differences between estimates of the effect of black carbon on climate. To date, reasons behind these differences have not been extensively examined or understood. This assessment provides a comprehensive and quantitative evaluation of black carbon's role in the climate system and explores the effectiveness of a range of options for mitigating black carbon emissions. As such, this assessment includes the principal aspects of climate forcing that arise from black carbon emissions. It also evaluates the net climate forcing of combustion sources that emit large quantities of black carbon by including the effects of co-emitted species such as organic matter and sulfate aerosol precursors. The health effects of exposure to black carbon particles in ambient air are not evaluated in this assessment.

1.2 Major Findings

1.2.1 Black Carbon Properties

  1. [4]

    Black carbon is a distinct type of carbonaceous material that is formed primarily in flames, is directly emitted to the atmosphere, and has a unique combination of physical properties. It strongly absorbs visible light, is refractory with a vaporization temperature near 4000K, exists as an aggregate of small spheres, and is insoluble in water and common organic solvents. In measurement and modeling studies, the use of the term “black carbon” frequently has not been limited to material with these properties, causing a lack of comparability among results.

  2. [5]

    Many methods used to measure black carbon can be biased by the presence of other chemical components. Measured mass concentrations can differ between methods by up to 80% with the largest differences corresponding to aerosol with low black carbon mass fractions.

  3. [6]

    The atmospheric lifetime of black carbon, its impact on clouds, and its optical properties depend on interactions with other aerosol components. Black carbon is co-emitted with a variety of other aerosols and aerosol precursor gases. Soon after emission, black carbon becomes mixed with other aerosol components in the atmosphere. This mixing increases light absorption by black carbon, increases its ability to form liquid-cloud droplets, alters its capacity to form ice nuclei, and, thereby, influences its atmospheric removal rate.

1.2.2 Black Carbon Emissions and Abundance

  1. [7]

    Sources whose emissions are rich in black carbon (“BC-rich”) can be grouped into a small number of categories, broadly described as diesel engines, industry, residential solid fuel, and open burning. The largest global sources are open burning of forests and savannas. Dominant emitters of black carbon from other types of combustion depend on the location. Residential solid fuels (i.e., coal and biomass) contribute 60 to 80% of Asian and African emissions, while on-road and off-road diesel engines contribute about 70% of emissions in Europe, North America, and Latin America. Residential coal is a significant source in China, the former USSR, and a few Eastern European countries. These categories represent about 90% of black-carbon mass emissions. Other miscellaneous black-carbon-rich sources, including emissions from aviation, shipping, and flaring, account for another 9%, with the remaining 1% attributable to sources with very low black carbon emissions.

  2. [8]

    Total global emissions of black carbon using bottom-up inventory methods are 7500 Gg yr1 in the year 2000 with an uncertainty range of 2000 to 29,000. Emissions of 4800 (1200 to 15000) Gg yr−1 black carbon are from energy-related combustion, which includes all but open burning, and the remainder is from open burning of forests, grasslands, and agricultural residues.

  3. [9]

    An estimate of background black carbon abundances in a preindustrial year is used to evaluate climate effects. In this assessment, we use the term “industrial era” to denote differences in the atmospheric state between present day and the year 1750. We use a preindustrial value of 1400 Gg of black carbon per year from biofuel and open biomass burning, although some fraction was anthropogenic at that time.

  4. [10]

    Current emission estimates agree on the major sources and emitting regions, but significant uncertainties remain. Information gaps include the amounts of biofuel or biomass combusted, and the type of technology or burning, especially in developing countries. Emission estimates from open biomass burning lack data on fuel consumed, and black-carbon emission factors from this source may be too low.

  5. [11]

    Black carbon undergoes regional and intercontinental transport during its short atmospheric lifetime. Atmospheric removal occurs within a few days to weeks via precipitation and contact with surfaces. As a result, black carbon is found in remote regions of the atmosphere at concentrations much lower than in source regions.

  6. [12]

    Comparison with remote sensing observations indicates that global atmospheric absorption attributable to black carbon is too low in many global aerosol models. Scaling atmospheric black carbon absorption to match observations increases the modeled globally averaged, industrial-era black carbon absorption by a factor of 2.9. Some of the model underestimate can be attributed to the models lacking treatment of enhanced absorption caused by mixing of black carbon with other constituents. The remainder is attributed to underestimates of the amount of black carbon in the atmosphere. Burden underestimates by factors of 1.75 to 4 are found in Africa, South Asia, Southeast Asia, Latin America, and the Pacific region. In contrast, modeled burdens in North America, Europe, and Central Asia are approximately correct. The required increase in modeled BC burdens is compatible with in situ observations in Asia and space-based remote sensing of biomass burning aerosol emissions.

  7. [13]

    If all differences in modeled black carbon abundances were attributed to emissions, total emissions would be 17,000 Gg yr1 compared to the bottom-up inventory estimates of 7500 Gg yr1. The industrial-era value of about 14,000 Gg yr1, obtained by subtraction of estimated preindustrial emissions, is used as the best estimate of emissions to determine final forcing values in this assessment. However, some of the difference could be attributed to poorly modeled removal instead of emissions. Both energy-related burning and open biomass burning are implicated in underestimates of emission rates, depending on the region.

1.2.3 Synthesis of Black-Carbon Climate Forcing Terms

  1. [14]

    Radiative forcing used alone to estimate black-carbon climate effects fails to capture important rapid adjustment mechanisms. Black-carbon-induced heating and cloud microphysical effects cause rapid adjustments within the climate system, particularly in clouds and snow. These rapid adjustments cause radiative imbalances that can be represented as adjusted or effective forcings, accounting for the near-term global response to black carbon more completely. The effective forcing accounts for the larger response of surface temperature to a radiative forcing by black carbon in snow and ice compared to other forcing mechanisms. These factors are included in the climate forcing values reported in this assessment.

  2. [15]

    The best estimate of industrial-era climate forcing of black carbon through all forcing mechanisms is +1.1 W m−2 with 90% uncertainty bounds of +0.17 to +2.1 W m−2. This estimate includes cloud forcing terms with very low scientific understanding that contribute additional positive forcing and a large uncertainty. This total climate forcing of black carbon is greater than the direct forcing given in the fourth Intergovernmental Panel on Climate Change (IPCC) report. There is a very high probability that black carbon emissions, independent of co-emitted species, have a positive forcing and warm the climate. This black carbon climate forcing is based on the change in atmospheric abundance over the industrial era (1750 to 2005). The black-carbon climate-forcing terms that make up this estimate are listed in Table 1. For comparison, the radiative forcings including indirect effects from emissions of the two most significant long-lived greenhouse gases, carbon dioxide (CO2) and methane (CH4), in 2005 were +1.56 and +0.86 W m−2, respectively.

  3. [16]

    The fossil fuel direct effect of black carbon of +0.29 W m2 is higher than the value provided by the IPCC in 2007. This increase is caused by higher absorption per mass and atmospheric burdens than used in models for IPCC. The black-carbon-in-snow forcing estimate in this assessment is comparable, although more sophisticated. Our total climate forcing estimate of +1.1 W m−2 includes biofuel and open-biomass sources of black carbon, as well as cloud effects that the IPCC report did not explicitly isolate for black carbon.

Table 1. Black Carbon Climate Forcing Terms, Evaluated for Industrial Era (1750–2005) Unless Otherwise Stated
Climate Forcing Term Forcing Components Forcing (W m−2)
(90% Uncertainty Range)
Black carbon direct effect Atmosphere absorption and scattering +0.71 (+0.09 to +1.26)
Direct radiative forcing split Fossil fuel sources +0.29
Bio fuel sources +0.22
Open burning sources +0.20
Black carbon cloud semi-direct and indirect effects Combined liquid cloud and semi-direct effect −0.2 (−0.61 to +0.10)
Black carbon in cloud drops +0.2 (−0.1 to +0.9)
Mixed phase cloud +0.18 (+0.0 to +0.36)
Ice clouds 0.0 (−0.4 to +0.4)
Combined cloud and semi-direct effects +0.23 (−0.47 to +1.0)
Black carbon in snow and sea-ice effects Snow effective forcing +0.10 (+0.014 to +0.30)
Sea-ice effective forcing +0.03 (+0.012 to +0.06)
Combined surface forcing terms +0.13 (+0.04 to 0.33)
Total climate forcingsa Black carbon only (all terms) +1.1 (0.17 to +2.1)
Net effect of black carbon + co-emitted species:
All sources −0.06 (−1.45 to +1.29)
Excluding open burning +0.22 (−0.50 to +1.08)
All source (includes pre-industrial) forcings Direct radiative forcing +0.88 (+0.18 to +1.47)
Snow pack effective forcing +0.12 (+0.02 to +0.36)
Sea-ice effective forcing +0.036 (+0.016 to +0.068)
  • a Note that the total best estimate is the median of the combined probability distribution functions across all terms, which differs from the mean of the best estimates.

1.2.4 Black-Carbon Direct Radiative Forcing

  1. [17]

    Direct radiative forcing of black carbon is caused by absorption and scattering of sunlight. Absorption heats the atmosphere where black carbon is present and reduces sunlight that reaches the surface and that is reflected back to space. Direct radiative forcing is the most commonly cited climate forcing associated with black carbon.

  2. [18]

    The best estimate for the industrial-era (1750 to 2005) direct radiative forcing of black carbon in the atmosphere is +0.71 W m2 with 90% uncertainty bounds of +0.08 to +1.27 W m2. Previous direct forcing estimates ranged from +0.2 to +0.9 W m−2, and the median value was much lower. The range presented here is altered because we adjust global aerosol models with observational estimates of black carbon absorption optical depth as done in some previous studies.

  3. [19]

    Direct radiative forcing from all present-day sources of black carbon (including preindustrial background sources) is estimated to be +0.88 W m2 with 90% uncertainty bounds of +0.17 to +1.48 W m2. This value is 24% larger than industrial-era forcing because of appreciable preindustrial emissions from open burning and biofuel use.

  4. [20]

    Estimates of direct radiative forcing are obtained from models of black carbon abundance and location. The ability to estimate radiative forcing accurately depends on the fidelity of these models. Modeling of and observational constraints on the black-carbon vertical distribution are particularly poor.

1.2.5 Black-Carbon Cloud Effects

  1. [21]

    Black carbon influences the properties of ice clouds and liquid clouds through diverse and complex processes. These processes include changing the number of liquid cloud droplets, enhancing precipitation in mixed-phase clouds, and changing ice particle number and cloud extent. The resulting radiative changes in the atmosphere are considered climate indirect effects of black carbon. In addition, in the semi-direct effect, light absorption by black carbon alters the atmospheric temperature structure within, below, or above clouds and consequently alters cloud distributions. Liquid-cloud and semi-direct effects may have either negative or positive climate forcings. The best estimates of the cloud-albedo effect and the semi-direct effect are negative. Absorption by black carbon within cloud droplets and mixed-phase cloud changes cause positive climate forcing (warming). At present, even the sign of black-carbon ice-cloud forcing is unknown.

  2. [22]

    The best estimate of the industrial-era climate forcing from black carbon cloud effects is positive with substantial uncertainty (+0.23 W m2 with a −0.47 to +1.0 W m2 90% uncertainty range). This positive estimate has large contributions from cloud effects with a very low scientific understanding and large uncertainties. The cloud effects, summarized in Table 1, are the largest source of uncertainty in quantifying black carbon's role in the climate system. Very few climate model studies have isolated the influence of black carbon in these indirect effects.

1.2.6 Black-Carbon Snow and Ice Effects

  1. [23]

    Black carbon deposition on snow and ice causes positive climate forcing. Even aerosol sources with negative globally averaged climate forcing, such as biomass combustion, can produce positive climate forcing in the Arctic because of their effects on snow and ice.

  2. [24]

    The best estimate of climate forcing from black carbon deposition on snow and sea ice in the industrial era is +0.13 W m2 with 90% uncertainty bounds of +0.04 to +0.33 W m2. The all-source present-day climate forcing including preindustrial emissions is somewhat higher at +0.16 W m−2. These climate forcings result from a combination of radiative forcing, rapid adjustments, and the stronger snow-albedo feedback caused by black-carbon-on-snow forcing. This enhanced climate feedback is included in the +0.13 W m−2 forcing estimate.

  3. [25]

    Species other than black carbon are a large fraction of the absorbing aerosol mass that reduces reflectivity of snow and ice cover. These species include dust and absorbing organic carbon; the latter is co-emitted with black carbon or may come from local soils.

  4. [26]

    The role of black carbon in the melting of glaciers is still highly uncertain. Few measurements of glacial black-carbon content exist, and studies of the impact on glacial snow melt have not sufficiently accounted for natural impurities such as soil dust and algae or for the difficulty in modeling regions of mountainous terrain.

1.2.7 Impacts of Black-Carbon Climate Forcing

  1. [27]

    The black-carbon climate forcings from the direct effect and snowpack changes cause the troposphere and the top of the cryosphere to warm, inducing further climate response in the form of cloud, circulation, surface temperature, and precipitation changes. In climate model studies, black-carbon direct effects cause equilibrium global warming that is concentrated in the Northern Hemisphere. The warming response to black-carbon-in-snow forcing is greatest during local spring and over mid-to-high northern latitudes. In terms of equilibrium global-mean surface temperature change, the BC total climate forcing estimate over the industrial era would correspond to a warming between 0.1 and 2.0K. Note that not all this warming has been realized in the present day, as the climate takes more than a century to reach equilibrium and many co-emitted species have a cooling effect, countering the global-mean warming of BC.

  2. [28]

    Regional circulation and precipitation changes may occur in response to black-carbon climate forcings. These changes include a northward shift in the Inter-Tropical Convergence Zone and changes in Asian monsoon systems where concentrations of absorbing aerosols are large. Black-carbon cloud indirect effects are also expected to induce a climate response. However, global models do not simulate robust responses to these complex and uncertain climate-forcing mechanisms.

1.2.8 Net Climate Forcing by Black-Carbon-Rich Source Categories

  1. [29]

    Other species co-emitted with black carbon influence the sign and magnitude of net climate forcing by black-carbon-rich source categories. The net climate forcing of a source sector is a useful metric when considering mitigation options. Principal co-emitted species that can change the sign of short-lived forcing are organic matter and sulfur species. The direct radiative forcing is positive for almost all black-carbon-rich source categories, even when negative direct forcings by sulfate and organic matter are considered. Liquid-cloud forcing by co-emitted aerosol species can introduce large negative forcing. Therefore, high confidence in net positive total climate forcing is possible only for black-carbon source categories with low co-emitted species, such as diesel engines.

  2. [30]

    The best estimate of the total industrial-era climate forcing by short-lived effects from all black-carbon-rich sources is near zero with large uncertainty bounds. Short-lived effects are defined as those lasting less than 1 year, including those from aerosols and short-lived gases. This total is +0.22 (−0.50 to +1.08) W m−2 for fossil fuel and biofuel-burning emissions and −0.06 W m−2 with 90% uncertainty bounds of −1.45 to +1.29 W m−2 when open burning emissions are included.

  3. [31]

    The climate forcings from specific sources within black-carbon source categories are variable depending upon the composition of emissions. Some subsets of a category may have net positive climate forcing even if the whole category does not. Selecting such individual source types from each category can yield a group of measures that, if implemented, would reduce climate forcing. However, the positive forcing reduction would be much less than the total climate forcing of +1.1 W m−2 attributable to all industrial-era black-carbon emissions in 2005.

  4. [32]

    Short-lived forcing effects from black-carbon-rich sources are substantial compared with the effects of long-lived greenhouse gases from the same sources, even when the forcing is integrated over 100 years. Climate forcing from changes in short-lived species in each source category amounts to 5 to 75% of the combined longer-lived forcing by methane, effects on the methane system, and CO2, when the effects are integrated over 100 years.

1.2.9 Major Factors in Forcing Uncertainty

  1. [33]

    Observational constraints on global, annual average black carbon direct radiative forcing are limited by a lack of specificity in attributing atmospheric absorption to black carbon, dust, or organic aerosol. These constraints are required to correct demonstrated biases in the distributions of atmospheric black carbon in climate models.

  2. [34]

    Altitude and removal rates of black carbon are strong controlling factors. They determine black-carbon absorption forcing efficiency, microphysical effects on clouds, and the sign of semi-direct effects. Models do not accurately represent black-carbon vertical distributions. Removal rates, particularly wet removal, affect most facets of black carbon forcing, including its lifetime, horizontal, and vertical extent, and deposition to the cryosphere.

  3. [35]

    Black-carbon emission rates from both energy-related combustion and biomass burning currently appear underestimated. Underestimates occur largely in Asia and Africa. Uncertainties in biomass burning emissions also affect preindustrial black-carbon emission rates and net forcing.

  4. [36]

    Black carbon effects on clouds are a large source of uncertainty. Models of liquid-cloud and semi-direct effects disagree on signs and magnitudes of forcing. However, potentially large forcing terms and uncertainties come from black carbon effects on mixed-phase clouds, cloud-absorption, and ice clouds, which have been estimated in a very small number of studies.

  5. [37]

    Estimates of forcing rely on accurate models of the Earth system. Black-carbon cloud interactions rely on fidelity in representation of clouds without black carbon present, and likewise, reductions in snow and sea ice albedo by black carbon depend on accurate representation of coverage of snow and sea ice. Uncertainties due to model biases of these distributions have not been assessed.

1.2.10 Climate Metrics for Black Carbon Emissions

  1. [38]

    The 100 year global-warming-potential (GWP) value for black carbon is 900 (120 to 1800 range) with all forcing mechanisms included. The large range derives from the uncertainties in the climate forcings for black carbon effects. The GWP and other climate metric values vary by about ±30% between emitting regions. Black-carbon metric values decrease with increasing time horizon due to the short lifetime of black carbon emissions compared to CO2. Black carbon and CO2 emission amounts with equivalent 100 year GWPs have different impacts on climate, temperature, rainfall, and the timing of these impacts. These and other differences raise questions about the appropriateness of using a single metric to compare black carbon and greenhouse gases.

1.2.11 Perspective on Mitigation Options for Black Carbon Emissions

  1. [39]

    Prioritization of black-carbon mitigation options is informed by both scientific and non-scientific factors. Scientific issues include the magnitude of black carbon emissions by sector and region, and net climate forcing including co-emissions and impacts on the cryosphere. Non-science factors, such as technical feasibility, costs, policy design, and implementation feasibility, also play roles. The major sources of black carbon are presently in different stages with regard to technical and programmatic feasibility for near-term mitigation.

  2. [40]

    Mitigation of diesel-engine sources appears to offer the most confidence in reducing near-term climate forcing. Mitigating emissions from residential solid fuels also may yield a reduction in net positive forcing. The net effect of other sources, such as small industrial coal boilers and ships, depends on the sulfur content, and net climate benefits are possible by mitigating some individual source types.

1.2.12 Policy Implications

  1. [41]

    Our best estimate of black carbon forcing ranks it as the second most important individual climate-warming agent after carbon dioxide, with a total climate forcing of +1.1 W m−2 (+0.17 to +2.1 W m−2 range). This forcing estimate includes direct effects, cloud effects, and snow and ice effects. The best estimate of forcing is greater than the best estimate of indirect plus direct forcing of methane. The large uncertainty derives principally from the indirect climate-forcing effects associated with the interactions of black carbon with cloud processes. Climate forcing from cloud drop inclusions, mixed phase cloud effects, and ice cloud effects together add considerable positive forcing and uncertainty. The relative importance of black carbon climate forcing will increase following reductions in the emissions of other short-lived species or decrease if atmospheric burdens of long-lived greenhouse gases continue to grow.

  2. [42]

    Black carbon forcing concentrates climate warming in the mid-high latitude Northern Hemisphere. As such, black carbon could induce changes in the precipitation patterns from the Asian Monsoon. It is also likely to be one of the causes of Arctic warming in the early twentieth century.

  3. [43]

    The species co-emitted with black carbon also have significant climate forcing. Black carbon emissions are primarily attributable to a few major source categories. For a subset of these categories, including diesel engines and possibly residential solid fuel, the net impact of emission reductions can be a lessening of positive climate forcing (cooling). However, the impact of all emissions from black-carbon-rich sources is slightly negative (−0.06 W m−2) with a large uncertainty range (−1.45 to +1.29 W m−2). Therefore, uniform elimination of all emissions from black-carbon-rich sources could lead to no change in climate warming, and sources and mitigation measures chosen to reduce positive climate forcing should be carefully identified. The uncertainty in the response to mitigation is larger when more aerosol species are co-emitted.

  4. [44]

    All aerosol that is emitted or formed in the lower atmosphere adversely affects public health. Mitigation of many of these sources would increase positive climate forcing (warming). In contrast, reduction of aerosol concentrations by mitigating black-carbon-rich source categories would be accompanied by very small or slightly negative changes in climate forcing. These estimates of climate forcing changes from source mitigation are associated with large uncertainties.

  5. [45]

    Forcings by greenhouse-gases alone do not convey the full climate impact of actions that alter emission sources. Black-carbon-rich source sectors emit short-lived species, primarily black carbon, other aerosols and their precursors, and long-lived greenhouse gases (e.g., CO2 and CH4). The total climate forcing from the short-lived components is a substantial fraction of the total (up to 75%) even when both short-lived and long-lived forcings are integrated over 100 years after emission.

2 Introduction

[46] In the year 2000, a pair of papers [Jacobson, 2000; Hansen et al., 2000] pointed out that black carbon—small, very dark particles resulting from combustion—might presently warm the atmosphere about one third as much as CO2. Because black carbon absorbs much more light than it reflects, it warms the atmosphere through its interaction with sunlight. This warming effect contrasts with the cooling effect of other particles that are primarily scattering and, thus, reduce the amount of energy kept in the Earth system. Radiative forcing (RF) by atmospheric BC stops within weeks after emissions cease because its atmospheric lifetime is short unlike the long timescale associated with the removal of CO2 from the atmosphere. Thus, sustained reductions in emissions of BC and other short-lived climate warming agents, especially methane and tropospheric ozone (O3), could quickly decrease positive climate forcing and hence climate warming. While such targeted reductions will not avoid climate change, their value in a portfolio to manage the trajectory of climate forcing is acknowledged in the scientific community [e.g., Molina et al., 2009; Ramanathan and Xu, 2010].

[47] Discussions of black carbon's role in climate rest on a long history. Urban pollution had been a concern for hundreds of years [Brimblecombe, 1977], and blackness was used as an indicator of pollution since the early 1900s [Uekoetter, 2005]. Black carbon was first isolated in urban pollution, as Rosen et al. [1978] and Groblicki et al. [1981] found that graphitic, refractory particles were responsible for light absorption. Shortly after McCormick and Ludwig [1967] suggested that aerosols could influence climate, Charlson and Pilat [1969] pointed out that aerosol absorption causes warming rather than cooling. The magnitude of climate effects was first estimated hypothetically to examine post-nuclear war situations [Turco et al., 1983] and later using realistic distributions from routine human emissions [Penner et al., 1993; Haywood and Shine, 1995]. It was known that particles traveled long distances from source regions [Rodhe et al., 1972], reaching as far as the Arctic [Heintzenberg, 1980]. International experiments organized to examine aerosol in continental outflow were initiated in the late 1990s [Raes et al., 2000]. They confirmed that absorbing aerosol was prevalent in some regions and an important component of the atmospheric radiation balance [Satheesh and Ramanathan, 2000]. This coincident confirmation of atmospheric importance and proposal of policies for mitigation triggered further debate.

[48] In the decade since the initial proposals, the speed of Arctic climate change and glacial melt has increased the demand for mitigation options which can slow near-term warming, such as reductions in the emissions of short-lived warming agents. The impact of air quality regulations that reduce sulfate particles is also being recognized. Most particles, including sulfates, cool the climate system, masking some of the warming from longer-lived greenhouse gases (GHGs) and BC. Thus, regulating these particles to protect human health may have the unintended consequence of increasing warming rapidly. BC also plays a direct role in surface melting of snow and ice and, hence, may have an important role in Arctic warming [Quinn et al., 2008]; if so, targeted reductions could have disproportionate benefits for these sensitive regions.

[49] Particulate matter was originally regulated to improve human health. Evidence supporting the link between particles and adverse respiratory and cardiovascular health continues to mount [Pope et al., 2009]. High human exposures to particulate matter in urban settings are linked to sources that emit black carbon [Grahame and Schlesinger, 2007; Naeher et al., 2007; Janssen et al., 2011] and to intense exposures in indoor air [Smith et al., 2010]. Thus, reducing particulate matter is desirable to improve human welfare, regardless of whether those reductions reduce climate warming.

[50] For the past few years, the opportunity to reduce black carbon has received pervasive policy attention at high levels. The G8 declaration, in addition to promising GHG reductions, is committed to “…taking rapid action to address other significant climate forcing agents, such as black carbon.” [9 July 2009, L'Aquila, Italy]. The Arctic Council, recognizing that “…reductions of emissions have the potential to slow the rate of Arctic snow, sea ice and sheet ice melting in the near-term…,” established a task force in 2009 to offer mitigation recommendations [29 April 2009, Tromsø, Norway] and “encouraged” the eight member states to implement certain black-carbon reduction measures [12 May 2011, Nuuk, Greenland]. The United States has complemented this international interest with passage of a bill [H.R. 2996] requiring a study of the sources, climate and health impacts, and mitigation options for black carbon both domestically and internationally. A proposed revision to the Gothenburg Protocol [UNECE, 1999] states that parties “should, in implementing measures to achieve their national targets for particulate matter, give priority, to the extent they consider appropriate, to emission reductions measures which also significantly reduce black carbon.” [UNECE, 2011]. In February 2012, the Climate and Clean Air Coalition was formed with the aim of reducing climate warming and air pollutants through action on short-lived pollutants—in particular, BC, methane, and hydrofluorocarbons (http://www.unep.org/ccac).

[51] The prospect of achieving quick climate benefits by reducing BC emissions is tantalizing, but the scientific basis for evaluating the results of policy choices has not yet been fully established. This assessment is intended to provide a comprehensive and quantitative scientific framework for such an evaluation. In the remainder of this section, we briefly define black carbon, present our terms of reference for the assessment, and describe its structure.

2.1 What Is Black Carbon?

[52] Black carbon is a distinct type of carbonaceous material, formed only in flames during combustion of carbon-based fuels. It is distinguishable from other forms of carbon and carbon compounds contained in atmospheric aerosol because it has a unique combination of the following physical properties:
  1. [53]

    It strongly absorbs visible light with a mass absorption cross section of at least 5 m2g−1 at a wavelength of 550 nm.

  2. [54]

    It is refractory; that is, it retains its basic form at very high temperatures, with a vaporization temperature near 4000K.

  3. [55]

    It is insoluble in water, in organic solvents including methanol and acetone, and in other components of atmospheric aerosol.

  4. [56]

    It exists as an aggregate of small carbon spherules.

[57] The strong absorption of visible light at all visible wavelengths by black carbon is the distinguishing characteristic that has raised interest in studies of atmospheric radiative transfer. No other substance with such strong light absorption per unit mass is present in the atmosphere in significant quantities. BC has very low chemical reactivity in the atmosphere; its primary removal process is wet or dry deposition to the surface. BC is generally found in atmospheric aerosol particles containing a number of other materials, many of which are co-emitted with BC from a variety of sources.

[58] In this assessment, the term “black carbon” and the abbreviation “BC” are used to denote ambient aerosol material with the above characteristics. Note that this definition of black carbon has not been used rigorously or consistently throughout most previous literature describing absorbing aerosol and its role in the atmosphere. Section 3 gives further discussion of terminology.

2.2 How Does Black Carbon Affect the Earth's Radiative Budget?

[59] Figure 1 illustrates the multi-faceted interaction of BC with the Earth system. A variety of combustion sources, both natural and anthropogenic, emit BC directly to the atmosphere. The largest global sources are open burning of forests and savannas, solid fuels burned for cooking and heating, and on-road and off-road diesel engines. Industrial activities are also significant sources, while aviation and shipping emissions represent minor contributions to emitted mass at the global scale. The difficulty in quantifying emissions from such diverse sources contributes to the uncertainty in evaluating BC's climate role. Once emitted, BC aerosol undergoes regional and intercontinental transport and is removed from the atmosphere through wet (i.e., in precipitation) and dry deposition to the Earth's surface, resulting in an average atmospheric lifetime of about a week.

Details are in the caption following the image
Schematic overview of the primary black-carbon emission sources and the processes that control the distribution of black carbon in the atmosphere and determine its role in the climate system.

[60] Radiative forcing over the industrial era (1750–present) has typically been used (e.g., by the IPCC [Forster et al., 2007]) to quantify and compare first-order climate effects from different climate change mechanisms. Many of BC's effects on clouds and within the cryosphere are not easily assessed within this framework. These effects result in rapid adjustments involving the troposphere and land surface that lead to a perturbed energy balance that can also be quantified in units of radiative forcing. We employ the term “climate forcing” to encompass both traditional radiative forcing and the rapid adjustment effects on clouds and snow (Table 2); this is discussed further in section 2.3.2.

Table 2. Definition of Climate Forcing and Response Terms
Forcing Term Definition Model Calculation
Climate forcing Generic term encompassing all forcing types below, quantifying a perturbation to the Earth's energy balance in W m−2
Radiative forcing (RF) RF is the change in the net vertical irradiance at the tropopause caused by a particular constituent. Usually, RF is computed after allowing for stratospheric temperatures to readjust to radiative equilibrium but with all tropospheric properties held fixed at their unperturbed values. Radiative forcing without stratospheric adjustment is called instantaneous. Difference between simulations:
(1) with radiative effect of the constituent change
(2) without radiative effect of constituent change
Held constant: All other tropospheric quantities, including cloud
Rapid adjustment Globally averaged flux change from the adjustment of the troposphere and land surface to a radiative forcing while holding the globally averaged surface temperature constant.a Energy balance perturbations arise from temperature, cloud, and constituent changes in the troposphere and from land-surface temperature and moisture changes. These changes occur within 1 year after a forcing is applied, usually within a few days but up to a season in the case of snowpack changes. For BC, the rapid adjustment to the direct effect is also called the “semi-direct effect.” The radiative forcing plus the rapid adjustment gives the “adjusted forcing” (see below). Difference between models:
(1) with radiative effect of constituent change and changes in cloud and land-surface temperature
(2) with radiative effect of constituent change and no tropospheric response
Held constant: global mean surface temperaturea
Adjusted forcing Flux perturbation from a given mechanism, allowing for changes in the stratosphere, troposphere, and some surface properties but not allowing a full response of global surface temperatures. This is the sum of a radiative forcing plus its rapid adjustment. Difference between models:
(1) with atmospheric constituent change and full atmospheric response
(2) without atmospheric constituent change
Held constant: sea-surface temperatures and/or global mean surface temperature responsea
Effective forcing A radiative forcing or adjusted forcing multiplied by its efficacy (see below) to give a climate forcing that is comparable to an equivalent climate forcing from a pre-industrial to 2× pre-industrial carbon dioxide change in terms of its globally averaged temperature response. Effective radiative forcing includes rapid adjustments, and it also accounts for differences in globally averaged responses due to latitudinal dependence of forcing. The efficacy is calculated for the constituent of interest in a climate model (see below). This is then multiplied by the associated radiative forcing or adjusted forcing to give the effective forcing.b
Climate response Large-scale long-term changes in temperature, snow and ice cover, and rainfall caused by a specific forcing mechanism. One of the most important climate responses is that of equilibrium globally averaged surface temperature (ΔT). Diagnostics from a global atmospheric climate model, coupled to either a mixed layer ocean for equilibrium experiments or a full ocean model for transient experiments
Climate sensitivity (λ) Equilibrium globally averaged surface warming per W m−2 of forcing (F), either radiative forcing or adjusted forcing. λ = ΔT/F. Computed from the climate response and radiative forcing diagnostics of a equilibrium climate model integration (see above)
Efficacy (E) Ratio of the climate sensitivity for a given forcing agent (λi) to the climate sensitivity for pre-industrial to 2× pre-industrial CO2 changes (i.e., Ei = λi / λCO2). Efficacy can then be used to define an effective forcing (=EiFi), where Fi can either be radiative forcing or the adjusted forcing. Computed from equilibrium climate model experiments with the constituent of interest, compared these with equivalent climate model diagnostics for a 2×CO2 experiment. Global mean equilibrium temperature and radiative forcing diagnostics are needed.
Climate impact Regional or local changes in weather and or climate indicators such as heat waves and storms that impact human livelihoods. Diagnosed from a climate model integration
  • a The adjusted forcing can be computed in different ways. Either regression can be used to determine the forcing at zero global temperature change or a fixed SST model forcing can be modified to account for a change in land temperatures, after Hansen et al. [2005], to give an estimate of the zero global surface T response forcing, or the fixed SST flux change can be used directly. The semi-direct effect is computed as the difference between the whole atmosphere adjusted forcing and the radiative forcing when the aerosol direct effect is included. The adjusted forcing for the cryosphere terms employs a different methodology (section 8.2.)
  • b For all changes apart from the snow and sea ice terms, we assume that this effective forcing is the same as the adjusted forcing. This is justified from Hansen et al. [2005] and Shine et al. [2003] who showed that for most forcings, the rapid adjustment accounted for the non-unity efficacy of the radiative forcing terms. However, for snow and sea ice changes, this is not the case. Their forcings directly influence surface snow and ice, and because they occur at high latitudes, the resulting heating is confined to the near surface. These forcings accelerate snow and ice melt, leading to a strong positive surface albedo feedback. These feedbacks lead to a very high efficacy that their associated rapid adjustments do not account for. Hence, the snow and sea-ice forcings are scaled to account for their enhanced climate response to give an effective forcing.

[61] The best quantified climate impact of BC is its atmospheric direct radiative forcing—the consequent changes in the radiative balance of the Earth due to an increase in absorption of sunlight within the atmosphere. When BC is located above a reflective surface, such as clouds or snow, it also absorbs solar radiation reflected from that surface. Heating within the atmosphere and a reduction in sunlight reaching the surface can alter the hydrological cycle through changes in latent heating and also by changing convection and large-scale circulation patterns.

[62] A particularly complex role of BC and other aerosols in climate is associated with changes in the formation and radiative properties of liquid water and ice clouds. BC particles may increase the reflectivity and lifetime of warm (liquid) clouds, causing net cooling, or they may reduce cloudiness, resulting in warming. Aerosol particles can change cloud droplet number and cloud cover in ice clouds, or in mixed-phase clouds made up of both ice and liquid water. Changes in droplet number may also alter cloud emissivity, affecting longwave radiation.

[63] BC also produces warming when it is deposited on ice or snow because BC decreases the reflectivity of these surfaces, causing more solar radiation to be absorbed. The direct absorption of sunlight produces warming which affects snow and ice packs themselves, leading to additional climate changes and ultimately to earlier onset of melt and amplified radiative forcing.

[64] An important consideration in evaluating the climate role of BC emissions is the role of co-emitted aerosols, aerosol precursors, and other gases. Many of these co-emitted species arise in the same combustion sources that produce BC. The greatest emissions by mass include sulfur-containing particles or precursors, organic aerosols that are directly emitted, organic compounds that are precursors to aerosols and ozone, nitrogen oxides that play roles in ozone formation and methane destruction and are precursors to nitrated aerosols, and long-lived GHGs. Sources also emit smaller quantities of ionic species such as potassium and chloride. With the exception of “brown” organic carbon, non-BC particles absorb little or no light, so they often cool rather than warm climate. They also play a role in many, but not all, of the same cloud processes as BC.

[65] In contrast to BC, most other aerosols and precursors are chemically reactive in the atmosphere. Because of transport and chemical and microphysical transformation after emission, the atmospheric aerosol becomes a complex array of atmospheric particles, some of which contain BC. Pure BC aerosol rarely exists in the atmosphere, and because it is just one component of this mixed aerosol, it cannot be studied in isolation. Compared with pure BC, mixed-composition particles differ in their lifetimes, interaction with solar radiation, and interactions with clouds. The components of these mixed particles may come from the same or different sources than BC.

[66] The overall contribution of natural and anthropogenic sources of BC to climate forcing requires aggregating the multiple aspects of BC's interaction with the climate system, as well as the climate impacts of constituents that are co-emitted with BC. Each contribution may lead to positive climate forcing (generally leading to a warming) or negative climate forcing (generally cooling). As discussed in the body of this assessment, BC impacts include both warming and cooling terms. While globally averaged climate forcing is a useful concept, BC concentration and deposition are spatially heterogeneous. This means that climate forcing by aerosols and climate response to aerosols is likely distributed differently than the forcings and responses of well-mixed GHGs.

2.3 Assessment Terms of Reference

[67] We use the term “scientific assessment” to denote an effort directed at answering a particular question by evaluating the current body of scientific knowledge. This assessment addresses the question: “What is the contribution of black carbon to climate forcing?” The terms of reference of this assessment include its scope and approach. The primary scope is a comprehensive evaluation of annually averaged, BC global climate forcing including all known forcing terms, BC properties affecting that forcing, and climate responses to BC forcing. Climate forcing of BC is evaluated for the industrial era (i.e.,1750 to 2000). A secondary evaluation addresses the potential interest of BC sources for mitigation. Therefore, we discuss the analyses and tools required for a preliminary evaluation of major BC sources: climate change metrics, net forcing for combined BC and co-emitted species, and factors relating to feasibility.

[68] Our approach relies on synthesizing results of global models from the published literature to provide central estimates and uncertainties for BC forcing. This analysis was guided by the principles of being comprehensive and quantitative, described in more detail as follows:
  1. [69]

    Comprehensiveness with regard to physical effect. As discussed in the foregoing section, BC affects multiple facets of the Earth system, all of which respond to changes in emissions. In evaluating the total climate forcing of BC emissions, we included all known and relevant processes. The main forcing terms are direct solar absorption; influence on liquid, mixed-phase, and ice clouds; and reduction of surface snow and ice albedo.

  2. [70]

    Comprehensiveness with regard to existing studies. Multiple studies have provided estimates of BC climate forcing caused by different mechanisms. These studies often rely on dissimilar input values and assumptions so that the resulting estimates are therefore not comparable. In order to include all possible studies, we sometimes harmonized dissimilar estimates by applying simplified adjustments.

  3. [71]

    Comprehensiveness with regard to source contribution. Atmospheric science has historically focused on individual pollutants rather than the net impacts of sources. However, each pollutant comes from many sources, and each source produces multiple pollutants. Mitigation of BC sources will reduce warming due to BC, but it will also alter emissions of cooling particles or their precursors; short-lived warming gases, such as ozone precursors; and long-lived GHGs. Multi-pollutant analyses of climate impacts have been demonstrated in other work, and we continue that practice here for key sources that account for most of the BC emissions. We include forcing for other pollutants emitted by BC sources by scaling published model results. Although such scaling may yield imprecise estimates of impact, we assert that ignoring species or effects could result in misconceptions about the true impact of mitigation options.

  4. [72]

    Quantification and diagnosis. For each aspect of BC climate forcing, we provide an estimate of the central value and of the uncertainty range representing the 90% confidence limits. When understanding of physical processes is sufficiently mature that the factors governing forcing are known, observations and other comparisons can assist in weighting modeled forcing estimates. Model sensitivity studies based on this physical understanding allow estimates of uncertainty. When the level of scientific understanding is low, an understanding of the dominant factors is not well established. In this situation, the application of observations to evaluate global models is not well developed, and only model diversity was used to estimate the uncertainty. When possible, we identified the causes of variation and key knowledge gaps that lead to persistent uncertainties. In this pursuit, we highlighted critical details of individual studies that may not be apparent to a casual reader. This synthesis and critical evaluation is one of the major value-added contributions of this assessment. The terms of reference require that new calculations be conducted if and only if required to harmonize diverse lines of evidence, including differences between simulations and observations.

[73] The target audience for this assessment includes scientists involved in climate, aerosol, and cloud research and non-specialists and policymakers interested in the role of BC in the climate system. The document structure reflects this audience diversity by including an Executive Summary (section 1), individual section summaries, and introductory material that is required to support understanding of principles.

2.3.1 Assessment Structure

[74] The remaining eleven sections of this document reflect the scope of this assessment. They include seven providing in-depth analysis of the science surrounding BC alone (sections 3, 3.1, 3.2, 3.2.1-3.2.3, 3.3-3.7, 3.7.1-3.7.5, 3.8, 3.8.1, 3.8.2,,, 3.8.3, 3.8.4, 3.9, 3.9.1, 3.9.2, 4, 4.1, 4.2, 4.2.1, 4.2.2, 4.3, 4.3.1-4.3.3, 4.4, 4.4.1-4.4.5, 4.5, 4.6, 4.6.1-4.6.5, 4.7, 4.7.1,, 4.7.2,, 4.8, 4.9, 4.9.1-4.9.3, 5, 5.1-5.4, 5.4.1,,, 5.4.2,,, 5.5, 5.5.1, 5.6, 5.6.1, 5.7, 5.8, 6, 6.1-6.5, 6.5.1, 6.5.2, 6.6, 6.6.1, 6.6.2, 6.7, 6.7.1-6.7.3, 6.8, 6.9, 6.9.1-6.9.3, 6.10, 7, 7.1-7.3, 7.3.1-7.3.3, 7.4, 7.5, 7.5.1, 7.5.2,, 7.5.3,, 7.6, 7.7, 7.7.1, 7.7.2, 8, 8.1-8.3, 8.3.1-8.3.7, 8.4, 8.4.1-8.4.6, 8.5, 8.6, 8.6.1-8.6.3, 9), a climate forcing synthesis (section 10), additional necessary context for discussions of the net climate forcing from BC-rich sources (section 11), a climate metrics analysis (section 12), and mitigation considerations for BC-rich sources (section 13). They are described in more detail as follows:

[75] 3. Measurements and microphysical properties of black carbon. The assessment begins with a review of BC-specific properties, including the techniques used to measure BC. The interactions of BC with the climate system depend upon its microphysical properties, optical properties, and mixing with other aerosol components. These govern all impacts shown in Figure 1.

[76] 4. Emission magnitudes and source categories. The origins and emission rates of BC are basic components of understanding its total impact. This section identifies major sources of BC and those containing high fractions of BC (i.e., “BC-rich sources”). It also identifies other climate-active aerosols or aerosol precursors emitted from these sources. Finally, data from ambient measurements are reviewed to evaluate emission estimates and the contributions of particular source types.

[77] 5. Constraints on black-carbon atmospheric abundance. The burden of BC in the atmosphere and its geographic distribution are basic quantities that directly affect all climate forcing estimates. Observations that constrain the magnitude and location of modeled atmospheric burdens are discussed here.

[78] 6. Black-carbon direct radiative forcing. The direct interaction between BC and sunlight unquestionably results in a net positive radiative forcing of climate. This section discusses the basic components that affect direct radiative forcing and the best estimates for each of these components, and it explores the reasons for differences in published radiative forcing values.

[79] 7. Black carbon interactions with clouds. BC influences clouds by changing droplet formation and microphysical properties and by altering the thermal structure of the atmosphere. BC is not uniformly distributed with altitude in the lower atmosphere. It directly warms the atmosphere where it is located and alters atmospheric dynamics, the meteorological conditions affecting cloud formation, and the quantity of clouds. In addition, BC, as well as other particles, influences the size and number of water droplets and ice crystals in water and ice clouds through microphysical interactions. All of these changes produce climate forcing by altering cloud properties. This section evaluates the magnitude of changes in water clouds, mixed-phase clouds, and ice clouds. The section emphasizes changes caused by BC alone, instead of the more common examination of cloud changes by all particles.

[80] 8. Cryosphere changes: Black carbon in snow and ice. This section evaluates BC that is removed from the atmosphere both in precipitation and through dry deposition and, thereby, is incorporated into surface snow and ice, reducing reflectivity. This initial radiative forcing is amplified by a series of rapid adjustments. This section evaluates modeled cryosphere forcing estimates, including discussion of the microphysical factors that affect radiative transfer in snow and ice packs, and model choices that affect the amplification of that forcing through rapid adjustments. The section also compares the sources and magnitudes of modeled cryospheric BC concentrations with observations and uses this comparison to scale model estimates of forcing for a best estimate.

[81] 9. Climate response to black carbon forcings. Forcing is a common measure of radiative impact but of ultimate concern is the climate response to BC, especially if it differs from that of other forcing agents. This section discusses the adjustments in the climate system that affect the efficacy of the forcing by BC in the atmosphere and cryosphere. It also reviews the sparse knowledge about how regional and global climate respond to changes in top-of-atmosphere (ToA) forcing and atmospheric heating.

[82] 10. Synthesis of black-carbon climate effects. In this section, best estimates of climate forcing from direct atmospheric light absorption, microphysical cloud changes, the rapid adjustment to direct atmospheric absorption, and the darkening of surface snow and ice by BC are combined with estimates of the forcing efficacy to estimate the total climate forcing of BC in the industrial era (1750 to 2000). Total present-day (i.e., “all-source”) forcing is also given for direct radiative forcing and snow and ice forcings.

[83] 11. Net climate forcing by BC-rich source categories. While the preceding sections examine the total climate forcing for BC emissions alone, many other species are co-emitted from BC sources, even when they are BC rich. This section examines the total climate forcing of BC-rich sources by quantifying the forcing per emission of all species from a given source.

[84] 12. Emission metrics for black carbon. One method of evaluating mitigation of BC versus mitigation of other climate-active species like greenhouse gases is to compare forcing per emitted mass of different compounds in a common framework. This comparison involves scientific issues as well as value judgments. This section summarizes metrics commonly used in climate policy discussions and provides metrics for BC based on the forcing values summarized in section 10 for direct use in the policy community.

[85] 13. Mitigation considerations for BC-rich sources. The preceding sections estimate the contributions to climate forcing of BC emissions alone and of the net effect from BC and co-emitted species from BC-rich sources. Future decisions regarding mitigation of climate forcing will include other considerations, including cost and availability of alternatives, additional benefits, and feasibility of implementation. This section discusses these elements for the major BC-rich sources, including regional differences among emission types and potential reduction policies. In addition, this section describes an evaluation framework that extends beyond purely scientific considerations, consistent with our goal of a comprehensive discussion. However, a thorough evaluation of the factors required for a complete mitigation analysis is beyond the scope of this assessment.

2.3.2 Use of Radiative Forcing Concepts

[86] In synthesizing results for BC climate forcing from the published literature, inconsistencies are regularly encountered in the use of radiative forcing concepts, conventions, and terminology. Our definitions for these terms as used in this assessment are listed in Table 2 along with descriptions of their derivation from climate models. Most discussions about climate change refer to measures that are variously called “radiative forcing” or “climate forcing.” Our use of the term “radiative forcing” follows the IPCC definition given by Forster et al. [2007] that keeps tropospheric and surface temperatures fixed. Many aerosol effects (e.g., rapid adjustments in aerosol, cloud, or snow distributions in response to the initial radiative forcing) can also be measured as changes in fluxes at the tropopause, making them amenable to comparison with radiative forcings. The sum of the radiative forcing and forcing due to these rapid adjustments yields the adjusted forcing. For the atmosphere and most of the land surface, such rapid adjustments occur within a few days of applying the forcing. For the cryosphere, there is more of a continuum of adjustment processes and timescales, and rapid adjustments are usually considered to occur on seasonal timescales or less. For comparability to other forcing agents such as long-lived GHGs, these forcings can be scaled by their efficacy to yield the “effective forcing.” We give the sum of radiative forcing and all other forcing-like terms the name “climate forcing,” and this usage is similar to that in the IPCC assessments [IPCC, 2007].

[87] Radiative forcing employed in IPCC reports assumes that anthropogenic impact is well represented by the difference between present day and the year 1750, the beginning of the industrial era. This may not be true for BC, where there is evidence of considerable anthropogenic biofuel and open burning before 1750. For the purposes of this assessment, the global climate forcings of BC and co-emitted species are evaluated from the beginning of the industrial era (1750). This definition gives forcing that can be compared with the temperature change since that time, without requiring attribution to a particular cause. Rather than assuming that this value represents the present-day contribution of humans to climate forcing, we refer to the difference between year-2000 forcing and year-1750 forcing as “industrial-era forcing.” For direct and cryosphere forcing, we also estimate forcing from all sources, even those that might have been ongoing before 1750. This is referred to as the “all-source” forcing.

2.3.3 Contrast With Previous Assessments

[88] Motivation to undertake this assessment derived, in part, from the scope of previous and concurrent assessments, namely, the IPCC Fourth Assessment Report (AR4) [IPCC, 2007] and the UNEP/WMO Integrated Assessment of Black Carbon and Tropospheric Ozone [UNEP/WMO, 2011a, 2011b], respectively. A comprehensive and quantitative evaluation of BC sources and associated climate forcing terms was beyond the scope of IPCC AR4 (see section 10.6 for more details). The primary focus in AR4 was the direct radiative forcing of fossil fuel sources of BC, and that forcing was estimated by weighting all studies equally without examining variation. Other terms beyond direct forcing were not evaluated; without these, even the sign of the total forcing cannot be estimated. The focus of the UNEP/WMO assessment was identification of mitigation strategies for short-lived climate forcing agents that would significantly reduce short-term anthropogenic climate forcing through examination with two climate models. Evaluation of model results in the context of observations, the microphysical properties of BC, and findings from other models were beyond the scope of the UNEP/WMO effort.

3 Measurements and Microphysical Properties of Black Carbon

3.1 Section Summary

  1. [89]

    The term “black carbon” has not been used rigorously or consistently in measurement studies or in modeling studies that use measurements. For future work, we recommend that the term “refractory black carbon,” or rBC, be adopted for the distinctive material defined herein as black carbon.

  2. [90]

    Either during or soon after emission, BC becomes internally mixed with other aerosol components such that it and other chemical species exist together within the same particle. This mixing can alter the optical properties of BC and influence its atmospheric lifetime and ability to form cloud droplets and ice crystals. Hence, the climate impacts of BC must be evaluated in the context of changes in its physico-chemical properties due to interactions with other aerosol components.

  3. [91]

    Freshly emitted BC particles are small in diameter and hydrophobic and, therefore, make very poor cloud condensation nuclei. Aging of BC after emission and associated accumulation of soluble mass increases the size and hygroscopicity of the internally mixed BC and enhances its cloud condensation nuclei (CCN) activity.

  4. [92]

    The mass absorption cross section of BC (MACBC) is a fundamental input to models of radiative transfer. Measured values for freshly generated BC fall within a relatively narrow range of 7.5 ± 1.2 m2g−1 at 550 nm. MACBC increases by approximately 50% as BC becomes internally mixed with other aerosol chemical components.

  5. [93]

    Filter-based measurements of both absorption coefficient and BC mass concentrations can be biased by the presence of other chemical components in internally or externally mixed aerosol. Measured mass concentrations can differ between methods by up to 80% with the largest differences corresponding to aerosol with lower BC to organic carbon (OC) ratios. These measurement uncertainties may confound our understanding of trends, spatial and temporal variability, and impacts on climate.

3.2 Definitions

3.2.1 Black Carbon

[94] Black carbon (BC) is distinct from other forms of carbon and carbon compounds contained in atmospheric aerosol. As stated in section 2.1, it has a unique combination of properties. These properties are strong visible light absorption of at least 5 m2g−1 at 550 nm [Bond and Bergstrom, 2006], refractory with vaporization temperature near 4000K [Schwarz et al., 2006], aggregate morphology [Medalia and Heckman, 1969], and insolubility in water and common organic solvents [Fung, 1990]. Its absorption is consistent with a wavelength-independent refractive index across the visible spectrum [Marley et al., 2001]. The combination of these properties distinguishes BC from other light absorbing material, such as some organic carbon compounds. The individual spherules are formed in flames, and the aggregate nature is caused by rapid coagulation [Haynes and Wagner, 1981], differentiating black carbon from planar graphite.

[95] In this assessment, the term “black carbon” and the notation “BC” are used to denote ambient aerosol material that has the above characteristics. The term “black carbon” given here has not been used rigorously or consistently throughout all previous modeling and measurement literature, and as we discuss later in this section, most measurement techniques do not respond only to this substance. Measurements and nomenclature that uniquely identify this substance would assist in more rigorous evaluation.

3.2.2 Organic Matter and Other Carbon Aerosols

[96] “Organic aerosol” (OA) is a broad term indicating carbon-containing compounds that contain hydrogen and, usually, oxygen. All sources that emit BC also emit primary organic aerosol (POA), as well as gases that may become secondary organic aerosol (SOA) in the atmosphere. In atmospheric chemistry, the combination of BC and OA is often called “carbonaceous aerosol.” As used in this assessment, this term excludes primary organic particles directly emitted from plants [Heald and Spracklen, 2009] and organisms such as fungi and bacteria, which tend to be much larger in size than BC. It also excludes biogenic SOA, which may have small sizes but are emitted from natural, non-combustion sources [Guenther et al., 2006]. This usage is consistent with previous IPCC reports [Forster et al., 2007]. Gelencsér [2005] further describes the chemical nature of OA. Mineral dust also contains carbon as carbonate, which is also not included in the carbon aerosol considered in this assessment.

[97] The term “organic carbon” (OC) refers to the carbon mass within OA, excluding the associated oxygen and hydrogen content. Although OA is the quantity most relevant to climate, measurements and emission inventories have usually reported values of OC because it was more commonly measured. The ratio between OA and OC mass (OA : OC ratio) depends on the amount of oxygen incorporated in the organic molecules. It varies from about 1.1 to 2.2 [Russell, 2003], depending on the combustion source, with lower values of OA : OC from coal or diesel and higher values from biomass combustion. A default value for POA : OC of 1.3 or 1.4 is often assumed in global modeling [Penner et al., 1998; Dentener et al., 2006].

3.2.3 Other Particulate Light Absorbers

[98] Two other types of atmospheric particles absorb visible light: dust and “brown carbon.” Most of the dust in the atmosphere originates from deserts [e.g., Chin et al., 2009], and smaller amounts come from construction, on-road and off-road traffic, and agriculture. Dust particles are more weakly absorbing per mass than BC: about 0.009 m2g−1 at 550nm for Asian dust [Clarke et al., 2004]. They can be distinguished from BC particles because they are typically large (i.e., greater than 2 µm in diameter), crystalline, and composed of crustal elements. They also have relatively more absorption at shorter wavelengths compared with long visible wavelengths. Although dust is more weakly absorbing than BC, globally averaged total absorption by dust is significant due to its relatively high mass abundance [Sokolik and Toon, 1996].

[99] Brown carbon, a subset of OA, is a complex mixture of organic compounds lacking a formal analytical definition. Its light absorption is weak, with MAC less than 1 m2g−1 at 550 nm, and has a strong wavelength dependence [Kirchstetter et al., 2004; Chakrabarty et al., 2010]. This strong wavelength dependence can be used to distinguish its absorption from that of BC [Wonaschütz et al., 2009]. Unlike BC, brown carbon is soluble in some organic compounds and responds to analytical techniques that use solubility to isolate humic-like substances [Andreae and Gelencsér, 2006; Graber and Rudich, 2006]. Brown carbon particles and BC are similar in size.

3.3 Black Carbon Formation and Evolution

[100] BC is produced during the combustion of carbon-based fuels when oxygen is insufficient for complete combustion. Even if adequate oxygen is supplied overall, fuel-rich, oxygen-poor zones can occur when the reactants are not well mixed. A complex series of reactions involving polycyclic aromatic hydrocarbon molecules forms precursors of BC. These precursors coagulate to sizes large enough to serve as particle nuclei and grow through reactions on the surface. Electron microscopy images show that these spherules are unique among atmospheric particles, with wrinkled graphite layers forming a shell around a hollow or disordered interior [Heidenreich et al., 1968]. They have diameters on the order of tens of nanometers [e.g., Martins et al., 1998a; Pósfai et al., 1999; Li et al., 2003] and high carbon-to-hydrogen ratios. Soon after formation, the graphitic spherules coagulate to form aggregates or fractal chain-like structures consisting of hundreds or thousands of spherules [e.g., Medalia and Heckman, 1969; Li et al., 2003] (Figure 2). If the combustion exhaust is kept hot, and if sufficient oxygen is well mixed with the flame products, these carbon particles may be eliminated by oxidation reactions before they leave the combustion chamber [e.g., Lee et al., 1962]. Otherwise, they are emitted.

Details are in the caption following the image
(a) Scanning electron microscope image of BC aggregates in young smoke from the Madikwe Game Reserve fire, South Africa, on 20 August 2000; (b) transmission electron microscope (TEM) image of chain-like BC aggregates in flaming smoke from the dambo fire near Kaoma, Zambia, on 5 September 2000; (c) TEM image of a compact BC aggregate in regional haze near Skukuza, South Africa, on 22 August 2000 [Li et al., 2003]. (Reproduced with permission of the American Geophysical Union).

[101] The morphology of emitted chain-like aggregates changes rapidly after emission. Water vapor and other gas-phase species condense upon the aggregates, which collapse into more densely packed clusters [Huang et al., 1994; Ramachandran and Reist, 1995; Weingartner et al., 1997; Martins et al., 1998b]. Other particulate and gas phase species present in the surrounding atmosphere also coagulate with the combustion aerosol. Electron microscopy images and measurements of optical properties indicate that freshly emitted particles often exist as an external mixture in which organic and inorganic light-scattering components and strongly light-absorbing components (BC) reside in different particles [Pósfai et al., 2003; Li et al., 2003; Mallet et al., 2004]. After emission, condensation and coagulation cause individual chemical components to become internally mixed (i.e., different aerosol components existing together within a single particle). Such particles are no longer pure BC but contain sulfate and organic material [Lee et al., 2002; Shiraiwa et al., 2007]. We refer to these internally mixed particles as “BC-containing” particles. Mixing has been observed to occur within a few hours after emission at some locations [Moteki et al., 2007; Moffet and Prather, 2009], but there are insufficient measurements to estimate the extent of internal mixing throughout the atmosphere. Global aerosol models that simulate microphysical processes predict that most BC is mixed with other substances within 1 to 5 days [Jacobson, 2001a] and this prevalent internal mixing is found at all altitudes [Aquila et al., 2011].

[102] The top portion of Figure 3 summarizes particle properties that are used as input into global and radiative transfer models and that can be constrained by measurements. Each characteristic may affect the representation of atmospheric processes and resulting modeled concentrations, as shown in the lower portion of Figure 3. Pure BC and BC-containing particles are separated in Figure 3 because their microphysical, optical, hygroscopic, and cloud-nucleating properties differ [Abel et al., 2003; Slowik et al., 2004]. For example, as BC ages, it becomes coated or internally mixed with non-BC components and the resulting BC-containing particles become more hydrophilic, which can lead to a reduced lifetime and atmospheric loading [Stier et al., 2006b]. Internal mixing with other compounds can enhance absorption of solar radiation according to microphysical models [Fuller et al., 1999; Jacobson, 2001a; Lack and Cappa, 2010] and measurements, primarily laboratory based [Schnaiter et al., 2005; Slowik et al., 2007]. As a result, the additional material in BC-containing particles must be taken into account when modeling both radiative transfer and atmospheric lifetime [Stier et al., 2006a].

Details are in the caption following the image
Schematic of the connections between properties of BC and BC-containing particles. A combination of these properties determines the contribution of BC and BC-containing particles to climate forcing. The properties depend on those of other substances produced in the atmosphere or co-emitted with BC and on atmospheric processes such as nucleation and condensation. Mass and number of BC and BC-containing particles (extensive variables) depend, in part, on particle properties that affect the lifecycle of BC (not represented here).

[103] In the remainder of this section, we discuss the physical properties that affect estimates of climate forcing, measurements of these properties, and representation of these properties in global models. We begin with a discussion of BC mass concentration in section 3.4. Important optical properties of BC, including absorption, are discussed in section 3.5, followed by a review of absorption measurements in section 3.6. Section 3.7 reviews microphysical properties affecting the mass absorption cross section and single-scattering albedo, while section 3.8 discusses measurements and modeling of these two parameters. Finally, section 3.9 discusses properties of BC and BC-containing particles that are relevant to nucleating liquid cloud droplets. The discussion of microphysical properties, which also affect the interaction of BC-containing particles with ice clouds, appears in section 7.5.

3.4 Measurement of BC Mass

[104] Ambient air samples of particles that contain BC always contain other constituents as well. The mass fraction of BC in atmospheric aerosol is typically less than 10%. Thus, BC mass concentration (in g m−3) cannot be measured directly by collecting aerosol in an air sample and weighing it. Instead, BC mass must be determined indirectly, usually through optical methods, thermal heating combined with optical methods, or laser-induced incandescence (Table 3). BC is associated with high sp2-bonded carbon content [Hopkins et al., 2007] and with Raman spectroscopic responses similar to graphite [Rosen et al., 1979; Dippel et al., 1999], but these analyses do not respond uniquely to all materials identified as BC.

Table 3. Techniques for Measuring BC Mass, Method-Defined Terminology for BC, and Sources of Systematic Biases
Technique Common Name Suggested Name Common Instruments Sources of Bias Direction of Biasa
Optical absorption with in situ detection BC Equivalent BC (BC + other absorbing material) Photoacoustic Change in MAC with aerosol agingb +, −
Presence of other absorbers (dust, organic carbon)c +
Optical absorption by collection on filters BC Equivalent BC Aethalometer; Particle Soot Absorption Photometer (PSAP); Same as optical absorption above, plus:
Optical interactions between particles and filter matrixd +
Multi-angle Absorption Photometer (MAAP) Modification of particle morphology by organic carbon +
Optical absorption with heated inlet BC Equivalent BC Continuous Soot Monitoring System (COSMOS) Charring of low volatility organic species +
Thermal heating and optical absorption Elemental carbon (EC) Apparent elemental carbon (ECa) Thermal Optical Reflectance (TOR); Thermal Optical Transmittance (TOT) Failure to accurately correct for charred organic carbon; +
Catalytic oxidation of BC in presence of metals or metal oxides;
Absorption by charred material affects split between OC and ECe
Detection of less volatile organic carbon +
Laser-induced incandescence BC or Refractory BC (rBC) Refractory BC (rBC) Single Particle Soot Photometer (SP2) Lack of detection of small particlesf
  • a (+) = BC measured too high; (−) = BC measured too low.
  • b If calibrated with fresh, unmixed BC, measurement overestimates BC mass after mixing with non-absorbing species and increase in absorption.
  • c Depends on wavelength; measurements around 630 nm often are unaffected unless dust concentrations are high.
  • d Positive bias if uncorrected filter-based measurements are used; otherwise, it can be either positive or negative.
  • e More extreme for transmission-based monitoring of charring.
  • f Particles with mass below about 0.5 fg, corresponding to a spherical particle of about 80 nm diameter.

[105] The measurement of BC mass concentrations by some methods can be biased when BC is sampled with other aerosol components. This mixing can take place either within ambient aerosol particles before sampling or in the sample itself if BC and other particles are collected and measured simultaneously. Biases occur either because the mixing increases absorption or extinction (section 3.6), because BC is incorrectly classified as another material, or because another material is incorrectly classified as BC (i.e., lack of specificity). Most techniques measure similar BC mass concentrations when they are applied to pure BC or when other aerosol components are removed by applying heat [Knox et al., 2009; Kondo et al., 2009] or by solvent rinsing [Subramanian et al., 2006]. For that reason, measurements of untreated samples are usually in closer agreement for diesel exhaust, which contains little non-BC material, than they are for aged (i.e., internally mixed) aerosol [Schmid et al., 2001; Chow et al., 2004; Hitzenberger et al., 2006] or biomass burning aerosol (BB), which has a high content of organic matter and other inorganic substances that could cause interferences [Novakov and Corrigan, 1995; Reid et al., 1998]. The major uncertainty in some measurements of BC mass is associated with isolating BC from the other constituents with which it is internally or externally mixed. A summary of measurement techniques for BC mass is listed in Table 3 along with common terminology and any directions and causes of bias. In addition, more explicit names for the measured quantities are suggested; some of these follow Andreae and Gelencsér [2006]. Details of the measurement of BC mass concentrations are given below.

[106] The most common separation between BC and OC is accomplished by volatilizing and combusting material collected on a filter and by detecting the CO2 produced as a function of temperature. The discrimination between these two classes of aerosol is based on the idea that BC is non-volatile or refractory, whereas OC is volatile. This thermal method does not detect BC directly, and the amount of refractory material detected depends on the details of the method. The analytical result is therefore an operational definition and is traditionally called elemental carbon (EC).

[107] In thermal methods based on volatility, the sample filter is first heated in inert gas to volatilize OC and then heated again with oxygen to combust the EC; cooling sometimes occurs before the second heating [e.g., Chow et al., 1993; Birch and Cary, 1996; Watson et al., 2005]. This technique measures all OC, not just combustion-derived primary aerosol. Inaccuracies in this method result from interpreting OC as EC, or vice versa, when properties of the sample cause EC to be released too early or OC to be held too long. One complication is “charring” of OC (i.e., conversion of OC to EC) at high temperatures, which reduces its volatility. Variations of thermal methods include different temperature ramping programs and correcting for the charring of OC during pyrolysis by monitoring the optical reflectance of the sample filter [Huntzicker et al., 1982] or light transmission [Turpin et al., 1990]. Comparisons among protocols with differing temperature and optical corrections show that derived EC concentrations can differ by over an order of magnitude [Schmid et al., 2001] and that large differences can be caused by the lack of correction for charring, which leads to considerable overestimates of EC. There are also significant differences between methods that correct for charring using optical reflectance or light transmission [Chow et al., 2001; Chow et al., 2004]. The two corrections yield comparable EC concentrations if the filter contains a shallow surface deposit of EC. If EC and OC are distributed throughout the filter, the two corrections yield different EC values, and the variability also depends on the temperature protocol used. Hence, the difference between the two methods depends, in part, on the OC/EC ratio in the sample. As a result, the correction schemes yield similar results for diesel exhaust, which is dominated by EC, but can differ widely for complex atmospheric mixtures. Optimization of temperature programs seeks to minimize charring and release EC and OC during the expected period of the analysis [Conny et al., 2003; Cavalli et al., 2010]. The accuracy of thermal measurements can also be increased by extracting OC with organic solvents before analysis, since BC is insoluble [Fung, 1990; Subramanian et al., 2006].

[108] Detection of refractory BC (rBC) mass by laser-induced incandescence (i.e., visible thermal radiation) has become increasingly widespread [Schwarz et al., 2006, 2008a; Moteki and Kondo, 2010]. The Single Particle Soot Photometer (SP2) is able to provide continuous, real-time, size and coating information for individual particles containing rBC over a wide dynamic range of mass concentration. In the SP2, the rBC component of individual particles is heated to vaporization temperatures (i.e., about 4000K) with an infrared intra-cavity laser, and incandescence proportional to rBC mass is detected. Essentially all rBC particles between 80 and 700 nm mass-equivalent diameter are measured, assuming void-free rBC has a density of 2 g cm−3. The response of the SP2 to rBC mass and the lack of interference in measuring rBC mass when non-rBC aerosol components are present have been evaluated in the laboratory [Schwarz et al., 2006; Slowik et al., 2007; Moteki and Kondo, 2007]. The dominant uncertainty in the measured mass lies in calibrating its sensitivity to ambient rBC material. The range of this sensitivity is around 15% [Moteki and Kondo, 2010; Laborde et al., 2012]. However, the limitations of the size range detected by the SP2 may introduce additional uncertainty, depending on the air mass. Typically, in remote regions, the size range captured by the SP2 contains most of the rBC mass and about 50% of the rBC number [Schwarz et al., 2008b; Shiraiwa et al., 2008]. In urban environments, a correction to total measured mass concentration of about 25% may be required. Pulsed laser-induced incandescence has also been used for environmental rBC measurements [Chan et al., 2011].

3.5 Optical Properties of Black Carbon

[109] Models of radiative transfer require the amount of particulate absorption and scattering in the atmosphere, known as the absorption and scattering coefficients (m2 m−3 or simply m−1). Atmospheric models convert modeled mass concentrations to these optical coefficients using intensive properties (i.e., optical cross section per mass in m2g−1) known as the mass absorption cross section (MAC) and mass scattering cross section (MSC). Estimates of these values are needed at all wavelengths, and the directional distribution of scattering (dimensionless) is also needed for radiative transfer models.

[110] From atmospheric measurements, MAC can be calculated in reverse: the light-absorption coefficient divided by mass concentration. Throughout this assessment, we often refer to MAC that is determined for BC alone (MACBC). MACBC is calculated by dividing the absorption coefficient attributable to BC by the BC mass concentration. The simple term MAC indicates the value determined by dividing by the total mass concentration of BC-containing particles, which is smaller than MACBC. All other properties are usually measured for BC-containing particles, not for pure BC. Optical properties depend on refractive index, density, size distribution, mixing state, and particle shape. The propensity for water uptake affects the MSC and MAC of BC-containing particles, as well as their ability to form cloud droplets and their atmospheric lifetime due to removal by precipitation. For sub-saturated conditions (i.e., relative humidity below 100%), this water uptake is characterized in terms of hygroscopicity or growth factor. When air is supersaturated, the climate-relevant quantity is the fraction of particles that act as CCN [Hallett et al., 1989].

[111] The MAC was mentioned earlier as a distinguishing feature of BC. Values of MAC and MSC are fundamental inputs to radiative transfer models, required for all aerosols or aerosol components. These quantities are necessary to translate mass concentrations simulated by chemical transport models to their effects on radiative transfer. The wavelength dependence of MAC must also be represented in models for the full solar spectrum. As its name implies, BC strongly absorbs light at all visible wavelengths. In contrast, other atmospheric aerosols that absorb light (OA, soil, and dust) are more yellow, brown, or red, meaning they absorb more blue light than red light. The quantity generally used to characterize the spectral dependence of light absorption is the absorption Ångstrom exponent:
where MAC(λ1) and MAC(λ2) are the mass absorption cross sections at wavelengths λ1 and λ2, respectively. Two wavelengths spanning the visible range are commonly used, such as 450 and 650 nm. Alternatively, Åabs can be calculated from absorption coefficients at the two wavelengths. The value of Åabs for particles is usually greater than that of the bulk material. If the refractive index of the bulk material has no wavelength dependence, as does graphite [Borghesi and Guizzetti, 1991], then Åabs=1 for particles much smaller than the wavelength of light [Rosen et al., 1979; Moosmüller et al., 2011].

[112] Measurement studies have confirmed the value of Åabs = 1 in regions where externally mixed BC dominates absorption [Rosen et al., 1978; Bergstrom et al., 2002, 2007; Kirchstetter et al., 2004; Clarke et al., 2007]. When BC becomes coated, Åabs can theoretically be as low as 0.8 or as high as 1.9 [Lack and Cappa, 2010]. In contrast, Åabs for OA has been observed to be between 3.5 to 7 [e.g., Kirchstetter et al., 2004; Sun et al., 2007; Lewis et al., 2008; Yang et al., 2009]. Åabs for dust is typically about 2 to 3 but can be higher for very red (iron-rich) dust [e.g., Fialho et al., 2006; Alfaro et al., 2004; Bergstrom et al., 2007]. This difference in the wavelength dependence of absorption of BC versus other absorbing aerosol has been used to approximate relative fractions of BC versus other light-absorbing constituents.

[113] The single-scattering albedo, ω0, is scattering divided by extinction (i.e., the sum of scattering and absorption) or

[114] Values of ω0 near 1 indicate that the aerosol is mainly scattering. Values below about 0.8 indicate that the particles could have a net warming effect [Haywood and Shine, 1995]. The value that divides warming from cooling also depends on the albedo of the underlying surface or clouds and the fraction of light that is scattered upward by the particles [Chýlek and Wong, 1995]. When the addition of aerosol causes a local increase in the planetary albedo, more shortwave radiative energy is reflected back to space, and aerosol exerts a negative forcing. In contrast, when aerosols locally decrease the planetary albedo, the forcing is positive. MAC and ω0 are the aerosol properties most relevant to the balance between negative and positive forcing, so we emphasize these two parameters in this section instead of MSC. Forcing is not very sensitive to ω0 for strongly absorbing aerosol with values below 0.4.

3.6 Measurement of Absorption Coefficients

[115] Comprehensive reviews of absorption measurements have been provided by Horvath [1993] and by Moosmüller et al. [2009]. Here we summarize only the major challenges in these measurements. The most widely used technique to measure the absorption coefficient involves collecting aerosol on a filter and inferring atmospheric absorption from the resulting change in transmission of light through the filter [Gundel et al., 1984], often at one mid-visible wavelength but sometimes at multiple wavelengths. Common instruments using this approach are the Particle Soot Absorption Photometer (PSAP) [Bond et al., 1999; Virkkula et al., 2005], which has been used to obtain a worldwide data base through the Global Atmospheric Watch program; the Hybrid Integrating Plate System (HIPS), which has been used to collect data by the Interagency Monitoring of Protected Visual Environments (IMPROVE) in U.S. National Parks [Malm et al., 1994]; and the Aethalometer [Hansen et al., 1982]. Other filter-based absorption instruments include the Integrating Plate [Lin et al., 1973] and the Multi-Angle Absorption Photometer (MAAP) [Petzold et al., 2005b]. These filter-based methods overestimate absorption if light transmission is also affected by particulate light scattering [Horvath, 1997; Bond et al., 1999]. With the application of empirical corrections to overcome this artifact, accuracies of the PSAP, IP, HIPS, and Aethalometer range between 20 and 30% [Bond et al., 1999; Weingartner et al., 2003, Virkkula et al., 2005], and accuracy of the MAAP is about 12% [Petzold et al., 2005b]. However, these correction schemes are based on laboratory-generated aerosols that may limit their application and accuracy for the measurement of atmospheric aerosols. In addition, coating of BC with volatile compounds can greatly contribute to variation in filter-based measurements of light absorption [Lack et al., 2008; Cappa et al., 2008; Kondo et al., 2009]. For example, PSAP absorption coefficients can be biased high (50 to 80%) when the ratio of organic aerosol to BC is high (15 to 20). Lack et al. [2008] postulated that this high bias was due to the redistribution of liquid-like OC that affected either light scattering or absorption. This difficulty can be overcome by removing most of the mass of volatile aerosol components before the BC particles are collected on filters. This removal can be accomplished without significant charring of organic compounds with a heated sampling inlet. This technique is used for the Continuous Soot Monitoring System (COSMOS) instrument [Kondo et al., 2011b]. Under those conditions, comparisons of ambient BC mass concentrations measured by light absorption (COSMOS), thermal-optical measurements, and laser-induced incandescence (i.e., SP2) measurements have been found to agree within 10% [Kondo et al., 2011b].

[116] Filter-based optical measurements of absorption are sometimes used to derive “effective” BC mass concentrations by using an assumed MACBC to convert measured absorption to BC mass [e.g., Sharma et al., 2002]. Given the artifacts and uncertainties associated with filter-based measurements and the choice of MACBC, the resulting BC mass concentrations also can be highly uncertain.

[117] Techniques that do not use filters are also available for the measurement of absorption by BC. In the photoacoustic spectrometer (PAS) [Petzold and Niessner, 1996; Arnott et al., 1997; Lack et al., 2006], particles are drawn into an acoustic cavity and irradiated by power-modulated laser light. The heat that is produced when the particles absorb laser light is transferred to the surrounding gas creating an increase in pressure. Sensitive microphones are used to detect the standing acoustic wave that results from the pressure change, and this signal is interpreted to infer the absorption coefficient. Gas phase absorbers can interfere with BC detection in PAS systems. The overall uncertainty of the PAS with respect to aerosol absorption has been reported at about 5% [Lack et al., 2006].

[118] Absorption techniques, whether filter based or non-filter based, are not specific for BC. Any light-absorbing aerosol other than BC that absorbs at the measurement wavelength is detected. The degree to which other light absorbing species interfere with the measurement of BC absorption depends on the relative abundance of light absorbing species and the size range and wavelength of the measurement. Measurement of the chemical composition of the aerosol and measurement of absorption at several wavelengths can help determine the interference from all atmospheric species. BC-containing aerosol generally fall into the submicron size range, so measurements of the submicron aerosol only can also reduce the interference of dust absorption.

3.7 Microphysical Properties Affecting MAC and ω0

3.7.1 Optical Models and Required Inputs

[119] Values of MAC and ω0 can be predicted with several theories that describe how the particles interact with light. Particle density and refractive index of the bulk material are required to calculate MAC for all models. The assumed particle shape and size affect the model chosen. Common optical-modeling treatments that account for the fractal nature of fresh BC particles include the Rayleigh-Debye-Gans approximation [Nelson, 1989; Dobbins and Megaridis, 1991], superposition T-matrix theory [Mishchenko et al., 2004], and discrete-dipole approximation [Draine and Flatau, 1994]. Sorenson et al. [2001] reviews many of the key optical relationships for aggregates. These models are useful for exploring how particle properties affect particle optics [e.g., Liu and Mishchenko, 2005].

[120] Global-model calculations of radiative transfer usually do not account for complex particle shapes. Instead, they use Mie theory, which can describe homogeneous or core-shell spherical particles. Although BC particles, especially freshly emitted ones, are not spherical, modeling radiative transfer accurately requires only that MAC, ω0, and angular scattering are correct.

[121] BC non-sphericity creates difficulties in inferring other properties used in optical modeling from measurements. Air drag is different between spherical and non-spherical particles and affects inferences of particle size based on mobility [Lall and Friedlander, 2006]. The SP2 measurement detects BC mass and provides equivalent spherical diameters. Inferences of refractive index are often taken from optical measurements with the assumption of spherical particles or from particles that are collected and compacted into a pellet. In either case, the material is not pure BC but contains an unknown fraction of voids. Ideally, the refractive index, density, and size used for modeling would be obtained for the pure material, but observed properties may be obtained for material with unknown and inconsistent void fractions. Despite these uncertainties, measurements of material properties and sizes are described below.

3.7.2 Density

[122] Fuller et al. [1999] compiled reported densities for several types of graphitic material and found values ranging from 0.625 to 2.25 g cm−3. Based on the type of BC emitted from diesel combustion, they adopted the highest density, which was measured for paracrystalline graphite, as representative of strongly absorbing atmospheric BC. BC is not perfectly crystalline, so its microstructure, density, and refractive index differ from those of graphite. The density of pure graphite is 1.9 to 2.1 g cm−3 [Hess and Herd, 1993]. Densities for pressed pellets of BC with corrections for air volume fractions in the surface layer are slightly lower at 1.8 to 1.9 g cm−3 [Medalia and Richards, 1972; Janzen, 1980]. Park et al. [2004] reported a density of 1.8 g cm−3, and Kondo et al. [2011b] found 1.718 ± 0.004 g cm−3 for fullerene soot. Some global models still use the density of 1 g cm−3 recommended by OPAC (Optical Properties of Aerosols and Clouds), which would result in an overestimate of absorption if all other factors are correct.

3.7.3 Refractive Index

[123] The refractive index of a material is critical in determining the scattering and absorption of light, with the imaginary part of the refractive index having the greatest effect on absorption. Refractive index values are derived by assuming a theory for the interaction of BC with light (i.e., reflectance or absorption) and adjusting optical parameters (including refractive index) until predictions match measurements. Three possible methods include (1) fitting Fresnel's formula to reflectance and transmittance data measured for a compressed pellet, (2) fitting Mie theory to light-scattering data for individual spherical particles, and (3) fitting either Mie theory or an approximation formula to scattering and extinction data for particle ensembles. Method 1 has been used widely to estimate refractive indices of solid aerosols like combustion-generated BC [Mullins and Williams, 1987]. However, direct evidence of the optical flatness of pellet surfaces, which is necessary for application of Fresnel's formula, has never been shown for wavelengths shorter than the infrared region [Janzen, 1979]. Method 2, which is limited to spherical particles, uses measurements of resonance structures in Mie scattering to determine the refractive index [Chýlek et al., 1983a], but its application to strongly absorbing particles is limited. In method 3, the refractive index and a size distribution function are inferred simultaneously from extinction or scattering data for an ensemble of particles. Solving this inversion problem requires a theory to connect microphysical and light-scattering properties; Mie theory has been used for spherical particles [Lack et al., 2009] and the Rayleigh-Gans approximation for non-spherical particles [Charalampopoulos et al., 1989; Van-Hulle et al., 2002]. For polydisperse or non-spherical particles, the inversion results may have large errors if the assumed size distributions or shapes differ from those of the actual particles. Moteki et al. [2010] developed a method to estimate refractive indices that accounted for non-spherical particles by measuring the relationship between the scattering cross section and the particle volume.

[124] A variety of values for the refractive index of BC has been used in global climate models including the OPAC value of 1.74 − 0.44i [Hess et al., 1998]. As reviewed by Bond and Bergstrom [2006], reported values of the refractive index of light absorbing carbon vary widely; the real part, n, appears to vary from that of water to that of diamond, and the imaginary part, k, varies from that of negligibly absorbing material to that of graphite. Bond and Bergstrom [2006] hypothesize that strongly absorbing carbon with a single refractive index exists and that some of the variation in reported values results from void fractions in the material. Based on agreement between measured real and imaginary parts of the refractive index of light absorbing carbon, Bond and Bergstrom [2006] recommended a value of 1.95 − 0.79i at 550 nm. They caution, however, that this value may not represent void-free carbon. Stier et al. [2007] found that this value led to better agreement with observed atmospheric absorption, compared with the OPAC value. Moteki et al. [2010] found that the refractive index of ambient BC in the Tokyo urban area was about 2.26 − 1.26i at 1064 nm. The OPAC assumption for imaginary refractive index is not taken from combustion-generated particles, is lower than either of the latter two recommendations, and would lead to a MAC prediction about 30% lower if all other factors were equal.

[125] Radiative inversion methods can also infer values of refractive indices from remote-sensing measurements [Dubovik and King, 2000], but only for the entire mixed aerosol, which includes water and many other constituents other than BC. Currently, such data are available from about 500 globally distributed Aerosol Robotic Network (AERONET) surface sites [Holben et al., 1998].

3.7.4 Particle Size

[126] Size distributions of the BC component in ambient aerosol are affected by the size of BC at emission and by subsequent coagulation. Condensation of non-BC material changes the overall aerosol size distribution, but the underlying size distribution of BC does not change except through coagulation. BC size can be diagnosed separately. The first measurements of ambient BC size distributions relied on impactors, which identify the size at which greatest mass appears [Mallet et al., 2003; Riddle et al., 2008]. SP2 measurements have provided BC number distributions in fresh urban plumes dominated by fossil fuel (FF) combustion [e.g., Kondo et al., 2011c; Schwarz et al., 2008b], in aged plumes in Asian outflow [Shiraiwa et al., 2008], and in the remote upper troposphere and lower stratosphere [Schwarz et al., 2006; 2008b]. In the urban areas of Tokyo and Nagoya (Japan) and Seoul (Korea), the mass median diameter (MMD) and count median diameter (CMD) of fresh BC ranged from 120 to 160 nm and 50 to 80 nm, respectively. In plumes associated with wildfires, the MMD was measured to be about 200 nm and the CMD to be 120 nm [Kondo et al., 2011a]. The distinct difference in the size between BC particles from fossil fuel combustion and biomass burning seems to be a general feature (Figure 4). The CMD and MMD of BC observed in the Asian outflow were significantly higher than those in fresh urban plumes [Shiraiwa et al., 2008], suggesting the growth of BC size by coagulation during transport after emissions. BC particles are largely found in the Aitken mode (i.e., less than 100 nm diameter) and the accumulation mode because of their formation mechanism. Large concentrations of BC in the Aitken mode can result from high combustion temperatures and efficient fuel burn. For example, BC particles produced by aircraft jet engines have mean number diameters of about 30 nm [Petzold et al., 2005a].

Details are in the caption following the image
Mass and number size distributions of BC particles observed in three fresh urban (red) and two fresh biomass burning (black) plumes as identified in the legend. The measurements are made on board an aircraft using an in situ, single-particle detection instrument (SP2). The variable coatings on the BC particles is not shown. The observed (a) mass and (b) number amounts are plotted as symbols versus volume equivalent diameter based on assuming a spherical particle shape. The mass distributions are normalized to the same peak value. The observations are fit by a lognormal function between 90 and 600 nm (solid lines). The number distribution fits are those consistent with the fit to the respective mass distribution and are scaled to represent the same BC mass. From Schwarz et al. [2008b].

3.7.5 Configuration of BC-Containing Particles

[127] Internal mixing between BC and other compounds increases absorption of visible light, in part because the non-absorbing material can refract light toward the absorbing particle [Ackerman and Toon, 1981]. Because BC is insoluble, it is always distinctly separated from the other material in an internally mixed particle. It is often assumed that non-BC material surrounds the BC completely and approximately symmetrically, known as a “core-shell” configuration. However, BC may also exist near the surface of the particle [Sedlacek et al., 2012]. The presence, relative quantities, and location of non-absorbing material in BC-containing particles require a complex characterization commonly summarized in the single term “mixing state.”

3.8 Measured and Modeled MAC and ω0

[128] As discussed above, optical models can either represent aggregates or rely upon Mie theory with the assumption of spherical particles. Here we review measured values of MACBC and ω0, and discuss whether models can simulate these values.

3.8.1 MACBC Derived From Measurements

[129] Empirical values of MACBC have been obtained from measurements of light absorption by filter or photoacoustic methods, divided by measurements of EC mass from thermal evolution methods [e.g., Martins et al., 1998a; Moosmüller et al., 2001]. Measured values for freshly generated BC, where care has been taken to eliminate non-BC material, fall within a relatively narrow range of 7.5 ± 1.2 m2g−1 [Clarke et al., 2004; Bond and Bergstrom, 2006]. The MACBC may be reduced as the particles collapse into more compact forms with higher fractal dimensions [Schnaiter et al., 2003; Lewis et al., 2009]. Instrumental artifacts, as shown in Table 3, can hamper accurate inferences of MACBC.

[130] Laboratory measurements show that absorption increase is low for very thin coatings [Slowik et al., 2007] and reaches a factor of 1.8 to 2 for thicker coatings [Schnaiter et al., 2005; Khalizov et al. 2009; Shiraiwa et al., 2010; Cross et al., 2010; Bueno et al., 2011]. Filter-based and photoacoustic measurements report different degrees of enhancement in the measured absorption [Slowik et al., 2007; Knox et al., 2009].

[131] Values of MACBC enhancement are harder to confirm for atmospheric aerosol and require measuring MACBC before and after removal of coatings. Knox et al. [2009] found enhancement by a factor of 1.2 to 1.6 near source regions. Lack et al. [2012] observed a factor of 1.4 enhancement for BC in biomass burning plumes. In contrast, similar measurements of absorption enhancement using a PAS in large urban centers revealed an average absorption enhancement of only 6%, although the MACBC of aged aerosol was 20 to 40% higher than that of freshly generated BC [Cappa et al., 2012]. Cappa et al. hypothesize that the low values of enhancement are caused by BC inclusions at the edge of the sampled particles. Measurements made downwind (up to hundreds of kilometers) of BC source regions also give a wide range of enhancements. Chan et al. [2011] found MACBC values ranging from 10 to 50 m2g−1, showing a rough relationship with the ratio between non-BC and BC components. This study, which derived time-resolved MACBC from a photoacoustic spectrometer and laser-induced incandescence instrument system, appeared to contradict earlier work by the same group [Chan et al., 2010] that found no apparent increase in MACBC using a PSAP to measure absorption and thermal-optical technique to measure EC. MACBC values inferred from filter-based techniques are more plentiful but may suffer from artifacts; Cheng et al. [2011] summarized MACBC from several studies ranging from 2 to 17 m2g−1 but used empirical correction factors to adjust some of the methods prior to calculating MACBC. A series of intensive experiments with identical sampling and analysis protocols gave MACBC values ranging from 6 to 20 m2g−1 [Quinn and Bates, 2005].

[132] Most of the above measurements of MACBC after mixing with non-BC material were made under dry conditions or at relatively low humidity. However, water is also a component of mixed aerosols, both in clear air at high relative humidity and in clouds. Particles that have taken up water become even larger than mixed, dry atmospheric particles, but measurements at such elevated humidities are difficult. Mikhailov et al. [2006] measured a threefold increase in absorption of BC at 100% relative humidity. Brem et al. [2012] also found an amplification factor of up to 2.7 for an absorbing organic material at 95% relative humidity.

3.8.2 Microphysical Model Ability to Simulate MAC MACBC of Unmixed BC

[133] Particle size distribution has almost no influence on calculated MAC when particle diameter is below about 80 nm [Bergstrom, 1973]. The simplest theory that accounts for aggregate particles indicates that MAC depends only on the size of the component spherules. Therefore, modeled MACBC should not be very sensitive to particle size if either small, spherical particles are assumed or if modeled aggregates are composed of spherules below 80 nm. However, measured MACBC values, at 7.5 ± 1.2 m2g−1, are about 30% higher than values calculated with either method, using best estimates of refractive indices and density. This comparison remains unchanged even with the higher refractive index values found by Moteki et al. [2010]. Bond and Bergstrom [2006] suggest that the 30% higher absorption of actual particles could be caused by interactions between neighboring BC spherules [Iskander et al., 1991; Fuller, 1995]. Liu and Mishchenko [2005] calculated a 15 to 25% increase for many changes in aggregate parameters, compared with no interaction but up to 20% decrease for spherule sizes larger than 30 nm. Kahnert [2010] showed that spherule interaction alone could not account for the discrepancy of 30%, and MAC modeled by Adachi et al. [2010] was similar for aggregates and spheres. The inability of microphysical models to reproduce measured MAC from best estimates of refractive index and density remains unresolved.

[134] Although MACBC is sensitive to the large differences in both refractive index and density used by different models, Bond and Bergstrom [2006] note some compensating errors. For models using OPAC refractive index and density, the low refractive imaginary index would underestimate MACBC, while the low assumed density (1 g cm−3) would overestimate it. Therefore, the MACBC simulated with the OPAC refractive index and density is comparable to MAC using the values of refractive index and density recommended by Bond and Bergstrom [2006], although the model inputs are quite different. However, it may not be possible to simulate MAC and ω0 of mixed particles using these erroneous values. MAC of BC Mixed With Other Substances

[135] Internal mixtures of BC and non-absorbing material, including water, are usually modeled either as a core-shell configuration or as a homogeneous particle for which the effective refractive index is obtained using “mixing rules.” Some frequently used mixing rules are volume mixing, in which refractive indices are proportional to substance volume; Bruggeman or Maxwell-Garnet effective medium approximations [Heller, 1965]; or the dynamic effective medium approximation [Chýlek et al., 1996].

[136] Different representations of mixing often produce comparable estimates of absorption. For example, using the Bond and Bergstrom [2006] refractive index and density, Adachi et al. [2010] calculated MACBC for uncoated spheres as 6.4 m2g−1, with increases for volume mixing (13.6 m2g−1), core shell (13.3 m2g−1), Maxwell-Garnet effective medium approximation (12.0 m2g−1), and realistic coated BC particles (9.9 m2g−1) at 550 nm wavelength. The assumption of a perfectly concentric core within a shell gives the highest absorption among core-shell particles [Fuller, 1995]. However, Jacobson [2006] showed that the dynamic effective medium approximation suggested for cloud droplets produced a much higher MACBC than a core-shell treatment. The absorption efficiency of BC may also be higher when it is coated by water, such as in cloud droplets (section 7.3.3).

[137] The optics of mixed particles vary widely in space and time because they depend on the relative concentrations of BC and non-BC components, which may include water, liquid, and solid components. A single value of MAC for an internally mixed aerosol is not appropriate due to the widely varying amounts of non-absorbing material surrounding BC cores. Moffet and Prather [2009] estimate that the increase above unmixed aerosol could be 80 to 200% for individual particles. This increase depends upon the relative sizes of the shell and the core, and the location of the core within the particle [Fuller, 1995]. Bond et al. [2006] suggest a smaller enhancement, concluding that MACBC is 80% higher for mixed than for unmixed spheres. However, because the absorption of aggregate BC appears to be 30% higher than that of spheres, the increase between uncoated, aggregate BC to coated, mixed spherical particles might be only 50%. The MACBC of 7.5 m2g−1 for freshly emitted BC, plus an enhancement of 50%, agrees approximately with observed MACBC at dry conditions. A more precise evaluation of the agreement between models and observations is hampered by artifacts in the measurements.

3.8.3 Measured and Modeled Single-Scattering Albedo

[138] Single-scattering albedo of freshly generated BC has been measured as 0.10 to 0.28, as summarized by Bond and Bergstrom [2006]. Khalizov et al. [2009] and Cross et al. [2010] report similar values of ω0 for fresh BC. Models of aggregate, pure-BC particles predict ω0 of 0.1 to 0.3 [Liu and Mishchenko, 2005]. This value depends strongly on component spherule size but not on overall particle size [Sorensen, 2001]. Scattering and ω0 for spherical BC depend greatly on the size distribution chosen. For example, ω0 for particles of 50 and 250 nm diameter is 0.02 and 0.44, respectively, for a refractive index of 1.95 − 0.79i. Forcing by BC scattering is small compared to that by BC absorption, so small variations in ω0 produce only small changes in total forcing.

[139] As BC becomes mixed with other components, a large increase in scattering and, hence, in ω0 occurs [Khalizov et al., 2009; Cross et al., 2010]. For mixed particles, values of ω0 range from that of fresh particles to 0.9, depending upon the amount of added material. For these particles, forcing can be quite sensitive to ω0. However, much of the added scattering and the tendency toward negative forcing are attributable to other chemical components, not to BC. Therefore, the forcing by BC alone should be estimated with the difference of two radiative-transfer scenarios, not just the change in ω0.

3.8.4 Aerosol Properties in Global Models

[140] In global simulations of radiative transfer, BC particles are assumed to be either externally or internally mixed with other aerosol components. For external mixing, the material properties of pure BC are used. If internal mixing is assumed, the non-BC material is either treated as a spherical shell around a BC core, or the refractive index and density of BC and non-BC material are averaged and a homogeneous particle is modeled. Koch et al. [2009b] summarize predicted MACBC and treatments of aging, removal, and optical properties for many global models.

[141] All global models consider how aging affects removal rates of BC. Early models of atmospheric BC expressed the aging of BC by prescribing a timescale for converting hydrophobic BC to hydrophilic BC, typically about 1 day [Koch et al., 2009b]. Many models have evolved to express aging explicitly in terms of coagulation with sulfate particles and condensation of sulfuric acid and secondary organic vapor on BC. Some models represent BC-containing aerosol with discrete size-resolved bins [Jacobson, 2001a]. Others, known as “multi-modal” models, represent different aerosol classes, such as unmixed BC and BC coagulated with other material [Whitby and McMurry, 1997]. Representations of removal rates vary from constant empirical values, to parameterization based on composition and particle number [Abdul-Razzak and Ghan, 2000], to explicit dependence on aerosol and cloud droplet size and composition [Jacobson, 2002].

[142] Although global models of BC consider mixing when determining removal rates, some do not incorporate enhanced MACBC due to mixing. Values of MACBC used in global models range from 2.3 to 18 m2g−1 [Koch et al., 2009b; Jacobson, 2012]. The diversity in MACBC arises either from whether the mixing state is assumed or calculated, from whether it is determined at ambient RH or constant RH, or from other choices of aerosol properties. Many of the values are similar to or lower than the value for unmixed BC and are, therefore, lower than the average value of BC in the atmosphere.

3.9 CCN Activity of Black Carbon

3.9.1 Measurements of CCN Activity

[143] Particle size, hygroscopicity, and mixing state also affect the interaction of particles and clouds [Pruppacher and Klett, 1997; McFiggans et al., 2006]. While BC-induced cloud changes are mainly discussed in section 7, we review the processes of CCN activation here because they are closely related to the microphysical properties of BC-containing aerosol.

[144] Aerosol particles serve as nucleation sites for forming cloud droplets through a process known as activation. The ability of an aerosol particle to act as a cloud condensation nucleus depends on its size, composition, and mixing state, and the supersaturation with respect to water vapor within the cloud. The critical supersaturation for a given particle is the lowest supersaturation at which that particle activates and produces a cloud droplet. Any particle can activate in extremely supersaturated air. Particles that activate more easily (i.e., that have a lower critical supersaturation) have a greater chance of affecting cloud droplet number concentration and cloud reflectivity, and they may also be more easily removed by wet deposition. Early studies found that soluble species were incorporated in cloud droplets more often than BC [e.g., Hallberg et al., 1994] but could not confirm whether this was caused by differences in activation or in scavenging.

[145] If all other factors are equal, small particles require greater supersaturation to activate than do large particles. Less hygroscopic particles have larger critical supersaturations than more hygroscopic particles. Because freshly emitted BC particles are small in diameter and hydrophobic, they have very large critical supersaturations and make very poor CCN [Dusek et al., 2006a]. Aging of BC after emission lowers the critical supersaturation of the BC-containing particle, as the addition of soluble mass increases both particle volume and hygroscopicity. Figure 5 shows the dependence of the critical supersaturation on particle diameter and BC mass fraction as simulated by a particle-resolved model [Riemer et al., 2009]. For a given particle size, a higher BC mass fraction increases the critical supersaturation, although particle size has a stronger effect on CCN than does chemical composition [Dusek et al., 2006b].

Details are in the caption following the image
Critical supersaturation required to activate particles of varying size and BC content into cloud condensation nuclei. Higher supersaturation values mean that particles are less likely to become cloud droplets. Results are from a particle-resolved model that simulates coagulation and condensation in ambient air onto individual particles. The two branches show aged particles from diesel exhaust (70% BC) and gasoline exhaust (20% BC), assuming that Köhler theory describes activation. Particle diameter has the greatest effect on activation, with BC content inhibiting activation to a lesser extent. Figure is based on simulations [Deville et al., 2011] of particle-resolved model [Riemer et al., 2009].

[146] “Closure” studies compare measured CCN concentrations with values predicted from particle size and composition, assuming that only soluble components serve as CCN. Predicted CCN concentrations are usually greater than measured values, with large variability in the degree of agreement [Medina et al., 2007]. This overprediction is worse with a mixture of hydrophobic and hygroscopic particles [Wex et al., 2010]. Near source regions, externally mixed BC plays little role in CCN [Rose et al., 2011] and closure is improved when externally mixed BC is assumed not to be CCN active [Lance et al., 2009]. However, closure studies to date have insufficient precision to confirm the contribution of coated BC to CCN. A comparison of heated and unheated particle size distributions can indicate the quantity of non-refractory material in atmospheric particles [Sakurai et al., 2003; Philippin et al., 2004; Kuwata et al., 2007]. Figure 6 shows the relationship between the condensed mass per particle and the CCN activity of 100 nm BC particles observed in Tokyo [Kuwata et al., 2009]. The number fraction of CCN-active BC particles increases with increasing condensed mass, indicating that 0.18 × 10−15 g of condensed coating material is required to activate these BC-containing particles at 0.9% supersaturation. In this case, the condensed compounds were primarily organic.

Details are in the caption following the image
The number fraction of CCN-active BC particles versus mass of coating material for supersaturation of 0.9%. The data represent the response of 100 nm BC particles sampled in Tokyo in April 2007. The black points with 1 σ error bars are averages of the individual runs shown as gray data points. The solid curve is a sigmoidal fit to the averaged data points. A sigmoidal function is used because the ordinate should fall in the region between 0 and 1 by definition, although some deviations are apparent likely due to the measurement errors as well as some deviations from the assumptions employed for the calculation. Adapted from Kuwata et al. [2009, Figure 11].

3.9.2 Global Model Treatment of CCN Activity

[147] Treatment of BC in global climate models (GCMs) does not always reflect the theoretical dependence on size and soluble fraction. In some GCMs, the activation of BC to form cloud droplets is not considered at all. The simplest formulation, a mass-based parameterization, assumes that the number of cloud droplets activated is empirically related to the submicron aerosol mass. In turn, that mass is determined from the hydrophilic aerosol species sulfate, submicron sea salt, and hydrophilic carbonaceous aerosol, where carbonaceous aerosol is the sum of OA and BC [Rotstayn et al., 2009]. This simple parameterization ignores differences in activation for particles of different sizes and composition. Many models have advanced to represent this behavior. Cloud droplet number concentration is sometimes empirically related to aerosol number concentration of certain sizes and to updraft velocities [Lohmann et al., 2007]. Parameterizations of cloud droplet formation derived from Köhler theory have been developed by Abdul-Razzak and Ghan [2000] and Nenes and Seinfeld [2003]. In these schemes, dependence on size and hygroscopicity can be taken into account, including the competition between different types of particles. For example, large hygroscopic particles take up water, reduce supersaturation, and therefore result in less activation of smaller or less hygroscopic particles such as those containing BC [Ghan et al., 1998]. Models of greater complexity represent size-resolved incorporation of aerosols in cloud droplets and their removal [Jacobson, 2006]. Section 7 describes studies of BC effects on clouds using some of these advanced representations.

4 Emission Magnitudes and Source Categories

4.1 Section Summary

  1. [148]

    Global emission estimates use the “bottom-up” method of multiplying emission factors by activity data. With this method, a bottom-up estimate of total global emissions in the year 2000 is about 7500 Gg BC yr−1, with an uncertainty range of 2000 to 29,000 Gg yr−1. About 4800 Gg BC yr−1 is from energy-related burning, with the remainder of about 2800 Gg BC yr−1 from open biomass burning. Total primary organic aerosol (POA) emissions excluding biogenic matter are 47,000 Gg POA yr−1 for the global total, with an uncertainty range of 18,000 to 180,000 Gg POA yr−1. Energy-related burning and open burning produce 16,000 and 31,000 Gg POA yr−1, respectively; the largest uncertainties are in open burning.

  2. [149]

    Industrial-era emissions are the difference between present-day and the preindustrial background year, 1750. These values are 6100 Gg BC yr−1 (4400 Gg BC yr−1 from energy-related burning and the remainder of about 1700 Gg BC yr−1 from open burning) and 33,000 Gg POA yr−1 (14,000 Gg POA yr−1 from energy-related sources and the rest from open burning).

  3. [150]

    Black-carbon emission sources are changing rapidly due to greater energy consumption, which increases emissions, and cleaner technology and fuels, which decreases them. Bottom-up inventories indicate that Asian emissions may have increased by 30% between 2000 and 2005.

  4. [151]

    Sources whose emissions are rich in BC can be grouped into a small number of categories, broadly described as diesel engines, industry, residential solid fuel, and open burning. Dominant emitters of BC from energy-related combustion depend on the location. Asia and Africa are dominated by residential coal and biomass fuels (60–80%), while on-road and non-road diesel engines are leading emitters (about 70%) in Europe, North America, and Latin America. Residential coal contributes significantly in China, the former USSR, and a few Eastern European countries.

  5. [152]

    Estimates of energy-related emissions agree broadly on the major contributing sectors and approximate magnitudes of BC emission. However, current inventories in many world regions lack information regarding the factors governing emissions. These include the type of technologies or burning, and the amounts of biofuel (BF) combusted. Major differences in estimates of energy-related emissions result from a few knowledge gaps. For energy-related emissions, these include sparse emission measurements of sources with the highest emissions. In industrialized countries, these may be a small fraction of the emitting sources. Only energy-use inventories in North America, Europe, and urban East Asia provide a high level of detail. Measurements in developing countries are scarce for all source types.

  6. [153]

    Current emission factors from biomass burning, and thus emission estimates, might be biased low by a factor of at least two. Emission estimates from open burning are generally uncertain due to insufficient data on burned area and fuel consumed. Quantification of BC emissions from this source is also difficult because they strongly depend on the burning behavior and because of inherent problems with sampling and analyzing BC in smoke plumes from vegetation fires.

  7. [154]

    The majority (80%) of open fire emissions occurs in tropical latitudes, and inter-annual variability of BC emissions from forest or savanna fires can exceed one order of magnitude in some regions.

  8. [155]

    Because the net aerosol effect on climate depends on the ratio of absorbing to reflective particles such as OA and sulfate, we group emission sources into categories based on their combustion type and co-emissions. Major sources of BC, ranked in order of increasing POA : BC ratio, are diesel vehicles, residential burning of coal, small industrial kilns and boilers, burning of wood and other biomass for cooking and heating, and all open burning of biomass. A few of these sources also emit significant quantities of SO2.

  9. [156]

    Receptor modeling studies of BC in urban areas that use chemical composition to identify dominant emission sources find source categories that are qualitatively similar to those in bottom-up inventories based on activity data.

4.2 Introduction

[157] Emission inventories, or global and regional tabulations of emission quantities, have a dual role. They are required inputs for atmospheric models that assess the environmental consequences of these emissions, and they also provide necessary information for the development of air quality and climate policies by indicating the largest or most easily manageable sources of emission. For BC and POA, inventories used in global models are typically “bottom-up” tabulations constructed from estimates of activity (e.g., mass of fuel burned or number of kilometers driven) combined with emission factors (e.g., grams BC emitted per mass of fuel burned or per kilometer driven). To address major fractions of particulate air pollution, or climate forcing related to BC, identification of the sources that contribute to total emissions is needed. The identification of emitting sectors is important not only for mitigation policies but also for historical reconstructions and future projections, as each sector has a different temporal evolution. The major BC emitters in each country or region depend on technological development and on practices common in each society.

[158] In this section, we first provide definitions of the terms, groupings of emission sources, and groupings of countries used throughout this assessment. Section 4.3 outlines general procedures for producing bottom-up inventories. We then present global totals and major source sectors, and regional contributions in section 4.4 , and several regional estimates are tabulated in section 4.5. Major causes of differences among inventories are reviewed in section 4.6. Section 4.7 discusses major sources of uncertainty in emission estimates, focusing on the sectors identified as most important. Section 4.8 briefly discusses trends in BC emissions and causes of change. In section 4.9, we discuss studies that have inferred the major sources of particulate matter based on the chemical composition of ambient aerosol, known as receptor modeling. These studies have been conducted primarily in urban areas.

[159] The forcing by species co-emitted with BC is important in determining the net radiative forcing by BC sources (section 11). In this section, we also discuss emissions of POA, the aerosol species most commonly emitted with BC. For BC sources, we also summarize co-emissions of SO2, which is a precursor to sulfate. Dust and sea salt are other major aerosol components, but they are generally not co-emitted with BC and they usually have larger particle sizes than BC or its co-emissions. Aged aerosol, however, may contain all of these components. The discussion of uncertainties in this section emphasizes BC emissions.

4.2.1 Geographic Aggregation

[160] In this assessment, we summarize emissions and concentrations based on 10 groups of countries (called “regions”), depicted in Figure 7. These groups were chosen based on proximity, development status, and basic meteorological similarity, although there is heterogeneity within each region. Countries in each region are listed in the auxiliary material (Table S1).

Details are in the caption following the image
Regions (country groups) used for summarizing emissions and concentrations in this assessment. EECCA = Eastern Europe, Caucasus and Central Asia. The regional country groups are listed in Table S1.

4.2.2 Definitions and Aggregation of Emission Sources

  1. [161]

    Activity. The term “activity,” as it is commonly used in the emission community, indicates a quantitative measure of an event that leads to emission, such as the quantity of fuel burned, product manufactured, or kilometers driven.

  2. [162]

    Emission factor. The term “emission factor” gives mass of BC emission per activity, as opposed to total emissions. Emission factors for BC vary by region and end-use, even for the same fuel.

  3. [163]

    Source categories. Because this assessment focuses on BC sources and their impacts, we isolate sources that contain large fractions of BC relative to other aerosol components or precursors. We group sources with some similarities into aggregates called “source categories,” although there are some heterogeneities within these categories. For example, we lump both modern diesel engines with emission reduction technology and high-emitting, poorly maintained diesel engines into the category “on-road diesel engines.” These categories are discussed further in section 4.4 .

  4. [164]

    BC-rich sources. This assessment examines sources that have non-negligible BC emissions and that, therefore, have a positive component of forcing. We differentiate between sources that are rich in BC with respect to co-emitted species and those whose aerosol forcing is dominated by inorganic species. For our purposes in this assessment, a “BC-rich” source is defined as having an estimated SO2 : BC emission ratio below 200 : 1.

  5. [165]

    Sectors. The term “sectors,” as used throughout emission literature, refers to broad activity categories used for reporting by the International Energy Agency and the United Nations. These sectors are energy or transformation (which includes electricity generation), industrial activity, transportation, agriculture, and residential. The sector frequently termed “residential combustion” may also contain commercial, agriculture, and miscellaneous activities. Energy-related combustion is conducted deliberately in pursuit of economic activity or to meet personal or industrial energy demands. Open biomass burning also produces large quantities of atmospheric pollutants, and this activity has also become known as a sector in atmospheric literature. Because sectors consist of many types of activity, they include both BC-rich sources and sources that emit little BC. For that reason, we present emissions from source categories rather than from sectors.

  6. [166]

    Aggregated emissions. We use some additional terms to describe aggregated emission categories. “Energy-related” emissions include power plants, industrial activity, transportation, and residential fuel use. “Open burning” includes combustion of forests and grasslands or savannah, regardless of the cause of the fire. We also include open burning of waste for disposal, including crop residue or urban waste, in the latter category. The term “fossil fuel” indicates emissions from combustion of all fossil fuels, including diesel fuel and coal used in residential and industrial sectors. The term “biofuel” denotes biomass burned intentionally to meet energy needs and includes solid, unprocessed biofuel such as wood and agricultural waste.

  7. [167]

    Industrial-era versus anthropogenic emissions. A distinction of importance to the IPCC, among others, is the identification of “anthropogenic” emission sources. As discussed in section 2.3.2, we follow IPCC practice and use a time-based definition of climate forcing that is more accurately described as “industrial-era” forcing: the difference between the near-current year 2000 and the background year 1750. Industrial-era emissions are required to evaluate climate changes since 1750. “All-source” emissions include the component that occurred both in 1750 and in the present day. Industrial-era emissions are more uncertain than all-source emissions because of the large uncertainties in activity during the background year, especially for open biomass burning. Anthropogenic emissions may not be the same as industrial-era emissions, as some emissions in 1750 may have been human caused. Nominally, energy-related sources are all anthropogenic, as are sources associated with waste disposal. Vegetation fires are the only BC emission source that sometimes occurs without human activity.

4.3 Bottom-Up Inventory Procedures

[168] Emissions from energy-related combustion and from open vegetative burning are derived from different types of input data and often created by separate communities. In this section, we summarize information needed to create inventory estimates to clarify the sources of uncertainties.

4.3.1 Energy-Related Emissions

[169] Emission estimates for activities related to energy use were first developed to evaluate air quality in urban areas, where both high concentrations and high population led to severe exposure to health risks. Urban regulations have historically targeted total particulate matter mass concentrations, not individual chemical species. For that reason, early source characterization focused on mass emissions, and many source measurements did not provide emission rates of individual components such as BC.

[170] Bottom-up inventory estimates from energy use are based on the following simple equation:

[171] In the equation above, Ai represents activity of a particular type (e.g., fuel consumption or commodity production, conducted in a specific way), EFi is an emission factor in grams per activity, and effi is the pollutant removal efficiency by a particular type of abatement. The subscript i represents different types of activity that result in emissions of the same pollutant. Both emission factors and removal efficiency depend on the type of activity and the pollutant. For global or regional emission inventories, activity is almost always given as mass of fuel burned.

[172] BC emission inventories from energy-related emissions are available for all countries. However, the level of detail used in each inventory varies greatly. Quantification of emissions and the identification of major contributing sources could be substantially refined by disaggregating the activity definitions used in equation 4.1. For example, activity for a country might be given as total coal consumption in the residential sector, with a single emission factor used for all coal burning installations. This lumping ignores the fact that large boilers might have very different emission factors or better control devices than small coal stoves.

[173] Bottom-up inventories can be graded based on the level of refinement used in emission estimation. Table 4 proposes such criteria for grading energy-related emissions. Major uncertainties in current bottom-up inventories include insufficient knowledge of activity rates and emission characteristics. Emission factors are not as well understood for sources largely found in rural areas, such as off-road diesel engines. Furthermore, measurement resources have been concentrated in developed countries, so there are still relatively few measurements of key emission sources in developing countries. Section 4.7 gives further details on uncertainties, which vary by emission sector and by region.

Table 4. Recommended Grade Levels for Inventories of Energy-Related Emissions
Level Activity Data Emission Factors
Low National fuel use for each fuel and sector or commodity Sectoral averages
Low−: Global average
Low+: Regionally specific average
Medium National fuel use for each fuel and sector or commodity, apportioned by technology Specific to each technology and fuel
Medium−: Some extrapolation from other countries
Medium+: Full participation of country in specifying activity
High Country tabulations of technology divided by substantially different emission characteristics; identification of major individual point sources Measured at each installation or closely matched to emitting technology, including control

4.3.2 Open Burning Emissions

[174] In contrast to energy use, open burning emissions are usually not included in national activity reports. There are thousands of open fires burning globally each day, and most of these are caused by humans, either purposefully or involuntarily. Open fires are ignited for many purposes, and their emissions differ by region and ecosystem type. In contrast to residential and industrial emission sources, which are predominantly located in the northern hemisphere mid-latitudes, open fires occur largely in tropical regions, with 80% of emissions occurring there. National regulations about fire use and their enforcement vary among countries. Another important distinction between emissions from open fires and emissions from energy use is the very large inter-annual variability of the former. This is caused by variations in the accumulation of wooded or grass fuels, in fuel characteristics such as dryness, and in other factors influencing fire spread and fire severity. An extreme case in point is fires in Indonesia that are strongly influenced by the El Niño-Southern Oscillation in combination with the draining of peatlands. According to Schultz et al. [2008], BC emissions from Indonesian fires ranged between 40 and 1400 Gg yr−1 during the 1990s. Mack et al. [2011] report a single tundra fire whose fuel consumption was similar in magnitude to the annual net carbon sink for the entire Arctic tundra biome.

[175] The method for estimating emissions from large open fires is similar to equation (4.1) but using different input data.

[176] Here BAi is the burned area (km2), FLi the available fuel load (kg dry matter per km2), CCi the combustion completeness (fraction), and EFi the emission factor (g compound per kg dry matter). The first three parameters combine to produce activity, or total mass burned. The index i usually stands for one ecosystem type in each inventory grid cell. As fire activity has a pronounced seasonal cycle, BA, FL, and CC must also consider temporal variability of fires, although this is not yet state of the art in global inventories for FL and CC. Similar to estimates of energy-related combustion, emissions for open burning can be accomplished at different levels of refinement, as summarized in Table 5.

Table 5. Recommended Grade Levels for Inventories of Open-Burning Emissions
Level Forest, Grassland, or Woodland Burned Area Fuel Load (FL) and Combustion Completeness (CC) Emission Factors Agricultural Waste
Low Global satellite product Aggregated ecosystem-dependent estimates at grid cell level Default (three ecosystem average values) Global data compilations of burned mass
Medium− Satellite product, plus fire hot spot data and retrieval uncertainty assessment Ecosystem-dependent estimates at fire pixel level, FL and CC highly aggregated (<12 ecosystem classes) Three ecosystem average values with adjustments for peat soils Bottom-up inventory using regional-scale information
Medium+ Satellite product, plus fire hot spot data and retrieval uncertainty assessment Finer aggregation of FL and CC Three ecosystem average values with adjustments for peat soils
High High-resolution (spatial and temporal) satellite product or complete aerial surveillance, fire hot spot, and fire radiative power; Information on available fuel load and combustion parameters at tree stand level Ecosystem-based emission factors with adjustments dependent on fire behavior (at least 12 global ecosystem classes) Cross-validation of satellite-derived and bottom-up estimates for agricultural waste burning emissions

[177] Burned area is either derived from aerial surveillance or retrieved from satellite instruments that measure surface temperature and reflectance [Stocks et al., 2002; Kasischke and French, 1995]. Space-borne retrievals of burned area are generally based on the changes in surface reflectance as a result of the dark burn scar after a fire, but they occasionally make use of active fire detection as well [Giglio et al., 2009]. These retrievals can be confounded by apparent changes caused by the viewing geometry, cloud shadows, snow melt or temporary flooding, and other factors [Roy et al., 2002; Simon et al., 2004]. Some studies have also related fire radiative power from active burns to the amount of biomass combusted [e.g., Wooster et al., 2005; Kaiser et al. 2009; Kaiser et al., 2012] or to the amount of aerosol emitted [e.g., Ichoku and Kaufman, 2005; Sofiev et al., 2009].

[178] Fuel loads and combustion completeness are normally extrapolated from smaller-scale field studies. Grassland fires often consume nearly 100% of the above-ground biomass [e.g., Shea et al., 1996; Keene et al., 2006]. In contrast, the combustion completeness in forests varies strongly depending on the fuel and burning conditions. If the fuel is sufficiently dry, the dead plant material and plant litter are often consumed almost completely. Larger branches and stems rarely burn entirely, and living tree biomass is generally affected only in severe fires. The actual consumption of biomass in a given fire depends strongly on the ecosystem type and the fire size, severity, and persistence. Fires can consume substantial amounts of soil material, which can dominate fire emissions, for example, in peat areas of Indonesia or Siberia [e.g., Goldammer and Seibert, 1990; Soja et al., 2004].

[179] Emission factors are obtained from laboratory experiments or field measurements in the smoke plumes of actual open burns. Different measurement methods for BC in each field study, systematic sampling biases, and the use of ecosystem mean emission factors to represent both flaming and smoldering combustion introduce uncertainties in the estimated emissions (section

4.3.3 Waste Burning Emissions

[180] Combustion may be used to dispose of agricultural, household, or industrial waste. Well-controlled combustion systems, such as incinerators, emit little particle mass, and the discussion here focuses on uncontrolled burning. Activity data are among the most difficult to estimate, as they are not of economic interest and, therefore, not quantified by any organization. Large agricultural fires are detected with remote sensing, but the smaller fires are not and may be excluded from open burning emission estimates.

[181] Agricultural wastes, such as cereal straws, woody stalks, and sugarcane leaves and tops, are generated during harvest periods. Some of this biomass finds use as animal fodder, thatching for rural homes, and fuel for residential cooking and agricultural and rural industry. A fraction of agricultural waste is burned in fields, often to clear them for a new sowing season, or sometimes as part of the harvesting process. The practice of agricultural waste burning has strong regional and crop-specific differences and has large seasonal variations. Emissions from agricultural waste burning are typically calculated by multiplying crop production of a particular type, a fraction of residue per product, a fraction burned in the field, a fraction of dry matter, and the combustion completeness. Reported residue-to-product ratios [Koopmans and Koppejan, 1997] are larger for straw and stalks of crops like cereals (rice, wheat, millets) and legumes but smaller for husks and hulls from rice and groundnut. Reported ranges of dry matter fraction are 0.71 to 0.85, and combustion efficiencies range from 0.68 to 0.89 [Streets et al., 2003a]. However, the assumed fraction burned in field is subject to large uncertainties and is sometimes computed based on local practice and knowledge of competing uses of the agricultural waste [Venkataraman et al., 2006].

[182] Emissions from garbage burning are estimated using per-capita waste generation rates, along with fraction burned and emission factors. Both waste generation rates and fraction burned are location specific [e.g., Christian et al., 2010]. Waste generation is higher in industrialized countries and urban areas, but the fraction burned is higher in developing countries. Burning of industrial waste is quantified in industrialized countries, but it is highly controlled so that emissions are small. In contrast, informal disposal of industrial waste has not been quantified in developing countries, and these emissions are not included in these estimates.

4.4 Total BC Emissions and Major Source Categories

[183] Table 6 summarizes the best estimates and their ranges for BC and POA emissions in the year 2000 as derived in this assessment from a variety of information sources described in this section. Figure 8 summarizes sources of global BC emissions from two global inventories, along with estimates of their uncertainty; it also shows emission ratios for co-emitted, cooling aerosol species or precursors. Figure 9 shows the same estimates tabulated by region. Estimates from SPEW (Speciated Pollutant Emissions Wizard) [Bond et al., 2004, 2007; Lamarque et al., 2010] and GAINS (Greenhouse Gas and Air Pollution Interactions and Synergies) [Kupiainen and Klimont, 2007; Cofala et al., 2007; Amann et al., 2011; UNEP, 2011] are used as a reference for energy-related emissions throughout this assessment because they have the technological detail required to explore mitigation. SPEW estimates contain bottom-up uncertainties and are therefore used as the basis for Figure 8. GAINS estimates have the advantage of providing all co-emitted species, including gaseous emissions, and are used in the discussion of source category impacts in section 11. Section 4.6 discusses differences between these inventories and compares other global and regional emission estimates.

Details are in the caption following the image
Emission rates of BC in the year 2000 by source category and ratios of co-emitted aerosols (e.g., primary organic aerosol, POA) and aerosol precursors (e.g., SO2) to BC. For reference, it is often assumed that the ratio of OA to primary organic carbon (OC) varies from 1.1 to 1.4, depending on the source (section 3.2.2). SPEW emissions are shown as colored bars and are described by Lamarque et al. [2010]. GAINS estimates are from UNEP/WMO [2011a, 2011b], and RETRO emissions for open burning are described by Schultz et al. [2008]. Sulfur emissions from Streets et al. [2009] were used for ratios to SPEW. Regions are shown in Figure 7.
Details are in the caption following the image
Emission rates of BC in the year 2000 by region, indicating major source categories in each region. SPEW, GAINS, and RETRO emission data are the same as in Figure 8. Regions are shown in Figure 7.
Table 6. Best-Estimate Bottom-Up Values for BC and POA Emissions in Year 2000a (Gg yr−1)
BC BC Range POA POA Range
All sources
Energy relatedb 4770 1220 to 15,000 15,900 8800 to 23,800
Open burningc 2760 800 to 13,800 31,100 9000 to 156,000
Total all-source 7530 2020 to 28,800 47,000 17,800 to 179,000
1750 background
Energy related 390 1560
Open burning 1020 12,800
Total background 1410 14,360
Industrial era
Energy related 4380 14,300
Open burning 1740 18,300
Total industrial era 6120 32,600
  • a Energy-related emissions are from year 2000; open-burning emissions are a climatological average of years around year 2000.
  • b Refer to section 4.4 for a derivation of the totals and ranges.
  • c Refer to section 4.4 for a derivation of the totals and section for a derivation of the ranges.

[184] Figure 8 and Table 7 also give bottom-up estimates from open burning. Both RETRO (REanalysis of the TROposphere over the last 40 years) [Schultz et al., 2008] and GFED (Global Fire Emissions Database) [van der Werf et al., 2006] incorporate remote sensing information on fires to provide seasonal and inter-annual emission variation and are frequently used by atmospheric models. Figure 10 shows the seasonal and inter-annual variability of open burning in five regions. However, because remote sensing poorly detects small agricultural fires, we rely on estimated activity data (SPEW and GAINS) for agricultural waste burning emissions. The figure and table also compare SPEW open-burning emissions, which are based on country information about total quantities burned.

Details are in the caption following the image
Seasonality of BC emissions from forest, grassland, and woodland burning. Average monthly emissions estimated by RETRO for the period 1996–2000 are shown [Schultz et al., 2008]. Error bars indicate minima and maxima during each period. Regions correspond to those in Figure 7. Africa, which includes the small Middle East emissions, has two burning seasons because the equator bisects it. The group of East, Southeast, and South Asia (E, SE, and S Asia) includes emissions from Oceania. The very large error bars in that region results from the very high fire season in 1997. Note the different vertical scales in the two panels.
Table 7. Central Estimates of All-Source Global Black-Carbon Emissions From Open Biomass Burning (Gg yr−1)
2000 clim. 2000 clim. 2000 clim.d
Grassland and woodland firese
Grassland and open savanna 820 850 310 290
Woodland 330 350 1220 1220
Total grassland and woodland 1150 1200 1530 1510 1710
Forest fires
Deforestation and degradation 230 440 n.s. n.s.
Forest (excluding deforestation and degradation) 300 370 n.s. n.s.
Peat 2 130 n.s. n.s.
Total forest fires 530 940 830 1240 1240
Agricultural waste burningf 50 60 n.e. n.e. 290 330
Grand total 1730 2200 2370 2750 3280
  • a Version 3.1 from http://www.falw.vu/~gwerf/GFED/GFED3/tables/emis_BC_absolute.txt; clim.: mean climatological values from 1997 to 2006.
  • b n.s. = not specified; n.e. = not estimated.
  • c clim.: mean climatological values for 1996 to 2000.
  • d SPEW totals provide a climatological average rather than representing any particular year.
  • e GFED and RETRO have different classifications of grasslands and woodlands.
  • f GFED agricultural waste burning is inferred from remote-sensing data and includes only large fires, while GAINS and SPEW are based on activity estimates.

[185] The left panel of Figure 8 presents total BC emissions for each source category, also indicating the regions of emission. This panel shows ranges calculated from uncertainties in both activity data and emission factors in SPEW, and a comparison with the same sectors in the GAINS database. Although totals for each category vary, there is general agreement that the three largest contributors are open burning, diesel engines, and residential solid fuels. A small number of industrial sources in developing countries also make a significant contribution.

[186] The two right panels of Figure 8 show emission ratios between POA and BC, and sulfur dioxide and BC. Higher ratios indicate that more aerosol species are co-emitted that could offset direct warming by BC. For BC in snow and sea ice, co-emission of non-absorbing aerosol (e.g., sulfate) does not affect BC forcing, and co-emission of absorbing aerosol (e.g., some POA) adds to the forcing. Not represented in the figure are gaseous species such as carbon monoxide and O3 precursors, many of which contribute to warming (section 11).

[187] Table 8 provides numeric values for energy-related emission estimates in Figure 8, including additional disaggregation. For an estimate of energy-related BC emissions in the year 2000, we average the GAINS and SPEW totals without flaring, aircraft at cruise altitudes, and international shipping yielding 4430 Gg yr−1. We then add those three sources for a total of 4770 Gg yr−1. Relative uncertainties in BC emissions are taken from the SPEW bottom-up calculations, giving 90% uncertainty bounds of 1220 to 15,000 Gg yr−1. A similar estimate for POA gives a central estimate of 15,900 Gg yr−1 with uncertainty bounds of (8800 to 23,800) Gg yr−1.

Table 8. Bottom-Up Central Estimates of All-Source Global Black Carbon Emissions From Energy-Related Combustion in the Year 2000 (Gg yr−1)a
Diesel engines
On-road diesel engines 840 780
Off-road diesel engines 480 370
Total diesel 1320 1150
Industrial coal
All coal in industry 740 340
Total industrial coal 740 340
Residential solid fuel
Wood, cooking regions 1000 1580
Other biofuel, cooking regionsb 290 310
All biofuel, heating regionsb 260 200
Coal, cooking, and heating 330 420
Total residential solid fuel 1880 2510
Other sources
Non-coal industry, including biofuel 170 30
On-road gasoline engines 110 80
Residential, including diesel generation 100 170
Aviation 20 1c
Shipping 100 40d
Flaring -e 260
All other BC-rich sources 70 20
Power plants f 20 20
Other low-BC sources -e 60
Total other sources 590 690
Grand total 4510 4690
  • a Data are based on year-2000 energy data and technology. Totals do not match exactly due to rounding.
  • b “Cooking regions” are those where solid fuel is primarily used for cooking, even if some heating occurs; the opposite is true for the definition of “heating regions.”
  • c This inventory only includes landing and takeoff (i.e., no cruise) emissions.
  • d This inventory only does not include international shipping.
  • e Not estimated.
  • f Considered to be low-BC sources.

[188] Table 7 provides numeric values for biomass-burning emission estimates in Figure 8, including additional disaggregation. For an estimate of all-source, bottom-up emissions in 2000 from open burning, we average the RETRO and GFED climatological values for forests, grasslands, and woodlands, yielding approximately 2450 Gg BC yr−1. We add estimated activity-based data (average of SPEW and GAINS) for estimates of agricultural waste burning emissions (310 Gg BC yr−1). The total is 2760 Gg BC yr−1. A parallel estimate for POA gives 29,200 Gg POA yr−1 from forests, grasses, and woodlands, and 1910 Gg POA yr−1 from agricultural waste burning. Bottom-up uncertainty estimates are discussed in section 4.7.

[189] For energy-related emissions in the background year of 1750, we assume the values of 390 Gg BC yr−1 and 1560 Gg POA yr−1 given by Dentener et al. [2006]. These emissions are entirely from biofuel use. Industrial-era emissions are the difference, 4380 Gg BC yr−1 and 14300 Gg POA yr−1. We also assume the Dentener et al. [2006] values for background open-burning emissions: 1230 Gg BC yr−1 and 12800 Gg POA yr−1. Industrial-era emissions are the difference, 1740 Gg BC yr−1 and 18300 Gg POA yr−1. With no constraints in the year 1750, the background values are crude assumptions, scaled by population and land cover. Other work has estimated BC emissions in the middle to late 1800s [Novakov et al., 2003; Ito and Penner, 2005; Bond et al., 2007; Junker and Liousse, 2008], but none has provided estimates for earlier years.

[190] In the following sections, we briefly review contributions from individual source categories. Section 4.6 discusses some reasons for differences between the estimates presented in the figures, as well as other emission estimates.

4.4.1 Diesel Engines

[191] On-road diesel engines include diesel cars and trucks, while off-road engines include engines used in agriculture, construction, and other heavy equipment. The diesel-engine category in this assessment specifically excludes shipping emissions, which are summarized separately. Diesel engines contributed about 20% of global BC emissions in 2000. These sources have the lowest co-emissions of aerosols or aerosol precursors of all the major BC sources. In order to enable use of the most advanced exhaust controls, sulfur must be removed from the diesel fuel during refining. Therefore, in regions with fewer controls, primary particulate matter emission factors are higher, but SO2 emissions are also higher.

4.4.2 Industrial Coal

[192] Industrial coal combustion is estimated to provide about 9% of global emissions, mainly in small boilers, process heat for brick and lime kilns, and coke production for the steel industry. Although coal combustors can be designed to produce little BC, coal can also be highly polluting when burned in simple combustors, which are still present in small industries, particularly in developing countries. Co-emitted SO2 is estimated from coal sulfur content and exhaust control. The SO2/BC ratio for industrial coal is much higher than that for the other emission categories, where the fuel has little sulfur or more efficient flue-gas controls are in place. Emissions from coal-fired power plants, which emit much less BC because of their better combustion efficiency, are not included here.

4.4.3 Residential Solid Fuels

[193] Wood, agricultural waste, dung, and coal are used for cooking or heating in homes, providing another 25% of BC emissions. Most of the emissions occur in single-family devices, which are often of simple design. When infrastructure and income do not allow access to low-emission residential energy sources such as electricity and natural gas, solid fuels are used extensively for cooking. Otherwise, they are used more often for heating. Coal and, less frequently, wood are also used for heating in multi-family building boilers. The designation “cooking” in Figure 8 refers to regions where wood is primarily used for cooking, even if some heating occurs. Similarly, “heating” includes all uses in regions where heating is dominant. This sector also includes emissions from both production and consumption of charcoal. The poor combustion and mixing in these simple devices result in relatively high POA : BC ratios, but SO2 emissions are low except for coal.

4.4.4 Open Burning

[194] Open burning of biomass in the location where it is grown is a very large contributor to global BC emissions, with bottom-up estimates predicting that it contributes about 40% of the total. Table 7 compares estimates of open-burning BC emissions for the periods 1997 to 2006 (GFED 3.1) and 1996–2000 (RETRO) with the climatological values from SPEW. Current global emissions estimated from open burning range between 2000 and 11,000 Gg yr−1 for BC and between 18,000 and 77,000 Gg yr−1 for OC in average years, as summarized by Schultz et al. [2008] and confirmed by subsequent work [Lamarque et al., 2010; van der Werf et al., 2010]. Most studies fall into the range of 2000 to 6000 Gg yr−1 for BC and 20,000 to 27,000 Gg yr−1 for OC. The highest BC and OC emission estimates by Chin et al. [2002] resulted from the use of larger emission factors. A smaller contribution originates from open burning in agricultural fields, which is often done to clear residues after harvest. This source contributes about 300 Gg yr−1 BC and 1500 Gg yr−1 OC. In regions like south Asia, this source can contribute about 20% of carbonaceous aerosol emissions [Venkataraman et al., 2006].

[195] The average of 2760 Gg BC yr−1 in Table 7 (including emissions from agricultural waste burning) agrees well with the mean value of global BC estimates from the review of Schultz et al. [2008] after they adjusted the literature estimates to standard emission factors. However, modeling studies that constrain emissions using satellite observations indicate substantially larger biomass burning emissions of particulate matter. Section 5.5.1 discusses these studies further.

[196] Because fuel-air mixing is completely unmanaged in open burning, large quantities of organic matter can escape and average POA : BC ratios are highest of all sectors. This ratio can vary greatly depending on the burning conditions; Andreae and Merlet [2001] give an inter-quartile range of 4 to 14. If the burning is natural or accidental, vegetation may burn while it is still rooted and standing, thus allowing for some greater ventilation and oxygen supply. Deliberate burning may involve land clearing before collecting, piling, and burning the material, which tends to favor oxygen-poor conditions.

4.4.5 Other Emission Sources

[197] BC emissions from the four major source types above are about 90% of the global total. Significant contributors of the remainder of BC emissions are listed in Table 8. Two of the largest groups are industrial and residential emissions that are not included in one of the preceding categories. In industry, biofuels are responsible for the higher BC emissions estimated by SPEW. Like coal-burning sources, these small sources are challenging to characterize. About half of the miscellaneous residential emissions come from middle distillate oil, and stationary diesel engines used for distributed power generation are estimated to produce about 85% of that. Charcoal, kerosene, and liquefied petroleum gas make up the remainder of residential emissions.

[198] Although aviation emits BC at altitudes that otherwise have low aerosol concentrations, its contribution to BC mass is quite small. Likewise, shipping emits aerosol into regions that otherwise have low concentrations, but it ranks with minor sources of BC. These statements are supported by more refined emission estimates than those in Table 8. Lee et al. [2010] estimated “soot” from aviation as 6 Gg yr−1 for 2004. Eyring et al. [2010] summarized shipping emission studies, giving a best estimate of 130 Gg yr−1 and a range of 5 to 280 Gg yr−1.

[199] Other BC-rich sources include gasoline engines and flaring in the oil and gas industry. Gasoline engines have much lower particulate matter emission rates than diesel engines, so total emissions from gasoline are less than 10% of diesel BC emissions, although gasoline vehicles are more numerous. Flaring in the oil and gas industry is a poorly understood source both in terms of activity and emissions. Considering data from Elvidge et al. [2009], Johnson et al. [2011a], and Johnson and Coderre [2011], GAINS estimated BC emissions at nearly 4% of the anthropogenic global total with the majority originating in Russia, Nigeria, and the Middle East.

[200] All BC-rich sources together constitute 99% of the global inventory. Low-BC sources include coal power plants for generating electricity. These are not considered a large source of BC because the high temperatures and well-managed combustion promote burnout of any BC that is formed. However, BC emission rates from power generation in developing countries are not well known (section 4.6.1).

4.5 Regional Emissions

[201] The level of refinement in current inventories, limited by the availability of data, varies greatly among world regions (Table 9). Figure 9 presents emission estimates tabulated by region, rather than by source category. Substantial BC emissions occur in all regions, with the largest contributions in regions where open burning is high (Africa and Latin America). Dominant sources of BC emissions from energy-related combustion change with development from residential coal and biomass fuels (60–80%) in Asia and Africa, to on-road and non-road diesel transport (about 70%) in Europe, North America, and Latin America. Industrial uses of coal are also important in East Asia.

Table 9. Black Carbon Emission Inventory Gradesa
Energy Use
Region Urban Rural Open Burning
North America High High Medium+
Latin America Medium+ Low Low
Europe High High Medium+
EECCA Low Low Medium−
Middle East Low Low Low
Pacific Low Low Low
Africa Low Low Medium−
East Asia High Medium+ Medium−
South Asia Medium+ Low Low
Southeast Asia Low Low Low
  • a Highest possible inventory grades based on current available disaggregation of activity and emission data for different world regions. Regions are shown in Figure 7. Grades defined in Tables 4 and 5.

[202] The largest open burning emissions occur in Africa, Latin America, and Southeast Asia. Tropical savanna and forest burning contribute about 45% and 40% of global open fire emissions, respectively. Globally, burned area and fire emissions are mostly decoupled because most of the burned area occurs within savanna ecosystems with relatively low fuel loads and emissions per unit area, while forest lands have less burned area but higher and more variable fuel loads. As a consequence, inter-annual variability is higher in equatorial Asia than in Africa. Open burning at northern middle to high latitudes does not constitute a large fraction of global, annual BC emissions, but the timing and location of Eurasian open burning emissions in particular (i.e., high-latitude spring in Figure 10) mean they have especially large contributions to BC cryosphere climate forcing, which has very high efficacy (section 8). Figure 11 shows the distribution of the major source types by latitude. The Northern Hemisphere contains about 70% of the emissions (i.e., 85% of energy-related emissions and 45% of open-burning emissions).

Details are in the caption following the image
Latitude plot of the data shown in Figures 8 and 9 using 5° latitude bins.

4.6 Comparison Among Energy-Related Emission Estimates

[203] Thus far, emission estimates from GAINS and SPEW have provided a perspective on the major sources of global BC emissions. Other estimates of global and regional emissions are summarized in Table 10. These differ substantially, by up to factors of 3 for specific regions. We begin by reviewing global emission estimates. Differences in the totals can usually be ascribed to choices of emission factors or other characteristics within the major source categories, which we discuss individually. We then use this background to discuss differences among Asian emission estimates.

Table 10. Bottom-Up Estimates of All-Source Global and Regional BC and OC Emission Rates From Energy-Related Sources (Fossil Fuel and Biofuel Combustion)a
Emission Rate (Gg yr−1)
Source BC OCb Inventory Levelc Year of Estimate
GAINS [Cofala et al., 2007]d 4700 12,300 Med−/+ 2000
SPEW [Bond et al., 2007]d 4530 8700 Med− 2000
Junker and Liousse [2008] 4800 7300 Low+ 1997
Ito and Penner [2005]e 4800 10,100 Med− 2000
Cooke et al. [1999] and Liousse et al. [1996] 6100f 16,000f Low+ 1984
Penner et al. [1993] 12,600 - Low− 1980
Reddy and Venkataraman [2002a, 2002b] 400 1150 High/Med 1997
Sahu et al. [2008] 1300 - Low+ 2001
Parashar et al. [2005] 900g 2300g Low+ 1995
Ohara et al. [2007] 800 3300 Low+/Med− 2000
Streets et al. [2003b] 600 2800 Med− 2000
Klimont et al. [2009] 750 1600 Med+ 2000
Lu et al. [2011] 680 1700 Med+ 2000
920 2200 2010
GAINS 580 1900 Med+ 2000
SPEW 500 1600 Med− 2000
Cao et al. [2006] 1500 4200 2000
Streets et al. [2003b] 1000 3400 Med− 2000
Zhang et al. [2009b] 1600 2800 Med+ 2001
1800 3200 Med+ 2006
Klimont et al. [2009] 1200 2800 Med+ 2000
Lu et al. [2011] 1200 2400 Med+ 2000
1700 3400 2010
GAINS 1100 3500 Med+ 2000
SPEW 1200 2800 Med− 2000
United States
Reff et al. [2009] 440 960 High 2000
Battye et al. [2002] 430 - High 1999
SPEW 350 500 Med− 2000
GAINS 260 370 Med− 2000
  • a Subsets of global inventories are given under each country for comparison. All emission rate values are rounded.
  • b Because most inventories report OC emission rates rather than POA, OC values are given here.
  • c See Table 9 for definitions of inventory levels.
  • d GAINS and SPEW emission factors and technology divisions have been updated since the original publications describing the methodology. Updates to SPEW are described in Lamarque et al. [2011].
  • e Digitized from figures to obtain fossil fuel plus biofuel only.
  • f Combined estimate is fossil fuel emissions from Cooke et al. [1999] (5100 Gg yr−1 BC, 7000 Gg yr−1 OC) and biofuel emissions from Liousse et al. [1996]; the two inventories are often used together. Bulk BC emissions in Cooke et al. [1999] (rather than only submicron emissions) are estimated to be 6400 Gg yr−1.
  • g Average of range: 400–1400 for BC, 1200–3300 for OC.

[204] Penner et al. [1993] developed one of the first global BC emission estimates based on applying constant global emission factors to broad sectors. Cooke and Wilson [1996] were the first to apply emission factors that depend on development level. An updated version of that inventory [Cooke et al. [1999], abbreviated as 1999Cooke99 for the following discussion] is commonly used in atmospheric modeling. Liousse et al. [1996] developed an early estimate of biofuel-burning emissions based on consumption estimates from the Food and Agriculture Organization. Bond et al. [2004] discussed the main sources of differences between SPEW and 1999Cooke99. As discussed below, the largest differences are for power generation and diesel engines. A composite of national inventories reflecting growth between 2000 and 2006 was developed for the field campaign Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS). The global estimate totaled 5200 Gg yr−1 [www.cgrer.uiowa.edu/arctas/emission.html].

4.6.1 Differences in Power Generation Emissions

[205] GAINS and SPEW agree on the magnitude of BC from power generation. However, 1999Cooke99 has much higher emissions from coal-fired power plants (an increase of 1600 Gg yr−1 BC and 2400 Gg yr−1 OC above both GAINS and SPEW). This difference resulted from assumptions of very high BC fractions in Cooke99, while the other estimates relied on existing measurements. New measurements [Zhang et al., 2008b] and subsequent harmonization of emission factors [Lamarque et al., 2010] support the use of lower emission factors and values nearer the SPEW and GAINS emissions for this sector. However, there remains a persistent but informal perception (e.g., G. Beig, personal communication, 2009) that BC fractions of particulate matter in developing countries, especially in power plants and industrial installations, could be higher than that represented by existing measurements. Even if older, poorly operating power plants have high BC emission factors, there is no evidence that newly installed, modern power plants do. Thus, rapid growth in electricity generation is unlikely to cause sharp increases in BC emissions.

4.6.2 Differences in Industrial Coal Emissions

[206] SPEW estimates of industrial coal emissions are about double those of GAINS. This difference is almost entirely due to assumptions in emission factors and BC fractions for the two major emitting categories, brick kilns, and coke ovens. A few particulate matter (PM) measurements were available for brick kilns, but no measurements of total PM from uncaptured coking were available at the time of inventory development. Composition measurements for both sources were estimated based on expert judgment, as no composition measurements were available for either source.

4.6.3 Differences in Diesel Engine Emissions

[207] GAINS and SPEW have similar emissions for year 2000 from on-road and off-road diesel engines, with SPEW being somewhat higher due to different assumptions about emission factors and the fraction of high-emitting vehicles. In previous versions of SPEW [e.g., Bond et al., 2007], diesel emissions were higher than those given here due to different assumptions about the implementation of standards. The Cooke99 inventory had much higher emissions from diesel engines, resulting from very large assumed emission factors in developing countries that were not based on measurements. Measurements have confirmed higher emission factors in locations with delayed emission standards [Subramanian et al., 2009; Assamoi and Liousse, 2010], although no measurements are as high as the assumptions in the Cooke99 inventory. However, an evaluation of trends in California [Kirchstetter et al., 2008] indicates that emissions prior to regulation could have been higher by a factor of 10, greater than either the emission factors used in SPEW and GAINS, or in any reported measurements.

4.6.4 Differences in Residential Solid Fuel Sector

[208] Small-scale residential combustors constitute the largest difference between SPEW and GAINS. Estimates from SPEW are lower by 30% for biofuel used in cooking and all uses of coal, and higher by 35% for biofuel used in heating. The discrepancies in heating biofuel and coal emissions are largely due to activity data. For cooking stoves, activity in SPEW is lower by about 15%, and emission factor choices cause the remainder of the difference. GAINS uses the highest emission factor from fireplace measurements for cooking stoves (about 1.1 g BC [kg fuel]−1), while SPEW uses an average of cooking stove measurements (about 0.8 g BC [kg fuel]−1). Both are consistent with observed emission factors.

4.6.5 Asian Emission Inventories

[209] Detailed emission inventories have been developed for the Asian region and for individual countries within Asia. Here we compare these inventories and identify some reasons for differences. Similar studies are not available in other world regions.

[210] Three emission inventories have been developed for Asia alone. Streets et al. [2003b] provided inventories to support modeling during the TRACE-P field campaign [Jacob et al., 2003]. Emission factors in that inventory were an earlier version of those in SPEW [Bond et al., 2004]. Two primary differences were the addition of small industries (brick kilns and coking) in SPEW and adjustment of emission factors for residential fuel, where BC emission factors were reduced slightly and those for organic matter were greatly reduced. In addition, the TRACE-P inventory relies on some projected energy use because it was developed before year 2000 data became available. Ohara et al. [2007] developed an Asian emission inventory and also provided time trends for 1980–2003. They used a single emission factor for each sector drawn from Streets et al. [2003b], except for transportation where they developed a representation based on differing vehicle types. Klimont et al. [2009] drew on additional national data collected during the GAINS-Asia project to estimate emissions of BC and OC for the period 1990–2005 and projections to 2030. They developed technology specific factors, although they did not distinguish among some uses (e.g., cooking versus heating), with the result that estimates were about 25% higher estimates than those of Streets et al. [2003b], Ohara et al. [2007], SPEW, and the current GAINS global model.

[211] For India, central values of BC emissions differ by a factor of about 3. The lowest central value [Reddy and Venkataraman, 2002a, 2002b; Venkataraman et al., 2005] is from a difference (compared to SPEW) in the residential sector, where the India-only mean emission factor is about 16% lower than the global mean used by Bond et al. [2004]. Estimates by Sahu et al. [2008] are a factor of 3 larger, resulting from the use of the high power-plant and diesel emission factors from Cooke et al. [1999]. Parashar et al. [2005] estimated emissions a factor of 2 larger than those of Reddy and Venkataraman, largely caused by their new measurements of very high emission factors for biofuel (especially dung). Estimates from Ohara et al. [2007] and Streets et al. [2003b] are based on the same emission factors, so the large difference between them is surprising. Most of the 200 Gg yr−1 difference is attributable to biofuels in the residential sector (616 versus 420 Gg yr−1, respectively). Activity data are frequently given in terms of total fuel calorific content rather than mass, and in these units, the Ohara and Streets inventories are only 16% different. The difference, therefore, must be caused by differing assumptions in the conversion between fuel calorific content and mass. Klimont et al. [2009] used higher emission factors for cooking and national energy use statistics and estimated nearly 30% higher BC than current global GAINS implementation or SPEW, while OC was well within the range of GAINS and SPEW. Lu et al. [2011] obtained similar estimates to Klimont et al. [2009] and also estimated that BC emissions rose 35% between 2000 and 2010. The Lu et al. [2011] trends in Asian aerosol emissions compare favorably with remote-sensing data.

[212] For China, regional inventories are in broad agreement with global inventories on magnitudes and sources of BC emissions. Cao et al. [2006] developed an emission inventory based on Chinese data and some new emission factors. SPEW and the inventory of Cao et al. [2006] are quite similar. The Streets et al. [2003b] inventory lacks treatment of small industry. Zhang et al. [2009b] included small industry, particularly brick and cement kilns, following the approach of Streets et al. [2006] and estimated significantly higher total BC emissions from China. Lu et al. [2011] estimated generally higher growth in BC emissions between 2000 and 2010 (+46%) than previous studies. Organic matter emissions in Cao et al. [2006] are higher than those of the other studies due to the use of different emission factors. Klimont et al. [2009] estimates for BC are higher by about 15% than Streets et al. [2003b] and global inventories but lower than Zhang et al. [2009b]. While comparing total emissions for China yields in general fairly good agreement across the studies, there are significant sectoral shifts resulting primarily from uncertainties in coal versus briquette use in the residential sector and assumptions for the brick-making sector. For example, Zhang et al. [2009b] and GAINS use comparable assumptions for coal, while Klimont et al. [2009] assumes more briquettes, leading to lower emissions from this sector in Klimont et al. On the other hand, for brick making both Zhang et al. and Klimont et al. rely on Bond et al. [2004] emission factors and estimate higher emissions than GAINS, which considers anecdotal evidence about the transition in this industry that leads to significantly lower emission factors.

4.7 Major Sources of Uncertainty in Emissions

4.7.1 Energy-Related Combustion

[213] As discussed in section 4.6, large differences between emission estimates are attributable to a few choices regarding emission factors made by inventory developers. Some of the diversity is not supported by measurements at emission sources, while other disagreements reflect true uncertainty in knowledge. A bottom-up estimate of uncertainty is about a factor of two using either a simple uncertainty combination for global emissions [Bond et al., 2004] or a Monte Carlo approach for individual countries [Lu et al., 2011]. Activity Data

[214] For fossil fuel combustion, activity estimates for many inventories in Table 10 have the same source (International Energy Agency (IEA), or United Nations fuel-consumption data), although many inventory developers adjust these data when inconsistencies are found or when finer allocation of activities is required (e.g., road and off-road vehicles or vehicle type). For several larger countries, regional statistics are also available and using them allows for better spatial resolution; however, national and international statistics are not always consistent. Finally, some models chose to use data sets specifically developed for particular projects. For example, GAINS often uses national fuel-consumption estimates for most of the European countries, China, India, and Pakistan. These may differ from IEA statistics, especially with regard to sectoral allocation. Total fuel consumption can be affected by the use of regional information on the calorific value of fuels.

[215] Consumption of residential solid fuels, especially biofuel, is not well constrained. Biofuel consumption data are frequently drawn from disparate sources. For example, GAINS relies on IEA and national data, while SPEW uses the tabulation of Fernandes et al. [2007]. Fuel production and sales are not centralized in this sector, and many fuels are collected by the consumers or by small sellers. Activity data are therefore much more uncertain than for liquid or gaseous fuels or consumption in large installations. Estimates of fuel consumption are often based on per-capita consumption estimates multiplied by the number of people using solid fuels. Comparison of the residential biofuel consumption for the past years in India and China [Streets and Aunan, 2005; Venkataraman et al., 2005; Ohara et al., 2007; GAINS] shows significant differences, typically ranging within about ±25%. Emission Factors

[216] Emission rates of both BC and organic carbon depend on the combustion process, including fuel composition, flame temperature, mixing between fuel and air during combustion, and post-combustion treatment of the exhaust. Carbonaceous aerosols can be destroyed if the exhaust is kept hot and well mixed with air. Large, properly operating combustors, such as power plants and some modern installations using biofuel, tend to achieve this burnout, resulting in little emitted BC. Mixing between fuel and air before combustion also limits BC formation, so that gasoline engines emit much less BC than do diesel engines. Finally, BC may be removed through end-of-pipe controls that capture fine particles, as it is in particulate filters after diesel engines.

[217] The strong dependence of BC emissions on combustion processes means that the disaggregation of activity to represent combustion quality explicitly is an important component of inventories. The dependence also demands the development of emission factors under realistic operating conditions. Because most sources have several operating modes, it is important to obtain emission factors by measuring during a realistic sequence. This sequence of conditions may be called “driving cycles” for vehicles, or “burn cycles” for stoves, small boilers, or open combustion. Although design of existing cycles seeks to represent real-world operation, unrealistic choices of operating conditions may yield emission factors that are poorly representative. For example, very cold conditions or startup phases, which promote poor efficiency and high emissions, may be omitted from vehicle and stove emission testing or tests may not include the poorest quality fuels.

[218] Uncertainties are particularly acute outside the United States and Europe. A few measurements are becoming available, but the limited number of studies has not dispelled concern that the highest emitters have been missed or significantly underrepresented.
  1. [219]

    Diesel engines. There is a good understanding of on-road emissions from normal vehicles in developed countries, although questions exist about how in-use vehicle emissions compare with those from tests in laboratories because of differences in driving conditions. Both averages and ranges of vehicle emissions in developing countries are not well quantified. Compared with on-road engines, there are many fewer measurements of off-road equipment, including engines used for agriculture and construction. A large uncertainty in determining total emissions is the contribution of poorly functioning vehicles with very high emissions, or “super-emitters.” The fraction of such vehicles and their emission factors are not well known, and these assumptions result in major discrepancies between inventories, because a single super-emitter can produce many times more BC than a properly operating engine with an emission control system. Super-emitters are likely to be more widespread in developing countries, but data on the fraction of vehicles with such high emissions are extremely limited.

  2. [220]

    Industrial solid fuels. Emission estimates from the industrial sector are dominated by small, simple kilns in traditional production processes and old boilers. Activity data and emission factors for these sources are particularly difficult to obtain. Magnitudes and composition of emissions are the most uncertain of any of the major source categories due to lack of measurements.

  3. [221]

    Residential solid fuels. Although the residential sector is highly heterogeneous with regard to the types of fuels and devices used, a relatively small number of emission measurements have been made and many of these are from laboratory rather than in-use measurements. Limited data on emission magnitudes and composition are available to characterize this sector. Summary of Uncertainties in Energy-Related Emissions

[222] Major contributors to uncertainty in BC and OC emissions from energy use are (1) measured particle emission factors obtained under laboratory conditions (i.e., rather than in-use or in-field) for residential combustion, traditional industry, and vehicles with high emissions due to malfunction; (2) speciation of PM from high-emitting technologies into BC and OC; (3) quantification of individual emitters in sectors that contain even a small fraction of highly polluting devices; and (4) amounts offuel burned in sectors where fuel and output are not formally monitored and tabulated. While most of the estimates tabulated here are given for the year 2000, rapid economic growth in countries with large BC emissions may have caused energy-related BC emissions to increase dramatically over the last decade. The year used for the inventory estimate is important when comparing atmospheric models with observations.

[223] In addition to the factors that affect total emission quantities, the location of emission may be poorly known for sources such as small industry and passenger or freight vehicles. These inaccuracies affect comparisons between measured and modeled concentrations.

4.7.2 Open Biomass Burning Burned Area

[224] Considerable uncertainties remain in the quantification of burned area. Current satellite retrievals cannot detect burn scars much smaller than 1 km2, and the size of the burned area is sometimes wrongly determined. Active fires can be detected when they are larger than about 0.1 ha, but many fires cannot be observed during their flaming stage because of incomplete coverage of satellite orbits or clouds obscuring the scene. Given the current resolution of satellite instruments that are used for burned area retrievals (typically 0.25 to 1 km2 at the sub-satellite point and 5 to 10 times larger at swath edges), only burn scars with a size of at least 12 to 40 ha can be detected from space. Field data and satellite retrievals of fire radiative power [Wooster et al., 2005] show that the majority of fires are smaller, particularly in tropical regions [cf. Schultz and Wooster, 2008]. Uncertainties arising from the limited spatial resolution of current instruments could be as large as −30% to +40%, but in reality, they are smaller due to compensating errors. Validation of individual fire scenes from the MODIS burned-area data set with high-resolution Landsat data indicates that burned areas are probably underestimated by about 10% on average (M. Wooster, personal communication, 2009). Fuel Load and Combustion Completeness

[225] Field studies, such as those in the savanna regions of South Africa [Shea et al., 1996] or in boreal North America (summarized by McKenzie et al. [2007]), reveal that fuel loads can easily vary by a factor of 3–20 within a region of limited ecosystem diversity. This uncertainty is consistent with estimates of fuel load derived from analysis of satellite data [Ito and Penner, 2004]. Combustion completeness strongly depends on the weather conditions because bulk fuel, which constitutes a large fraction of fuel mass in wooded ecosystems, burns only when the fuel is sufficiently dry and when it is windy. Uncertainties in vegetation modeling for calculation of available fuel load and combustion completeness are highest in deforestation regions and in regions where peat fires occur [van der Werf et al., 2006] (e.g., Southeast Asia) where problems in modeling the combustion of organic soil layers containing peat leads to an uncertainty of about a factor of 5. In some regions, like southern hemisphere Africa, models incorrectly predict seasonality, with peak emissions in bottom-up estimates occurring about 1–2 months earlier than the peak in satellite-detected AOD [van der Werf et al., 2006]. This is attributed to an increase in emissions as the fire season progresses, caused by a shift from grassland fires early in the dry season to woodland fires later in the dry season, associated with different fuel moistures and burning behaviors. Ecosystem modeling frameworks are moving toward capturing such temporal variations. Assuming that errors in fuel load estimates are uncorrelated across different regions, the global uncertainty in fuel load densities would be substantially lower than the variance encountered in individual field studies but greater than the 50% model diversity reported by Knorr et al. [2012]. Our conservative estimate of the global uncertainty caused by fuel load densities and combustion completeness is a factor of 2. Emission Factors

[226] As shown in section 5.5, current models of the atmosphere underestimate absorption aerosol optical depth (AAOD) in Africa and Latin America, which are major biomass burning regions [Koch et al., 2009b]. Some of the extra absorption in Africa could be caused by dust, but an explanation is needed for the underestimate in Latin America. The underestimate could be caused by low biases in burned area or fuel loads (see above), but inverse models [Arellano et al., 2004; Arellano et al., 2006; Stavrakou and Müller, 2006] indicate little bias in carbon monoxide (CO) emissions from biomass burning in these regions. The remaining suspects are fire emission factors, the relationship between emitted mass of BC and optical absorption in fire plumes, or incorrectly modeled aerosol lifetimes.

[227] Although biomass burning is a large component of global BC emissions, particulate and BC emission factors for this source are poorly constrained. As discussed in section 3, measurements of BC depend on the analysis method. These method-dependent biases are more critical for biomass-burning emissions than for other sources, because these emissions pyrolyze and also contain materials that catalyze BC emission. The comprehensive literature review of Andreae and Merlet [2001] has become widely used, in particular for the compilation of global inventories [e.g., van der Werf et al., 2006; Schultz et al., 2008]. However, this review included only values based on thermal oxidation techniques and excluded optical absorption measurements. Even the highest BC emission factor in Andreae and Merlet [2001] is lower than values inferred from absorption measurements, which have been used in other studies [Patterson and McMahon, 1984; Liousse et al, 1996; Chin et al., 2002; Liley et al., 2003]. In part due to higher emission factors, Liousse et al. [2010] estimated African biomass burning emissions that were about 2.5 higher than GFED. Martins et al. [1998a] showed that thermal oxidation measurements underestimated BC mass, leading to unrealistically high MACBC values. Therefore, it is possible that the use of imperfect thermal methods yields BC emission factors that are too low.

[228] Comparisons between chemical and optical measurements would increase confidence in biomass-burning emission factors for BC. A review by Watson et al. [2005] showed differences of up to a factor of 7 between different BC field measurements and discussed the various uncertainties related to both thermal and optical measurements. In contrast to Martins et al. [1998a], thermal measurements did not always yield lower BC mass estimates than absorption measurements.

[229] Representativeness of measured emission factors is another concern. Open fires have a high inherent variability. Some emission factors and characteristics are inferred from small, better-controlled fires in laboratory settings. The combustion intensity and the burning and airflow characteristics of these small fires may differ from those of fires in the real world [Reid et al., 2005]. Oxygen-rich flaming combustion is generally associated with more BC and more heat release, while lower-temperature smoldering fires have higher overall particulate matter emission factors and CO emissions than flaming fires [Lobert et al., 1991; Ward et al., 1992; Yokelson et al., 1997]. If a sample is dominated by emissions from one of these phases, then ratios of BC to total PM, or BC to CO, do not represent the overall emission profile.

[230] The relationship between emitted mass of BC and absorption of the fire plume can be biased because of aging processes. As smoke ages, organic material condenses, but no further BC is created. BC/PM ratios in aged plumes are lower than those in fresh plumes, and single-scattering albedos are higher (more scattering relative to absorption). Emission ratios are generally determined in fresh smoke plumes, while models are evaluated with data from long-term regional averages, which are mostly of aged aerosol. If the relatively low BC ratios are then applied to fresh plumes, either BC emissions or absorption by BC could be underestimated.

[231] Data are presently insufficient to support firm conclusions about what updates should be made to BC emission rates from open burning of vegetation. However, thermal measurements may underpredict absorption, and using regional-average optical properties as constraints could also underpredict absorption at the time of emission. This body of evidence suggests that current emission factors from biomass burning might be biased low. Based on the data from Andreae and Merlet [2001], we estimate the lower uncertainty of BC emission factors to be a factor of 0.6 multiplied by the central value. Our estimate for the upper bound is a factor of 4. Summary of Uncertainties in Open Burning Emissions

[232] The dominant uncertainty term for open burning emissions is the emission factor (error range: factor of 0.6 to 4). Fuel load and combustion completeness are uncertain by about a factor of 2, while burned area is probably known within 10% on the global scale, although regional differences can be larger (cf. auxiliary material of Schultz et al. [2008]). Neglecting the independent error estimates from agricultural waste burning and assuming that errors are independent, error propagation yields an uncertainty range of a factor of 0.29 to 5. Based on the emission estimates in section 4.4 , bottom-up uncertainty ranges are 740 to 12,800 Gg yr−1 for BC, and 8400 to 144,000 Gg yr−1 for POA. These ranges are quite asymmetric.

4.8 Trends in BC Emissions

[233] Novakov et al. [2003], Ito and Penner [2005], Bond et al. [2007], and Junker and Liousse [2008] all estimate large changes in BC emissions during the industrial era. All of these studies demonstrate that BC emissions are related but not directly proportional to fuel consumption. Typically, emission rates become greater as population and economic activity increase and then decrease as cleaner technology is deployed. The resulting trend is an emission increase followed by a decrease and is broadly consistent with measurement records downwind of industrializing countries [McConnell et al., 2007].

[234] This discussion has focused primarily on the situation in the year 2000, but historical trends suggest how emissions may have changed since the year 2000 and how they will continue to change (see also section 13.5). In regions where emission factors decrease faster than fuel consumption is growing, BC emissions and concentrations decline. Murphy et al. [2011] observed a 25% decrease across the United States between 1990 and 2004, and Bahadur et al. [2011] found a 50% decrease over a similar period in California, a state with more stringent standards. Ice-core measurements in Europe [Legrand et al., 2007] also indicate a decrease in BC deposition since the 1970s. In contrast, when growth is rapid and clean technology has not yet been implemented, BC emissions and concentrations rise. Lei et al. [2011] suggests a 30% increase in Chinese BC emissions between 2000 and 2005. Lu et al. [2011] conclude that Asian BC emissions increased strongly since 2000, with emissions from the largest contributors, China and India, growing by about 40%. Increasing BC deposition is recorded in Himalayan ice cores [Ming et al., 2008], although more measurements would help constrain the high spatial variability in deposition in this region.

[235] Besides increases in fuel use and decreases in emission factors, other trends in fuel use affect net emissions. Cleaner fuels are chosen because of environmental regulations. Households switch to more convenient fuels as income rises, and those fuels tend to be cleaner. Fuels may be adopted for different uses (e.g., energy production from agricultural waste).

[236] Historical records of charcoal deposits and anecdotal evidence suggest a decline of global emissions from open burning between the end of the nineteenth century and present day, with strong regional differences [Moulliot and Field, 2005; Power et al., 2008; Marlon et al., 2008]. These changes have been attributed to the expansion of intensive grazing, agriculture, and fire management [Marlon et al., 2008]. Fire records from North America suggest that fire severity, and hence carbon loss, increased during the last three decades [Turetsky et al., 2010]. This is consistent with climate model simulations indicating that fire activity will increase in the future because of increased temperatures and reduced rainfall [Pechony and Shindell, 2010; Liu et al., 2010]. However, there have not yet been estimates of BC emissions from future increases in open burning.

4.9 Receptor Modeling to Evaluate Source Contributions

[237] Most BC is emitted from sources that are either small and numerous or large but episodic, so monitoring data for individual sources are not available to validate emission inventories. The quality of emission inventories and the contributions of dominant sources can be determined only by inferences drawn from atmospheric measurements. The use of measured atmospheric chemical composition to deduce the influence of emission source types is generally known as receptor modeling. Another type of study that uses the magnitude and spatial distribution of atmospheric constituents to infer emission source strengths is called inverse modeling. The latter type of study requires atmospheric measurement networks, which are discussed in section 5.4. A discussion of inverse modeling results is given in section 5.5. In this section, we summarize information from receptor studies that have used chemical composition to identify particular source categories.

[238] Receptor methods based on the chemical composition of particles include examining the ratios of target chemical compounds, such as isotope ratios measured in time-averaged aerosol samples [Gustafsson et al., 2009] or in single particles [Guazzotti et al., 2003]. Other approaches use an expanded suite of chemical species to elucidate additional sources, including elemental and ionic composition and organic and elemental carbon [Watson et al., 1994] and additional organic molecular markers [Schauer et al., 1996; Zheng et al., 2002]. A limitation common to all receptor models is the inability to distinguish sources whose emissions have a very similar chemical composition.

[239] Among receptor models, the chemical mass balance model [Friedlander, 1973; Watson et al., 1984; Chow and Watson, 2002, Watson et al., 2002] and positive matrix factorization [Paatero, 1997; Hopke, 2010] have seen wide application in air quality assessment. Receptor modeling may also exploit ensembles of atmospheric trajectories [Ashbaugh et al., 1985]. The outcome is the identification of the pollution source types and estimates of the contribution of each source type to the observed concentrations. Elemental carbon (as measured by thermal techniques) is often used as a tracer of certain emission sources. As discussed in section 3, this measurement may not be equivalent to light-absorbing carbon. However, it is referred to as BC in this section for congruency with other sections.

4.9.1 Receptor Modeling in Urban Areas

[240] Table 11 summarizes source apportionment studies conducted to determine the sources of particulate matter pollution, usually in urban areas. The most detailed information exists in the United States, based on measurements from the U.S. Environmental Protection Agency's network of speciation samplers. Source contributions shown are specific to BC, in decreasing order of influence. For United States urban areas, it is possible to separate approximately 6–10 major sources of BC, including diesel, gasoline, biomass, residual oil, and local traffic. Several other sources also contribute to BC concentrations, depending on the location: steel mills, railroad emissions, and metal processing facilities.

Table 11. Major Sources Contributing to Urban BC (USA) and PM2.5 (Other World Regions)a, b
Region Sources Cities Sample References
North America TR, IN, SA, RE Anchorage, Atlanta, Baltimore, Buffalo, Burlington, Camden, Chester, Chicago, Cleveland, Detroit, Dover, Elizabeth, Indianapolis, Los Angeles, New Brunswick, New York City, Portland, Rochester, San Jose, San Diego, Seattle, St Louis, Wilmington Chen et al. [2010b]; Sheesley et al. [2010]; Zhou et al. [2009]; Watson et al. [2008]; Gildemeister et al. [2007]; Kim and Hopke [2007]
Latin America TR, IN, RE Mexico City, Santiago, Sao Paolo Johnson et al. [2006]; Johnson et al. [2011b]
Europe TR, IN, SA Areao, A Coruña, Ballinasloe, Birkenes, Skrealaden, Coimbra, Lisbon, L'Hospitalet, Milan, Gent, Waasmunster, Amsterdam, Dublin, Cork, Birmingham, Duisburg, Erfurt, Helsinki, Dresden, Huelva Viana et al. [2008]; van Dingenen et al. [2004]; Hazenkamp-von Arx et al. [2004]; Manoli et al. [2002]; Gotschi et al. [2005]; Houthuijs et al. [2001]
Africa OB, RE, IN, TR Cairo, Oalabotjha, Addis Ababa (qualitative) Abu-Allaban et al. [2002]; Engelbrecht et al. [2002]; Etyemezian et al. [2005]
East Asia IN, RE, TR Shanghai (TC), Beijing, Xi'an Song et al. [2006]; Zheng et al. [2005]; Cao et al. [2005]
South Asia TR, IN, OB, RE Delhi, Mumbai, Kolkata, Chandigarh, Hyderabad, Dhaka, Rajshahi Chowdhury et al. [2007]; Johnson et al. [2011b]
Southeast Asia TR, IN, RE Bangkok, Hanoi, Bandung, Manila

Hien et al. [2004]

  • a Estimated from receptor modeling, in order of decreasing importance. TR = transport (vehicle exhaust including gasoline and diesel); IN = industry including coal and oil and biomass burning; coal burning power plants; RE = residential energy; OB = open burning of biomass and refuse; SA = secondary aerosols; O = Others.
  • b In receptor modeling, quantity assessed is usually elemental carbon from thermal-optical analysis rather than BC. Here, the major sources of both substances are assumed to be the same.

[241] For Europe and other world regions, no separate apportionment of BC has been done. Instead, reviews [e.g., Viana et al., 2008; Johnson et al., 2011b] were used to identify the largest contributors to the mass of particles smaller than 2.5 µm diameter (PM2.5). Major sources of BC are also major sources of PM2.5, but the converse is not always true; major sources of PM2.5 may produce little BC if their emissions are primarily inorganic. Sources that are BC and OC emitters are shown in the table. Resuspended dust, secondary pollutants like sulfate and nitrate, or sea salt, could also be contributors to PM2.5 at some locations but are not included in Table 11.

[242] Globally, BC from gasoline combustion is only about 10% that from diesel (Table 8), but in the urban atmosphere, this source may constitute a significant fraction of particulate emissions. There are considerable problems in separating emissions from diesel and gasoline vehicles. For example, Shah et al. [2004] showed that slow-moving and stop-and-go diesel vehicles emit organic and elemental carbon in patterns that are very similar to those of gasoline powered vehicles. Thus, their mass contributions of diesel engines might be mis-allocated to gasoline engines. In a growing number of urban centers, especially in the developing countries like India and Bangladesh, Guttikunda and Jawahar [2012] showed that a significant fraction of BC emissions also originate from the brick kilns surrounding the city administrative boundaries, which consume a mix of coal, agricultural waste, and bunker fuel (in coastal cities).

[243] For European urban areas, the main sources of BC are vehicles, oil or solid-fuel combustion, and industrial and shipping emissions. Emissions from burning of biomass or biofuel were originally reported to be significant only in Denmark and Spain. Later, this source was shown to contribute relatively large fractions in rural and even urban areas [Szidat et al., 2006; Alfarra et al., 2007; Puxbaum et al., 2007] suggesting that this source was not discriminated in the earlier studies.

[244] Information on urban areas in other world regions is largely qualitative, because many studies in these regions did not measure the complete suite of pollutants used for receptor modeling. Furthermore, they often use source profiles that were not locally measured and therefore may be unrepresentative [Johnson et al., 2011b]. Table 11 shows that in Latin America, traffic, oil combustion, and small industry, including copper smelters, are important sources of fine particle mass. Mugica et al. [2009] attributed 42% of fine particulate matter in Mexico City to vehicles, but they were unable to separate wood burning due to their similarity with diesel sources. In Africa, open burning of biomass and refuse, residential coal combustion, and traditional industries (brick making, lead smelters, and foundries) are significant contributors. In East Asia, coal burning industries are dominant sources followed by traffic and biomass burning. Residential coal burning is a large wintertime contributor. In Southeast Asia, traffic is the dominant source in more urbanized locations, while local burning (for cooking and brick making) is important in less urban locations. In South Asia, traffic and burning of refuse and biomass are important sources, followed by industrial sources. Brick kilns are important in some locations, while lumped small industrial sources contribute 10 to 30% of fine particle mass. Overall, where traffic is an important source, an aging motor vehicle fleet containing high emitters is of concern. Coal and biomass combustion for residential cooking and heating and small-scale industrial applications is widespread, likely under poor combustion conditions.

4.9.2 Receptor Modeling in Continental Plumes

[245] Urban studies constrain sources that affect cities, but many emissions, such as residential solid-fuel burning or agricultural use of diesel engines, occur preferentially outside urban areas. Source apportionment studies have been applied to carbonaceous aerosol in the continental plume from South Asia. Novakov et al. [2000] suggested that fossil fuel was responsible for 80% of the BC in the continental outflow, while studies in Dhaka [Salam et al., 2003] indicated a negligible contribution of biomass burning in South Asian cities. However, these findings relied on ratios between total carbon and BC that are representative of open biomass burning but not biofuel burning. Studies that do consider differences between emissions from biomass and biofuel burning estimate an approximately equal contribution of biofuel and fossil fuel [Stone et al., 2007]. In continental outflow, a strong biofuel influence on total carbonaceous particles is indicated by single-particle measurements (74%) [Guazzotti et al., 2003] and radiocarbon measurements (66%) [Gustafsson et al., 2009]. In Indian cities, fossil fuel burning dominates fine particulate matter concentrations, but biofuel burning is not negligible, according to organic marker studies [Chowdhury et al., 2007]. Fossil fuel and biofuel burning contribute 20 to 60% and 7 to 20%, respectively.

4.9.3 Summary of Findings From Receptor Modeling

[246] Results from receptor modeling agree qualitatively that major BC sources identified by global or regional emission inventories are similar to the major sources of fine particulate matter. Traffic is the largest source in North and Latin America and Europe, and contributions from residential solid-fuel burning and open burning are found in Asia and Africa. This broad agreement lends confidence to the identification of the largest BC sources. However, receptor modeling of regional aerosols with specific source markers capable of resolving similar sources is needed before the approach can be extended beyond this qualitative assessment. Further examination of both urban and rural sources is required to constrain the dominant sources in continental plumes.

5 Constraints on Black-Carbon Atmospheric Abundance

5.1 Section Summary

  1. [247]

    Black carbon concentrations, like those of all short-lived species, are variable in space and time and are largest around source regions. Constraints on black carbon abundance are provided by in situ measurements of BC concentration and by ground- and satellite-based remote sensing of AAOD. Aerosol absorption inferred from satellite-based remote sensors is nearly global in coverage, but these data are less quantitative than ground-based data.

  2. [248]

    Ground-based remote sensing provides information on the atmospheric aerosol column burden and optical properties, which is directly relevant to radiative forcing. However, BC AAOD is inferred indirectly from these measurements. Aerosol can be sensed only when the sky is cloud free, absorption sensing has large uncertainties when aerosol loading is low, and the interpretation of BC amount is confounded by the presence of other light-absorbing aerosol in the column.

  3. [249]

    National and global networks of measurements are used to evaluate chemical transport models that provide estimates of BC surface concentrations (in situ monitoring) and column AAOD (ground-based remote sensing). Measurements are most sparse in some of the regions with the highest BC loadings: Africa and most of Asia. Measurements from field campaigns, which typically sample only one region over periods shorter than a year, can also be used to test processes within models more comprehensively but within a limited domain and time.

  4. [250]

    Comparisons with in situ observations indicate that many models simulate near-surface BC concentrations approximately correctly for North America and East Asia, have a slight high bias in Europe, and have a strong low bias in parts of Asia.

  5. [251]

    On the other hand, many model estimates of column BC AAOD over continents have a low bias in all regions. In many regions, this underestimate can be explained at least in part by the fact that these models do not account for aerosol internal mixing, and thus, their modeled mass absorption cross section (MACBC) is too low. However, even if this factor is taken into account, models would still underestimate AAOD in some regions.

  6. [252]

    Comparisons with satellite observations indicate that bottom-up estimates of aerosol emitted from biomass burning are too low by factors of 2–4, and emission estimates should be revised upward.

  7. [253]

    Airborne campaigns measuring vertical profiles of BC now allow for comparisons of vertically resolved concentrations away from continents. These measurements suggest that models overpredict BC in remote Pacific regions—on average by a factor of five—and that the overestimate is generally greatest in the upper troposphere. An exception is in the Arctic, where models appear to underpredict upper tropospheric BC. While concentrations in these regions are generally low, the large spatial area and bright underlying surfaces means the contribution of Arctic BC to globally averaged radiative forcing may be significant.

  8. [254]

    These combined results suggest that modeled removal of BC is an important source of model error. Further, inter-model comparisons indicated that differences between modeled BC concentrations and AAOD can only be attributed in part to differences in assumed emissions; differences in vertical transport and aerosol removal also play a large role, and MACBC differs between models.

5.2 Introduction

[255] The impact of BC on climate depends on its atmospheric abundance. Aerosol concentrations vary in space and time, and coverage by observations is insufficient to capture all such variations. Therefore, atmospheric models must be used to determine global concentration fields, and these models can be partially evaluated by comparisons with available observational data. Figure 12 shows modeled atmospheric absorption by two absorbing aerosols: BC and dust. The figure shows that BC, like all short-lived species, is most concentrated around source regions. For energy-related combustion, sources and concentrations are largest where population density is highest, as can be seen by comparing the east and west coasts of the United States, or Eastern Asia with Central Asia. Large BC concentrations also occur in and around biomass burning regions, especially South America and central Africa. Continental outflow also contains high concentrations, especially to the east of Asia and the west of Africa. Although BC is not well mixed throughout the atmosphere, some of it is carried to remote regions. Most energy-related BC emissions occur in the Northern Hemisphere, and BC is found in remote regions there, including in the deep Arctic. In contrast, there is very little BC throughout large remote regions in the Southern Hemisphere.

Details are in the caption following the image
Aerosol absorption optical depth (AAOD) at 550 nm attributable to BC and dust. BC-AAOD fields are from the AeroCom median model fields [Schulz et al., 2006] of all-source BC, including AAOD that would have been present before the pre-industrial era. Dust distribution is from Luo et al. [2003], with AAOD calculated from particle size in Mahowald et al. [2006] and optical properties in Yoshioka et al. [2007].

[256] Section 5.3 discusses atmospheric absorption and extinction by absorbing species. This section highlights the fact that if all absorption is attributed to BC, then concentrations of BC using measured absorption would be overestimated, especially in dusty regions. In section 5.4, we review measurements that report the spatial and temporal variation of BC concentrations. Section 5.5 examines how well global models simulate BC concentration over broad regions, and section 5.6 compares modeled vertical BC distributions with observations. These evaluations are used in section 6 to derive a best estimate of BC direct radiative forcing. Throughout this assessment, modeled BC concentrations are often drawn from an intercomparison project known as AeroCom (Aerosol Comparisons between Observations and Models) [Kinne et al., 2006; Schulz et al., 2006]. The AeroCom results described here reflect the first round of model experiments, known as Phase I.

5.3 Atmospheric Absorption and Extinction by Absorbing Species

[257] Section 3.4 discussed measurements of BC mass concentration (with units of g m−3). Another measure of particle abundance, albeit indirect, is AOD, which is the vertically integrated extinction of aerosols in an atmospheric column. Aerosol light extinction and AOD values are given for a specific wavelength; small (submicron) aerosols have higher AOD at shorter solar wavelengths. For illustration, if we neglect losses of sunlight to molecular scattering by gases, an aerosol AOD of 1.0 at a given wavelength indicates that only 37% [e−1] of the direct solar beam at that wavelength reaches the surface without being scattered or absorbed, if the sun is directly overhead.

[258] Of particular relevance for inferring absorbing aerosol amount is the column single-scattering albedo (ω0), which relates the amount of light scattered to the amount attenuated (similar to equation 3.2). The value of ω0 is influenced by microphysical properties (section 3.7) and by compositional properties, which are quantified by refractive indices. These optical coefficients are usually determined through laboratory or in situ measurements. The fraction of AOD attributable to absorption, or AAOD, is
where ω0 denotes a column-averaged single-scattering albedo. For any species, the product of mass concentration (with units of g m−3) and mass absorption cross section (MAC) (m2g−1) is the absorption coefficient (m−1). AAOD (dimensionless) is the result of integrating absorption coefficient over the entire atmospheric column and is the quantity shown in Figure 12. AAOD is more closely related to BC column abundance than AOD because BC is responsible for a much larger fraction of total absorption than of total extinction.

[259] As discussed in section 3, atmospheric concentrations of BC are often inferred from observations of absorption. Absorption measured in the atmosphere is attributable to all light-absorbing aerosols: BC, mineral dust, soil, and light-absorbing organic carbon (i.e., brown carbon) from combustion. If all AAOD is attributed to BC, then inferences of BC may be overestimated, especially in dusty regions. Measured absorption may also include contributions from nitrogen dioxide and ozone. The relative contributions of each component to AAOD are required to infer BC column abundance.

[260] Figure 12 shows that although the regions most affected by BC and dust are somewhat different, both species make significant contributions to globally averaged absorption, and in some regions, the contributions are equal. Dust has a much greater total AOD than does BC, but a smaller fraction of that AOD is absorption (i.e., ω0 is larger), so globally averaged AAOD of BC and dust are of similar magnitudes at visible wavelengths.

5.4 Observations of Atmospheric Black Carbon Concentrations

[261] Monitoring studies provide information about BC concentrations in the atmosphere. These studies incorporate either in situ observations, in which sampled air is drawn into an instrument for measurement, or remote observations, which measure the intensity of light coming from the atmosphere, and from this infer information on the atmospheric distributions and properties of aerosols. In this section, we illustrate the contributions of different types of studies to the understanding of BC and other aerosol concentrations and properties (Table 12).

Table 12. Observations That Contribute to the Understanding of Black-Carbon Distributions
Type of Monitoring Spatial Coverage Major Contributions
Intensive field campaigns Single point, surface, or aircraft In-depth examination of aerosol properties and transformation processes
In situ monitoring (includes networks) Single point, surface Long-term average aerosol concentrations, properties, and fluctuations in remote or urban locations
Ground-based remote sensing (includes networks) Single point, column Long-term averages and fluctuations of aerosol optical depth and inferences about other optical properties, often with unattended operation
Space-based remote sensing Global coverage, column Global coverage of aerosol optical depth, with limited information on properties

[262] Data gathered from intensive field campaigns or observational networks are subject to the limitations of the measurement techniques used, as discussed in section 3. Furthermore, when measurements are made at a single point, the measured values may not represent an average of the surrounding area, so comparisons between these point measurements and the average of a large model grid box should be done cautiously [e.g., Tegen et al., 1997; Vignati et al., 2010]. Thus, model simulations of the BC burden are evaluated by comparison with observations, which themselves may be biased. If the bias were known for each observation, or if it were the same for all observations, it might not affect our understanding of BC's life cycle, but this is not the case. Some biases may depend on aerosol age, source type, or location, and these relationships are poorly known.

5.4.1 In Situ Monitoring

[263] The in situ BC measurement techniques described in section 3, especially thermal-optical measurements on filter samples and optical measurements, have been employed in several intensive field campaigns (section and in routine monitoring networks (section at national or continental scales. The uses and limitations of these measurements are described in the following subsections, and inferences drawn from these measurements are discussed in sections 5.5 and 5.6. Intensive Field Campaigns

[264] Multi-investigator field campaigns have examined the nature of atmospheric aerosol in many world regions. Such campaigns typically take place over the course of a few weeks and provide “snapshots” of regional aerosol. They often combine measurements from several platforms, beginning with heavily instrumented surface stations but also adding aircraft and shipboard platforms to measure vertical and horizontal distributions in continental plumes. Intensive campaigns provide information about aerosol composition, microphysical properties including scattering and absorption, and reaction rates. Because they incorporate large arrays of measurements, they provide a wealth of information on aerosol and gaseous precursors, and can often be used to evaluate some aspects of emission inventories. This type of observational approach generally has limited temporal and spatial coverage and, alone, does not provide the information required to determine the average influence of BC or other aerosols on climate. The value of this type of effort in the context of evaluating BC concentrations and climate impacts lies in the richness of simultaneous measurements that they provide. These studies provide detailed physical understanding for evaluation and the improvement of the modeling of aerosol processes.

[265] Examples of early field campaigns that studied carbonaceous particles included a mission to the Arctic [Rosen et al., 1984] and the search for the light-absorbing component in the Denver Brown Cloud [Groblicki et al., 1981]. Coordinated campaigns to study aerosol-climate interactions in particular regions began with the Aerosol Characterization Experiments (ACE) series of experiments [Quinn and Coffman, 1998; Raes et al., 2000], and these have frequently examined the outflow regions from major source areas [Russell et al., 1999; Ramanathan et al., 2001b; Mayol-Bracero et al., 2002; Jacob et al., 2003; Huebert et al., 2003]. Large field campaigns in more recent years have focused on aerosol evolution and properties in large regions with high aerosol loadings [Bates et al., 2005, Querol et al., 2008, Zhang et al., 2008a]. Equally intensive efforts have examined aerosol in urban areas or heavily source-influenced regions [Watson et al., 2000; Neususs et al., 2002; Solomon et al., 2003; Cabada et al., 2004; Parrish et al., 2009; Wang et al., 2010].

[266] Intensive field campaigns have provided information on emissions in addition to aerosol properties. Experiments to evaluate emissions from open biomass burning [Lindesay et al., 1996; Kaufman et al., 1998; Eck et al., 2003] have determined aerosol characteristics and emission ratios. Chemical and transport models have also been used in conjunction with measurements to evaluate regional emission inventories [Rasch et al., 2001; Dickerson et al., 2002; Carmichael et al., 2003; Minvielle et al., 2004]. The interaction between atmospheric models and campaign measurements can be iterative, as models have also been used to guide aircraft flights to intercept urban and biomass burning plumes.

[267] The long history of intensive measurements now allows analyses across many campaigns and pollutants. For example, Clarke and Kapustin [2010] combined aerosol and trace gas measurements from 11 campaigns to relate anthropogenic aerosol concentrations to concentrations of CCN, AOD, and CO. This type of relationship can be used to test modeled emissions and transport.

[268] One recent campaign relied on a single well-instrumented aircraft performing near-continuous vertical profiling over global scales in remote areas over a period of a few weeks [Wofsy et al, 2011]. The sampling approach of reaching both high northern and southern latitudes was repeated 5 times in different seasons, so the spatial and temporal coverage is greater than intensive field campaigns. This type of measurement, along with vertical profile measurements in a single location over several years [Andrews et al., 2011], can provide constraints on background aerosol loadings and removal processes. Here they are used to estimate biases in modeled vertical distributions of BC (section 5.6.1) and to correct biases in modeled BC forcing efficiency (section 6.6.1).

[269] Intensive field campaigns usually include both in situ and remote-sensing measurements from multiple surface (either land- or ship-based) platforms and from one or more aircraft, as well as integrating satellite remote sensing measurements. Research aircraft often fly over surface stations and ships, so measurements can be compared between platforms and used to determine vertical profiles of the full aerosol column. This rich set of data allows comparisons of different techniques for measuring BC and other aerosol properties and applications of the same technique in different environments [e.g., Livingston et al., 2000; Schmid et al., 2003; Doherty et al., 2005]. Understanding of the applicability and limitations of each type of measurement is fostered by comparisons between properties measured at the surface, columnar properties measured by satellite-based remote sensing, and vertically resolved information from aircraft measurements [e.g., Redemann et al., 2000; Magi et al., 2003]. Long-Term In Situ Monitoring

[270] Long-term monitoring networks usually combine relatively simple measurement techniques with tens or even hundreds of stations to create a widely spaced network of observations that operate under uniform techniques. Surface sites may be influenced by orography and local sources, although such effects are usually considered when selecting sites. Because network sites are spatially sparse and have relatively simple data products, they do not provide the same level of insight into the fine details of aerosol processes as intensive field campaigns. However, they do provide valuable information on aerosol trends and broad spatial concentration patterns, and these long records contribute strongly to evaluation of the modeling of emissions, transport, and removal.

[271] Table 13 lists the major networks contributing information on aerosol absorption and BC concentrations. Continuous in situ measurements to monitor pollution levels were initiated as early as 1962 [Novakov and Hansen, 2004]. The IMPROVE network in the United States [Malm et al., 1994] and the European Monitoring and Evaluation Programme network in Europe [Kahnert et al., 2004] are among the most extensive and have historically focused on chemical speciation rather than on aerosol properties. Stations participating in the WMO Global Atmosphere Watch monitor aerosol properties relevant to climate, including light absorption, in a sparse, but worldwide, network of rural and remote sites that are predominantly operated in the EUSAAR- and NOAA-coordinated sub-networks. While they have achieved broad coverage within their regions, similar coverage is not available in many other world regions, such as South America and Africa. Monitoring stations are being developed throughout Asia [Oanh et al., 2006; Cao et al., 2007; Ramana and Ramanathan, 2006; Marcq et al., 2010], although coverage is still poor.

Table 13. Largest Global or Regional Ground-Based Observations Providing Long-Term Measurement Information Related to Aerosol Absorption or Black-Carbon Abundances in the Atmosphere
Network Location and # of Sites Dates Products Comments (Reference or URL)
AERONET World 400+ 1996–present AAOD (column) AOD*(1-SSA), less accuracy at lower AOD (http://aeronet.gsfc.nasa.gov/)
IMPROVE US 100+ 1985–present Absorption coefficient, EC measurement (ground) http://vista.cira.colostate.edu/improve/
EMEP Europe 70 1989–present PM and EC/OC (ground) http://www.emep.int/
WMO/GAW World 39 1994–present Absorption coefficient (ground) http://www.gaw-wdca.org/
(EUSAAR 19 Europe, NOAA: 15 World) http://www.eusaar.net http://www.esrl.noaa.gov/gmd/aero
ABC 10 2004–present Absorption coefficient, EC (ground) Ramanathan et al. [2007]

Bonasoni et al. [2010]

AIRPET Asia 6 2001–2004+ EC/OC, BC by reflectance, PM2.5 (ground)

Oanh et al. [2006]

CARIBIC 1 commercial aircraft 2005–Present Impactor with single-particle carbon content

Nguyen et al. [2008]

[272] As multiple measurement locations are required to evaluate emissions, transport, and removal processes in global and regional models, measurements from all of these networks have been used for model evaluation [Cooke et al., 2002; Park et al., 2003; Solmon et al., 2006; Koch et al., 2009b]. Most national networks have rigorous quality control procedures that ensure comparability of their measurements among similar network stations. However, as discussed in section 3, important differences in sampling or measurement practices may affect comparisons between networks.

5.4.2 Remote Sensing

[273] Remote sensing methods measure changes in the amount of sunlight reaching the Earth's surface, or the amount of radiation leaving the Earth's atmosphere, rather than the actual atmospheric BC or aerosol content. Through a process known as inversion, the nature and quantity of aerosol is inferred from its observed physical effect on the light. The procedure is also called a retrieval, because aerosol properties are retrieved from observed changes in radiance. Radiance properties that may be measured include wavelength dependence, angular distribution, and polarization. These can be used to infer the columnar amount of aerosol, the average aerosol size, the presence of non-spherical particles (such as dust), and even estimates of columnar light absorption.

[274] Advantages of remote sensing are the multi-year nature of the measurements and the ability to obtain measurements without continuous maintenance. Remote measurements sense column properties of aerosol, rather than concentrations at the surface as are typically quantified by in situ measurements. Column properties are more directly relevant for quantifying climate impacts. The relationship between surface in situ observations and column data depends on many factors, including the stability of the boundary layer, the presence of atmospheric layers, and measurement conditions that affect in situ and remote measurements differently. When in situ measurements are made from an aircraft that profiles the depth of the atmosphere, however, robust comparisons with column-integrated remote sensing measurements are possible. Disadvantages of passive remote sensing include a lack of specificity to individual chemical species and an inability to retrieve the properties of aerosols in distinct layers independently, such as a dust layer overlying a pollution layer.

[275] Information from retrievals may be used to infer the column burdens of chemical species that have distinctive physical or optical characteristics. However, many assumptions are required for these inferences, resulting in large uncertainties, especially at low optical depths (low concentrations). Furthermore, most remote sensing techniques can obtain valid data only in cloud-free conditions. Ground-Based Remote Sensing

[276] The utility of ground-based remote sensing for constraining BC atmospheric abundance is similar to that of long-term in situ sampling. Networks of measurements can be combined to evaluate model performance, but they do not provide fine-scale or global coverage. However, unlike ground-based in situ measurements, they do measure the whole atmospheric column which is more directly relevant to radiative forcing.

[277] Passive, ground-based remote sensing with upward-looking sun or sky photometers can be used to estimate AAOD. AERONET is the largest global network of these photometers [Holben et al. 1998; Dubovik et al., 2002]. These instruments use information measured at multiple wavelengths and multiple angles by sun- and sky-photometers [Dubovik and King, 2000]. Properties derived from the sky radiance measurements include the aerosol size distribution, AOD, and refractive indices at four solar wavelengths; from this, single-scattering albedo can be determined. However, the retrieved refractive indices are obtained for the total aerosol. No distinction is made between the fine and coarse mode aerosol components, which are compositionally different and have different refractive indices. Thus, when both dust and BC are present in an aerosol column, the retrieved imaginary refractive index (which dictates aerosol absorption) does not accurately reflect that of BC alone.

[278] We have calculated values of BC AOD from retrievals of AAOD at AERONET sites and use these values in section 6 to produce a scaled estimate of BC direct radiative forcing. The process of extracting BC AAOD from total AAOD requires certain assumptions. Limitations in the derivation of AAOD and specifically BC AAOD are discussed next.

[279] First, AAOD values are overestimated by the averaged AERONET Version 2.0 data product. Values of AAOD are based on aerosol extinction optical depth and single-scatter albedo, and single-scatter albedo can only be retrieved reliably for rather polluted conditions. For this reason, the AERONET Version 2.0 values of AAOD are published only when AOD is greater than 0.33 at 550 nm. The exclusion of low-AAOD conditions introduces a significant sampling bias that Reddy et al. [2005b] estimate to be a factor of two. This issue, and its implication for inferred BC quantities, is explored further in section 6.5.1 and Appendix B.

[280] Second, not all solar absorption by aerosol is caused by BC; some is due to dust and some to organic matter. Dust is less absorbing per mass than BC, but its AOD is often much larger. The net effect is that global average dust AAOD is comparable to that of BC AAOD but differently distributed in space and time. Thus, the BC contribution to AAOD needs to be isolated to use AERONET absorption for model evaluation. Dust particles tend to be larger than BC particles, so retrieved size distribution data might be used to attribute absorption by super-micron particles to dust, and absorption by submicron particles to BC and light-absorbing organic matter. However, AERONET retrievals provide a single ω0 and refractive index for the entire size distribution. Although this refractive index can be used to estimate AAOD of the total aerosol, the composition and, hence, the refractive indices and ω0 of the separate fine and coarse modes may differ greatly [Quinn and Bates, 2005]. As a consequence, a method is needed to apportion AAOD among the two modes. Separating the sub-micron AAOD into contributions of organic matter and BC would require a further inference not based on size distribution. The spectral absorption properties of the total aerosol have been combined with assumptions about the wavelength dependence of absorption for dust, BC, and organic matter to estimate their contributions to AAOD [Arola et al., 2011; Chung et al., 2012].

[281] Third, sample biases may occur because AERONET AAOD retrievals are made only during daylight hours and in cloud-free conditions. Two potential sources of bias oppose each other. Clear conditions can be expected to favor larger concentrations and column burdens of BC because fires are more likely and because scavenging by clouds (i.e., the primary removal mechanism for BC) cannot occur under clear skies. To estimate biases in retrievals of BC AAOD, we averaged modeled BC AAOD at AERONET locations only during days when AERONET retrievals were possible. Bias was determined by comparing with the average BC AAOD. For four AeroCom models, biases were regionally dependent, but the average was only 1%. This is consistent with a small global clear-sky bias in AOD found by Zhang and Reid [2009]. Another factor is that solar absorption by BC is enhanced when BC is coated by water, which may be much more likely in the vicinity of clouds. The model of Jacobson [2010] estimated solar absorption enhancement to be a factor of 2 to 4, depending on the size of the BC particle. Clear-sky measurements, which exclude partially cloudy scenes, likely miss this near-cloud absorption enhancement. To explore this sensitivity, we used the GATOR model [Jacobson, 2010], the only aerosol model that treats the full effect of optical focusing on absorption, to compare the AAOD for clear-sky versus all conditions. Although the global mean MACBC for cloudy conditions (14.4 m2g−1) was higher than that for clear-sky (13.9 m2g−1) conditions, the global mean burden was greater by 35% and BC AAOD was 20% higher for clear sky conditions than for cloudy conditions, so it is unlikely that AERONET observations are missing large enhanced absorption under cloudy conditions. However, this finding is not specific to AERONET sites or days of retrieval.

[282] A method to infer BC column mass loading is to determine the volume concentrations of components of aerosol mixtures required to fit the AERONET retrieved total column refractive index. These can then be used, along with assumptions about densities, to determine the column mass of each component, including BC [Schuster et al., 2005]. The BC column mass uncertainty is estimated at −15% to +40%, although the imaginary refractive indices used affect the retrieval and the AAOD. For analyses in this assessment, we use AAOD rather than inferred BC column loadings, as the former requires fewer assumptions. Remote Sensing From Space

[283] Knowledge of BC concentrations is needed in regions with few or no ground sites, such as over oceans. Remote sensing from space-borne sensors on satellites has near-global coverage and can, therefore, fill observational gaps in ground-based networks. AOD can be inferred from retrievals from satellite data, but the procedure also requires general assumptions regarding environmental properties (e.g., surface reflectance), and aerosol properties [Torres et al., 1998; Levy et al., 2007; Kahn et al., 2010], so that uncertainties are greater than those in ground-based sensing. Satellite-based instruments that provide data on column aerosol absorption and, therefore, information specific to constraining light-absorbing aerosol, are discussed below.

[284] The longest satellite data record on aerosol absorption, dating to 1978, is provided by the TOMS sensor (Total Ozone Mapping Spectrometer, http://jwocky.gsfc.nasa.gov/aerosols/aerosols_v8.html). The TOMS Aerosol Index is a retrieved quantity that indicates whether the column aerosol is largely absorbing or scattering. Over the last decade, TOMS-type retrievals have been continued with measurements from OMI (Ozone Measurement Instrument, http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/OMI). This instrument senses reflection and its spectral variability in the ultraviolet region (UV) of the solar spectrum [Torres et al., 2002]. Aerosol retrieval at UV wavelengths has two advantages: first, surface contributions to the signal are small, especially over land, and second, small aerosols have stronger signals in the UV than at visible wavelengths. On the other hand, in addition to BC, both dust and organic carbon contribute to UV absorption. Another complication is a reduced sensitivity towards the surface, so that BC near the surface (approximately the lowest kilometer) may not be detected. Therefore, a separate estimate of the vertical distribution of aerosol is an essential element of the retrieval. In the past, the vertical profile of aerosol concentration was prescribed using estimates from global modeling; more recently, aerosol profiles inferred from the CALIPSO space-borne lidar have been applied [Winker et al., 2010]. Even with this constraint, this type of retrieval is mostly useful for identifying atmospheric aerosol from large events such as biomass burning [Zhang et al., 2005].

[285] Satellite-based aerosol sensors that provide spectrally resolved information allow general estimates of aerosol size and separate contributions from fine and coarse particles. Moderate Resolution Imaging Spectroradiometer (MODIS) sensors have the greatest temporal resolution [Remer et al., 2005]. Additional information on aerosol shape and difficult retrievals over bright surface come from polarization sensing (e.g., POLDER) [Deuzé et al., 2001] and multiple views of the same scene (e.g., Multi-angle Imaging Spectro-Radiometer (MISR)) [Kahn et al., 2009b]. Active remote sensing with a space-borne lidar (CALIPSO) [Winker et al., 2010] offers information on aerosol vertical distribution. This information may provide estimates of total aerosol amount, or even total fine-mode aerosol, but they do not constrain the amount of absorbing aerosol.

[286] The satellite-based MISR instrument [Kahn et al., 2009b] provides another constraint on aerosol absorption. The MISR instrument detects radiance at multiple angles and multiple wavelengths. This additional information is used to estimate surface radiance, and it allows inversion for additional aerosol properties, including estimates of aerosol absorption. The MISR standard retrieval algorithm identifies two to four classes of absorbing aerosol, rather than providing exact values of absorption. The atmosphere must be relatively uniform and cloud free, and the mid-visible AOD must be at least 0.15 or 0.2 [Chen et al., 2008; Kahn et al., 2009c].

[287] While all these satellite-based instruments provide some constraint on atmospheric absorption, several factors limit their utility in determining global, annual average BC AAOD. Signals reaching satellite-based sensors are influenced by both atmospheric constituents and by the reflectance properties of the Earth's surface. This increases the uncertainty in AOD and, especially, AAOD, particularly over spatially heterogeneous or highly reflective surfaces. Thus, quantitative atmospheric aerosol absorption cannot be retrieved from existing satellite-based sensors. Conversion of AOD to AAOD requires use of an assumed column aerosol single-scatter albedo, such as from a model [e.g., as employed by Chung et al., 2012]. Although satellite retrievals may infer the separation between fine and coarse particles, they cannot distinguish AOD from BC versus that from other fine mode aerosols like OA or sulfate. The division must be estimated either from the relative magnitude of modeled concentration fields or from ground-based observations. The relatively coarse resolution at which most satellite sensors can provide retrieved aerosol fields also makes them susceptible to cloud contamination. Finally, while polar-orbiting satellites have nearly global coverage, data coverage at any given location is limited and does not always allow for sufficient sampling for good statistics.

5.5 Comparison Between Modeled and Observed BC Concentrations and AAOD

[288] Estimates of forcing by individual aerosol species, including BC, are all based on aerosol distributions generated by global chemistry transport models (CTMs). Even estimates that incorporate observations require model results to expand concentration fields beyond observed locations or to separate observed aerosol influences among chemical species. Evaluation of BC concentrations in these models is therefore critical in assessing our understanding of climate forcing. A first-order evaluation of these models is accomplished by comparing either simulated aerosol concentrations or simulated optical depths with independent observations. This comparison requires long-term, spatially distributed aerosol measurements, so that observations from nationwide or global networks, such as the ones discussed above, are particularly useful.

[289] Comparisons between models and observations shed light on our understanding of the quantity of BC in the atmosphere, but they often cannot distinguish between various possible causes of model error. For example, a model could underestimate atmospheric concentrations if its emissions are too low or if its overall removal rate is too high.

[290] Koch et al. [2009b] evaluated the AeroCom suite of global models by comparing modeled BC concentrations with observations from several intensive campaigns and long-term in situ measurements. Figure 13 summarizes this comparison, showing the ratio between observed and modeled surface concentrations of BC. The figure shows that the median model predicts surface BC concentrations in the North America fairly closely, slightly overestimates BC in Europe, and clearly underestimates it in Asia. The global annual average difference among models in this suite was about 30% of the median concentration; variations in individual seasons can be greater.

Details are in the caption following the image
Evaluation of model performance and model sensitivities for simulating BC concentrations in surface air for specific regions and the rest of the world. Box plots show ratios between observations and models (i.e., the factor by which modeled values are multiplied to obtain observed values). The middle bar in each box shows the median model value. Box boundaries show 25th and 75th percentiles. Data points outside 1.5 times the interquartile distance from the box are marked as outliers; otherwise, they are included in the whiskers. Symbols show sensitivity experiments using different emission databases (red triangles) and removal rates (blue squares) in a single model (GISS) with the base case shown as filled symbols. For emission experiments, SPEW is shown with red fill and GAINS with black fill; unfilled symbols show results using two other emission databases. The number of surface observations is shown below each region name. Data from Koch et al. [2009b].

[291] Koch et al. [2009b] also compared modeled fields with values of AAOD inferred from AERONET. Modeled AAOD was lower than observed values in all regions. Column BC AAOD is the sum over altitude of the product of BC concentration and MACBC. Because most of the models used in the comparison do not represent internal mixing, the modeled MACBC (section 3.7) was underestimated. This bias could explain the disagreement in North America and Europe but not in other areas of the world. However, AAOD data were not separated into contributions from BC and dust. A refined analysis, which considers this separation and incorporates more AERONET stations, appears in section 6 and is more fully described in Appendix B.

[292] The sources of differences between model predictions are not completely understood. For any given location, the inter-quartile range of the concentrations predicted by the 15 models participating in AeroCom was a factor of about 3, and the range excluding outliers was a factor of about 5 [Kinne et al., 2006; Schulz et al., 2006]. Koch et al. [2009b] provided sensitivity experiments to explore the sources of this variability by using one model from the suite (GISS, shown as symbols in Figure 13). Simulations with different emission inventories are given in red, and concentrations with removal rates increased or decreased by a factor of 2 are shown as blue squares. Filled symbols indicate base-case results. Varying removal rates and emissions do cause significant variations in a single model, but these uncertainties do not explain most of the inter-model differences. Textor et al. [2007] also found that inter-model differences were only partially explained by differences in emission inventories. Large differences in modeled horizontal and vertical transport are largely responsible for the diversity. Once BC is lifted into the free troposphere, removal processes are slower, and its atmospheric lifetime may be extended.

[293] Figure 14 compares modeled and observed values of BC AAOD. The left panel shows BC AAOD inferred from AERONET observations (Appendix B). Although these values include absorption by organic matter, we use them as our observational estimate for BC AAOD. Section 11.4 discusses implications of this combination. Modeled BC AAOD from the 15 AeroCom models are shown in the middle panels of the figure. We use this particular collection because it contains a large number of consistent modeled fields, but this presentation excludes subsequent improvements in many of these models. The AAOD fields shown in Figures 12 and 14 are from the AeroCom “median model,” for which a median of the local (1° × 1° grid resolution) values from all AeroCom models was calculated every month. These median fields, unlike averages, lose their additivity, but the use of median values excludes the influence of individual outlier models. Figure 14 also presents differences in BC AAOD between co-located AERONET and model median values, showing that these models tend to underestimate BC AAOD. The regional dependence of model biases are discussed further in section 6.

Details are in the caption following the image
Seasonal absorption aerosol optical depth (AAOD) at 550 nm due to BC as inferred from observations and from models. The left panels show seasonal averages for BC AAOD at AERONET sites, as they were derived from sky-inversion data that were sampled during the last decade (2000–2010). Each symbol represents an AERONET site. The middle panels show seasonal BC-AAOD median maps of AeroCom models (listed in Table 14) for year 2000 conditions (see text for methods). The right panel shows differences between the BC-AAOD values inferred from AERONET observations (left column) and AeroCom modeling efforts (middle column) at each location, illustrating deficiencies in global modeling of BC.
Table 14. Global Industrial-Era or All-Source Black-Carbon Mass-Related Properties as Simulated by Several Modelsa
Model Reference EMIb Lifetimec Load Above 5 kmd Ind-Era Fraction BC AODe Loadf Mixingg MACh BC AAODi
Tg yr−1 d % % mg m−2 m2 g−1 *1000

Schulz et al. [2006]

6.3 7.2 21.2 80 0.24 Ext 8.4 1.83

Schulz et al. [2006]

6.3 7.3 19.1 80 0.25 Ext 8.0 1.98

Schulz et al. [2006]

6.3 7.5 40.6 73 0.25 Ext 4.4 1.11

Schulz et al. [2006]

6.3 4.9 6.1 0.16 Int 7.7 1.23

Schulz et al. [2006]

6.3 10.6 19.1 78 0.37 Ext 9.8 3.50

Schulz et al. [2006]

6.3 5.5 14.5 88 0.19 Ext 7.2 1.34

Schulz et al. [2006]

6.3 5.5 32.7 89 0.19 Int 10.5 1.95

Schulz et al. [2006]

6.3 5.8 17.5 79 0.19 Ext 6.8 1.29
Other models

Zhang et al. [2012]

7.7 3.31 16 0.14 Ext 4.3 0.60

Kim et al. [2008]

14.4 3.75 0.29 Ext 10.6 3.06
CAM5 3-mode

Ghan et al. [2012]

5.7 3.6 68 0.11 Int 15.0 1.65

Jacobson [2010]

4.7 9.9 0.25 Ext + Int 14 3.11

Chung and Seinfeld [2002]

11.5 6.4 91 0.39 Ext 7.8 3.01j
11.5 6.4 91 0.39 Int
GISS-MATRIX Bauer and Menon [2012] 2.3 6.3 0.08 Int 8.2 0.66

Textor et al. [2006]

13.9 7.16 32 0.53 Ext 8.1 4.1

Ramanathan and Carmichael [2008]

13.9l n/a 5.8m

Chung et al. [2012]

6.3 5.5 91 0.19 Ext n/a 7.7

Myhre et al. [2009]

6.3 5.5 91 0.19 Int 7.3 1.39j
  • a AeroCom models UMI-SPRINTARS represent difference between present day and preindustrial simulations: AeroCom experiments B and PRE.
  • b EMI = Emissions. For comparison to emission totals in the rest of the document, 1 Tg = 1000 Gg.
  • c Lifetime computed from load and emissions.
  • d Load above 5 km refers to the fraction of BC mass above 5 km altitude [from Textor et al., 2006].
  • e Ind-era fraction BC AOD: Industrial-era fraction of BC aerosol optical depth.
  • f Load is global area weighted column mass of BC.
  • g Int = Treatment or explicit assumption of internal mixing; Ext = External mixing, or no explicit treatment of internal mixing, even if a relatively high value of MAC was uniformly applied to all aerosol. Some models used external mixing for the AeroCom study [Schulz et al., 2006] but are capable of internally mixed treatments too.
  • h MAC: Mass absorption coefficient due to BC.
  • i Absorption aerosol optical depth due to industrial-era BC, except for MACR-Assim, GOCART, GATOR, and BCC_AGCM, which use all-source BC.
  • j Inferred from column burden of BC and MACBC given for pure BC. This calculation cannot be done for internal mixtures, as MACBC depends on the aerosol composition.
  • k All-source BC is reported for these models rather than the industrial-era component.
  • l Emissions in underlying GOCART model but modified through assimilation.
  • m C.E. Chung, personal communication [2010] and Chung et al. [2005].
  • n/a = not applicable.

[294] The quantity and representativeness of observation sites is fundamentally important in interpreting comparisons between models and measurements. North America and Europe have the best coverage of all long-term measurement sites, followed by Asia. The number of sites is not in proportion to the strength of emissions, and concentrations in regions with large emissions and, therefore, highest aerosol forcing, are relatively poorly constrained.

[295] Remote regions with low concentrations also have few measurement sites and hence are also poorly constrained. Although aerosol forcing in these regions is small, their large spatial coverage means that contribution to the global-average forcing may be important. Compared with BC concentrations measured during a field campaign over the remote Pacific Ocean (Figure 15) [Schwarz et al., 2010], the median model overpredicted remote BC concentrations by a factor of 5. Some models, such as the lower quartile of the AeroCom model suite or GATOR [Jacobson, 2012], show better comparisons with these remote measurements.

Details are in the caption following the image
Airborne, in situ measurements of BC made in the remote Pacific Ocean between 80°N and 67°S in January 2009 using an SP2 instrument. Upper left panel: average measured BC mass concentrations observed in five latitude bands with whiskers representing atmospheric variability. In the additional panels, colored lines repeat the observed average profiles and include the range of BC mass mixing ratios from the AeroCom model suite. The legend explains lines and shading used to represent model results. The whiskers are numerically symmetric about the average values but are omitted on the left side to simplify the presentation on a log scale. Adapted from Schwarz et al. [2010].

[296] Koch et al. [2009b] note that model diversity in atmospheric BC concentrations is particularly high for the Arctic. Models almost universally underestimated concentrations in the free troposphere during spring, and they did not match the observed decrease in concentrations from spring to summer. This finding is based on a very limited data set from intensive field campaigns, but it is consistent with multi-year observations from surface sites in the Arctic, which show that most models underestimate BC concentrations in the winter and spring and generally do not capture strong seasonal variations in boundary-layer BC concentrations [Shindell et al., 2008; Huang et al., 2010; Liu et al., 2011].

5.5.1 Estimating Emissions by Integrating Models and Observations

[297] The comparisons between modeled and observed BC concentrations and column absorption described above are a first step in evaluating model performance. More detailed evaluations have been conducted, usually focusing on inferring the most likely sources of error. The simplest of these compare the temporal or spatial dependence of atmospheric concentrations with that given by models. More complex approaches use a mathematically strict evaluation of sources and sinks using a formulation of Bayes' theorem, which has been demonstrated for CO2, CO, and aerosols [e.g., Chevallier et al., 2005; Dubovik et al., 2008; Chevallier et al., 2009; Huneeus et al., 2012]. Evaluation methods that use measurements of atmospheric abundance in combination with modeled fields are broadly called inverse methods. Inverse modeling of gaseous concentrations on continental scales, particularly CO and oxides of nitrogen (NOx), can also provide information about emission sources. Although this technique has usually been applied to infer emissions, it should not be forgotten that model processes such as transport and removal can also contribute to errors and that the initial estimate of emissions has a strong influence on the inferred emissions. Here we summarize the results of inverse modeling studies that have focused on particular regions.

[298] 1. North America. An inverse model by Hu et al. [2009] found that USA BC emission estimates using data mainly from a rural network (IMPROVE) was about 30 to 35% lower than an estimate that also included urban sites. This implies that in order to constrain concentrations on continental scales, measurements in high-concentration areas should be included if models have sufficient resolution, and that this inclusion may affect the total continental burden of aerosol. There are also differences in the thermal evolution methods for measuring BC (section 3.3) between urban and rural sites in the United States. The former tends to use a transmittance-corrected thermal-optical method, while the latter corrects using reflectance and heats the sample to a lower temperature. Before performing an inverse estimate, these authors had to multiply the transmittance-corrected values by factors ranging from 1.7 to 2.6, depending on the season. The resulting emission estimates were similar to the bottom-up emission estimate in Figure 9. On the other hand, an inverse model by Park et al. [2003] estimated a BC emission rate of about 750 Gg yr−1, which is about 70% higher than the bottom-up estimate.

[299] Bhave et al. [2007] used organic markers to show that USA emission estimates of total aerosol carbon (OC plus BC) from vehicle exhaust and biomass combustion were not biased. They did find a large unexplained source of total aerosol carbon unassociated with combustion, but this would not affect BC totals.

[300] Inverse modeling of carbon monoxide has indicated that inventories overestimate fossil fuel sources in North America by a factor of 3 in summer and two in spring [Miller et al., 2008]. Kopacz et al. [2010] found a large underestimation especially in winter, ascribing this difference to heating and poor vehicle efficiency at cold temperatures. Although BC model results in the United States do not appear highly biased, these findings indicate that some small sources that could affect BC emissions are not well understood.

[301] 2. Europe. Tsyro et al. [2007] found that a model based on the IIASA inventory underestimated BC concentrations by about 20%. Significant overestimate and underestimate of BC were correlated with concentrations of levoglucosan, a tracer of wood smoke, indicating that biofuel or open biomass burning could be responsible for the difference. Modeled concentrations were generally too high in Northern Europe and too low in Southern Europe. Mobile source emissions were thought to be underestimated in Austria and some Eastern European countries.

[302] 3. East Asia. Hakami et al. [2005] used BC concentrations observed in field campaigns to constrain emission over eastern Asia. They found that the magnitude of the total emission inventory given by Streets et al. [2003b] did not change significantly as a result of the assimilation. However, anthropogenic emissions over southeastern China were reduced, while those in northeast China and Japan were increased. The increase in the industrialized regions is consistent with the low bias in estimates of CO emissions [Kasibhatla et al., 2002; Heald et al., 2004], which is more strongly associated with the more northern industrialized regions [Yumimoto and Uno, 2006]. The observed CO bias has been largely resolved by improving the fraction of small combustors [Streets et al., 2006], as discussed in section 4.6.5. Tan et al. [2004] suggested that modeled particulate carbon emissions in East Asia should increase in order to be consistent with observations, but they did not separate BC and OC nor account for the formation of secondary organic aerosol, which could increase atmospheric concentrations. Studies that compare ground and aircraft measurements with modeled BC values over the East China Sea estimate an annually averaged BC emission flux of 1920 Gg yr−1 [Kondo et al., 2011a], close to bottom-up emission estimates for the late 2000s. Inverse modeling of NOx emissions in East Asia indicates rapid growth in high-temperature sources [Kurokawa et al., 2009], although high-NOx sources are presently a small part of the East Asian BC inventory.

[303] 4. South Asia. Rasch et al. [2001] estimated that increases of about 10 to 20% in fine-mode aerosols were needed in the model to match satellite observations over South Asia, but this study did not specifically identify biases in BC. Dickerson et al. [2002] suggested a BC emission rate of 2000–3000 Gg yr−1, much higher than bottom-up estimates, based on atmospheric ratios of BC and CO that are much higher than is given by bottom-up estimates. Menon et al. [2010] also find that modeled concentrations in South Asia are much too low compared with in situ measurements.

[304] The inverse modeling studies discussed above rely on in situ data from a small number of sites to represent continental-scale concentrations of aerosols. Satellite measurements have much broader spatial coverage and could produce global emission estimates. This approach has been applied in biomass burning regions, where optical depths are large and seasonal. Freitas et al. [2005] found that emission estimates derived from observed AOD were larger than bottom-up estimates. Zhang et al. [2005] used aerosol index from the TOMS instrument to estimate BC from biomass burning as 5700 to 6900 Gg yr−1, depending on the assumptions used, and total smoke (BC plus OM) emissions as 60,000 to 73,000 Gg yr−1. Reid et al. [2009] estimated that bottom-up emission estimates of particulate matter from biomass burning need to be enhanced by factors of 1.5 to 3. The global emission flux derived for 2006 to 2008 was about 140,000 Gg yr−1, as compared with the value of 30,000 to 44,000 Gg yr−1 BC plus POA given in section 4.4.4, if OA is assumed to be 1.4 times OC. Similarly, Kaiser et al. [2012] derived a global average enhancement factor of 3.4 for the OA and BC resulting from biomass burning, summarizing several other top-down studies that recommended emissions 2–4 times larger than the bottom-up estimates. These studies indicate that BC emissions from biomass burning should be much larger than represented in models, although they may not scale linearly with total aerosol emissions. However, the fraction of the discrepancy attributable to BC is uncertain. Formation of SOA may contribute to the observed OA and particulate matter loading.

[305] In summary, inverse modeling studies have produced inconsistent results for North America, indicating that either BC emissions are estimated well or are too low by 70%. European BC concentrations appear slightly underestimated. East Asian concentrations appear reasonable, especially after the addition of small combustors. Although observations are limited, measurements in South Asia suggest that concentrations are greatly underestimated there. All of the studies described have used single models and await further confirmation. On the other hand, there is a consensus that aerosol concentrations, including BC, may be too low in biomass burning regions. Considering also the likely bias in emission factors (section, emission values that are greater than bottom-up estimates appear warranted.

5.6 Vertical Distribution of BC

[306] As discussed in sections 6 and 7, the vertical distribution of absorbing aerosol affects its forcing. Vertically resolved measurements below 8 km of the refractory aerosol concentration, which correlates strongly with BC concentration, have been carried out since the 1990s [Clarke and Kapustin, 2010]. Remotely piloted airborne sensors have measured optical properties and total aerosol absorption below 3 km [Corrigan et al., 2008].

[307] BC vertical profiles extending through the troposphere have been made with a variety of measurement methods, but the development of the SP2 instrument (section 3.5) has allowed measurements of vertical profiles of BC in very clean air where they were previously unobtainable [Schwarz et al., 2006, Schwarz et al., 2010]. These profiles extend vertically from near the surface to the lower stratosphere, as shown in Figure 15. Although BC vertical profiles are quite variable, in polluted regions they generally show a declining BC mass mixing ratio from the surface to about 4 km altitude, and relatively constant values up to well above the tropopause. However, in remote regions influenced by the transport of pollution from source regions, BC loadings tend to peak in the free troposphere or above. BC that has been lifted above the low altitude ranges where aerosol removal is fast due to precipitation is more likely to have a longer lifetime in the atmosphere.

[308] Important BC properties, such as mixing state (section 3), also change with altitude. One set of vertical profile observations in the tropics indicates that the coatings of BC-containing particles become thicker and more prevalent with increasing altitude between the surface and the lower stratosphere [Schwarz et al., 2008a]. The fraction of coated BC particles and the acquired coating mass depend on both the BC source and the history of the particle and the air it is contained in. Thus, MACBC (section 3.8) can be expected to depend on altitude.

5.6.1 Comparison Between Modeled and Observed Vertical Distributions

[309] Correctly modeling the vertical distribution of AAOD is important for calculations of direct radiative forcing, as discussed in section 6.6.1. Schwarz et al. [2010] compared the AeroCom set of models with a global-scale “snapshot” of SP2 BC measurements from remote locations (Figure 15). The AeroCom model ensemble captured the vertical trends of BC concentration in the southern mid to high latitudes, but in northern middle and equatorial latitudes, the models underestimated the decrease in concentration with altitude. This comparison in regions far from sources suggests that modeled removal of BC is an important source of error. However, the aircraft data in this comparison were obtained over a limited time range; in particular, the profile for the northernmost band (60–80°N) was from a single flight. Koch et al. [2009b] also compared BC vertical distributions between AeroCom models and SP2 measurements from several intensive field campaigns, mainly over North America and the Arctic. That comparison indicated that models overestimate middle and upper troposphere BC at mid-latitudes, consistent with Schwarz et al. [2010], but underestimate BC in the Arctic. In urban areas, model predictions of column loadings were consistent with observations [Schwarz et al., 2006], but the vertical distribution was not well predicted.

[310] Observations of the vertical distribution of AOD have become available in recent years with active remote sensing by the space-borne CALIOP lidar [Winker et al., 2010], for which multi-year averages are shown in Figure 16. AOD includes scattering and absorption by BC, dust, and other aerosol types (e.g., sulfates, organics, and sea salt) with sizes greater than 0.1 µm diameter [Omar et al., 2009]. Although chemical speciation is not possible, limited inferences about aerosol type can be obtained through depolarization data that identify non-spherical particles and permit estimates of dust contributions. Geographical distributions of full-column CALIOP AOD (Figure 16) are very similar to BC and dust AAOD from AeroCom models (Figures 12 and 14), with high loadings in South and East Asia, the Middle East, Africa, and Latin America. Maximum total AAOD values are a few percent of maximum AOD values. The panels in Figure 16 show that AOD is confined largely to altitudes below about 4 km. Although interpretation of CALIOP data is still undergoing refinement, these data also indicate that models tend to place aerosol at higher altitudes than is observed [Koffi et al., 2012].

Details are in the caption following the image
Multi-year averages of the global vertical distribution of mid-visible (532 nm) AOD from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) spaceborne lidar on the NASA CALIPSO satellite. The panel colors show average AOD derived from CALIOP observations from June 2006 to December 2011 (version 3) for specific altitude intervals referenced to sea level (see legends). The color bar has separate scales for the 0 to 6 km and full-column data panels as indicated. The full-column AOD is shown in the bottom right panel. The gray shading indicates “no observations” usually because of elevated land surfaces. CALIOP statistics were provided by D. Winker of NASA Langley Research Center (Hampton, VA, USA).

5.7 Causes of Model Biases

[311] The comparisons between models and observations and the inverse model studies discussed above provide some consistent messages regarding modeled atmospheric distributions of BC. In the following sections, these observations are used to adjust modeled BC AAOD and direct radiative forcing (section 6) and cloud forcing (section 7). The source of these model biases is less clear. Large underestimates in modeled BC concentrations strongly argue that there are low biases in emission estimates. However, it is difficult to attribute these biases wholly to errors in model emissions, because transport patterns and removal rates can also affect spatial distributions. Transport and removal rates are not independent factors, because aerosol transported to higher altitudes is less likely to be removed and, therefore, has a longer lifetime and greater horizontal transport.

[312] At least one model sensitivity study [Koch et al., 2009b] (section 5.5; Figure 13) indicates that varying removal rates by a factor of 2 cannot fully explain the inter-model range in concentrations or the bias between models and observations. Significant overestimates in BC concentrations over the remote Pacific are consistent with vertical transport that is too strong or scavenging rates that are too low, but significant underestimates in concentrations in the Arctic free troposphere indicate opposite biases in these processes. Across-the-board adjustments such as altering wet scavenging rates may improve biases in one region but make them worse in another. It is likely that different regions have different sensitivities to each of these processes, requiring focused studies in regions where biases have been observed. For example, Garrett et al. [2010] argue that scavenging plays a dominant role in controlling Arctic BC distributions, but this may not be true in other regions such as the remote Pacific. Concentrations near sources are most sensitive to emissions, while concentrations in remote regions depend on transport and removal as well as emissions. Although concentration biases could be reduced by first constraining emissions with near-source concentration measurements and then focusing on transport and removal using measurements downstream from sources, such an activity is beyond the scope of this study, which instead resorted to an after-simulation adjustment that can be applied to all previous simulations that save the requisite information.

5.8 Summary of Limitations in Inferring Black Carbon Atmospheric Abundance

[313] Like all aerosols, BC concentrations are spatially and temporally variable. No measurement strategy provides global coverage of BC concentrations or of AAOD. Space-based sensors have varying limitations: they detect aerosol abundance but not composition (MODIS), provide qualitative measurements of absorption (MISR), or preferentially detect BC at greater altitudes (OMI). Estimating global distributions of BC or absorbing aerosol must therefore rely on modeled aerosol fields, which in turn are constrained by ground-based observations. Ground-based AERONET stations are widespread and produce reasonably accurate AAOD, but coverage is not complete and techniques to separate BC, OM, and dust absorption from these observations are not yet mature. In situ surface measurements have less coverage than AERONET but greater specificity in identifying individual species. These measurements have been used to broadly evaluate modeled global fields and to assess modeled emissions in selected regions. However, observations are sparse in some of the regions with the greatest emissions. A systematic evaluation of model biases caused by emissions, removal rates, and vertical transport, leveraging near-source and remote measurements, has not yet been accomplished.

6 Black-Carbon Direct Radiative Forcing

6.1 Section Summary

  1. [314]

    Black carbon in the atmosphere reduces the planetary albedo by increasing the absorption of sunlight. The perturbation in the radiative balance resulting from this change is called black-carbon direct radiative forcing (BC DRF).

  2. [315]

    The best estimate for the BC DRF in the industrial-era is +0.71 W m−2 with an uncertainty range of +0.08 to +1.27 W m−2. The best estimate of the global annual mean all-source BC DRF is +0.88 W m−2 with an uncertainty of +0.17 to +1.48 W m−2. Locally, BC-DRF forcing can be larger, on the order of +10 W m−2, for example, over regions of East and South Asia. These values are estimated by scaling modeled forcing to the observationally constrained estimate of BC absorption aerosol optical depth (BC AAOD) and reducing the estimated forcing by 15% to account for the incorrect vertical distribution of BC. Our estimate of BC AAOD includes absorption by organic aerosol (OA).

  3. [316]

    Most BC-DRF estimates originate with simulations of BC lifecycle and radiative transfer using global aerosol models. To improve confidence in these estimates, model-derived fields, such as aerosol optical depth, BC AAOD, or BC distributions, are often adjusted to match observations of AOD, BC AAOD, or concentration, either by direct scaling or by tuning removal rates.

  4. [317]

    BC DRF can be considered the product of four factors: emissions, lifetime, mass absorption cross section (MACBC), and radiative forcing per unit absorption optical depth (termed forcing efficiency). Diversity in previous global annual estimates of industrial-era BC DRF (+0.2 to 0.9 W m−2) is caused by variations in modeling assumptions, as evident in the modeled diversity of these four factors: 6000 to 18,000 Gg yr−1 for industrial-era BC emissions, 3.8 to 11.4 days for BC lifetime, 4.4 to 13.4 m2g−1 for MACBC, and 90 to 270 W m−2 per AAOD for forcing efficiency.

  5. [318]

    BC emissions, lifetime, and mass absorption cross section—three of the governing factors—combine to produce BC AAOD, which can be constrained by measurements. Comparison with remote-sensing fields indicates that BC AAOD from most models appears too low. Applying regionally dependent scale factors to match observations increases globally averaged, industrial-era BC AAOD by a factor of 2.9, from 0.0017 to 0.0049. Some of the underestimate can be attributed to the lack of modeled enhanced absorption due to internal mixing, which would increase BC AAOD by about 50%.

  6. [319]

    If the remaining model bias is caused by burdens that are too low, then model burdens should be increased by factors of 1.75 to 4 in five regions (Africa, South Asia, Southeast Asia, Latin America, Middle East, and the Pacific region). Simulated burdens in North America, Europe, and Central Asia are approximately correct considering uncertainty in BC AAOD and MAC.

  7. [320]

    The altitude of BC in the atmosphere and its location relative to clouds affects its interaction with solar radiation and hence the forcing efficiency. Solar absorption by BC above low clouds is particularly important and is poorly constrained in models. Limited observations suggest that most models overestimate high-altitude BC at tropical and mid-latitudes and, thus, BC DRF. Our central estimate of forcing includes a 15% reduction to account for this bias.

  8. [321]

    If all of the differences in modeled burden were attributed to emissions, global emissions would be 17,000 Gg BC yr−1, or an 87% increase above modeled emissions and a 100% increase above the bottom-up emissions in section 4.4. Industrial-era emissions would be 13,900 Gg BC yr−1. These values are associated with concentrations in approximately the year 2005. We use the latter value as a best estimate of emissions throughout the remainder of this assessment, although poorly modeled aerosol lifetimes could also explain the discrepancies.

  9. [322]

    The interpretation of remote-sensing data involves assumptions that affect the amount of absorption and, hence, radiative forcing, that is attributed to BC versus dust. Uncertainties in this division are responsible for nearly half the uncertainty in BC forcing.

  10. [323]

    If OA does contribute significantly to absorption at 550 nm, then our estimate of observationally constrained BC DRF is biased high, but total forcing by OA would correspondingly become less negative or more positive. About 10% of our BC AAOD may be due to absorption by OA.

  11. [324]

    The uncertainty range (+0.08 to +1.27 W m−2) for the best estimate of industrial-era BC DRF includes relative uncertainties of about 40% associated with the extrapolation of global BC AAOD, about 50% in the radiative forcing efficiency per unit BC AAOD, and asymmetric uncertainty due to the difficulty in separating BC AAOD from dust AAOD.

6.2 Introduction

[325] Direct radiative forcing by BC refers to the change in the energy balance at the top of the atmosphere due to absorption and scattering of sunlight by BC in the atmosphere. All estimates of BC DRF are based on the difference between radiative transfer calculations for the atmosphere with present-day BC and with a background level of BC. The BC spatial distribution is generally taken from global models of the BC atmospheric lifecycle, which includes emissions, transport, aging, and removal. As described in section 2.3.2, the background level of BC for industrial-era radiative forcing is set by models using year-1750 emissions. For all-source radiative forcing, the background level of BC is zero. Thus, “industrial-era” in this section, when used for any quantity including BC burden, means the difference between modeled quantities with present-day emissions and the same quantities modeled with 1750 emissions.

[326] As with aerosol concentrations, radiative forcing is highly variable in space and time. Global modeling of BC concentrations [Liousse et al., 1996; Cooke and Wilson, 1996] and radiative forcing [Penner et al., 1998] began in the middle to late 1990s. Since then, many groups have estimated both concentration and radiative forcing fields. An example of modeled direct forcing is illustrated in Figure 17, which shows radiative forcing simulated by the Community Atmosphere Model CAM5 [Ghan et al., 2012]. As discussed in section 2.3.2, any estimated industrial-era forcing depends on the selection of the “preindustrial background.” Industrial-era BC DRF is estimated to be negative in regions such as the eastern United States and the United Kingdom, due to declines in biomass burning since 1750. However, the emission inventories used for the background are associated with considerable uncertainties, especially for open burning. The largest forcing is over the Congo Basin in Africa and southeast China, and Bangladesh, with significant forcing also in the Amazon Basin, over northern India and Indonesia, and off the eastern coasts of South America and Africa where BC is transported over low clouds from distant fires.

Details are in the caption following the image
Annual mean direct radiative forcing by BC from 1850 to present day, as estimated by the CAM5 AeroCom model and adjusted according to the global annual-mean BC-AAOD scale factor of 3.0 derived in the comparison with AERONET observations (Table 15 and section 6.7.1). Forcing is negative in regions where biomass or fossil fuel (e.g., coal) emissions were large in 1850. Year 1750 emissions from these regions would be lower, so the total radiative forcing would be more positive. Based on Liu et al. [2012] and Ghan et al. [2012].
Table 15. Global Averages of Aerosol Absorption Optical Depth (AAOD) Industrial-Era (ind.-era) or All-Source BC Direct Radiative-Forcing (DRF) Measures as Simulated in Several Modelsa
Model Values Normalized Values Scaled BC-DRF
Model BC Quantity BC AAODb BC-DRF ToAc BC-Atm RFd AFEe Normalized DRFf Ind.-era ToAg All-Source ToAh
Units *1000 W m−2 W m−2 W m−2/AAOD W g−1 W m−2 W m−2
AeroCom models
GISS Ind. era 1.83 0.22 120 920 0.59 0.73
LOA Ind. era 1.98 0.32 159 1280 0.78 0.96
LSCE Ind. era 1.11 0.30 0.85 270 1200 1.32 1.63
MPI-HAM Ind. era 1.23 0.20 165 1250 0.81 1.00
SPRINTARS Ind. era 3.50 0.32 1.00 91 870 0.44 0.55
UIO-CTM Ind. era 1.34 0.22 164 1160 0.80 0.99
UIO-GCM Ind. era 1.95 0.36 184 1900 0.90 1.11
UMI Ind. era 1.29 0.25 0.89 194 1320 0.95 1.17
Other models
BCC-AGCM All 0.60 0.10 166 710 0.82 1.01
CAM3 ECA Ind. era 3.06 0.57 1.57 186 1970 0.91 1.13
CAM5 3-mode Ind. era 1.65 0.30 182 2730 0.89 1.10
GATOR All 3.11 0.55 177 1830 0.86 1.07
GISS-GCM II ext Ind. era 3.01 0.51 170 1310 0.83 1.03
GISS-GCM II int Ind. era 0.79 2030
GISS-MATRIX Ind. era 0.66 0.15 227 1.11 1.38
GOCART All 4.1
MACR-Assim All 5.80 0.90 2.6 155 0.76 0.94
MACR All 7.70 0.75 2.75 97 0.48 0.59
UIO-CTM ext Ind. era 1.30 0.26 187 1370 0.98 1.21
UIO-CTM int " 0.33 1740
Average ± 1σ 171 0.84 ± 0.21i 1.03 ± 0.26i
  • a All model values are those reported by the listed model for either industrial-era (Ind. era) or all-source (All) BC emissions. See Table 14 for references associated with each model.
  • b BC AAOD: Absorption optical depth due to modeled BC, either industrial era or all-source emissions.
  • c BC-DRF ToA: Top of atmosphere all-source (MACR-Assim and GOCART) or industrial-era (all other models) shortwave radiative forcing from unscaled AeroCom models.
  • d BC-Atm-RF: This is the warming by BC of the atmospheric column, which is much greater than the ToA radiative forcing.
  • e AFE: Normalized radiative forcing per unit absorption aerosol optical depth.
  • f Normalized DRF: Direct radiative forcing normalized to BC column load.
  • g Industrial-era BC-DRF ToA scaled: Top-of-atmosphere industrial-era shortwave radiative forcing after scaling to 0.0049 industrial-era absorption optical depth.
  • h All-source BC-DRF ToA adjusted: Top-of-atmosphere all-source shortwave radiative forcing by BC after scaling to 0.0060 all-source absorption optical depth.
  • i Final central estimate of direct forcing is 15% lower than these values to account for BC vertical distribution (section 6.7.2).
[327] To understand the factors that affect radiative forcing in models, it is useful to express the global-mean direct radiative forcing by BC, BC DRF, as the product of four factors [Schulz et al., 2006]:
where E is the global mean BC emission rate, L is the global mean lifetime of BC governed by removal, MACBC is the global-mean mass absorption cross section discussed in section, and AFE is the global mean absorption forcing efficiency (forcing per aerosol absorption optical depth). This relationship is summarized in Figure 18. Three-dimensional models simulate many complex processes that are captured only to first order by the factors in equation 6.1.
Details are in the caption following the image
Schematic showing the relationships between principal diagnostics of models used to simulate BC direct radiative forcing (DRF) and the role of models in deriving these diagnostics. The arrows show which diagnostics can be multiplied to yield another diagnostic. The schematic expands upon equation 6.1 which expresses DRF as a product of factors DRF = E L MACBC AFE. The diagnostics shown within double lines can be directly inferred from observations.

[328] Additional diagnostics shown in Figure 18 can be used to compare model results and identify the reasons for differences in simulated radiative forcing. Modeled L can be determined from the ratio of global mean column burden of BC to E, and global average MACBC is the ratio of the global mean absorption optical depth of BC to the global mean burden. Modeled AFE can be determined from the ratio of the global mean BC radiative forcing to BC AAOD, which is the global mean product of E, L, and MACBC. BC AAOD is a particularly powerful diagnostic that can be compared with atmospheric measurements and is therefore emphasized in this section. Although AOD can also be used to evaluate modeled aerosol quantities, atmospheric AOD is dominated by scattering and thus has large contributions from other non-absorbing aerosol components.

[329] Table 14 summarizes modeled global mean estimates of each term in equation 6.1 as well as other model diagnostics. Many of these models were used as the basis for IPCC radiative forcing estimates [Forster et al., 2007]. Estimates of industrial-era BC DRF range from +0.2 to +0.9 W m−2. The large diversity is due to different model choices that affect one or more of the factors in equation 6.1. Global emissions of BC range from 5700 to 18,000 Gg yr−1. Lifetime ranges from 3.3 to 10.6 days. Global mean MACBC ranges from 4.3 to 15 m2g−1. AFE ranges from 91 to 270 W m−2AAOD−1 (Table 15). Because the correlation among these four factors can be negative [Schulz et al., 2006], and because no model has maximum or minimum values of all four factors, the diversity in the radiative forcing in Table 15 (+0.2 to +0.9 W m−2) is much smaller than the diversity estimated by multiplying the minima and maxima of the four factors (+0.05 to +3.7 W m−2). The lower net diversity in simulated BC DRF is to some extent due to tuning simulations to match observational constraints on BC concentration and AOD. The best estimates of each component are summarized in sections 6.3-6.5, 6.5.1, 6.5.2, 6.6 before providing a best estimate of BC DRF in section 6.7.

6.3 Emission, Lifetime, and Burden

[330] As discussed in section 4.4, BC emission rates used in estimates of radiative forcing vary regionally and globally [Haywood and Boucher, 2000; Bond et al., 2004]. Commonly used values of global industrial-era emissions are 6300 Gg yr−1 for AeroCom [Schulz et al., 2006], but emissions as high as 18000 Gg yr−1 have been used [Chin et al., 2002]. For the Climate Model Intercomparison Program simulations for the IPCC Fifth Assessment Report, the value was 5700 Gg yr−1 for present-day emissions [Lamarque et al., 2010].

[331] Column burden is proportional to BC emissions and to modeled lifetime. Lifetime is determined by removal, which is represented differently in each model and varies by location and season. Global average lifetime for BC varies by a factor of more than 3 for the models reported in Table 14. Removal in precipitation plays a large role in aerosol lifetime [Ogren and Charlson, 1983], which is difficult to represent accurately at the coarse resolution of global modeling [Rasch et al., 2000; Textor et al., 2007; Croft et al., 2010]. Differences in dry deposition contribute to uncertainty as well [Easter et al., 2004; Bauer et al., 2008; Vignati et al., 2010]. Because removal processes depend on altitude, aerosol lofting also affects lifetime. For example, Koch et al. [2007] showed that lofting of biomass-burning aerosols from South America led to longer lifetimes compared with African emissions. Most developers of aerosol models have applied surface observations of BC concentrations and surface and satellite retrievals of AOD to constrain emissions or the treatment of scavenging, so one would expect the product of emissions and lifetime (burden) to agree broadly with observations [Vignati et al., 2010].

[332] Early models used a limited number of parameters to account for sub-grid variability in precipitation and the efficiency with which particles are scavenged. These scavenging treatments used crude parameters that were applied globally. More advanced models treat the coating of BC with hygroscopic material to form particles that are more readily activated to form cloud droplets and scavenged more efficiently. The representation of scavenging affects the magnitude and seasonality of BC concentrations, especially at remote locations [Croft et al., 2005]. Model representation of the BC seasonal cycle in the Arctic improves with more accurate microphysical treatments [Vignati et al., 2010; Lund and Berntsen, 2012; Park et al., 2011].

[333] The global mean column burden of BC ranges from 0.11 to 0.53 mg m−2 in Table 14. The studies with the largest radiative forcing estimate (MACR-Assim) [Ramanathan and Carmichael, 2008; Chung et al., 2012] did not calculate a burden, but the BC simulation [Chin et al., 2002] that the studies relied on for absorption far from AERONET sites [Chung et al., 2005] had a global mean BC column burden of 0.6 mg m−2. Some of the models did not constrain the combination of emissions and lifetime with BC concentration measurements [Vignati et al., 2010], including the model that estimated the largest BC burden (i.e., SPRINTARS; Table 15) [Takemura et al., 2005]. Some of the diversity in the column burden of BC in Table 14 can be attributed to differences in emissions. Yet many of the estimates used the same BC emission inventory and still yielded BC burdens differing by more than a factor of 2 (0.16 to 0.38 mg m−2) due to differences in lifetime [Schulz et al., 2006]. As discussed in section 5.5, most models simulate surface BC concentrations within a factor of 2 of measurements in most regions, with weak over-predictions in Europe, underpredictions in biomass burning regions, and large underpredictions for eastern and southern Asia [Koch et al., 2009b].

[334] Vertical lofting of BC affects its prospects for wet scavenging and hence its lifetime. Vertical profiles of BC show that simulated concentrations have large biases in the free troposphere [Koch et al., 2009b]. High biases of a factor of 3 to 15 (average 8) are found in tropospheric BC concentrations above the boundary layer at tropical and subtropical latitudes, while most models underestimate the BC concentrations in the Arctic troposphere, on average by a factor of 2.5 [Koch et al., 2009b, Table 8]. Adjustments to simple scavenging treatments cannot reproduce these observed distributions; increased removal in tropical regions would reduce the over-prediction at high altitudes there (section 5.5) but would enhance the low bias in the Arctic troposphere where abundances may be controlled by scavenging processes [Garrett et al., 2010]. However, improved treatments of scavenging may result in greater fidelity.

6.4 Mass Absorption Cross Section

[335] Section 3.5 discussed measured and modeled MAC of BC and particles that contain BC. Section 3.5 also summarized how mixing of BC with other aerosol components would increase MACBC, and field measurements have found that BC particles are largely mixed with other components. Since MAC depends on refractive index, water content, particle size, and mixing with other aerosol components, measured MACBC data provide a useful diagnostic check for models when information about each variable is also known.

[336] The global mean MACBC listed in Table 14 ranges from 4.3 to 15 m2g−1, with the AeroCom median around about 6.5 m2g−1. Many models have MACBC values similar to the Bond and Bergstrom [2006] estimate of 7.5 ± 1.2 m2g−1 for freshly emitted, externally mixed BC, and a few others are close to the estimates of about 12.5 m2g−1 for internally mixed aerosol. Two models (LSCE and BCC-AGCM) are much lower than this range. If models that assumed external mixing were excluded or adjusted for the increase in MACBC, the model average MACBC and the average estimated DRF would increase [Jacobson, 2000; Myhre et al., 2009]. The internally mixed value of MACBC may be an upper limit on the global mean value because the actual global mean depends on how much of the BC is mixed with other components throughout the atmosphere.

[337] All global models that represent internal mixing report a large increase in absorption and positive forcing for internally mixed particles. Haywood and Shine [1995] reported doubled forcing due to internal versus external mixing in a simple model. Jacobson [2001a] was the first to confirm that finding with realistic BC spatial distributions and accounting for the evolution of mixing state. Chung and Seinfeld [2002] found a forcing increase of 36% for BC with homogeneous internal mixing compared with external mixing. Bauer et al. [2010] included a core-shell representation in a multi-modal model. Flanner et al. [2007] and Myhre et al. [2009] associated internally mixed values of MACBC with aged BC, and Adachi et al. [2010] explored several mixing assumptions in a global model. Model treatment of aging and mixing affects both absorption and lifetime, often with compensating effects. Ghan et al. [2012] found that two treatments of BC produced similar forcing estimates, as internally mixed BC had higher MACBC but was removed faster. Stier et al. [2006b] found regional differences in which effect of BC mixing state dominates.

6.5 Scaling Modeled BC AAOD to Observations

[338] BC AAOD is the product of three of the factors that determine BC DRF (i.e., E, L, and MACBC in equation 6.1), and constraining this product can reduce uncertainty in radiative forcing. In this section, we present an estimate of observationally constrained BC AAOD that is used to produce scaled forcing estimates in section 6.7.

[339] Measurements of AAOD are sparsely distributed, and many models underestimate BC AAOD, as shown in Figure 14. In section, we cautioned that both sampling bias and dust absorption could produce a high bias in BC AAOD inferred from AERONET AAOD. Comparisons between model and observationally derived total AAOD in Koch et al. [2009b] include contributions from both BC and dust. As a result, either or both BC and dust could be responsible for model discrepancies. However, AAOD comparisons for regions and seasons not affected by dust still show that many models simulate too little absorption in the atmosphere and, therefore, underestimate BC DRF.

[340] To obtain an estimate of BC AAOD constrained more closely by observations, a method was developed to scale modeled BC AAOD fields using BC AAOD inferred from the AERONET ground-based network. Appendix B gives the details and background of this method. Briefly, sampling biases in the AERONET retrievals of total AAOD were removed, and BC AAOD was estimated from total aerosol AAOD by subtracting the contribution to AAOD by dust. This process allocates all non-dust absorption to BC. To the extent that organic aerosol contributes to AAOD, our estimate of BC AAOD is biased high. Biases in the AeroCom median model BC AAOD were quantified by calculating the ratio of the observationally constrained and modeled values at the same month and location. The scale factors so determined were then extended to other grid boxes within the region, including outflow from that region over the ocean. This expansion of scale factors relies on the spatial distribution of BC in the AeroCom median model.

[341] BC AAOD estimated this way is found to be much greater than modeled BC AAOD, so the procedure increases the globally averaged estimate of total BC AAOD from 0.0021 before scaling to 0.0060 after scaling. As discussed in section 6.9, causes of the resulting BC AAOD changes would likely have affected preindustrial BC AAOD as well, so we applied the same scaling factors to preindustrial BC AAOD. Hence, the global mean industrial-era BC AAOD increases by the same factor, from 0.0017 before scaling to 0.0049 after scaling. These increases in BC AAOD are large, but previous estimates that employ observations have also found BC AAOD much greater than model values (section 6.9).

6.5.1 Sensitivities in AAOD Scaling

[342] A summary of modeled BC AAOD values before scaling is presented in Table 15 and summarized in Figure 19. Industrial-era, global-mean BC AAOD from models without any scaling ranges from 0.0006 to 0.0035, with two models that included internal mixing giving values above 0.0030 and the remainder lying within 30% of the median value. Figure 19 also shows BC AAOD estimates scaled to observations for industrial-era sources and for all sources. For both scaled and unscaled fields, all-source BC AAOD is higher than the industrial-era average by about 23%.

Details are in the caption following the image
Summary of the sensitivity of global BC-AAOD derived from AeroCom models. The scaling of AeroCom BC-AAOD values is based on AERONET observations following the method in Appendix B and discussed in section 6.5. Separate unscaled and scaled AeroCom model values are shown in Table 15. Data points (circles) are AeroCom median values. Ranges are shown with lines and uncertainty with whiskers. Not shown on the figure due to limited scale is an estimate of scaling to average BC AAOD inferred from AERONET version 2.0 averages (0.0143). The sensitivity of scaled all-source BC-AAOD to key assumptions in using the AERONET observations is shown for three cases (see text). Three other model study results are shown at the bottom for comparison. Central estimates, value ranges, and uncertainty ranges are shown in the right-hand column.

[343] Error bars in the scaled estimates in Figure 19 reflect uncertainty in the spatial distribution of BC in the model used to extend scaling to regions without observations. This uncertainty is estimated by applying the full scaling procedure to three additional modeled fields (SPRINTARS, GOCART, and UMI) that were chosen to represent high and low average BC AAOD. The scaled value of global BC AAOD ranges from 0.0046 for SPRINTARS to 0.0064 for UMI, so the highest initial model field results in the lowest scaled value.

[344] The BC AAOD values derived in Appendix B are very sensitive to assumptions imposed in the use of the AERONET data fields, as discussed in Appendix B. To demonstrate this, Figure 19 shows estimates of all-source BC AAOD with three different changes in basic assumptions used in deriving BC AAOD from AERONET observations. The revised AERONET fields are then used to rescale the AeroCom model fields, and a new global estimate is calculated. First, BC AAOD is separated from total AAOD by combining the AERONET size distribution for submicron particles and the retrieved AERONET refractive index. This method of separation reduces the mean BC AAOD from the baseline estimate of 0.0060 to 0.0034 but does not account for the differing properties of submicron and super-micron aerosol. This method ascribes a relatively small AAOD to BC and a large AAOD to dust. Second, the selection of AERONET stations introduces an uncertainty of 30%. When four different subsets of stations were used for the scaling of all-source BC AAOD, global averages ranged from 0.0057 to 0.0071. The analysis presented here includes this uncertainty. Third, if all absorption were attributed to BC, the modeled BC-AAOD field would increase by a factor of 1.33 to 0.0086. Fourth, the AERONET Version 2.0 data exclude low-AOD conditions, so that average AAOD values in that dataset are biased high. If the AERONET Version 2.0 data were used, estimated BC AAOD would increase by a factor of 2.2, to 0.0143 (not shown in Figure 19 due to limited scale). The interpretation applied here does not suffer from this bias because of the method used to interpret the AERONET data. BC-DRF estimates relying on averages of published AERONET AAOD data cannot account for this bias and are expected to be much greater than those derived in this assessment. These sensitivity calculations highlight the importance of clearly stating assumptions and their consequences when using AERONET observations to scale modeled fields of absorbing aerosol. Other uncertainties in this analysis, used to set bounds on BC DRF, are discussed in section 6.7.3.

6.5.2 Reasons for Low Bias in Modeled AAOD

[345] Because BC AAOD is the product of MACBC and BC burden (section 6.2), the mismatch between modeled and observed BC AAOD may derive from errors in modeling both factors. Apportionment is considered separately for AAOD resulting from open biomass burning and energy-related combustion in the ten regions shown in Figures 7-9.

[346] We first assume that the MACBC scale factor for all months and regions is 1.5 (i.e., the approximate ratio between the MAC of internally and externally mixed BC). Most models in the Phase I AeroCom group described here did not account for internal mixing, and the AeroCom median BC AAOD at each grid box likely represents aerosol with this lower MACBC. We also assume that AAOD attributable to open burning and energy-related sources are proportional to their emissions; that is, aerosol lifetimes from the two sources are similar in each month. The scaled BC AAOD values for each source are then averaged over the year to produce an annual BC AAOD and then divided by the original BC AAOD to give an annually averaged scaling factor.

[347] Figure 20 displays the annual scale factors for each region and source group. Although our assumption of similar lifetimes between the source groups may be questionable, the method should produce dissimilar scale factors if more correction is needed during months with high or low biomass burning. The bars for each scale factor show the portions attributable to BC burden (closed) and MACBC (open). The middle panel of Figure 20 also shows the annual fraction of BC AAOD attributed to energy-related emissions versus biomass burning emissions and also shows estimates of the amount attributed to dust in each region. This fraction is not a true estimate of regional dust AAOD, as it does not account for the difference in the dust spatial distribution. Instead, it shows the amount of AAOD that would be added if BC-AAOD fields were scaled to total AAOD instead of BC AAOD. Regions with a high apparent dust fraction have greater uncertainty in BC AAOD because of the difficulties in separating the two. The narrow blue box shows the dust fraction that would be inferred using an alternate technique: applying the AERONET-retrieved refractive index to the fine fraction of the retrieved size distribution. Finally, the rightmost panel of Figure 20 shows the contribution of each land region to scaled global BC AAOD. These percentages do not sum to 100% because 35% of the BC AAOD occurs over oceans. Over half of this BC AAOD, or 20% of the global total, occurs in outflow regions for which scaling was affected by adjacent continents.

Details are in the caption following the image
Summary of regional BC aerosol absorption optical depth (BC AAOD) scaling and contributions. Left: Summary of regional scaling factors required for AeroCom median model values to match BC AAOD constrained by AERONET observations. The regions are shown in Figure 7. Scaling factors are effective annual averages apportioned between open biomass burning and energy-related burning. Each factor is an annual average of monthly scaling factors in each region and includes a prescribed MACBC scaling of 1.5 for all months and regions (open bar), the approximate ratio between the MACBC of internally and externally mixed BC. The remainder of the scaling is attributed to burden. Center: Estimate of BC AAOD fraction in each region due to energy-related combustion, biomass burning, and dust; the dust contribution is not a true estimate of dust AAOD but of dust AAOD in AERONET. The small blue rectangles indicate where the left boundary of the dust AAOD would lie if BC AAOD were determined by applying AERONET refractive indices to the fine mode of the retrieved size distribution. Right: Contribution of modeled, scaled BC AAOD over land to global total BC AAOD. These percentages do not sum to 100% because BC AAOD over oceans is excluded.

[348] Burden scale factors greater than unity (solid bars) in Figure 20 indicate that the AeroCom median model usually underestimates the burden. Differences between energy-related and biomass burning scaling factors are small, indicating that corrections are not wholly attributable to either one. The greatest burden underestimates, by factors of 3–4, are found in South and Southeast Asia. Most other regions require significant increases of 60 to 160%. Modeled burdens are more realistic, within 40%, in North America, EECCA, and Europe; this analysis is consistent with surface observations in Europe (Figure 13 and section 5.5). Some regions requiring large scaling factors are dominated by energy-related emissions, and the underestimate must occur in those emissions. In other regions, particularly Africa and Southeast Asia, open burning is a large fraction of BC AAOD. Uncertainties in attribution between BC and dust are greatest when apparent dust fractions are large, which could affect the global burden greatly in Africa. Section 6.9 discusses the implications for emission rates, and compares the findings presented here with previous model-measurement comparisons.

6.6 Forcing Efficiency

[349] Radiative effects of BC are estimated by comparing radiative transfer calculations with and without BC emissions, accounting for scattering and absorption by the surface and by gases, clouds, and other aerosol components throughout the atmosphere. Aerosol optical properties are important inputs to determine the magnitude and location of aerosol absorption and scattering. However, once the optical depths and single-scattering albedo are known, variations in global-average radiative forcing per absorption optical depth are largely driven by environmental variables, especially the reflectivity of the underlying surface. BC aerosol alters the ToA net energy balance much more over a bright reflective surface or cloud layer than above a dark one [Haywood and Shine, 1995; Chýlek and Coakley, 1974]. This sensitivity produces the threefold diversity in the global mean absorption forcing efficiencies (AFE) listed in Table 15, with values ranging from 91 to 270 W m−2AAOD−1. Figure 21 shows the spatial distribution of annually averaged AFE from the AeroCom median model. It tends to be higher for aerosol located over land and stratus clouds, and the highest values occur over snow.

Details are in the caption following the image
Global distribution of absorption forcing efficiency (AFE) defined as direct radiative forcing divided by aerosol absorption optical depth (AAOD). Both forcing and AAOD used are AeroCom median values as described in section 5.3.

6.6.1 Vertical Location of BC

[350] Top-of-atmosphere AFE is enhanced considerably when BC is located over clouds compared with over the dark ocean surface. Thus, some of the diversity in AFE is caused by differences in the BC vertical distribution because elevated BC is more likely to overlie low clouds [Haywood and Ramaswamy, 1998]. BC over low clouds increases forcing disproportionately compared with BC in clear skies, while BC under clouds has a decreased contribution [Haywood and Ramaswamy, 1998; Zarzycki and Bond, 2010]. Greater solar flux at high altitudes can also increase forcing by high-altitude BC [Samset and Myhre, 2011]. Comparing forcing from models that all use the same emissions, the model with the lowest AFE of 91 W m−2AAOD−1 places 14% of the total BC mass above 5 km, and the model that simulates the highest forcing efficiency of 270 W m−2AAOD−1 has 37% of its BC above 5 km. Textor et al. [2006] find that the fraction of the simulated BC mass above 5 km ranges from 7 to 37% for 16 AeroCom models. Direct forcing by BC can be greatly intensified when aerosol-cloud collocation on sub-grid scales is considered [Chand et al., 2009], and this is not treated in global models or in this estimate.

[351] How realistic are simulated vertical distributions of BC? As discussed in section 5.5, recent measurements have shown that aerosol models generally simulate too much BC in the upper troposphere, especially at tropical and mid-latitudes [Schwarz et al., 2010]. Models also underestimate the decrease in aerosol extinction with altitude [Yu et al., 2010]. Adjustments to reduce these biases are likely to put more of the BC below rather than above clouds and, hence, reduce estimates of the ToA radiative forcing by BC. Thus, the highest values of AFE shown in Table 15 are probably unrealistic. Zarzycki and Bond [2010] estimate that radiative forcing would decrease by approximately 15% if a model with an intermediate BC vertical profile was scaled to match measured profiles. Koffi et al. [2012] compared AeroCom models and CALIPSO measurements and found that aerosol was too high by about 400 m (range of 50 to 1080 m) in source regions. This finding also suggests a forcing decrease of approximately 15% when combined with the sensitivities given by Samset and Myhre [2011].

6.6.2 Horizontal Location of BC

[352] Some of the diversity in AFE could also be due to differences in the BC horizontal distribution, particularly with respect to the albedo of the underlying surface or the amount of sunlight at different latitudes. Kinne et al. [2006] find the greatest diversity in the simulated BC burden is in polar regions, Textor et al. [2006] find that the fraction of BC mass between 80° and 90°N varies greatly in the AeroCom models, and Shindell et al. [2008] find that all BC models in their study simulate far too little BC at Barrow and Alert (both in the Arctic), particularly during winter and early spring.

6.7 All-Source and Industrial-Era BC DRF

[353] Figure 22 summarizes the two major determinants of direct radiative forcing—BC AAOD and AFE—for the model results reported in Tables 14 and 15. Figure 22(top) shows that modeled radiative forcing has a strong relationship with modeled AAOD. AFE is the ratio between forcing and AAOD and varies among models. Figure 22(bottom) shows larger AFE for models with more BC above 5 km altitude, which suggests that AFE can be strongly affected by the amount of BC above 5 km altitude.

Details are in the caption following the image
AeroCom model results for BC DRF in the industrial era. Top: Unscaled Aerocom BC DRF versus global average BC AAOD in each model (circles) and scaled BC-DRF values (short lines) plotted at the scaled industrial-era AAOD value (0.0049). BC AAOD is proportional the amount of BC in the atmosphere. Unscaled models with higher BC AAOD generally have higher BC DRF. When BC DRF from each model is scaled to the BC AAOD global mean value of 0.0049, the resulting BC DRF has a mean and standard deviation of 0.84 ± 0.21 W m−2 (blue triangle slightly offset for clarity). Bottom: The percentage of BC above 5 km versus absorption forcing efficiency (AFE) in several Aerocom models. Data presented here are also provided in Tables 14 and 15. The scaling of BC DRF in AeroCom models is discussed in section 6.7.1. The forcing average shown here is reduced for the best estimate of forcing, as discussed in section 6.7.2.

6.7.1 Scaled Estimates of BC DRF

[354] Many of the models with results reported in Tables 14 and 15 simulate too little BC AAOD compared to atmospheric observations and, therefore, too little BC DRF. If modeled values of AFE were trustworthy, then a best estimate of all-source radiative forcing could result from multiplying those values by a best estimate of BC AAOD for every grid cell and season and averaging over the globe. A map of the increase in radiative forcing obtained by applying this method with the AeroCom median model is shown in Figure 23. Because we do not have all the modeled forcing fields to perform this scaling, we capture the diversity in AFE by scaling each of the model-based estimates of industrial-era radiative forcing using the observation-based estimate of industrial-era BC AAOD, which has a global annual mean of 0.0049. These estimates are only approximate because the scaling ignores regional and seasonal variations. However, this scaling method does improve absorption fields that are too low when models do not include internal mixing or sufficient emissions.

Details are in the caption following the image
Adjustments to the annual mean, direct radiative forcing (W m−2) by BC in the median AeroCom model required for consistency with the AERONET retrieved aerosol absorption optical depth (AAOD).

[355] Table 15 shows the industrial-era BC-DRF values from each model, scaled by the ratio of the observationally based BC AAOD to the modeled BC AAOD. These scaled values range from +0.48 to +1.32 W m−2, with a mean and standard deviation of +0.84 ± 0.21 W m−2. Radiative forcing is increased for all models because all models simulate a lower BC AAOD. Similarly, this scaling can be used to estimate DRF by all BC sources. Model BC-DRF values scaled to the total BC AAOD of 0.0060 are also listed in Table 15. Estimates range from +0.59 to +1.63 W m−2, with a mean and standard deviation of +1.03 ± 0.26 W m−2.

6.7.2 Central Estimate of BC DRF

[356] For a central estimate of BC DRF, we use an average of the scaled radiative forcing estimates in Table 15. However, section 5.6 showed that most models overpredict BC at high altitudes, leading to an overestimate of BC DRF. Therefore, we estimate a central value of 15% less than the mean of all estimates to compensate for the tendency of models to place BC at too high altitudes and, hence, produce an unrealistically high radiative forcing efficiency, as discussed in section 6.6.1.

[357] With this scaling, the central estimate of global annual DRF by industrial-era BC is +0.71 W m−2. The central estimate of DRF by all-source BC is +0.88 W m−2. These values may be considered an observationally constrained model average. This estimate includes forcing due to atmospheric absorption by organic carbon since our estimate of BC AAOD includes absorption by both BC and OA. In section 11, total aerosol DRF from sources that emit BC is estimated, and there we do not account for atmospheric absorption by OA, so while BC DRF may be biased high, the total forcings by emissions from sources that include both BC and OA do not include this bias.

[358] The modeled forcing discussed above was calculated only at visible wavelengths, except for the work of Jacobson [2010]. Aerosols can also interact with infrared radiation, especially in source regions [Lubin and Simpson, 1994]. Reddy et al. [2005a] estimated infrared forcing by BC as +0.006 W m−2. Jacobson [2001b] presented a graphical summary in which ToA direct infrared forcing by BC was small relative to forcing at visible wavelengths. Therefore, we assume that the central estimate given above encompasses infrared forcing, even if most models have neglected it.

6.7.3 Uncertainties in BC DRF

[359] A 90% confidence range for BC DRF is based on independent uncertainties in the industrial-era BC AAOD and the radiative forcing efficiency, AFE. The final estimate of uncertainty is higher than that reflected in the diversity of scaled BC DRF estimates listed in Table 15, which do not account for uncertainty in BC AAOD.

[360] Appendix B provides an estimate of the uncertainty in BC AAOD. Uncertainties are caused by the spatial patterns of BC AAOD used in filling gaps between AERONET sites, ambiguity in how the scale factor is defined, the limited number of AERONET sites, clear-sky biases in the BC AAOD, the influence of BC transport from sources to oceanic regions, and the impacts of fine-mode dust and OA on the estimate of BC AAOD. The industrial-era BC-AAOD estimate has an additional 18% uncertainty due to a 50% uncertainty in the estimate of the preindustrial background. We assume that the overall uncertainty of the total AAOD retrieval contributes to the above-mentioned sources of uncertainty, although the degree of this contribution is unknown, and we thus do not account separately for it. One factor that could decrease the BC DRF is the attribution of more AAOD to dust. As discussed in Appendix B, we allow this factor to introduce an asymmetric uncertainty. The 90% uncertainty range is from 0.0023 to 0.0087 for all-source BC AAOD and from 0.0014 to 0.0078 for industrial-era BC AAOD.

[361] The uncertainty in AFE is estimated from the standard deviation of the model AFE values listed in Table 15. This uncertainty is attributable to the vertical and horizontal location and the temporal covariance of the clouds and BC. Other uncertainties included are the vertical location of BC relative to clouds (25%) [Zarzycki and Bond, 2010], the effect of surface albedo (20%), and the choice of radiative transfer code (20%) [Boucher et al., 1998]. Schulz et al. [2006] attributed a variation in total aerosol cloudy-sky radiative forcing of ±0.26 W m−2 among nine AeroCom models mainly to diversity in forcing above clouds. However, this diversity in AFE has not been apportioned to individual sources of uncertainty.

[362] Our final estimate of 90% uncertainty bounds comes from combining the uncertainties in AAOD and the 50% uncertainty in AFE. The global annual DRF by industrial-era BC is estimated to have an uncertainty range of +0.08 to +1.27 W m−2 about the central estimate of +0.71 W m−2. This range spans all scaled estimates listed in Table 15. The uncertainty range for all-source BC is +0.17 to +1.48 W m−2 about the central estimate of +0.88 W m−2.

[363] A final uncertainty not estimated here is caused by interpretation of AERONET remote-sensing data. Such data sets are subject to periodic revision, and changes in the interpretation would shift the estimate of AAOD and forcing presented here.

6.8 Previous Observationally Scaled Radiative-Forcing Estimates

[364] Three previous studies have reported similar estimates of all-source BC DRF by scaling to AERONET data. Sato et al. [2003] scaled modeled fields to the AERONET Version 2.0 product to obtain modeled forcing of +1.0 W m−2. As discussed previously (section 6.5.1), this product eliminates low-AOD observations, with the result that the average AAOD has a high bias, and any scaling of BC AAOD using that product is also biased high. The Sato et al. BC AAOD relied on data from earlier years, so their value of 0.006 is not as high as our value extracted from AERONET Version 2.0 averages (0.0143).

[365] Ramanathan and Carmichael [2008] estimated forcing of +0.9 W m−2 by combining modeled fields with AERONET data, MODIS data, and estimated ω0. Modeled fields were based on Chung et al. [2005], which reports BC AAOD of about 0.0068. Both BC AAOD and AFE are similar to those derived here, and the DRF estimate for all-source BC (+0.9 W m−2) is similar to the value given here. Chung et al. [2005] also point out that correcting MODIS AOD with those of AERONET decreases the estimated global AOD by 25%.

[366] The most recent observationally based estimate, by Chung et al. [2012], used a combination of AERONET retrievals, partitioning to distinguish BC AAOD from contributions by dust and organic carbon, satellite retrievals of AOD, and model estimates of single-scattering albedo. Their estimate of BC AAOD (0.0077) is 28% larger than our best estimate. However, this study also estimates a relatively low AFE value (97 W m−2 AAOD−1); applying this value to our best estimate of BC AAOD produces a BC DRF that is much smaller than that of most estimates in Table 15. Chung et al. [2012] estimate an uncertainty range for all-source BC AAOD of 0.006 to 0.009. The lower bound is much higher than ours, because we have more strongly weighted the possibility that more AAOD is attributable to dust.

6.9 Implications of Increased BC AAOD

6.9.1 Apportionment of Bias in BC AAOD

[367] It is of interest to attribute the model bias in BC AAOD to causes beyond the division between MACBC and burden presented in Figure 20. The required burden scaling could be caused by emissions, by atmospheric lifetime, or a combination of the two. Inverse modeling that explored adjustment of both removal processes and emissions might help to apportion the bias with more confidence, but these studies are beyond the scope of this assessment. However, because the scaling factors were mainly determined using land-based AERONET observations, they are less strongly affected by modeled removal than they would be if the observation sites were very distant from sources. Sensitivity studies reported by Koch et al. [2009b], as shown in Figure 13, show that the ratios between observed and modeled concentrations could change by 10% if aging rates were altered by a factor of 2 in either direction. An exception to this finding is North America, for which altered lifetimes would alter the comparison more greatly, but modeled burdens in that region are modeled reasonably well. It is likely that underestimates of emissions are a major cause of the observed discrepancies.

6.9.2 Revised Estimate of BC Emissions

[368] Emission rates that would be required to match observations were estimated by applying the monthly regional scale factors described above to modeled emissions in each region. Then annual emissions were determined by summing the monthly emissions. Table 16 shows the annual scaled totals for energy-related and open-burning emissions and compares them with the bottom-up emission estimates in section 4.4. This division between energy-related and open-burning scale factors depends on the modeled seasonality of these major source groups, as well as the assumption that monthly scale factors for each group are equal. The scaling of emission totals is more robust than the apportionment of scaling between the two categories.

Table 16. Modeled, Scaled, and Current Bottom-Up Annual BC Emissions by Region in Gg yr−1
Energy-Related Emissions Open Biomass Emissions Total Emissions
Region Modeleda Scaledb Currentc Modeledd Scaledb Currentc Scaledb Currentc
North America 340 400 330 80 100 40 500 370
Latin America 400 800 360 590 1200 720 2000 1080
Middle East 80 140 140 0 0 0 140 140
Africa 500 1200 680 1520 3200 1140 4400 1820
Europe 450 500 470 10 10 10 510 480
EECCA 270 500 310 280 100 120 600 430
South Asia 610 2500 670 20 80 40 2580 710
East Asia 1440 2800 1310 10 20 50 2820 1360
Southeast Asia 380 800 400 400 1400 460 2200 860
Pacific 130 500 120 190 700 170 1200 290
All-source 4600 10,140 4790 3100 6810 2750 16,950 7540
Pre-industrial 390 390 390 1030 2640 1030 3030 1420
Industrial-Era 4210 9750 4400 2070 4170 1720 13,920 6120
  • a Based on SPEW emissions, which were used in many AeroCom models.
  • b Modeled regional emissions adjusted by burden scale factors as described in section 6.5.1.
  • c As reported in Tables 7 and 8, and Figures 8 and 9, for comparison with the initial and scaled modeled values given here. Current energy-related emissions are a composite of SPEW and GAINS as described in section 4.4. Current biomass burning emissions given here are from RETRO only and do not match the central estimate for that reason. Biomass burning estimates have decreased substantially since the initial AeroCom exercise.
  • d Based on GFED 2.0 emission fields, reported as the source of open-burning emissions for AeroCom models [Dentener et al., 2006]. Regions scaled identically to total emission rate reported for AeroCom exercise.
  • e Pre-industrial emissions estimated by Dentener et al. [2006] and assumed to be constant to estimate industrial-era forcing.

[369] Table 16 summarizes these scaled annual emissions. Three emission estimates are given in the table: modeled emissions, which are an estimate of the median used by AeroCom models; scaled emissions, which are inferred using the burden scale factors applied to monthly emissions; and current emissions, which are the bottom-up estimates summarized in section 4.4. Current emission estimates are different than those used by the AeroCom models that produced the fields for analysis. Although energy-related emissions are similar, modeled open-burning emissions in AeroCom models were about 35% higher than the most current values.

[370] Scaled estimates for all-source annual emissions are 17,000 Gg BC yr−1, a factor of 2.2 increase over current emission estimates and modeled values. Because AERONET observations from years 2000 to 2009 were used for scaling, this emission estimate is approximately associated with the year 2005. In this simple scaling exercise, energy-related and open-biomass emissions are scaled each month resulting in similar scaling for annual totals which are 10,100 Gg BC yr−1 and 6800 Gg BC yr−1, respectively. Table 16 also gives estimates of industrial-era emissions, inferred by subtracting the scaled preindustrial emission values. The industrial-era total is 9750 Gg BC yr−1 for energy-related emissions and 4200 Gg BC yr−1 for open biomass emissions.

[371] The industrial-era emission estimate of about 13,900 GgC yr−1 is the value obtained with a simple scaling of AAOD. However, several caveats apply to the method described here. First, this scaling method neglects the fact that burden in some regions may be caused by emissions in other regions. According to tagged-tracer modeling given in Bond et al. [2011], Middle Eastern concentrations are affected by South Asian emissions, Southeast Asian concentrations have significant contributions from both South Asia and East Asia, and the Pacific region is affected by all Asian emissions. However, the model used tends to over-transport aerosol, as evidenced by atmospheric burdens at high altitudes. A true inverse modeling approach would give a better estimate of emissions in each region, but modeled transport and removal should be more thoroughly evaluated before embarking on such a study. We caution that inferred emissions in Southeast Asia and the Pacific are particularly uncertain.

[372] Second, as discussed above, uncertainties in modeled lifetimes could also contribute to errors in burdens. Changes in modeled lifetime were shown to alter modeled BC by about 10% at the observation sites (Figure 13), which would alter inferred emissions by the same amount. Third, the MACBC scale factor used here is also simplified because it assumes that BC is always internally mixed and, therefore, amplifies absorption at the observation sites. If BC were externally mixed at observation sites, the MACBC scale factor would decrease and the burden scale factor would increase. If the appropriate MACBC scaling were a 20% increase instead of 50%, all-source emissions would be estimated as 21,000 Gg BC yr−1. For an assumed increase of 80%, the estimate would be 14,000 Gg BC yr−1. Finally, observations are sparse in many of the regions with the largest emissions and the largest estimated increases. In the absence of more definitive studies that explain biases in aerosol absorption, the central value is used as a best estimate of industrial-era BC emissions throughout this assessment.

6.9.3 Scaling of Modeled AAOD in the Context of Other Evidence

[373] Inverse modeling studies and other comparisons between models and measurements were discussed in section 5.5.1. As these studies have attributed errors to emission biases rather than modeling of any other factor, we compare them with the emission increases in Table 16.

[374] Scaled emissions for total biomass burning BC of 6800 Gg yr−1 are in good agreement with the upper limit derived by Zhang et al. [2005]. The increase by a factor of 2.5 falls within the large uncertainty ranges for biomass smoke given by Reid et al. [2009] and Kaiser et al. [2012]. Carbon monoxide measurements do not indicate such a large discrepancy, so either modeled emission factors or modeled aerosol lifetimes are suspect. Section discussed some reasons why modeled biomass burning emission factors could be too low. These proposed adjustments should also apply to the preindustrial background used here, and hence, the scaling described here is also represented in our estimate of preindustrial AAOD.

[375] The finding that small to moderate emission increases are required in North America and Europe is in reasonable agreement with inverse modeling studies. Scaled emissions in Figure 20 suggest that emissions for East Asia should increase as suggested by Kondo et al. [2011a], but by 90% instead of 30%. Since inverse modeling by Hakami et al. [2005] suggested that BC emissions in East Asia were approximately correct at an earlier date, the increased emissions may be related to growth, estimated as 30 to 50% [Lu et al., 2011]. The higher emission estimates for South Asia are 30% greater than those estimated by Dickerson et al. [2002] for a much earlier year, and they agree with observations that modeled burdens in that region are too low by about a factor of 4 to 5.

[376] Total emissions are in reasonable agreement with predictions of a Kalman filter inverse estimate [Jason Blake Cohen, personal communication, 2012]. Scaled global energy-related emissions are estimated at 10,100 Gg yr−1, with 80% of the increase appearing in Africa, South, East, and Southeast Asia. Although the apportionment between energy-related emissions and biomass burning is not certain, it would be difficult for the small amount of biomass burning in South and East Asia to cause such a large discrepancy. Applying the higher BC emission factors recommended by Cooke et al. [1999] for power generation would cause much greater discrepancies in North America and Europe than currently exist. The evidence points to underestimation of emissions in regions with low regulation, so that sources such as poorly operating vehicles, industrial installations, or high solid-fuel combustion may be responsible.

[377] Land regions of Latin America, Africa, Southeast Asia, and the Pacific region (Figure 7) are, together, responsible for 40% of the global BC AAOD and suggested scaling factors in these regions are 2 to 3.5. Especially uncertain are findings for Africa, where 25% of global AAOD is found, and partitioning between BC and dust affects inferences about the magnitude of BC AAOD. Inverse modeling studies or continent-wide comparisons with long-term measurements have not been reported for these regions.

[378] In summary, we find that global AAOD from AeroCom models is low compared to that suggested by observations, that this underestimate occurs in particular regions, and that it cannot be fully explained by the absorption enhancement produced by internal mixing or by uncertainties in retrievals. Modeled absorption is reasonable in regions with the best measurement coverage, relevant emission measurements, and low biomass-burning emissions: North America and Europe. The dominance of different source categories and larger uncertainties in the other regions (Figure 9 and section 4.7.1) that require increased scaling to match observations suggests that poorly modeled emissions are a potential cause. Studies using space-based remote sensing indicate that modeled aerosol concentrations and absorption from biomass burning are too low (section 5.5.1), and an independent line of argument indicates that biomass-burning emission factors could lead to underestimates of absorption (section Surface measurements in South and East Asia indicate that models using the emissions summarized in section 4.4 would underestimate observed BC concentrations (section 5.5.1). In Asia, where emissions have increased since the global emission fields were developed (section 4.8), some of the underestimation is to be expected. Although we cannot fully exclude modeled lifetimes, instrumental uncertainties, and the division of remotely-sensed AAOD into BC and dust components in some regions (particularly Africa) as explanatory factors, multiple lines of evidence support the position that AeroCom models using year-2000 BC emissions substantially underestimate absorption and direct radiative forcing. Despite the inherent limitations of these models, they are required to produce estimates of global forcing, because global space-based observations cannot separate radiative impact by chemical component. Model fields scaled to observations are therefore used to develop our best estimate of direct radiative forcing.

6.10 Summary of Uncertainties in BC DRF

[379] The discussion in this section shows that several factors affecting BC radiative forcing are not well constrained. First, the most quantitative and widespread data set for determining atmospheric absorption is remote sensing provided by AERONET, but use of these observations requires an estimate of the division between AAOD attributable to BC and dust. Different plausible assumptions can greatly alter the inferred BC AAOD, up to a factor of 2. Parameters retrieved from AERONET observations, such as size distributions or wavelength dependence of absorption, have been used in interpreting AAOD data, but these have not been thoroughly evaluated with in situ measurements. Second, the dust refractive index may have large variability within regions [Moosmüller et al., 2012; Wagner et al., 2012] and could greatly affect the estimate of dust AAOD and the inference of BC AAOD. Third, despite the uncertainties, observations indicate more BC AAOD than is modeled by about a factor of 2.5. A simple scaling procedure indicates that a large increase in emission estimates is warranted for biomass burning and for energy-related sources in developing regions. Emission factors may explain low biases in modeled biomass burning emissions, but the cause of errors in energy-related emissions is unknown. Fourth, despite the fact that emission rates are implicated in underestimates of BC AAOD, a full attribution of differences in BC distributions to errors in model emissions, transport, or removal rates is not yet possible. No simple adjustment to modeled transport can explain the range of model biases (section 5.7). Fifth, the contribution of absorbing organic aerosol to AAOD has not been elucidated. Attribution of some of the absorption to OA rather than BC would reduce the direct forcing estimate for BC and make that of OA less negative or more positive. Finally, model evaluations are limited in regions that contribute large fractions of the global AAOD, because few long-term observations are available.

[380] Although present-day observations can be used to adjust modeled forcing terms, a lack of understanding of the causes of bias limits the accuracy of modeled forcing for the past and future. Further, vertical transport and therefore location of BC is poorly modeled, so that modeled absorption forcing efficiency also contains errors. While biases in BC distributions in the Arctic produce only small errors in global, annual average BC AAOD (and thus forcing), they produce large biases in forcing within the Arctic, where BC may contribute to greater-than-expected warming and declines in sea ice [Quinn et al., 2008, 2011]. In subsequent sections, horizontal and vertical distributions are shown to affect the interaction with clouds (section 7) and forcing in snow and sea ice (section 8). Thus, a lack of understanding of process fundamentals and the resulting modeled biases are large sources of uncertainty in all BC forcing terms.

[381] The amount of aerosol in the preindustrial atmosphere is also poorly known. Biomass burning produced much of this aerosol, and preindustrial estimates used here were based on the emissions given by Dentener et al. [2006]. Although biomass burning has increased overall during the industrial era, charcoal records suggest that activity in some regions was similar between 1750 and today [Marlon et al., 2008]. Although this preindustrial background does not affect the estimate of total radiative forcing, it does affect industrial-era forcing and temperature change since preindustrial times. We retain conventional nomenclature and refer to this background value as that of 1750, but we caution that it represents a reference atmosphere rather than a true preindustrial value.

7 Black Carbon Interactions With Clouds

7.1 Section Summary

  1. [382]

    Black carbon may change cloud cover, emissivity, or brightness in four ways: (1) by changing the vertical temperature structure of the atmosphere, which could shift cloud distributions (semi-direct effect); (2) by changing the number concentration of liquid cloud droplets and the lifetime of liquid clouds; (3) by changing phase partitioning and precipitation in mixed-phase clouds; and (4) by changing ice particle number concentration. These effects may cause either positive or negative radiative forcing. Very few model studies isolate the influence of BC on each effect. All aerosol effects on clouds are highly uncertain, and the isolated effects of BC have even greater relative uncertainty.

  2. [383]

    Many of the cloud effects described here may be considered rapid adjustments, either to direct radiative forcing or to the presence of BC. However, all effects can be quantified in units of climate forcing, a practice we continue here for consistency with previous studies and for convenience.

  3. [384]

    In the semi-direct effect, cloud cover can increase or decrease, depending on region and conditions. Some studies examining regional cloud reduction have suggested a positive forcing, but global model studies indicate that the BC semi-direct effect averages −0.1 ± 0.2 W m−2 for industrial-era forcing if all BC is emitted at the surface. An additional −0.28 W m−2 is added to the lower uncertainty range to account for the potential reduction of high-level clouds from biomass-burning BC, making the 90% uncertainty range –0.44 to +0.1 W m−2 over the industrial era. Major uncertainties include poorly modeled BC vertical distributions and the fidelity of modeled cloud responses.

  4. [385]

    BC has two competing indirect effects on liquid clouds. First, adding BC increases the aerosol number concentration. If the number of cloud droplets also increased, negative forcing by clouds would increase, but BC-rich particles are inferior cloud-forming particles. Second, BC particles may also serve as sites to collect soluble material, reducing overall cloud droplet number concentration and producing a positive forcing. BC alone is estimated to have an indirect climate forcing of about −0.1 W m−2. Many studies do not separate this effect from the semi-direct effect discussed above, so an isolated estimate is difficult to obtain. Therefore, we estimate the combined industrial-era climate forcing from liquid-cloud and semi-direct effects as −0.2 (with a 90% uncertainty range of −0.61 to +0.1 W m−2).

  5. [386]

    Greater absorption by BC within cloud droplets decreases cloud albedo, heats clouds, and dissipates them. This is a special case of the semi-direct effect that we estimate separately as a climate forcing of +0.2 W m−2, with a 90% uncertainty range of −0.1 to +0.9 W m−2 over the industrial era. Model results for this effect are very sensitive to the representation of optical properties for the mixed BC and cloud droplet.

  6. [387]

    The liquid-cloud indirect effect is sensitive to BC particle size and to mixing with other particles. Many model studies indicate that particles emitted from biofuel combustion appear to have more negative forcing per emission than do particles from fossil fuel combustion, possibly because of size or co-emitted POA.

  7. [388]

    The BC mixed-phase indirect effect acts on clouds that are part liquid and part ice. BC may act as ice nuclei (IN) that enhance ice formation and increase ice fall-out. An estimate for this effect is +0.18 ± 0.18 W m−2 for industrial-era climate forcing and 90% uncertainty range, but the magnitude would be reduced for BC mixed with sulfate.

  8. [389]

    In the ice-cloud indirect effect, greater IN concentrations either increase or decrease the concentrations of ice particles and the lifetime of cirrus clouds, depending on conditions. Cirrus clouds affect both shortwave and longwave radiative fluxes. The cloud cover change and forcing may, therefore, be positive or negative. Two model studies estimate effects of opposite signs: −0.4 and +0.4 W m−2. Laboratory and field observations suggest that both BC concentration and BC's ability to act as IN are probably less than assumed in these model estimates. We estimate an industrial-era climate forcing of 0 ± 0.4 W m−2 (90% uncertainty range) for this highly uncertain effect, which excludes the effects of aviation. These effects are sensitive to assumptions about BC's role as IN and to the number concentration and mixing state of BC particles in the free and upper troposphere. Ice-cloud effects also depend on the assumed background conditions, including concentrations of other IN and updraft velocities.

  9. [390]

    Few modeling and measurement studies are able to constrain cloud-absorption, mixed-phase and ice-cloud indirect effects. Consequently the uncertainties are large. Model diversity, rather than true uncertainty propagation, provides uncertainties in cloud radiative forcing.

7.2 Introduction

[391] Clouds and their responses to aerosol addition introduce a large uncertainty in the understanding of total climate forcing and climate response [Heintzenberg and Charlson, 2009]. The general term “indirect effects” refers to the suite of climate forcings that aerosols impose through the modification of cloud properties [Forster et al., 2007; Denman et al., 2007]. Aerosol indirect effects on clouds have been extensively studied, but the influence of BC alone has received less attention. Although BC makes a small contribution to aerosol mass load in the atmosphere, it may play an important role in determining the CCN or IN particle number concentration that in turn alters indirect effects. Aerosol indirect effects and BC's role in the indirect effect differ depending on the cloud phase: liquid, ice, or “mixed” containing both liquid and ice, and these are discussed individually. Furthermore, absorption by BC embedded within cloud droplets is greater than that of BC alone or with coatings.

[392] Figure 24 summarizes the mechanisms by which BC can influence clouds. A brief overview is given here before the detailed discussions in the remainder of this section. The first general class of effects involves perturbations to the atmospheric temperature structure, which affect cloud distributions. These mechanisms are commonly termed “semi-direct effects,” a term that is sometimes used to describe evaporation of cloud droplets when absorbing aerosols heat a cloud layer [Hansen et al., 1997; Ramanathan et al., 2001a]. Additional BC radiative effects have also been documented whereby cloud cover may be either enhanced or reduced, depending on factors such as its altitude relative to cloud cover [e.g., Hansen et al., 2005; Johnson et al., 2004], and meteorological conditions. BC's semi-direct effects are a rapid adjustment to BC direct radiative forcing, through local warming of the atmosphere. Some model calculations treat this adjustment as a distinct forcing term. Following these studies, and consistent with other treatments in this assessment, we review semi-direct estimates in forcing units.

Details are in the caption following the image
Schematic of the causes and effects that lead to cloud effects from BC emissions. Each row begins with an identifying label followed by a cause, namely, an atmospheric or microphysical parameter representing a potential perturbation to cloud properties. To the right of each cause are the response(s) to the perturbation. On the far right is the associated effect, namely, the cloud parameter that changes in response to the perturbation. The single color of the components in a row indicates a climate system response of warming (red) or cooling (blue). The split coloring in LC3 indicates that the response can be either warming or cooling (section 7.2). Although MC2 is not attributable directly to BC, it is included here because it alters the effects of BC.

[393] Another general category, commonly termed “indirect effects,” involves changes in concentrations of cloud droplets or ice crystals that alter cloud brightness, emissivity, and lifetime, which produce a radiative forcing. This category includes the groups titled “liquid-cloud effects,” “mixed-phase cloud effects,” and “ice-cloud effects” in Figure 24. These cloud effects are driven by changes in the number of CCN or IN, aerosols on which cloud droplets or ice particles may form. The process of forming a stable liquid droplet or ice particle is known as activation. In principle, any particle can activate in either liquid or ice clouds if high enough supersaturation is reached. However, when many aerosol types are present, the particles that activate the most easily do so first and have the greatest potential influence on the resulting number concentration of cloud particles. Microphysical characteristics of a particular type of aerosol are thus very important in determining its cloud effects.

[394] Warm cloud indirect effects include two components. The cloud albedo effect (i.e., the first indirect or Twomey effect) [Twomey, 1959] refers to the change in radiation caused by a change in cloud albedo or brightness resulting from a change in the cloud droplet size distribution. Increased CCN lead to more and smaller cloud liquid droplets for a given cloud water content (case LC1 in Figure 24). This effect can be observed in ship-track studies [e.g., Ferek et al., 1998]. However, if BC attracts condensing gases that would otherwise form particles, the net result is a decrease in CCN and cloud droplets and is, therefore, a positive radiative forcing (case LC2) [Bauer et al., 2010].

[395] The cloud lifetime effect (LC3, second indirect effect) refers to the fact that in the presence of increased aerosol concentrations, more and smaller cloud droplets form, and these collide less efficiently. Factors such as drop evaporation rates and depletion of water vapor can also alter cloud lifetimes. While cloud resolving models suggest that this effect could lead to either an increase or decrease in liquid water [Ackerman et al., 2004; Sandu et al., 2008], the GCMs are parameterized such that this effect can only lead to longer cloud lifetimes with greater overall cloud reflectivity [Lohmann and Feichter, 2005]. Thus, although this forcing could be either positive or negative, GCM global estimates of LC3 are always negative. Altered lifetime also plays a role in mixed-phase clouds, where temperatures typically fall between 0 and −35°C. Two competing effects have been suggested: first, glaciation, which refers to an increase in ice nuclei causing more frequent glaciation of supercooled clouds, and second, de-activation. In the glaciation effect, ice crystals grow at the expense of water droplets because of the difference in vapor pressure between water and ice, referred to as the Bergeron-Findeisen effect. An increase in IN would enhance this process, causing clouds to precipitate more readily (case MC1 in Figure 24) and thereby reducing cloud amount [Lohmann, 2002]. However, deactivation occurs when sulfur or secondary organic aerosol coat IN and make them less efficient [Girard et al., 2005] in mixed-phase clouds by changing the mode of freezing from contact to immersion freezing. In global climate models, this leads to less sedimentation and more cloud cover [Storelvmo et al., 2008; Hoose et al., 2008] (case MC2 in Figure 24).

[396] BC may also affect cirrus clouds that occur at high altitude in the upper troposphere [Kärcher and Spichtinger, 2009]. Aerosol particles at these high altitudes have long atmospheric lifetimes [e.g., 4–30 days; Williams et al., 2002] and, therefore, extended opportunities for mixing with other aerosol components. Thin, high cirrus clouds are believed to have net positive cloud forcing [Chen et al., 2000]. Absorption of infrared radiation and re-emission at colder temperatures dominates scattering of solar radiation, which results in a net positive forcing. Efficient IN may either increase or reduce cirrus cloud forcing. Cirrus cloud particles are mostly formed through homogeneous ice nucleation. Even a small number of particles can initiate heterogeneous freezing well below the homogeneous freezing threshold, depleting some of the supersaturated water vapor, partially preventing homogeneous formation of solution droplets and reducing the number of particles [e.g., Kärcher et al., 2006]. Early onset of nucleation from a few efficient BC particles affects forcing less than enhanced sedimentation, yielding case IC1 in Figure 24. If high concentrations of IN already exist, heterogeneous freezing dominates the cloud and additional IN increase the number concentration of cloud particles (case IC2 in Figure 24).

[397] Some challenges are common in obtaining forcing from models of aerosol-cloud interactions. First, models use different emission levels and input assumptions, and variation in values of direct forcing may result from these differences alone. Second, each model study may have a different set of impacts, such as treating aerosol interactions with only liquid or ice clouds, or allowing certain climate responses. The net effect on radiation is usually inferred from two sets of simulations—one with and one without perturbed pollution conditions [Lohmann et al., 2010]—so that the individual effects contributing to the changes are difficult to distinguish without well-designed diagnostics. The combination of varying inputs and inconsistent selection of effects makes global cloud-forcing estimates difficult to compare. Third, some aerosol-induced changes are best simulated at scales smaller than a global model grid box. Physical confirmation of the factors governing cloud dynamics and cloud microphysical interactions must also occur on relatively small scales. Cloud-resolving models (CRMs) and large-eddy simulations (LES) may be most appropriate for modeling these scales, but these simulations do not have the spatial extent to calculate globally averaged forcing, and do not always agree with the results of global models that use coarser resolutions. The reasons for these disagreements is not always understood, making it difficult to infer a global, annual average forcing from either set of model studies. Fourth, some effects that cause cloud redistribution on large scales are best simulated with global models, but their broad spatial or temporal extent makes their existence and magnitude difficult to confirm. Fifth, the fidelity of aerosol-cloud simulations depends on accurate representation of the cloud amount and location, yet these factors are not often verified in the model studies. Finally, inter-annual and spatial variability of modeled cloud forcing is large compared with the magnitude of the aerosol-induced changes. Therefore, extracting statistically significant changes is challenging.

[398] While aerosol effects on clouds are complex, accounting for them is critical because they may induce changes similar to or greater than aerosol direct forcing. For a perspective on all aerosol effects, Denman et al. [2007] summarized studies ranging from a mean of −1.2 W m−2 with a range of −0.2 to −2.3 W m−2 for the total effect of all anthropogenic aerosols on climate, including direct forcing, cloud albedo, cloud lifetime, and semi-direct effects. Using a combination of 10 models and input from relationships observed by satellite, Quaas et al. [2009] estimated the total aerosol effect on stratiform clouds as −1.5 ± 0.5 W m−2. The global annual mean climate forcing for just the first indirect effect of all aerosols has been estimated from GCMs with some guidance from satellite observations as −0.7 W m−2 with a range between −0.3 and −1.8 W m−2 [Forster et al., 2007]. Notably, Lohmann et al. [2010] showed that estimates of total anthropogenic aerosol effects have become smaller with time. GCMs have added processes, such as aerosol effects on mixed-phase clouds, or treating rain as a prognostic variable, which place more emphasis on accretion instead of autoconversion. These more advanced GCMs arrive at smaller total anthropogenic cloud forcing compared with earlier versions. GCMs that are constrained by satellite data also predict a smaller total anthropogenic aerosol effect.

[399] This assessment mainly presents global, annually averaged climate forcing by semi-direct and indirect effects. However, forcing within a given region may differ significantly from this average. For example, in the Arctic the combined semi-direct and indirect effect of aerosols may produce a much smaller negative forcing than on the global average, or possibly even a positive forcing [Koch et al., 2009a; Jacobson, 2010; Alterskjær et al., 2010], because of the Arctic's high surface albedo and because changes in cloud emissivity due to aerosol microphysical effects may produce significant positive forcing in late winter and spring [Garrett and Zhao, 2006].

[400] In summary, BC may affect clouds by perturbing the atmospheric thermal structure, or by changing liquid cloud droplet number concentration, ice crystal number concentration, or some combination of the two in mixed-phase clouds. Each of these processes affects the distributions or reflectivity of clouds, which, in turn, alters the radiative balance of the Earth. As such, these are rapid adjustments to the climate system that can be quantified as adjusted forcings (section 2.3 and Table 2). The role of BC in each of these cloud effects is presented below, including increased absorption by BC inclusions in cloud droplets (section 7.3.3), semi-direct effects (section 7.3.2), indirect effects on liquid clouds (section 7.4), and mixed and ice phase clouds (section 7.5). Although modeled cloud forcing depends on assumed emission rates or simulated atmospheric burden, this dependence may not be linear, so scaling modeled forcing to the new emission rates or forcing determined in section 6 is also discussed for each effect. Best estimates of adjusted forcings are calculated for the combined semi-direct and liquid-cloud effects for mixed-phase clouds and ice clouds. These estimates are followed by a discussion of uncertainties (section 7.7). All cloud forcings discussed here are industrial-era forcings; all-source forcings are not estimated. Clouds in a clean atmosphere are much more susceptible to change than clouds in an atmosphere with even a small aerosol background [Boucher and Pham, 2002], so cloud forcing is always calculated against a preindustrial background.

7.3 Black-Carbon Semi-Direct Effects on Cloud Cover

[401] Atmospheric BC absorbs solar radiation, perturbs the temperature structure of the atmosphere and, therefore, influences the cloud distribution. The first of these effects to be documented, the original semi-direct effect, is the evaporation and dissolution of clouds by BC suspended near or within clouds [Hansen et al., 1997]. Since then, numerous studies have demonstrated this effect, as well as additional mechanisms by which BC either increases or reduces cloud cover. Some of these studies use cloud-scale models, some are observational—typically with focus on a particular region—and some use global models. Koch and Del Genio [2010] reviewed many studies on this topic, which are summarized here. The top segment of Figure 24 summarizes a framework proposed by Koch and Del Genio [2010] to classify previous studies.

[402] Each of the effects shown in the top portion of Figure 24 is the result of a rapid adjustment to the initial direct radiative forcing by BC. Following the terminology defined in section 2, the sum of the direct effect and the semi-direct effect can be interpreted as a single “adjusted forcing.” However, model studies to date have either calculated the semi-direct effect as an independent forcing, derived a combined forcing for the semi-direct and liquid cloud effects, or report a reduced efficacy for BC direct radiative forcing, attributing this decrease to semi-direct effects. Because semi-direct effects are diverse, all changes other than cloud microphysical changes might be grouped under this label [e.g., Ghan et al., 2012]. For consistency with the existing literature, we also discuss the semi-direct effect here as a climate forcing term and include the efficacy studies in our estimate of that forcing.

7.3.1 Cloud Scale and Regional Studies

[403] The altitude of BC relative to a cloud or potential cloud layer plays an important role in determining the cloud response. For aerosols embedded near cloud, cloud evaporation is enhanced due to their heating and reduction of relative humidity (Figure 24, case SD4). This effect was demonstrated in the LES experiments of Ackerman et al. [2000] for trade cumulus and Hill and Dobbie [2008] and Johnson et al. [2004] for marine stratocumulus clouds.

[404] Absorbing aerosol aloft increases atmospheric stability. Increased stability over stratocumulus clouds reduces cloud-top entrainment of overlying dry air and tends to strengthen the underlying clouds (Figure 24, case SD1). LES experiments by Johnson et al. [2004] showed that absorbing aerosols above cloud increased cloud cover, because they increased the difference in potential temperature across the inversion, decreased entrainment rate, and caused a shallower, moister boundary layer with higher liquid-water path. The same model had demonstrated cloud reduction when absorbing aerosols were within the cloud layer (Figure 24, case SD4). Brioude et al. [2009] analyzed satellite cloud observations and modeled biomass-burning tracers for a field study near the coast of California. They found that biomass-burning aerosols enhanced cloud cover, especially for high-humidity conditions and for low lower tropospheric stability conditions, when the aerosols increased lower tropospheric stability. Similarly, Wilcox [2012] analyzed satellite data of subtropical, South Atlantic biomass-burning smoke overlying marine stratocumulus clouds and found that the smoke enhanced the cloud liquid-water path, countering more than 60% of the smoke direct radiative effect.

[405] While the stabilizing effect of absorbing aerosols aloft can enhance stratocumulus cloud cover, they may suppress cumulus cloud development (Figure 24, case SD2). Fan et al. [2008] performed experiments in a cloud-resolving model for the Houston area and demonstrated that absorbing aerosols aloft decreased the temperature lapse rate, leading to a more stable atmosphere and decreased convection. MODIS observational studies for the Amazon biomass burning season by Koren et al. [2004, 2008] also demonstrated cumulus cloud cover reduction due to increased smoke. They argued that smoke plumes stabilized the boundary layer, reducing convective activity and boundary layer cloud formation. The smoke also reduced radiation penetration to the surface, therefore reducing evaporation and atmospheric moisture. Ten Hoeve et al. [2011] used satellite observations of biomass-burning regions to show that cloud optical depth increased with AOD for low values, consistent with cloud brightening (case LC1). At higher AOD values, cloud optical depth decreased with AOD, possibly due to cloud evaporation (case SD4).

[406] On the other hand, in some land regions, lofted absorbing aerosols may enhance upper level convection, promoting low-level moisture convergence of oceanic air masses, which could increase continental clouds (Figure 24, case SD3). Monsoon enhancement due to lofted absorbing aerosols (known as the “Elevated Heat Pump” hypothesis) was shown in the global model studies of Lau et al. [2006], Randles and Ramaswamy [2008], and Chung et al. [2002]. However, global climate models are limited by relatively coarse spatial resolution, which may preclude accurate representation of aerosol transport over the Tibetan Plateau. In a study using observed vertically resolved aerosol distributions over the Tibetan Plateau and surrounding regions, Kuhlmann and Quaas [2010] use observed surface albedo and a radiative transfer model to show that aerosols do not produce the large elevated heating needed to drive the Elevated Heat Pump. In the non-monsoon season, wintertime pollution as observed in the Indian Ocean Experiment reduced the meridional temperature gradient in the global model study of Ramanathan et al. [2005], which included a coupled ocean response. These SST shifts were found to enhance precipitation over sub-Saharan Africa [Chung and Ramanathan, 2006]. Many studies predicting enhanced convergence over land indicate a shift in clouds and precipitation rather than an overall enhancement.

[407] BC below cloud generally promotes convective activity and can enhance cloud cover (Figure 24, case SD6). Cloud resolving model studies of McFarquhar and Wang [2006] for trade wind cumuli demonstrated that absorbing aerosols placed below cloud promoted vertical motion and increased liquid water path. Similarly, LES experiments of Feingold et al. [2005] found that Amazon smoke emitted at the surface could destabilize the surface layer and increase convection and cloud cover. However, smoke at cloud level decreased cloud cover and promoted dissipation in both of these studies.

7.3.2 Global Model Semi-Direct Estimates

[408] Consistent with the variety of responses found in regional studies, global model studies also find that regional variations and global-average forcing include positive and negative forcing effects over all regions. Although semi-direct effects cause positive forcing in some regions, most models indicate that the global average is negative. Table 17 tabulates the studies discussed here.

Table 17. Studies Used to Estimate Values of the Semi-Direct Effect
Study Reported Semi-direct Effect (W m−2) and Efficacy Adjustment Type (Value Used in Study)a Adjusted Effect (W m−2)
Semi-direct effect calculated directly

Penner et al. [2003]

−0.01, surface emissions EM (12.2 Tg yr−1) −0.01
−0.39, lofted BB emissions EM (BB) (5.6 Tg yr−1) −0.28

Wang [2004]

−0.16 EM (14 Tg yr−1) −0.16

Ghan et al. [2012]

−0.10 to +0.08 EM (7.6 Tg yr−1) −0.18 to +0.15
Semi-direct effect inferred from efficacyb

Hansen et al. [2005]

−0.08 (0.58) DRF (+0.19 W m−2) −0.30

Hansen et al. [2005]

−0.11 (0.78) DRF (+0.49 W m−2) −0.16

Chung and Seinfeld [2005]

−0.1 (0.70) DRF (+0.33 W m−2) −0.21
Yoshimori and Broccoli [2007] −0.40 (0.59) DRF (+0.99 W m−2) −0.29

Jones et al. [2007]

−0.11 (0.71) DRF (+0.39 W m−2) −0.21
  • a Adjustments: EM = scaled from BC emissions used in study to our estimate of BC emissions (13.9 Tg yr−1 or 4.0 Tg yr−1 for biomass); DRF = scaled from BC direct forcing found in study to industrial-era direct forcing of +0.71 W m−2; BB = biomass burning. For comparison to emission values in the remainder of the document, 1 Tg = 1000 Gg.
  • b Reported effect = −DRF × (1 − efficacy). The two values from Hansen et al. [2005] are together weighted equally to the other studies.

[409] Wang [2004] performed experiments in the NCAR model, with and without BC. One type of experiment included an ocean temperature response; a second experiment had fixed observed sea-surface temperatures. He found that BC in the experiments with ocean response had enhanced convective activity and cloud cover in the northern branch of the intertropical convergence zone, with a smaller magnitude reduction in clouds and convective activity in the south. The cloud forcing change (i.e., the difference between all-sky and clear-sky radiative flux change due to BC) was −0.16 W m−2 at the ToA. However, the simulation using observed sea-surface temperatures gave a cloud forcing of −0.06 W m−2. BC did not warm the climate in these experiments, due to compensating cooling from increased cloud cover.

[410] In sensitivity studies of a coupled transient climate model study, Koch et al. [2011b, section 6] showed that removal of BC from 1970 to 2000 would not have caused significant climate cooling. BC removal cooled the atmospheric column, but it also reduced low-level stability and caused a decrease in low-level clouds. Therefore, the surface air temperatures were not significantly decreased, probably due to the loss of low-level clouds. Koch et al. [2011a] did not isolate the forcing by the semi-direct effect from that of the direct effect, so it is not included in Table 17.

[411] Roeckner et al. [2006] performed future transient climate simulations in the ECHAM5 model, with one simulation including projected increases for carbonaceous aerosols (37% BC and 25% particulate organic matter relative to year 2000). This model included indirect as well as direct and semi-direct effects and did not separately diagnose the individual forcings, so it is not included in Table 17. Increased BC in a region caused cooling there, mostly near African biomass burning regions. In these regions, liquid water path and precipitation increased, possibly due to enhanced instability (Figure 24, case SD6); the additional cloud cover reduced surface solar radiation and cooled the surface. For this study, the increased organic carbon together with the indirect effects might also contribute to the increased cloud cover.

[412] Some global model studies show reduced high-level clouds from BC, also a cooling effect (Figure 24, case SD5) [Penner et al., 2003; Menon and Del Genio, 2007; Jacobson, 2010]. Small-scale models have also simulated this effect [Fan et al., 2008]. Penner et al. [2003] found a negligible semi-direct effect for fossil fuel and biomass burning in the GRANTOUR GCM when all aerosols were injected at the surface. However, when biomass-burning aerosols were injected aloft, they found a net negative semi-direct cloud climate forcing response to carbonaceous aerosols (both BC and organic carbon), mostly due to loss of high-level clouds. In most other studies, BC aerosols were injected at the surface and the response was smaller. Menon and Del Genio [2007] also reported a negative semi-direct effect of −0.08 W m−2 in their GISS simulations due to decreased long-wave cloud forcing and loss of high-level clouds, mostly in biomass burning regions. This study did not specify how the semi-direct effect was calculated nor what was used for emissions in the baseline and forcing scenarios, so it is not included in Table 17. Lohmann and Feichter [2001] obtained a negative direct plus semi-direct effect of −0.1 W m−2 in their ECHAM simulations. However, this value was smaller than the inter-annual standard deviation. The semi-direct forcing was not isolated, so this study is also not included in Table 17. Ghan et al. [2012] isolated liquid-cloud effects by setting aerosol refractive indices to zero, allowing an estimate of the semi-direct effect by difference. Changes in shortwave and longwave radiation were of similar magnitudes and totaled −0.10 to +0.08 W m−2 depending on the aerosol representation.

[413] Some studies report BC direct radiative forcing “efficacy” of less than one instead of calculating a negative semi-direct forcing as a rapid adjustment to the direct forcing. The definition of efficacy is temperature change per forcing relative to that for CO2, so that efficacy of less than one indicates that a mechanism acts to reduce the positive direct radiative forcing (Table 2). In some cases, these studies indicate that either increased low-to-middle level or decreased upper level cloud changes are responsible for radiative forcing efficacies less than one [e.g., Roberts and Jones, 2004; Hansen et al., 2005; Yoshimori and Broccoli, 2008]. It is noteworthy that BC radiative forcing efficacy estimates are consistently about 0.6 to 0.8. If we assume that the reduced BC radiative forcing efficacy is entirely due to cloud cover changes, we can infer BC semi-direct forcing estimates for these models. This is justified by the fact that adjusted forcing efficacies for absorbing aerosol are much closer to 1.0, implying that rapid adjustment accounts for most of the small radiative forcing efficacy [Shine et al., 2003; Hansen et al., 2005; Crook et al., 2011].

[414] Roberts and Jones [2004] found that the radiative forcing efficacy of BC was 0.62, due mostly to reduction of high-altitude clouds. Hansen et al. [2005] found BC radiative forcing efficacy of 0.78 for fossil fuel BC and 0.58 for biomass-burning BC and direct forcing values of +0.49 W m−2 and +0.19 W m−2, respectively. If their smaller than 1.0 radiative forcing efficacies are all due to rapid adjustment of clouds, the semi-direct effect forcings are −0.11 and −0.08 W m−2, respectively. Yoshimori and Broccoli [2008] had efficacy of 0.59 for direct BC radiative forcing of +0.99 W m−2, giving a maximum semi-direct effect forcing of −0.4 W m−2. Jones et al. [2007] report a BC radiative forcing efficacy of 0.71 for BC forcing of +0.39 W m−2, for an inferred semi-direct effect of −0.11 W m−2. Chung and Seinfeld [2005] calculated a 0.70 radiative forcing efficacy for 0.33 W m−2 direct forcing, giving a −0.1 W m−2 semi-direct effect. These efficacy studies give a range of −0.4 to −0.08 W m−2 of inferred semi-direct effect of BC. The finding of a narrow range of efficacies implies that this effect is approximately linear with direct forcing. The range is altered to −0.30 to −0.15 W m−2 when each study is scaled to either direct forcing of +0.71 W m−2 (for those studies that reported forcing) or to industrial-era emissions of 13,900 Gg yr−1. These scaled estimates are given in Table 17.

[415] Our estimate of the climate forcing from the BC semi-direct effect for emissions that do not have significant lofting, scaled to 13,900 Gg yr−1 emissions, is −0.1 ± 0.2 W m−2. The central value is the average of the Wang [2004] climate forcing estimate, scaled as described above, the near-zero sum of fossil fuel and biomass values from Penner et al. [2003], the Ghan et al. [2012] value, and the five efficacy estimates scaled to a BC climate forcing of +0.71 W m−2. Most of these estimates come from studies that isolate pure BC, but the Penner et al. [2003] estimate is for BC and OC. All estimates used in this average came from models where emissions were injected at the surface. The magnitude of our uncertainty estimate is larger than the standard deviation of the studies but reflects inter-annual variability and sensitivity analyses reported by individual studies.

[416] The semi-direct estimate is influenced by the five studies that cast the response in terms of radiative forcing efficacy. These studies may have included other climate responses in addition to changes in cloud amount, so we caution against comparing this value with pure estimates of the semi-direct effect.

[417] As noted above, Penner et al. [2003] also estimated a large negative semi-direct effect when all aerosols from open biomass burning were injected aloft (−0.39 W m−2). This estimate was given for a larger emission rate; scaled to our anthropogenic biomass-burning estimate of 4000 Gg BC yr−1, the additional semi-direct forcing would be −0.28 W m−2, with a 100% uncertainty given by the inter-annual variability. We do not have estimates of the quantity of biomass burning aerosol that is lofted, but it is less than 100%. We have added this value as an asymmetric uncertainty to the lower bound of the semi-direct forcing. When we estimate forcing by individual source categories (section 11), this uncertainty for lofted aerosol is attributed entirely to biomass-burning emissions. All other BC emissions are emitted near the surface.

[418] Semi-direct effects are strongly dependent on the amount of absorption [e.g., Johnson et al., 2004; Fan et al., 2008; Randles and Ramaswamy, 2008; Perlwitz and Miller, 2010; Wang, 2004]. For example, Perlwitz and Miller [2010] showed that in GISS model climate simulations, dust with sufficiently large and absorbing AOD caused global mean cloud cover to increase. For weakly absorbing dust, mean cloud cover decreased. Johnson et al. [2004] showed that absorbing aerosols above stratocumulus cloud level strongly increase cloud cover, an effect that did not appear with scattering aerosols. Two global model studies found net decreased cloud cover in response to total pollution aerosols, and positive semi-direct effect [Allen and Sherwood, 2010; Lohmann and Feichter, 2005], but BC effects were not separated in these studies.

[419] Although global models simulate a negative semi-direct effect, the models have substantial uncertainties. Because cloud enhancement is caused mostly by BC above cloud level, model BC altitude distributions must be accurate. As discussed in section 6.6.1, models are especially diverse in their simulated altitude distribution of BC, and measurements to verify vertical distributions are sparse (section 5.6). Many of the models that simulate negative semi-direct effects overestimate BC at high altitude, when compared with a few available BC measurements over North America [Koch et al., 2009b]. Further, global model cloud schemes may not be able to reproduce some small-scale features that would influence the semi-direct effect, such as cloud layer thickness, cloud-top entrainment, cloud fraction, and the tendency to drizzle, which are affected by the scale of interactions among radiation, turbulence, and moist physics on small horizontal and vertical scales. However, the models should capture many larger-scale features, such as stabilizing or destabilizing of the boundary layer, cloud burn-off, and impacts on larger circulation.

[420] Model cloud responses to absorbing aerosols have not been compared carefully with cloud-scale model and field studies for specific conditions. The framework presented here, and in greater detail in Koch and Del Genio [2010], requires ongoing revision as future studies provide information.

7.3.3 Increased Absorption by Cloud Droplet Inclusions

[421] Absorption by BC increases when it is covered with non-absorbing material [Graßl, 1975; section], including water. This increase affects both the amount of atmospheric absorption attributable to BC and the estimated absorption that is collocated with clouds, with both effects possibly leading to positive forcing. The lack of simulated absorption by mixed or cloud-borne BC in most models probably affects the general prediction of the global average semi-direct effect.

[422] Forcing due to altered cloud albedo was first estimated by Chýlek et al. [1984, 1996] as 1 to 3 W m−2 in a simple analysis that assumed fixed volume fractions of BC within cloud droplets. Chuang et al. [2002] used a chemical transport model to estimate the positive forcing as +0.07 W m−2, although it was not clear how they determined the fraction of BC within droplets. Models that simulate the dynamics of aerosols and clouds find that although more than 90% of BC in the atmosphere passes through and is removed by clouds, the short residence time of BC in clouds results in only a few percent of the global BC burden being present in cloud droplets at a given time [Jacobson, 2012; Ghan et al., 2012]. Stier et al. [2007] and Ghan et al. [2012] estimated direct forcing by BC in droplets as +0.02 W m−2 and less than +0.01 W m−2, respectively, using the Bruggeman mixing rule to represent BC-droplet absorption. The dynamic effective medium approximation predicts the greatest absorption increase for wetted particles [Jacobson, 2006]. It is consistent with the small number of laboratory studies measuring absorption at high relative humidity (section 3.7.1) and results in a direct forcing change of +0.05 to 0.07 W m−2. Another mechanism involves droplet heating by BC and subsequent deactivation of CCN, but this has been shown to be insignificant at appreciable supersaturation levels [Conant et al., 2002].

[423] A potentially greater forcing results from cloud burn-off when cloud-borne BC increases absorption within clouds. The forcing results summarized in the preceding paragraph do not account for this effect. The choice of model to represent absorption by BC residing within cloud droplets greatly affects the modeled absorption and, thus, the thermodynamic effect on clouds. In the model with lower absorption described above, cloud absorption caused a small negative semi-direct effect [−0.07 W m−2, Ghan et al., 2012]. However, predicted heating rates within clouds are 2 to 2.3 times greater when the dynamic effective medium approximation is used, compared with a core-shell treatment [Jacobson, 2012]. Jacobson [2010] found an increase in temperature of +0.18K for fossil fuel soot and +0.31K for fossil fuel plus biofuel soot and gases, when cloud absorption by BC and its resulting feedbacks were included versus excluded (see their Figure 1a). With the equilibrium climate-sensitivity of this model (0.6K (W m−2)−1), these temperature changes correspond to a forcing increase of +0.30 W m−2 for the effect of cloud absorption by fossil fuel soot effect alone. Cloud absorption by fossil fuel plus biofuel soot and gases has a greater forcing of +0.52 W m−2, but these simulations may include effects other than those of BC. All of these forcing estimates include the small forcing change caused by cloud albedo. When the soot-only values are scaled to our industrial-era emission rate of 13,900 Gg yr−1, this climate forcing would be +0.76 W m−2. However, some plausible optical treatments produce much lower in-cloud heating rates in Jacobson [2012] and lower forcing in Jacobson [2006].

[424] The cloud absorption effect is highly uncertain because there are few measurements of aerosol optical properties at cloud conditions or of the relationship between in-cloud aerosol absorption and cloud dissipation. Although it is difficult to assign a central estimate under these circumstances, we average the negative estimate of Ghan et al. [2012] and an average of the highest and a mid-range estimate from the Jacobson studies (+0.54 W m−2) to obtain a very uncertain central estimate of +0.2 W m−2 for this effect, with uncertainty bounds of (−0.1, +0.9) W m−2. The upper bound comes from using the highest forcing-per-emission obtained from Jacobson [2010] for the case of fossil fuel and biofuel soot.

7.4 BC Indirect Effects on Liquid Clouds

[425] Although aerosol liquid-cloud indirect effects have been extensively studied, only a few studies isolate the influence of BC alone. Many models do not have adequate sophistication to simulate aerosol microphysics, including mixing between BC and other aerosol species and the evolution of aerosol size distribution. Earlier model studies had minimal treatment of the aerosol microphysics that are important for capturing the gaseous and aerosol interactions that dominate CCN concentrations and, hence, the BC indirect effect. Below we review earlier literature, discuss the importance of aerosol microphysics, and summarize results from studies that include the microphysics of aerosol-cloud interactions.

[426] Two simple global model studies estimated the effect of BC on liquid-cloud indirect effects and found a negative BC indirect effect. Hansen et al. [2005] estimated that BC contributed 5% of the aerosol indirect effect in a model study with the net indirect effect magnitude prescribed and parameterized so that cloud cover and cloud albedo are augmented proportionately to the logarithm of the aerosol number concentration. Aerosol number concentration was derived from aerosol mass and assumed size, and aerosols were externally mixed. Chuang et al. [2002] estimated the liquid-cloud effects of carbonaceous aerosols, using a parameterization of cloud droplet number concentration (CDNC or the number of cloud droplets per volume of cloud). CDNC was parameterized based on Köhler theory but without explicit aerosol microphysics. They estimated that carbonaceous aerosols from biomass-burning and fossil fuels contributed 63% and 28% of the indirect effect, respectively; these aerosols contained both BC and organic carbon. Sulfate had a smaller effect because of its lower burden. The total indirect effect was approximately proportional to the atmospheric burden, but the effects of different species were slightly less than additive.

[427] BC could have a substantial influence on the indirect effect due to its potentially large contribution to aerosol number concentration rather than aerosol mass. However, number concentration of aerosol and BC mass are not linearly related. BC also provides a surface upon which volatile inorganic or organic compounds may condense. In the absence of BC, sulfate might preferentially nucleate fresh particles and additional sulfate would condense upon pure sulfate particles. In general, larger particles make better CCN or IN, although CCN activity increases when BC is internally mixed with soluble species such as sulfate or OA, while there is evidence that IN activity is optimal for unmixed or pure BC. CCN activation for BC is discussed in section 3.8, where methodology for applying Köhler theory to a BC-solute mixture is described. Activation depends upon particle size, moles of solute in the particle, and water supersaturation in the environment (Figures 5 and 6). The indirect effect dependence on particle number concentration is also nonlinear. Additional particles generally have a greater effect on clouds in clean conditions and relatively less in more polluted environments [e.g., Twomey, 1991; Lohmann and Feichter, 2005; Hoose et al., 2008]. Given these nonlinearities, it is difficult to model BC-cloud effects without models that include detailed aerosol microphysical schemes.

[428] Several model studies used aerosol microphysical schemes in simulations of the warm-cloud indirect effect of BC (Table 18). These model studies were done by comparing a simulation with all aerosols with another simulation in which BC, or BC and OA, have been removed. For consistency, we report all results as the response to the addition of BC. Below, we discuss the forcing changes actually reported by each study. However, each investigation used different changes in emissions, and Table 18 shows values scaled to the relevant BC emission rate.

Table 18. Studies That Provide Indirect Effect Estimates
Study Reported Indirect Effect (W m−2) Emissions Changed in Study; (BC/OC (Tg yr−1))a Adjusted IE for BC (W m−2); Adjustment Methodb Effectsc; Diagnostic Methodd
Liquid clouds (microphysics and lifetime)

Kristjansson [2002]

−0.1 100% BC; 12.4/0 ALL: −0.11 WIE(1/2), SD (DE); M1

Chen et al. [2010a]

−0.13 ± 0.33 50% FF BC + OC; 1.5/2.2 FF: −0.13 WIE(1/2); M1
−0.31 ± 0.33 50% ALL BC + OC; 3.9/30 ALL: −0.13

Bauer et al. [2010]

+0.05 100% FF BCf; 3.0/0 FF: +0.18; EM, SD WIE(1/2), SD (DE); M1
−0.12 50% FF + BF BC; 2.3/0 BF: −0.10; EM, SD
−0.2 100% BF BC + OC; 1.6/6.4 FF + BF: −0.32; EM, SD
Koch et al. [2011b]e −0.08 to +0.31 100% FF BCf; 3.0/0 FF: +0.05 (−0.15 to +0.26); EM, SD, DE WIE(1/2), SD, DE; M3
−0.2 to +0.08 100% BF BC + OC; 1.6/6.4 BF: −0.05 (−0.12 to +0.02); EM, SD, DE

Spracklen et al. [2011]

−0.23 100% FF + BF BC + OC 16.9 total FF + BF: −0.13 EM WIE(1)
−0.34 100% ALL BC + OC 54.6 total ALL: −0.09

Storelvmo [2012]

0.01 100% FF ALL: 0.01 WIE
Mixed-phase clouds

Lohmann and Hoose [2009]

+0.12 g 100% BC; 6.3/0 ALL: +0.12; Noneh MIE (WIE, SD); M2
+0.2 g ALL: +0.2; Noneh

Storelvmo et al. [2011]

+0.16 ± 0.08 100% BC; 6.3/0 ALL: +0.16; Noneh MIE (WIE, SD); M2
Yun and Penner [2012]i +0.15 BC (PCT) ALL: +0.15h MIE; M4
+0.71 to 0.93 BC + OM (YCT)
Ice clouds

Penner et al. [2009]

−0.3, −0.4 100% BC + OC; 13.5/97 ALL: −0.27; None IIE; M1

Gettelman et al. [2012]

−0.06 100% BC ALL: −0.06; None IIE; M2

Liu et al. [2009a]

+0.22, +0.39 100% BC + OC; 13.5/97 ALL: +0.24; None IIE, SD (WIE, DE); M3
All effects, including direct and cryosphere
Jacobson [2010]j, k +0.47 100% FF BC + OC; 3.2/2.4 FF: +0.47 WIE, SD, DE, MIE, IIE, CRY, CLIM
+0.68 100% FF + BF BC + OC, BF gases FF + BF: +0.68
  • a FF = fossil fuel, BF = biofuel, BB = biomass burning; ALL = FF + BF + BB. When a percentage reduction is given, it refers to all-source emissions. For comparison to emission values in the remainder of the document, 1 Tg = 1000 Gg.
  • b Adjustments: EM = scaled from mass of emissions used in study to our estimate of BC emissions (FF 5.6, BF 4.3, OB 4.0, ALL 13.9 Tg yr−1); SD = estimate of semi-direct effect deducted assuming −30% of direct forcing; DE = estimate of direct effect in cloudy skies removed by assuming 65% of direct forcing; SZ = adjusted from particle size used in study to 100 nm diameter using dependence reported by Bauer et al. [2010].
  • c WIE = warm cloud indirect effect, with 1 = first indirect effect (cloud-albedo) and 2 = second indirect effect (lifetime); MIE = mixed-phase cloud indirect effect, IIE = ice cloud indirect effect, DE = direct effect, SD = semi-direct effect, CRY = cryosphere, CLIM = climate response. Effects in parentheses are included in the simulations but did not affect the indirect effect diagnostic and, therefore, are not included in the values in the “Reported indirect effect” column.
  • d M1 = Change in ToA cloud forcing with and without BC; M2 = Difference between model runs with and without mixed-phase indirect effects; M3 = Change in cloud forcing with and without BC + OC; M4 = Change in cloud forcing with and without BC + OC in mixed-phase clouds only.
  • e Values exclude those from the GISS model, which are reported separately in Bauer et al. [2010].
  • f Simulations were also done with reductions of 50% BC + OC emitted from diesel. The results were very similar to those for fossil-fuel BC emissions.
  • g The two estimates use different parameterizations of the Bergeron-Findeisen process.
  • h All estimates were first averaged without adjustment to obtain +0.15 W m−2 and then scaled for emissions as described in the text.
  • i Contact nucleation parameterizations: PCT = Phillips et al. [2008]; YCT = Young et al. [1974]. The main effect in PCT is constrained by findings of BC within ice nuclei; YCT includes all OM as contact nucleus, but because the response is quite nonlinear, OM cannot be extracted from the latter treatment and is not used as an estimate of BC effect.
  • j This study included a simulation with biofuel and biomass emissions removed. Emissions of short-lived and long-lived GHGs emitted from BF and BB were also removed, so the results are not comparable with the aerosol-only effects summarized here.
  • k Effective forcing values from this study were not separated by mechanism, so the total forcing is discussed in section 10.

[429] Kristjansson [2002] calculated a BC indirect forcing of −0.1 W m−2 (5–10 % of the total aerosol indirect effect), with regional values reaching about −0.25 W m−2 in more polluted regions where BC was an important component of the accumulation mode aerosols. In some remote oceanic regions, the sign was positive because BC reduced aerosol hygroscopicity.

[430] Chen et al. [2010a] modeled liquid-cloud indirect effects due to BC and OA emissions. Their results show that 50% of present-day BC and OA emissions from fossil fuel cause a −0.13 W m−2 indirect forcing, and 50% of all BC and OA emissions (including biomass-burning emissions) cause a −0.31 W m−2 indirect forcing. Fossil fuel BC particles were assumed to have quite small sizes of 25 nm diameter at emission (compare to section 3.6.2). The indirect effects more than offset the calculated direct effects for fossil fuel and all carbonaceous emissions of +0.07 and +0.12 W m−2, respectively. In the Chen et al. model experiments, indirect effects were isolated from semi-direct effects by running the simulation without aerosol-radiation coupling. The authors also estimated the standard deviation of the forcing over multiple years, which was as large as or greater than the forcings for 15 year simulations. While the absolute change in BC + OA emissions was larger for biomass emissions, so one would expect a larger effect, Chen et al. also attribute the larger forcing mostly to the larger size of the biomass and biofuel combustion particles as well as to the greater hygroscopicity of the non–fossil fuel emissions.

[431] Bauer et al. [2010] performed three BC reduction experiments. Adding either fossil fuel BC or fossil fuel plus biofuel BC decreased CDNC, probably because BC reduces the number of pure sulfate aerosols. The semi-direct effect in these cases was negative (more low-level clouds). The combined warm cloud indirect plus semi-direct effects resulted in positive forcing when fossil fuel BC only was added but negative forcing when a mix of biofuel and fossil fuel BC was added. Adding biofuel combustion emissions apparently produces a larger semi-direct effect, especially at high latitudes of both hemispheres, perhaps due to lofted BC in these regions (section 6.2). A third experiment tested the combined effect of BC and OA from biofuels only. In this case, aerosols caused increased CDNC and cloud cover with a larger negative forcing. The indirect effect was found to depend greatly on assumed particle sizes, with a factor of 2 decrease or increase in diameter producing a 45% increase and 30% decrease in magnitude, respectively.

[432] Koch et al. [2011a] presented a multi-model study of the effects of BC and OA from fossil fuel and biofuel burning. Six models examined alterations in cloudy-sky fluxes due to the same emission changes. The response to fossil fuel BC was −0.08 to +0.31 W m−2 with an average of +0.08 W m−2, while the response to 50% of biofuel BC and OA ranged from −0.20 to +0.08 W m−2 with an average of −0.10 W m−2. Cloud response in this study was diagnosed using the difference in cloudy-sky fluxes between simulations with and without BC, except for one model (GISS, reported separately by Bauer et al. [2010] and discussed above). Therefore, these values include not only cloud semi-direct effects and warm-cloud indirect effects but also direct radiative forcing in cloudy skies, which can be more than half of the direct radiative forcing [Zarzycki and Bond, 2010]. Removing the above-cloud direct forcing effect would result in more negative liquid-cloud effect.

[433] Many model studies do not separate microphysical effects from semi-direct effects, although such a division could be useful for comparing across studies that do estimate these effects separately. To isolate liquid-cloud microphysical effects in the Koch et al. [2011a] study, we estimated semi-direct and direct radiative forcing in cloudy skies for each model. Adjusted values, scaled to our emission estimates, are summarized in Table 18.

[434] Spracklen et al. [2011] modeled liquid-cloud effects, comparing CCN against observations. They found that the contribution of carbonaceous aerosols was required to explain observed CCN concentrations. Their estimate of forcing for BC plus OA emissions from fossil fuel and biofuel combustion was −0.23 W m−2 and reached −0.34 W m−2 with the addition of BC and OA emissions from biomass burning. Storelvmo [2012] found that BC had only a +0.01 W m−2 indirect effect in a study using four modes to represent BC and OA, while −0.25 W m−2 was attributed to OA.

[435] Jacobson [2010] studied the effects of fossil fuel and biofuel combustion on climate. Fossil fuel emissions, including BC, OA (some of which absorbs light), and primary sulfate, caused a net decrease in liquid cloud cover. Biofuel combustion emissions, including these species as well as co-emitted gaseous species, caused a net increase in liquid cloud cover. In this case, the cloud effects include the semi-direct effect, which is reported to be a cloud loss in this model. Jacobson [2010] argued that fossil fuel emissions contain fewer hygroscopic particles and, therefore, have lesser indirect effect, while the more hygroscopic biofuel combustion particles affect liquid clouds more significantly. This study includes all effects (direct, liquid, mixed, semi-direct, and ice clouds, as well as cryosphere forcing), and they were not separated.

[436] For the studies discussed above, the mean value of the isolated liquid-cloud indirect effect, when scaled to the magnitude of industrial-era BC, from the three studies that modeled all carbonaceous aerosol, is about −0.1 W m−2. These three models give similar forcings when scaled to emissions of 13,900 GgC yr−1. We use the model variation from the Koch et al. [2011a] study, about 0.2 W m−2, as the total uncertainty in this effect. However, models are frequently unable to distinguish separate mechanisms, so the separation between liquid-cloud indirect effect and semi-direct effect is somewhat arbitrary. We therefore estimate the combined indirect and semi-direct effect, excluding forcing by cloud droplet inclusions, as −0.2 ± 0.3 W m−2, where the central value is the sum of the separate estimates and the uncertainties have been added in quadrature. This value is more robust than the separate estimates. When the asymmetric uncertainty due to lofted biomass emissions is also added, the uncertainty bounds become −0.82 to 0.08 W m−2.

[437] A common result among the models summarized in Koch et al. [2011a] is that aerosol emissions from biofuel combustion cause larger indirect effects per mass than do emissions from fossil fuel. Table 18 shows how the studies discussed above were scaled for emission rate and, in one case, particle size. Studies that isolate fossil fuel or diesel emissions suggest a small positive liquid-cloud forcing, scaled to emissions, averaging +0.08 W m−2, while those that isolate biofuel combustion emissions have a small negative forcing averaging −0.08 W m−2. However, studies that examine the two in combination estimate a larger negative forcing that is greater than the near-zero sum. This finding suggests that the liquid-cloud system contains significant nonlinearities so that the total effect is not equal to the sum of the parts. In contrast, Spracklen et al. [2011] attribute greater negative forcing to fossil fuel and biofuel particles compared with particles emitted from open burning.

[438] Biofuel and fossil fuel combustion emissions are different in four ways that may affect cloud responses. The first is chemical; the ratio of organic to black carbon is larger for biofuel and organic carbon is generally more hygroscopic than BC. However, some models do not explicitly treat the difference in hygroscopicity, so that modeled BC and OA particles are identical. Second, co-emitted aerosol species or precursors also affect the number of particles. For example, altering biofuel sources to remove 1 Gg of BC also removes about 4 Gg of primary OC, so it could reduce primary particle number concentrations much more effectively than altering fossil fuel sources to achieve the same reduction. Third, models assume different emitted particle sizes. Both Chen et al. [2010a] and Bauer et al. [2010] assumed larger particle size in biofuel combustion emissions than in fossil fuel emissions. Since particles must grow large enough to act as CCN, biofuel combustion particles are closer to activation than fossil fuel particles. Conversely, an assumption of smaller diameter yields many more particles for the same mass emission. The modeled indirect effect depends on assumptions about source-specific particle size, number concentration, and composition, but such observations are limited. Fourth, cloud response has a regional dependence. Fossil fuel consumption is more prevalent at temperate latitudes, where cloud types and atmospheric dynamics differ greatly from tropical regions. Few models have conducted experiments that isolate sensitivities to these factors.

7.5 BC Indirect Effects on Mixed-Phase and Ice Clouds

[439] The influence of aerosols on both mixed-phase and ice clouds depends upon whether they can effectively nucleate ice. Typical IN are mineral dust, biological particles, and possibly BC. Although coating of BC with sulfate or organic material increases its ability to act as a CCN, coatings might reduce BC's IN activity, although there is less evidence for this change. In order to have an influence on ice particle formation, it is not enough that BC have the capability to serve as IN; it must also be equally or more efficient than other types of aerosols.