2.1 Model description

This study uses the Community Earth System Model (CESM), with the coupled Community Atmosphere Model version 5.4 (CAM5.4) (Hannay and Neale, 2015) and Community Land Model version 4 (CLM4) (Oleson et al., 2010). In Jiang et al. (2016) a similar model setup is used with the exception of its use of CAM5.3. CAM5.3 contains physics parameterization updates from the previous CAM versions, including a two-moment stratiform cloud microphysics scheme to predict mass and number mixing ratios of cloud liquid and cloud ice (Morrison and Gettelman, 2008), and a three-mode version of the Modal Aerosol Module (MAM3) to predict mass and number mixing ratios of aerosol components (X. Liu et al., 2012). Compared to the CAM5.3 version, some notable additions in the CAM5.4 version include a new prognostic precipitation scheme (Gettelman et al., 2015), new ice nucleation treatments (Y. Wang et al., 2014), and improved dust optical properties and emissions (Albani et al., 2014). This model also uses the four-mode version of MAM (MAM4) (X. Liu et al., 2016). MAM4 consists of the following four lognormal modes (shown with their median size ranges and standard deviations): Aitken (0.015–0.053 µm, σ=1.8), accumulation (0.058–0.27 µm, σ=1.6), coarse (0.80–3.65 µm, σ=1.8), and primary carbon (0.039–0.13 µm, σ=1.6). The median sizes of aerosol modes are changed due to the microphysical processes (e.g., condensation and coagulation) while standard deviations for each mode are fixed. The OA from accumulation and primary carbon modes, which is used to represent BrC depending on its source (mentioned later), has a density in the model of 1 g cm⁻³.

MAM4 differs from MAM3 in its inclusion of the primary carbon mode. This mode, which consists of the species BC and primary organic matter (POM), is emitted from the incomplete combustion of fossil and biomass fuels. In MAM4, this mode is transferred to the accumulation mode once it has aged. The primary carbon mode ages by acting as a nucleus for the condensation of sulfuric acid (H₂SO₄) vapor, ammonia (NH₃), and semi-volatile organics, as well as through collision and coalescence with Aitken and accumulation modes. This acts to both increase the size of the aerosol as well as increase the hygroscopicity of hydrophobic aerosols such as BC. The optical calculations in the model consider these aged particles to be internal mixtures of aerosols within the mode, and the RI of the accumulation mode, as well as the other three modes, is calculated as a volume-weighted mean of the refractive indices of all of the aerosols' components within the mode (X. Liu et al., 2012, 2016).

2.2 Model modifications

2.2.2 BrC refractive index

Another modification involved the inclusion of a parameterization for BrC in CAM5.4. The parameterization is described in Saleh et al. (2014) and determines the imaginary refractive index (RI) of POM based on the BC-to-OA ratio of the BB and BF emissions. A later experiment by Lu et al. (2015) derives a similar parameterization based on many different BB datasets, including that of Saleh et al. (2014). The Saleh et al. (2014) RI is calculated based on the following equations:

$$\begin{array}{}\text{(1)}& {\displaystyle }& {\displaystyle }\text{RI}=\mathrm{1.7}\left(\pm \mathrm{0.2}\right)+k_{\mathrm{OA}}i=\mathrm{1.7}\left(\pm \mathrm{0.2}\right)+k_{\mathrm{OA},\mathrm{550}}{\left({\displaystyle \frac{\mathrm{550}}{\mathit{\lambda }}}\right)}^{w}i,\text{(2)}& {\displaystyle }& {\displaystyle }{k}_{\mathrm{OA},\mathrm{550}}=\mathrm{0.016}\cdot {\log }_{\mathrm{10}}\left(\text{BC to OA}\right)+\mathrm{0.04},\text{(3)}& {\displaystyle }& {\displaystyle }w={\displaystyle \frac{\mathrm{0.21}}{\left(\text{BC to OA}+\mathrm{0.7}\right)}},\end{array}$$

where Eq. (1) describes the real (1.7) and imaginary (k_OA) parts of the RI. Equation (2) represents the imaginary RI at 550 nm (k_OA,550), and Eq. (3) represents the wavelength dependence of k_OA(w). The parameterization describes decreasing wavelength dependence and increasing BrC absorption as the emission BC-to-OA ratio increases. This relationship was derived for the wavelengths of 370–950 nm and a laboratory-derived BC-to-OA ratio range of 0.01–0.5. In the model, the parameterization is applied over this BC-to-OA range and for the CAM waveband midpoints of 304–1010 nm (wavebands 8–12 in the CAM RRTMG_SW model; Neale et al., 2012). Uncertainty in k_OA from this parameterization is associated with the lab measurements of the particle mass, the range in assumed complex refractive index for BC, the mixing state of BC and OA, the measured real part of the OA refractive index, and the measured absorption coefficients used in optical closure calculations (Saleh et al., 2014).

Figure 2The ratio of column averaged burden of black carbon to POM of (a) biomass burning (BB) and (b) biofuel burning (BF). Note the different scales.

Download

The parameterization in CAM only changes the RI of POM (emissions of POM and density (1 g cm⁻³) remain unchanged) and was applied in the part of CAM5.4 that calculates modal aerosol optical properties using a parameterization described in Ghan and Zaveri (2007). This parameterization calculates the specific scattering, specific absorption, and asymmetry parameter of aerosols in a mode as a general internal mixture of components, bilinearly interpreting these optical properties from offline Mie calculations. These calculations are a function of the wet surface mode radius as well as the wet refractive index of each mode. The code modification to implement a changing RI for BrC is implemented before the bilinear interpretation. When the model calls the RI for BB or BF POM from an input physical properties (phys_prop) file, the parameterization changes the imaginary part of the RI at the wavelengths 304–1010 nm based on the calculated BC-to-OA ratio at each time-step for each grid point and level. The BC-to-OA ratio is calculated based on each source's ratio (BB or BF) to preserve the initial emission ratios and minimize error due to inter-source mixing (see Fig. 2 for BB and BF BC-to-OA ratios). As shown in Fig. 2, the BB BC-to-OA ratio is uniformly distributed in the range of 0.07 to 0.095 with a global mean of 0.084. The BF BC-to-OA ratio is much higher, varying in the range of 0.1–0.3 with a global mean of 0.185.

A modified BrC experiment was run with the same Saleh et al. (2014) RI parameterization, but the BC-to-OA ratio was set to 0.08 (approximate global column average biomass burning BC-to-OA) at each time step, grid point, and level (see Table 1). This results in a constant imaginary RI for BB and BF POM.

A few assumptions in this model simulation introduce uncertainty in the representation of BrC in CESM. One of those assumptions is neglecting absorption by BB SOA (Lin et al., 2014; Saleh et al., 2015) or absorbing aromatic SOA (X. Wang et al., 2014, 2018; Jo et al., 2016), which is neglected due to the lack of SOA speciation in the model. This assumption, in conjunction with the use of GFED 3.1 instead of GFED 4, may act to underestimate the climate effect due to BrC. Another assumption is the model use of a volume mixing assumption, which may act to overestimate aerosol light absorption (Jacobson, 2000; Adachi et al., 2010). We also assume that the BC-to-OA ratio in transported smoke is similar to BC-to-OA from the source region, allowing for the use of a BC-to-OA ratio at each grid cell at every time step to calculate k_OA in each grid cell. The uncertainty in k_OA associated with this assumption is small (<10 % for BB emissions assuming transport from the Equator to the Arctic (not shown)) and is assumed to be negligible.

Another source of uncertainty when considering an absorbing aerosol in the model is the aerosol's vertical distribution. CAM5.4 uses six vertical injection heights for wildfire emissions described in Detener et al. (2006): 0–100 m, 500–1000 m, 1–2 km, 2–3 km, and 3–6 km. These fire emission heights depend on the geographic location of the fire and the vegetation type derived from GFED, with the highest plumes corresponding to boreal fires. If BrC is lofted over a more reflective surface such as a cloud, its shortwave radiative forcing will be more positive than if it stays below the cloud or remains lower in the atmosphere. A counterbalancing effect is a more negative longwave forcing at higher levels in the atmosphere (Penner et al., 2003). The vertically sensitive semi-direct effects of BrC (i.e., changes in atmospheric stability and cloud cover due to atmospheric heating by BrC) are discussed in more detail in Sect. 3.2. Comparisons between the total OA (POA+SOA) vertical distribution and aircraft observations in Shrivastava et al. (2015) show that the standard CAM5 aerosol treatment largely underestimates Arctic biomass OA, possibly due to the model neglecting important SOA contributions from biomass burning. This could lead to an underestimation of BrC radiative effects due to lower BrC concentrations at all levels of the model. Vertical profiles of aerosols, cloud fraction, and heating rates in the model are shown over six regions with a strong BrC radiative effect due to aerosol–radiation interaction (Fig. S2 in the Supplement).

2.3 Experimental design

The model was run at a horizontal resolution of 0.9^∘ latitude by 1.25^∘ longitude with 30 vertical levels. The simulation period covers 9 years from 2003 to 2011 with prescribed monthly sea surface temperature and sea ice. The emissions consist of daily BC, POM, and sulfur dioxide (SO₂) emissions from GFED 3.1 (Giglio et al., 2013), vertical distribution of fire emissions based on the AeroCom protocol (Detener et al., 2006), and anthropogenic aerosol and precursor gas emissions from IPCC AR5 (Lamarque et al., 2010). The model runs consist of a spin-up year (2003), followed by a 9-year run (2003–2011) using a concatenated version of the emissions files for each species and emission type. Four experiments, with five ensemble members each, were run following the aforementioned setup. Ensemble members are varied by applying different initial temperature perturbations of the order 10⁻¹⁴. “NOBRC” is the control experiment with the default model configuration, while “BRC” includes a parameterization for BrC that takes into account a varying RI that is dependent on the BC-to-OA ratio. “BRC_CNST” includes a parameterization for BrC that assumes a constant imaginary RI based on the approximate global average BC-to-OA ratio of 0.08, derived from biomass burning emissions. “BRC_BL” includes the BrC parameterization as well as a BrC bleaching parameterization. Table 1 describes the different model runs.

Table 1Description of the model simulations used in this study.

Download Print Version | Download XLSX

2.4 Radiative effect calculation

The direct radiative effect and indirect effect of BrC are calculated using the method recommended in Ghan (2013). This method breaks the total radiative effect into changes due to aerosol–radiation interaction (REari), aerosol–cloud interaction (REaci), and surface-albedo changes (REsac) (note that REsac are not analyzed in this paper because the model does not take into account snow albedo reduction via BrC deposition on snow).

In Eq. (5), Δ represents the difference between the model runs with and the model runs without BrC, in this case represented by BRC (BRC_CNST, BRC_BL) – NOBRC. The differences in each of the parentheses are conducted for each model run represented in Δ. The variable F represents all-sky, top of the atmosphere (TOA) radiative flux (longwave+shortwave); F_clean is the same but without aerosols, calculated from the diagnostic radiation call in the model run, and F_clean,clear removes both aerosols and clouds.

While this method calculates aerosol indirect effects – changes in cloud microphysics due to more/less numerous cloud condensation nuclei – and aerosol semi-direct effects – changes in cloud environments due to radiation interaction of the aerosols – (collectively REaci), when it is applied to the experiment with absorbing versus non-absorbing organics, REaci only represents the semi-direct effect of BrC (indirect effects of organics are already accounted for in these two experiments). This effect describes the ability of absorbing aerosols to change the lifetimes of clouds by changing the static stability of the troposphere and limiting incoming solar radiation to the surface; also, these aerosols may help to evaporate cloud droplets when heating the cloud levels (Ackerman et al., 2000; Koch and Del Genio, 2010).

Figure 3AERONET SSA compared to CAM5.4 model output SSA, not including brown carbon (NOBRC), including brown carbon (BRC, BRC_CNST), and including brown carbon and a bleaching effect (BRC_BL). Vertical lines are color-coded error bars and run from left to right as follows: observed, NOBRC, BRC, BRC_BL, and BRC_CNST. Values below the upper x axis indicate percentage of available data in the 9-year period.

Download

3.1 Model validation

Four model simulations (i.e., NOBRC, BRC, BRC_CNST, BRC_BL) are compared to Level 2.0 AERONET (http://aeronet.gsfc.nasa.gov, last access: 1 December 2018) observations from nine different sites, located in three different BB regions: North American and Asian boreal forests, South America (SA), and Africa. The AERONET data are averaged over the period 2003–2011 to match the model simulation period. The model directly outputs the 440 nm single scattering albedo (SSA), aerosol optical depth (AOD), and absorption aerosol optical depth (AAOD) used in this analysis.

The relationship between AERONET and SSA (Fig. 3) and AOD (Fig. S3) from the CAM5.4 NOBRC experiment is the same as in the Jiang et al. (2016) comparison with CAM5.3. As with Jiang et al. (2016), CAM5.4 does a reasonably good job of simulating AOD and SSA, although there are some regions of disagreement, possibly due to incorrect fire emissions, excessive aerosol scavenging by liquid-phase clouds, or inaccurate representation of aerosol optical properties and/or size (Jiang et al., 2016; Yu et al., 2016).

Another possibility for the disagreement between the model and AERONET may be due to error associated with AERONET. The AERONET cloud screening process may introduce error by screening aerosol plumes that have highly variable optical properties (Smirnov et al., 2000). Furthermore, the inversion algorithm used to calculate SSA from AOD (Dubovik and King, 2000) introduces error in retrieving BB SSA in low aerosol concentration situations (Dubovik et al., 2000). Also, the dynamics in the model are going to be different than those influencing the AERONET sites, which could lead to a difference in aerosol exposure at these locations.

When comparing the four model experiments to AERONET AOD (Fig. S3), there is good agreement between the model with and without BrC. This is due to the fact that model AOD is dominated by aerosol scattering, and any changes in absorption with the incorporation of BrC are small by comparison. Overall there is little obvious effect on AOD due to BrC. A much stronger effect can be seen when looking at AAOD, with peak BrC effects in biomass burning seasons (Fig. S4).

When looking at SSA (Fig. 3) the differences between the four model runs are more apparent. In all of the regions (Arctic, Africa, South America), incorporation of BrC (BRC, BRC_CNST, BRC_BL) decreases SSA with respect to the default CAM5.4 model (NOBRC). This is especially true of biomass burning seasons. One would expect this to be true given the large burdens of OA released at these times coupled with the modification to their RI. This seasonality is less apparent at Ascension Island, possibly due to less concentrated OA from diffusion and scavenging during transport to the island, leading to a smaller effect from BrC.

The bleaching effect in BRC_BL reduces its absorption during these seasonal peaks, placing it closer to the NOBRC simulation than either BRC or BRC_CNST. When looking at the African and South American sites, especially those further away from the BB source regions (Fig. 3b–c), the BRC_BL is more similar to NOBRC; in the Arctic, BRC_BL tends to be closer to the BRC simulation. This can be explained by the higher [OH] in the tropics resulting in a faster bleaching of BrC than in the Arctic. Table S1 in the Supplement shows [OH] in different regions and the half-life of BrC due to the bleaching effect in these regions, which ranges from 0.37 days (southeast Asia) to 2.09 days (Arctic).

Figure 4Absorption Ångström exponent (AAE) comparison between all of the AERONET sites from Fig. 3 and the four model runs: (a) NOBRC, (b) BRC, (c) BRC_CNST, and (d) BRC_BL. The lower row looks at the same model comparisons neglecting the Ascension Island site. Red indicates African sites, orange indicates South American sites, and blue indicates Arctic sites. The solid line indicates a 1:1 fit and the dotted line is a best fit to the data with the slope and correlation coefficient shown.

Download

Another effect that can be seen in the SSA plots is a subtly lower SSA (i.e., stronger absorption) in the Arctic regions (Fig. 3g–i) for BRC_CNST than for BRC. This may relate to the constant BC-to-OA of 0.08 in BRC_CNST, which is higher than the column burden average BC-to-OA at these high latitudes (Fig. 2). A higher BC-to-OA relates to stronger absorption based on the Saleh et al. (2014) parameterization. While the differences between BRC and BRC_CNST are quite small, this presents the possibility of overestimation of the BrC effect at northern high latitudes with BRC_CNST.

While the SSA comparison does not show improvement with BrC incorporation, agreement between the model and the AERONET absorption Ångström exponent (AAE) is better with BrC (Fig. 4). The AAE is calculated based on the two wavelengths 440 and 675 nm (λ₁ and λ₂, respectively) and the measured absorption coefficients at the two different wavelengths (b_abs(λ₁) and b_abs(λ₂)).

It describes the wavelength dependence of the measured emissions, with AAEs greater than 1 representing higher wavelength dependence (i.e., larger differences in b_abs(λ₁) and b_abs(λ₂)). As a result, higher AAEs can be used to identify the presence of BrC (Bond, 2001; Kirchstetter et al., 2004; Chen and Bond, 2010; Bond et al., 2013).

Table 2Comparisons between modeled AAE and the AAE from observations at Ascension Island during the LASIC (Layered Atlantic Smoke Interactions With Clouds (LASIC) campaign, the Las Conchas fire in New Mexico, and biomass burning and prescribed burns from the United States during SEAC⁴RS (Studies of Emissions and Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys).

Download Print Version | Download XLSX

Default CAM (NOBRC) tends to overestimate the AAE of the previously mentioned biomass burning sites when AERONET predicts low values, and vice versa (Fig. 4a). In the BRC, BRC_CNST, and BRC_BL model runs, the underestimation of the AAE by CAM is improved, especially in the South American and Arctic regions (Fig. 4b–d). This is consistent with Feng et al. (2013) who showed that including different values for absorbing OA drives the model AAE closer to that of AERONET.

The large African outliers in this comparison are all from the Ascension Island site (Fig. 4e–h). A comparison of the AAE and the scattering Ångström exponent (SAE) for the AERONET sites (Fig. S5) suggests that the Ascension Island aerosols are a large particle–low absorption mixture (Fig. 8 in Cappa et al., 2016). When the same comparison is made for the BRC model run (Fig. S6), the Ascension Island aerosols are identified by the Cappa et al. (2016) classification as a large particle–BC mixture. The large particle trend at Ascension Island can likely be attributed to sea salt aerosol, while the difference between the model and AERONET may be due to more smoke transport to the island in the model. Two other possibilities for the disagreement were previously mentioned: error introduced to AERONET due to cloud screening processes and model bias in the circulation patterns.

Figure 5Differences in (a) AAOD and (b) the AAE between the BRC model run and the NOBRC model run.

Download

Other comparisons between the model and observational AAEs are shown in Table 2. These observations are from Ascension Island (Zuidema et al., 2018), a southwestern US (New Mexico) wildfire (S. Liu et al., 2014), and continental US BB and prescribed burns from the SEAC⁴RS campaign (Mason et al., 2018). These comparisons are largely qualitative given the coarse resolution of the model (∼100 km) compared to the much finer wildfire spatial resolution. Both the Ascension Island and New Mexico observations are from the surface, while the continental US observations are made via aircraft. The Ascension Island and SEAC⁴RS US BB and BF observations were made during different years than represented in the model run, so our comparison operates under the assumption that these regions have consistent seasonal smoke exposure from 2003 to 2016. The bleaching effect (BRC_BL) drives the model closer to the observations over Ascension Island, but still acts to overestimate the AAE along with BRC and BRC_CNST. This could suggest either too much BrC transport to this region in the model, too weak a bleaching effect, or too much of some other type of aerosol with a higher wavelength dependence (i.e., dust). The models underestimate the AAE over the United States, especially in the BRC_BL and NOBRC model runs. This could be due to the influence of fossil fuel emissions in the model grid cell driving the AAE closer to 1.

Other observations of forest fires from Los Alamos National Laboratory (LANL) in New Mexico from 2018 show the variability of the measured fire AAEs. These include the Buzzard fire, the Ute fire, and the San Antonio fire (with average AAEs of 1.44, 1.42, and 2.38, respectively) (Romonosky et al., 2018). Due to their being outside the model run period and their intermittent nature, these fires are not directly compared to the model. However the Las Conchas model comparison from Table 2 is compared to the other fires assuming a similar regional fuel composition in the model. The model with the bleaching effect is closer to the Ute and Buzzard fires than to the Las Conchas and San Antonio fires, and overall the inclusion of BrC brings model AAEs closer to the New Mexico fire measurements.

Figure 6Differences in (a) the zonally averaged absorption coefficient (m⁻¹) and (b) the zonally averaged AAE between the BRC model run and the NOBRC model run.

Download

Figure 7Radiative effects from (a) aerosol–radiation interaction (REari) and (b) aerosol–cloud interaction (REaci) from the BRC model run. Units are W m⁻². Note different scales. Hatching indicates regions where the change over the ensemble years is significant to the 0.05 (REari) and 0.1 (REaci) levels.

Download

3.2 BrC radiative effect

One of the effects of adding BrC to the model is an increase in atmospheric absorption due to BB and BF emissions. This can be seen in Fig. 5a which shows the difference in AAOD between the BRC and NOBRC model runs. The max AAOD over southern Africa is about 0.024, which makes up approximately one-third of the total AAOD in this same region (Fig. S7a). As with Fig. 4, Fig. 5b shows a global increase in AAEs due to BrC. These effects are the strongest over the southern African BB region and in the Arctic, with the strongest AAE increases over the Arctic (Fig. 5b) correlated with weaker AAOD in Fig. 5a. Vertical cross sections of aerosol absorption coefficient and AAE changes due to BrC (Fig. 6) show the vertical extent of BrC. Zonal BrC absorption is dominated by the African and South American biomass burning regions, with visible aerosol transport to the Arctic from boreal fires (Fig. 6a). While absorption over the Antarctic is nearly zero, upper level transport of dilute BrC to the south can be inferred from the AAE changes in Fig. 6b.

Figure 7a shows REari of BrC from the BRC model run. The global annual average of REari is 0.13±0.01 W m⁻² (±1σ standard deviation), which indicates a global warming effect due to BrC. The maximum forcing (∼1.75 W m⁻²) occurs off the west coast of southern Africa; lofted BB emissions are transported over a low-lying semi-permanent stratocumulus cloud deck off the coast of Namibia and Angola. This maximum forcing can be explained by an enhancement of BrC absorption, as incoming solar radiation is not only absorbed at the top of the smoke plumes, but is also reflected off the clouds and is absorbed by the smoke plumes as outgoing radiation from the cloud top (Abel et al., 2005; Zhang et al., 2016). Significant absorption also occurs over South American BB regions (∼0.3 W m⁻²), southeast Asian BF sources (∼0.5 W m⁻²), and regions over the Arctic Ocean and Greenland (0.1–0.3 W m⁻²).

While the effects of REari correspond well to emission sources of absorbing aerosol, REaci presents a more complicated picture to diagnose (Fig. 7b). There is a weakly positive REaci global average (0.014±0.04 W m⁻²), suggesting that the BrC absorption has a negative impact on low cloud formation and lifetimes in the model, reducing the cooling effect of clouds globally. Some regions where the REaci is positive include South America (∼2 W m⁻²), the Gulf of Mexico (GM; ∼2.5 W m⁻²), Kazakhstan (∼1.5 W m⁻²), parts of southwest Alaska (∼1.5 W m⁻²), much of Australia (∼2 W m⁻²), and the Weddell Sea (WS; ∼2 W m⁻²). This positive effect is almost balanced by the negative REaci (up to −2 W m⁻²) over the ocean (e.g., northeast Pacific (NEP), Indian Ocean, and Antarctic coast (AC)) and in parts of South America and northeastern China (NEC). A negative REaci suggests an increase in cloud formation as a result of the BrC addition to the model. We note that most of this REaci is not statistically significant.

Johnson et al. (2004) used high-resolution large eddy simulation model experiments of absorbing aerosols in and above the boundary layer to show that REaci could be positive if absorbing aerosols within the boundary layer heated the air, evaporating cloud liquid water and cutting off the surface moisture source via a decoupling of the boundary layer. They also found that REaci could be negative via an enhancement of an elevated stable layer due to absorbing aerosols above the boundary layer. By creating a strong inversion and increasing the potential temperature above the cloud top, the presence of elevated absorbing aerosols allowed for more persistent low-level clouds by creating less favorable conditions for cloud top entrainment. Another study by Sakaeda et al. (2011) looked at the semi-direct effect of BB aerosols over Africa and found that REaci was typically negative over the ocean and positive over land. In their study they found that the low-level marine cloud cover is increased as a result of increasing low-level tropospheric stability, caused by the direct radiative heating above clouds and the resulting cooling of the surface. This increases the relative humidity in the lower troposphere. Over land they found positive REaci to indicate an increased static stability due to warm aerosols aloft, which reduces convection and slows the development of convective precipitation. Possible mechanisms for the significant regional REaci will be addressed in Sect. 3.4.

Interestingly, semi-direct effects are absent from the southern African region where BrC absorption dominates. Within REaci, indirect effects such as cloud droplet effective radius reduction due to the enhanced concentrations of BB aerosols have greater effects than semi-direct effects (Jiang et al., 2016; Lu et al., 2018). With this in mind, the lack of semi-direct effects over Africa could be due to the dominating indirect aerosol effects in this region from OA, which may be acting to mask the smaller semi-direct effects of BrC. Another possibility is that the absorption of BC (∼4 W m⁻²; Jiang et al., 2016), being stronger in this region than the absorption due to BrC (∼1.75 W m⁻²), is the most important factor in semi-direct effects. Both of these hypotheses may lead to fewer occurrences of significant cloud change due to BrC.

Figure 8The effect of BrC addition in the BRC model run on (a) atmospheric absorption (W m⁻²), (b) surface solar flux (W m⁻²), (c) low-level cloud fraction (%; clouds below 700 mb), (d) mid-level cloud fraction (%; clouds between 700 and 400 mb), (e) large-scale precipitation (mm day⁻¹), (f) convective precipitation (mm day⁻¹), (g) liquid water path (g m⁻²), and (h) surface air temperature (^∘K). Hatching indicates significance in ensemble year change to the 0.1 level.

Download

3.3 BrC effects on SW radiation, surface temperature, clouds, and precipitation

Looking at differences between BRC and NOBRC for specific climate variables helps clarify possible mechanisms for the previously discussed radiative effects. In Fig. 8a, the distribution of atmospheric absorption is largely synonymous with REari with a global mean of 0.29±0.01 W m⁻². The atmospheric absorption is most apparent over the southern African BB and southeast Atlantic outflow regions (∼3 W m⁻²), and over BF emission sources on the edge of the Tibetan Plateau (∼1.5 W m⁻²). Surface SW flux reduction due to BrC absorption can be seen in the southeast Atlantic outflow region and the African BB site (Fig. 8b). Also, surface SW flux change (Fig. 8b) corresponds well to the REaci, indicating an increase (decrease) in solar radiation reaching the surface with positive (negative) REaci. But what causes changes in REaci? To understand this, it is important to look at the driving forces behind these radiative effects: clouds and precipitation.

The vertically integrated low-level cloud fraction for clouds lower than 700 mb (LCF; Fig. 8c) shows an inverse correlation with REaci in some regions, indicating that a decrease in the LCF may be correlated partially with an increase in REaci, or vice versa. This is most apparent in South America, western Australia, the Middle East, and northeastern China, with smaller pockets of significant change in Kazakhstan, northern Canada, and Antarctica. The mid-level (between 700 and 400 mb) cloud fraction (MCF; Fig. 8d) is consistent with the LCF in the Arctic, parts of South America, western Australia, northeastern China, and Kazakhstan. Some of the regions where it differs from the LCF are off the northwest coast of the United States, in northern Russia, and off the southeast tip of Australia.

Globally there is little change in large-scale precipitation (PRECL; Fig. 8e) (0.0003±0.001 mm). Regionally, there are significant changes in PRECL corresponding to changes in the LCF. Examples include the Arctic Ocean (−0.15 to 0.1 mm day⁻¹), Iran (0.15 mm day⁻¹), off the east coast of the United States (0.2 mm day⁻¹), parts of South America (0.2 mm day⁻¹), near Kazakhstan (−0.15 mm day⁻¹), and off the west coast of the United States (−0.4 mm day⁻¹). Increases in PRECL in the Arctic may indicate increased snowfall due to colder temperatures at these latitudes. There is a significant decrease in PRECL off the west coast of the United States, and while this does not correspond to a LCF, it does agree with a decrease in the MCF (Fig. 8d).

Change in precipitation from convective clouds (PRECC; Fig. 8f) has a different distribution than PRECL. Significant regions of change are found in the tropics with large reductions in PRECC found along the Gulf Coast of Mexico (−0.4 mm day⁻¹), in western Australia (−0.2 mm day⁻¹), and in parts of South America (−0.1 mm day⁻¹). There are also significant changes following the North Atlantic storm track (−0.3 mm day⁻¹) and off the west coast of the United States (−0.2 mm day⁻¹). While there are cells of significant PRECC increase in parts of Bangladesh (0.2 mm day⁻¹), China (0.1 mm day⁻¹), and Africa (0.1 mm day⁻¹), the global mean of $-\mathrm{0.007}\pm \mathrm{0.0015}$ mm day⁻¹ indicates a slight decrease in PRECC with the addition of BrC.

Changes in liquid water path (LWP; Fig. 8g) are consistent with changes in LCF and PRECC. The most negative LWP changes are over land in South America (−4.5 g m⁻²), Australia (−3.5 g m⁻²), and the Gulf of Mexico (−5 g m⁻²), corresponding to reductions in PRECC (Fig. 8f). This is consistent with Sakaeda et al. (2011), who correlated large reductions in convection with increased static stability resulting from elevated BB aerosols. Significant increases in surface air temperature (TS; Fig. 8h) appear in South America (0.4 ^∘K), the Gulf of Mexico (0.4 ^∘K), and Australia (0.4 ^∘K). These regions correspond to decreasing PRECC, PRECL, and LCF, indicating surface warming in regions with decreasing cloud cover and precipitation. Significant increases in TS over Alaska (0.5 ^∘K) and northern Africa (0.25) are not correlated with cloud cover change and may be due to surface changes due to changing dynamics in the model with BrC incorporation. Similarly, TS significantly decreases over the Rocky Mountains and the Northern Great Plains of the United States (−0.6 ^∘K), but does not seem to correspond to significant changes in cloud cover or precipitation. This change is unclear but may indicate a change in surface albedo in the mountainous regions, possibly due to increased snowfall.

It is important to note that the model output does not give a complete picture of temperature or precipitation changes due to the prescription of sea surface temperatures. It would be worth revisiting this study with a fully coupled atmosphere and ocean model to allow BrC absorption to affect the air–sea interaction in the model.

Figure 9Locations of the vertical profile sites influenced by significant REaci in the BRC model run. Sites are referenced herein as (1) Weddell Sea (WS), (2) Antarctic coast (AC), (3) South America (SA), (4) Gulf of Mexico (GM), (5) northeast Pacific (NEP), and (6) northeastern China (NEC). Hatching indicates significance in the ensemble year change to the 0.1 level.

Download

3.4 Vertical changes due to BrC semi-direct effects

To understand how the BRC model BrC plays a role in aerosol–cloud interaction, we looked at vertical profiles of land (western Gulf of Mexico (GM), South America (SA), northeastern China (NEC)) and oceanic regions (Weddell Sea (WS), western Antarctic coast (AC), northeast Pacific (NEP)) with significant positive and negative BrC REaci (Fig. 9). The vertical profiles come from the averaged grids in these regions with greater than a 0.9 confidence level (Figs. 10 and 11).

Figure 10Vertical profiles (a–e) and vertical profile perturbations (f–j) for the sites in the Weddell Sea (WC) and the Antarctic coast (AC). In (a)–(e) the model run NOBRC is plotted, and the difference between BRC and NOBRC is plotted in (f)–(j). Red represents regions with a positive REari and blue represents regions with a negative REari.

Download

Figure 11Vertical profiles (a–e) and vertical profile perturbations (f–j) for the sites in South America (SA), the Gulf of Mexico (GM), and northeastern China (NEC). In (a)–(e) the model run NOBRC is plotted, and the difference between BRC and NOBRC is plotted in (f)–(j). Red represents regions with a positive REari and blue represents regions with a negative REari.

Download

In the Antarctic oceanic regions, there is little influence from BB and BF POA (i.e., sources of BrC) (Fig. 10a) and there is a positive vertical motion associated with the Ferrel cell convergence zone (Fig. 10e). The NEP site has a larger aerosol influence as well as negative vertical motion due to the descending branch of the Hadley circulation. WS low-level cloud fraction (Fig. 10b) and potential temperature (Fig. 10c) correspond with the greater annual average ice fraction in the WS (63 % versus 15 % off the AC). The heating rate in all three regions corresponds to the local cloud fraction maxima, indicating strong contribution from latent heat release (Fig. 10d). In all of the regions, the upper level concentration of BB and BF POA increases by about 10 %–15 % (Fig. 10f), indicating a change in atmospheric circulation with BrC implementation. This is backed up by positive changes in omega (ω; ∼20 % max) (Fig. 10j) and decreases of ∼4 % (increases of ∼2 %) in high-level (low-level) clouds (Fig. 10g). The NEP region shows a decrease in aerosol concentration at lower levels (5 %), possibly indicating a strengthening of the Hadley cell downwelling, which may change advection of aerosol into the region or inhibit advection due to the high-pressure enhancement. In the WS there is a decrease in low-level clouds (3 % max) and an increase in upper level clouds (1 %), and off the AC the low-level clouds are enhanced (1 %) (Fig. 10g). Cloud fraction changes at all three sites can be related to changes in potential temperature (Fig. 10h) and heating rate (Fig. 10i). Due to low concentrations of aerosol in these regions, the main driving force behind the significant REaci may be due to changes in vertical motion (Fig. 10j).

Increased cloud cover near the surface may lead to the negative REaci off the AC, which is consistent with strengthening upward velocities and can also be seen by a slight significant increase in the LCF (Fig. 8c). The factors leading to a positive REaci in the WS may be the increased high cloud cover as well as the decreased low cloud cover correlating with descending air near the surface (Fig. 10j). Why the air descends in this region is unclear, but it may be related to the larger ice fraction in the region contributing to an inversion that may decouple the lower levels from the upper level circulation changes, as observed in a WS field campaign by Andreas et al. (2000).

Figure 12500 mb ω in Pa s⁻¹ from the BRC model run: (a) the default annual average and (b) the difference with the incorporation of BrC. Hatching indicates significance in the ensemble year change to the 0.1 level.

Download

The land sites that are selected have a much stronger influence from BB and BF POA than the Antarctic Ocean sites (Fig. 11a). NEC is marked by slightly higher mid-level cloud fractions and lower temperatures than the GM or SA (Fig. 11b–c), and is in a region of descending air aloft (Fig. 11e). Both the GM and SA are located in latitudes near the descending branch of the Hadley circulation but are marked by different regional circulations: SA has an annual average upwelling (Figs. 11e, 12a) influenced by lee-side cyclogenesis (Mendes et al., 2007), while GM has annual average downwelling influenced by the Hadley circulation (Figs. 11e, 12a). Figure 11f shows changes in POA concentrations ranging from ∼4 % decrease to ∼1 % increase near the surface. Regions with positive REaci have a decreased cloud fraction (∼7 % max), while regions with negative REaci have an increased cloud fraction (∼5 % max) (Fig. 11g). In NEC the increased cloud fraction corresponds to cooler surface temperatures (Fig. 11h) and an increased heating rate (Fig. 11i; associated with latent heat release). In SA and GM, decreasing clouds at lower levels correspond to an increased heating rate (Fig. 11i), potentially due to the presence of absorbing BrC in the atmosphere overpowering the reduction in latent heating rate. Decreasing upper level clouds result in a cooling of the upper atmosphere, which may be attributed to a weakening of the high cloud atmospheric warming effect. Cloud fraction increase (decrease) corresponds to negative (positive) changes in ω in the selected regions (Fig. 11j). A strong enhancement of vertical velocity at the surface in the GM (Fig. 11j) may be attributed to the influx of BrC into the region below cloud level (Fig. 11f) and the high heating rate near the surface (Fig. 11i).

Figure 12b shows the change in 500 hPa ω with the incorporation of BrC. There is little significant change in ω over the regions selected here with the exception of AC, SA, and NEP, for which the incorporation of BrC acts to enhance both sinking (SA and NEP) and rising motion (AC). This may be due in part to an enhancement of the global rising and sinking motions due to an enhancement of convection by absorbing aerosol (Perlwitz and Miller, 2010). Of note is the large swath of sinking motion off the west coast of North America, which correlates to changes in MCF, PRECL, and PRECC in Fig. 8. The NEP cooling region corresponds to the significant increases in LCF (Fig. 8c) and LWP (Fig. 8g) within this region of increased ω.

Table 3Different RE contributions globally, in the tropics (25^∘ S to 25^∘ N), and in the Arctic (60^∘ N to 90^∘ N) from BRC, BRC_CNST, and BRC_BL model runs. These values are from a 9-year period, 2003–2011. Standard deviations from the five-ensemble means are included.

Download Print Version | Download XLSX

3.5 Comparison of BRC, BRC_CNST, and BRC_BL climate effects

There are two methods of BrC implementation that have been used in CAM5. Other than the code modification described in Sect. 2.2.2 (BRC model run), the other option has been to use offline calculated phys_prop files with a constant OA RI that is calculated based on a constant BC-to-OA ratio of 0.08 (BRC_CNST model run; Yan Feng, personal communication, 20 April 2016–22 March 2017). The assumption of a constant BC-to-OA ratio may be inaccurate in regions that have a BC-to-OA ratio higher or lower than 0.08. To see if this results in a significant change between the different model simulations, the global climate effects of both methods are examined to determine how they differ. We also show the effect of bleaching (BRC_BL) on these regions but do not discuss it in this section.

Table 3 shows the average of different components of BrC RE for the tropics (25^∘ S to 25^∘ N), the Arctic (60^∘ N to 90^∘ N), and the entire globe. REari is similar between the two methods in the tropics ($\sim \mathrm{0.14}\pm \mathrm{0.01}$ W m⁻²) as well as globally (0.13±0.01 W m⁻²). In the Arctic, REari for BRC_CNST (0.23±0.02 W m⁻²) is slightly larger than REari for BRC (0.21±0.04 W m⁻²). This difference in the Arctic may be due to a larger BC-to-OA ratio in BRC_CNST when 0.08 is assumed. In the model, the BB BC-to-OA ratio ranges from <0.07 to 0.072 (Fig. 2a) in the Arctic, which leads to weaker absorption in BRC than BRC_CNST. This may also explain the similarity with REari in the tropics where the model BB BC-to-OA ratio is similar to 0.08, with a range of 0.076 to 0.096 (Fig. 2a). While BFs undoubtedly play a role in these regions, the effect is likely less important in large global sectors than BB due to its lower emissions.

Looking at the global and Arctic regions in Table 3, the REaci associated with BRC_CNST is lower than that associated with BRC. This comes with the caveat that both have very large standard deviations, lowering the significance of this comparison.

Table 4Global averages over the period 2003–2011 for select climate variables from BRC, BRC_CNST, and BRC_BL. Standard deviations from the five-ensemble means are included.

Download Print Version | Download XLSX

Table 4 shows the global averages of selected variables associated with atmospheric SW transmittance, clouds, precipitation, and surface temperature. Like REari, global atmospheric absorption is very similar between BRC (0.29±0.01 W m⁻²) and BRC_CNST (0.28±0.03 W m⁻²). The largest difference between the two model runs seems to be related to differences in aerosol–cloud interaction. In Table 4, LWP change is higher in the BRC_CNST (0.06±0.09 g m⁻²) than in the BRC (0.004±0.1 g m⁻²) model run, which seems to correspond to higher LCF (0.06±0.03 %) and MCF ($-\mathrm{0.006}\pm \mathrm{0.02}$ %) changes in BRC_CNST. Consistent with increased cloud cover and LWP is a lower surface flux in BRC_CNST ($-\mathrm{0.18}\pm \mathrm{0.06}$ W m⁻²). While cloud cover and LWP changes are higher in BRC_CNST, both PRECL and PRECC changes seem to differ less between the two model runs. This emphasizes the nonlinearity between cloud cover and the microphysical conditions necessary to generate precipitation.

Lastly, the global surface temperature change between the two models is similar and small (BRC: $-\mathrm{0.008}\pm \mathrm{0.01}$ ^∘K; BRC_CNST: $-\mathrm{0.003}\pm \mathrm{0.01}$ ^∘K).

Figure 13Same as Fig. 6 but for the BRC_BL model run. Hatching indicates regions where the change over the ensemble years is significant to the 0.05 (REari) and 0.1 (REaci) levels.

Download

3.6 Effects of BrC bleaching on REari and REaci

Figure 13a shows REari and REaci due to BrC with a bleaching parameterization from the BRC_BL model run. BrC bleaching significantly reduces REari from 0.13±0.01 W m⁻² (Fig. 7a) to 0.06±0.008 W m⁻² (Fig. 13a). What is readily apparent when comparing Fig. 13a to Fig. 7a is that the extent of BrC absorption is greatly constrained when BrC bleaching is incorporated in the model, especially in the tropics (Table 3). This is related to greater insolation in the tropics, leading to more OH formation and a more rapid BrC bleaching process (Eq. 4). Also, the direct effects associated with BF BrC are greatly reduced with bleaching such that significant effects over most of Asia do not occur.

BRC_BL REaci (Fig. 13b) shows some similar characteristics to BRC REaci (Fig. 7b). Some similar regions with negative REaci include the northeast Pacific, the south Australian coast, western South America, and near the southern tip of South America. Similar positive REaci regions include Australia, the Weddell Sea, parts of South America, the Gulf of Mexico, and off the coast of southern Africa. In most cases REaci due to aged BrC is smaller in magnitude than REaci without BrC aging. A notable exception to this is a strong warming signal in Fig. 13b over eastern China (∼3 W m⁻²). This correlates with a reduction in LWP over this region (−5 g m⁻²; Fig. S10g), which may also be related to enhanced atmospheric downwelling in this region (Fig. S11b). This indicates some of the nonlinear effects of BrC absorption reduction in the model simulations.

Table 5Comparison of different model runs for the year 2003 with varying OA density. The values are BB and BF OA density from Feng et al. (2013), and the default CAM OA properties (Hess et al., 1998). REari and REaci are annual averages calculated for the BRC_CNST model run.

Download Print Version | Download XLSX

3.7 Effects of changing OA density on REari and REaci

Another experiment to determine model sensitivity is a comparison of BRC_CNST model runs with altered input OA density for BB and BFs (Table 5). Species density is accessed in the model from the offline phys_prop files and is used to calculate the modal refractive index (Ghan and Zaveri, 2007) as well as calculate the number concentration of aerosols activated as cloud condensation nuclei (CCN) (Abdul-Razzak and Ghan, 2000). This is one of the few aerosol species physical properties used directly in the model code, as most of the aerosol physical properties are calculated for each aerosol mode. The inspiration for this comparison lies in the different values used by other modeling studies for OA density, ranging from 1 to ∼1.5 g cm⁻³. This comparison only looks at the densities used in the default CAM phys_prop file (from Hess et al., 1998) and the density of BB and BF OA used in Feng et al. (2013). Each model run is for the year 2003 and the REari and REaci are both 2003 annual averages.

By decreasing OA density, there is an increase in BrC REari. This is likely due to the use of inverse aerosol density (specific volume) when calculating the radiative effects of the internally mixed aerosol, with a smaller density indicating a larger specific volume and larger BrC absorption attributed to the mixed aerosol. Furthermore, given the same mass concentration and size distribution, an increase in density would reduce the number of particles per volume of air. This would reduce the aerosol optical depth and the RE. In Table 5, the sign on REaci changes dramatically depending on the density used, and is attributed to the alteration of the size of the modal aerosol and its droplet activation efficiency, as well as the number of particles at a given size distribution. The driving factor behind this difference is likely the number of particles per volume of air, which would result in a reduced cloud optical depth with fewer particles (i.e., higher density). A higher density aerosol will also lead to a smaller modal aerosol size, requiring a higher critical supersaturation, which results in lower activation of CCN and less cooling attributed to clouds. What this brief analysis indicates is that there is a need to better constrain OA physical properties if there are to be direct comparisons between simulations of OA radiative effects.