2.1 The CESM model

We developed the brown carbon simulation based on the Community Earth System Model version 1.2.2 and its atmospheric component, the Community Atmosphere Model version 5 (CAM5) (Neale et al., 2010). The CAM5 model has a comprehensive mechanism for aerosols and aerosol–cloud interaction (Liu et al., 2012; Ghan et al., 2012; Gettelman et al., 2010). The CAM5 radiation scheme is the Rapid Radiative Transfer Method for Global Climate Models (GCMs) (RRTMG), which provides an accurate method for radiative flux calculation (Iacono et al., 2008; Mlawer et al., 1997). The atmospheric-chemistry module, turbulence scheme, convection scheme and cloud physics are coupled in the model (Zhang and McFarlane, 1995; Bretherton and Park, 2009; Park and Bretherton, 2009; Richter and Rasch, 2008; Morrison and Gettelman, 2008). We used a three-mode version of the modal aerosol model (MAM3) for aerosol modeling in CAM5. The three modes are the Aitken, accumulation and coarse modes (Neale et al., 2010). In our BrC simulations, we used CAM5 with a spatial resolution of $\mathrm{1.9}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$. The wet scavenging scheme of aerosols in CAM5 includes below-cloud scavenging and in-cloud scavenging, which was found to have a high bias (Wang et al., 2011; Liu et al., 2012; Yu et al., 2019). For the simulations used to compare with field observations, we nudged the CAM5 meteorological field (temperature, humidity, wind, surface pressure and heat) to the same meteorological year, month and day as the observations using Goddard Earth Observing System (GEOS)-5.2.0 meteorological data products (Rienecker et al., 2008) every 6 h in order to evaluate the model simulations with BrC observations (Ma et al., 2013; Chipperfield, 2006). We also conducted 5-year free-running model simulations using the climatology of 2010 to analyze the climate response to BrC and BC heating. The spin-up time is 3 months in the nudged CAM5 simulations and 1 year in the free-running simulations.

2.2 Emissions

We derived global BrC emissions from biomass burning, biofuel and secondary formation. As with Brown et al. (2018), we used the parameterization by Saleh et al. (2014) for biomass and biofuel burning, in which emitted BrC absorption is a function of the BC∕OA (organic aerosol) emission ratio. We included secondary aerosols from the oxidation of aromatics as the major source of secondary BrC (Hecobian et al., 2010; Sareen et al., 2013; Lin et al., 2015). Secondary BrC produced from aromatic oxidation absorbs more solar radiation in high-NO_x conditions (Laskin et al., 2015; Lin et al., 2015; Liu et al., 2012; Nakayama et al., 2010, 2013; Yu et al., 2014; Zhong and Jang, 2011). We did not consider the NO_x dependence of secondary BrC in this study. More details will be described in Sect. 3.

The biomass burning emissions we used are from the Global Fire Emission Database version 4 (GFED4) including small fires (GFED4s) (Giglio et al., 2013; Randerson et al., 2012). It contains global burned area distribution and biomass burning emission factors of related aerosols and gas species for different fire types, with a spatial resolution of $\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$. In CAM5, we aggregated it to a spatial resolution to $\mathrm{1.9}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$ and used GFED daily emission and diurnal cycle factors. The different emission factors for tropical forests, temperate forests, boreal forests, savannas, agriculture waste and peat burning are based on Akagi et al. (2011). Fire emissions can reach high altitudes (e.g., Neale et al., 2010). We used an observation-constrained global fire plume rise dataset in which MODIS fire hotspot and fire radiance power data were used in a 1-D fire plume rise model, and the resulting fire plume distribution is in good agreement with the Multi-angle Imaging SpectroRadiometer (MISR) observations (Ke, 2020). Biomass burning emissions have high uncertainties caused by burned area, emission factors, fuel loads and combustion completeness factors (Akagi et al., 2011; Giglio et al., 2013), and the complex interactions between fire, terrestrial ecosystem and climate systems amplify these uncertainties (Zou et al., 2019).

Anthropogenic emissions are from the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) emission dataset (Lamarque et al., 2010), and BC and OC emissions are updated based on the emission inventory of 2000 (Bond et al., 2007; Junker and Liousse, 2008). We increased anthropogenic emissions in China by 50 % according to Zhang et al. (2009). For the Arctic region (>66^∘ N), we used the Evaluating the Climate and Air Quality Impacts of Short-lived Pollutants (ECLIPSE) emission dataset, which includes an improvement for the Arctic BC emissions (Stohl et al., 2013; Klimont et al., 2015).

For the optical properties of BC, we used 10 and 8.1 m² g⁻¹ for 345–442 nm and 442–625 nm, respectively, as the MAE values of BC (Knox et al., 2009; Bond and Bergstrom, 2006). The MAE values are lower than the estimation by Bond et al. (2013) (11 m² g⁻¹) and Jacobson (2016) (16 m² g⁻¹ including high-relative-humidity conditions) and are higher than the estimation by Schulz et al. (2006) (7.9±1.9 m² g⁻¹). MAM3 assumes that primary carbon is internally mixed with secondary aerosols in the accumulation mode.

Table 1BrC emission sources (Tg C yr⁻¹) of this and previous studies.

Download Print Version | Download XLSX

Figure 1Spatial distributions of global emissions of BrC from biomass burning (a), anthropogenic biofuel combustion (b) and secondary formation (c) in 2010. The unit is µg C m⁻² s⁻¹. The total emission is 3.6, 3.1 and 4.1 Tg C yr⁻¹ for biomass burning, biofuel and secondary formation, respectively.

3.1 BrC optical property and photobleaching

BrC absorption depends on its Mass Absorption Efficiency (MAE), which is the ratio of light absorption in the wavelength λ to BrC mass concentration (m² g⁻¹).

where A(λ) represents the absorption of BrC at the wavelength of λ (m⁻¹) and C_BrC is the mass concentration of BrC (g m⁻³).

Similar to Jo et al. (2016), we used a constant MAE value for primary BrC, 1.0 m² g⁻¹ at 550 nm (McMeeking, 2008), and we used an MAE value of 0.19 m² g⁻¹ at 550 nm for secondary BrC based on the work by Nakayama et al. (2010). There are other MAE estimates in experiments such as 3.6–4.1 m² g⁻¹ (Alexander et al., 2008) and 0.58–0.64 m² g⁻¹ (Hecobian et al., 2010) and model-specified values such as 0.35 m² g⁻¹ by Wang et al. (2018). The MAE value may also change in different seasons (Cheng et al., 2011). At present, there are not enough observations to specify variable MAE values in the model. The MAE value at the other wavelengths was calculated using the following function (Bond and Bergstrom, 2006):

where AAE is the absorption Ångström exponent and λ₀ is 550 and 365 nm for primary and secondary BrC, respectively. We used AAE = 5.0 for λ<2 µm (Jo et al., 2016; Kirchstetter and Thatcher, 2012). BrC AAE varies depending on its source and the wavelength used (Kirchstetter and Thatcher, 2012; Liu et al., 2014). Jo et al. (2016) found BrC∕BC ratio decreases when the BrC AAE increases from 5 to 6.19; Saleh et al. (2014) also found that the BC∕OA ratio is negatively related to BrC AAE and positively related to BrC absorption. Therefore, the variation of BrC AAE leads to additional uncertainty of the BrC simulation. The imaginary part of the refractive index for BrC is estimated using the following equation (Liu et al., 2013):

where ρ is particle density (g m⁻³), A(λ) is the light absorption at wavelength λ and c is mass concentration.

Figure 2Monthly mean global BrC emission rates (Tg C yr⁻¹) in 2010. Green, blue and red bars represent the emissions from biomass burning, biofuel combustion and secondary BrC formation, respectively.

Download

The estimated k_BrC value is 0.045 at 550 nm for primary BrC and 0.043 at 365 nm for secondary BrC, respectively.

There is observational evidence that both primary and secondary BrC are affected by photochemical aging (or bleaching), which reduces BrC light absorption when exposed to light (Forrister et al., 2015; Sareen et al., 2013; Lee et al., 2014; Zhong and Jang, 2011; Wong et al., 2017, 2019). Previous modeling of the BrC photobleaching effect by Wang et al. (2018) and Brown et al. (2018) applied a 1 d e-folding time for BrC before reaching a threshold of 25 % of the original BrC absorption. Our approach to BrC photobleaching considers different bleaching effects depending on BrC source. We specify a decay half-life of 12 h when light is present for primary biomass and biofuel BrC in the daytime until 6 % is left and no further photobleaching occurs (Forrister et al., 2015) due to stable high-molecular-weight chromophores (Di Lorenzo and Young, 2015; Di Lorenzo et al., 2017; Wong et al., 2017, 2019). Different components of secondary organic aerosol (SOA) have different photobleaching lifetimes. Aromatic SOA has a half-life of 12–24 h (Liu et al., 2016; Lee et al., 2014; Zhong and Jang, 2011), and limonene SOA has a half-life of <0.5 h (Lee et al., 2014). Methylglyoxal SOA has a half-life of 90 min (Zhao et al., 2015; Wong et al., 2017). Therefore, the half-life for secondary aromatic BrC is specified at 12 h in daytime until it is completely removed (Liu et al., 2016). The other secondary BrCs that have shorter lifetimes contribute little to global radiative forcing and are not included in the model.

The analysis of aircraft BrC observations by Zhang et al. (2017) showed that BrC transported by deep convection plays a significant role in the radiative heating of the upper troposphere and that BrC warming is about one third of BC warming at the tropopause. The standard model simulations show a large low bias of BrC in the upper troposphere compared to the observations by Zhang et al. (2017). In addition, in-cloud heterogeneous BrC production is another possible reason for the high level of BrC in the upper troposphere. A fraction of biomass burning BrC from heterogeneous oxidation by ozone will stay free from photobleaching, and BrC from heterogeneous oxidations by OH has a long lifetime of days (Browne et al., 2019). Therefore, we conducted sensitivity simulations of BrC to investigate the effects of photobleaching and wet scavenging during deep convection.

One important finding by Zhang et al. (2017) is that wet scavenging of BrC during convection differs from BC and OC. Therefore, BrC is simulated using a different tracer from OC in this work unlike Brown et al. (2018). The BrC property of interest is absorption, and we assume that the tracer's optical property is light absorption only (no scattering). Consequently, there is no double counting of OC scattering. However, it should be noted that BrC is a class of organic aerosols that both scatter and absorb light. We analyzed in this study the effect of BrC light absorption. The model simulation of OC mass and scattering was not affected by the simulation of a BrC tracer that only absorbs light. In the following analysis, the DRE from BrC is for light absorption only such that it represents the DRE of the OC absorption and can be compared to the DRE of the BC absorption.

Figure 3Flight track of the five HIPPO missions. Colored lines in red, blue, black, brown and green represent flight track of HIPPO-1 to HIPPO-5, respectively. Flights over continental North America east of 140^∘ W were not included in this study.

Figure 4Comparison of HIPPO BC (ng C kg⁻¹) measurements (a), simulated BC data from the modified CAM5 model (b) and simulated BC data from the default CAM5 model (c) during HIPPO-1 to HIPPO-5. The five rows from top to bottom are HIPPO-1 (January 2009), HIPPO-2 (November 2009), HIPPO-3 (March–April 2010), HIPPO-4 (June–July 2011) and HIPPO-5 (August–September 2011), respectively. Measurement data along the flight tracks of Fig. 3 are 1 min averages. Model data are selected corresponding to the location and time of aircraft measurements.

Download

3.2 BrC emissions

We assumed that BrC is emitted in the accumulation mode (Liu et al., 2013). Sources of the more stable forms of BrC include primary emissions of biomass burning and biofuel as well as secondary formation from aromatic oxidation. Similar to previous model approaches (Wang et al., 2018; Brown et al., 2018), biomass burning BrC emissions were parameterized as a function of the BC-to-OA ratio of the emissions (Saleh et al., 2014).

where k_OA,550 is OA absorptivity at 550 nm and E_BC and E_OA are BC and OA emission rates (g m⁻² s⁻¹), respectively. We computed k_OA,550 in order to calculate BrC emissions. In the model, the absorption of the OC tracer was specified to be 0. All OC absorption was due to the BrC tracer. We scaled BrC emissions based on k_OA,550; MAE and OA emissions used the following equation by Liu et al. (2013):

where ρ is particle density (g m⁻³) and E_BrC is the BrC emission rate (g m⁻² s⁻¹).

Using the GFED emissions inventory, we estimated an annual average global BrC source from biomass burning of 3.6 Tg C yr⁻¹, ∼23 % of OC emissions (15.9 Tg C yr⁻¹) and about twice as large as BC emissions (1.9 Tg C yr⁻¹). The variability of the BrC emission rate among biomes therefore depends on the BC-to-OA emission ratios in the GFED emission inventory. Using the same equations and an average E_BC∕E_OA ratio of 0.41 (Junker and Liousse, 2008), we estimated an E_BrC∕E_OA ratio of 0.38 and an annual global BrC biofuel source of 3.1 Tg C yr⁻¹ on the basis of the biofuel emission inventory by Fernandes et al. (2007). The estimates of primary BrC emissions are comparable to previous studies (Table 1). BrC emissions from fossil fuel combustion are not yet characterized well enough to be included in a global model (Saleh et al., 2014; Xie et al., 2017).

Figure 5Comparison of observed and simulated BC vertical profiles during HIPPO missions for the latitude bins of 90–60^∘ S, 60–20^∘ S, 20^∘ S–20^∘ N, 20–60^∘ N and 60–90^∘ N. Black lines and shaded areas show the means and standard deviations of the observations binned in 1 km intervals, respectively. The colored vertical lines and horizontal bars show the means and standard deviations of the default (blue) and modified CAM5 results (red), respectively.

Download

Figure 6Comparison of monthly mean AOD and AAOD data in 550 nm for fire-dominated months and regions (Fig. S1a) of model simulations with the observations for (a) 2010 AERONET AOD, (b) 2010 MODIS AOD and (c) 2005–2014 AERONET AAOD. For (a) and (b), model data correspond to the same time and location of the observations. The data points in (a) and (b) are color-coded as a function of latitude. The solid line denotes a PC regression line, and the dashed line denotes the 1:1 reference line. For (c), monthly mean values of model data corresponding to AERONET AAOD observations are shown. The solid lines denote PC regression lines for model results with and without BrC absorption, and the corresponding regression slope (k) values are shown. The dashed line denotes the 1:1 reference line.

Download

The major fraction of secondary BrC that affects atmospheric heating is formed during the oxidation of aromatics (Jacobson, 1999; Nakayama et al., 2013; Zhong et al., 2012). As in previous studies (Jo et al., 2016; Wang et al., 2014), we assumed that secondary BrC is from aged aromatic SOA. In the CAM5–MAM3 aerosol mechanism, the SOA mass yield for aromatic oxidation is 15 % (Neale et al., 2010; Odum et al., 1997). We estimated a global secondary BrC source of 4.1 Tg C yr⁻¹ in agreement with previous studies.

BrC emissions used in this study and the comparison to previous studies are summarized in Table 1, and the emission distribution is shown in Fig. 1. Biofuel emissions are high in China and India, and secondary BrC sources are also large in Europe and North America. Figure 2 shows the annual cycle of BrC emissions in 2010. Biofuel and secondary BrC sources have little seasonal variation in the model, while biomass burning has significant seasonal variation. The BrC source is the highest in August at 18.9 Tg C yr⁻¹. Biomass burning emission accounts for more than 60 % of the BrC emissions in August.

Figure 7Flight tracks of the SEAC4RS (red) and DC3 (blue) field experiments.

4.1 Black carbon measurements from HIPPO

HIAPER (High-Performance Instrumented Airborne Platform for Environmental Research) Pole-to-Pole Observations (HIPPO) measured atmospheric composition approximately from the Arctic to the Antarctic over the Pacific Ocean (Wofsy, 2011). HIPPO executed five missions from January 2009 to September 2011. The flight path of each HIPPO mission is shown in Fig. 3. Measurements over continental North America east of 140^∘ W were not included in our model evaluation. BC measurements for particles with a size range of 90–600 nm were made from a single-particle soot photometer (SP2), and we increased measured BC data by a factor of 1.1 to account for larger-sized BC in the model evaluation (Schwarz et al., 2010, 2013). We make use of HIPPO BC measurements to constrain convective transport and wet scavenging.

Wet scavenging is uncertain in the 3-D global modeling of BC (Schwarz et al., 2010; Liu et al., 2011). Wang et al. (2013) tested the sensitivities of different physical mechanisms and found a high sensitivity of BC simulations to convective transport and wet removal. A comparison of CAM5 BC simulations with HIPPO observations in Fig. 4 shows large overestimates of BC in the tropics and the upper troposphere. Since the emissions of BC are from the surface, the model high biases in these regions suggest insufficient wet scavenging during convection. Wang et al. (2014) updated the model wet scavenging by scavenging hydrophobic aerosols in convective updrafts and scavenging hydrophilic aerosols from cold clouds. In all simulations of this study, we increased interstitial BC scavenging by a factor of 5 to increase wet scavenging and reduced stratiform liquid-containing clouds based on model evaluations using HIPPO observations. The high biases above 300 hPa at mid and high latitudes persisted particularly for HIPPO-1 in January 2009 and HIPPO-2 in November 2009. In winter, the simulated high BC concentrations were above the tropopause level at mid and high latitudes, indicating that convective transport reached too high an altitude. We therefore limited deep-convection mass transport to an altitude of 50 hPa below the model estimated tropopause, after which the high biases at mid and high latitudes above 200 hPa were corrected. During HIPPO-3 in March–April 2010, model-simulated free-tropospheric BC at northern mid and high latitudes was much lower than the observations, suggesting excessive scavenging in the model. We reduced cloud-born BC scavenging to 10 % when cloud ice is present and to 50 % for the other conditions, which improved the model simulations of free-tropospheric BC at mid and high latitudes in HIPPO-3 (and HIPPO-2). The modification slightly worsened the model high bias at northern mid and high latitudes in the summer for HIPPO-5 in August–September 2011 and to a lesser extent for HIPPO-4 in June–July 2011. Figure 5 shows the comparison between BC vertical profiles during all HIPPO campaigns with CAM5 simulations for five latitude bins (90–60^∘ S, 60–20^∘ S, 20^∘ S–20^∘ N, 20–60^∘ N and 60–90^∘ N), respectively. The modified CAM5 simulations agreed better with the observations in all regions, but they still overestimated BC in the middle and upper troposphere over the tropics, which may lead to a low bias in the model-simulated BrC∕BC heating ratio in the tropics (to be discussed in Sect. 5.3).

Table 2BrC sensitivity simulations.

Download Print Version | Download XLSX

4.2 Aerosol optical depth and absorption aerosol optical depth over fire-emission-dominated regions

Direct assessments of BrC sources using observations are difficult because of limited observations. We can, however, evaluate model simulations of fire aerosols with aerosol optical depth (AOD) and absorption aerosol optical depth (AAOD) measurements. For this purpose, we chose the months and regions in model simulations for which >50 % of monthly mean AOD data are from fire emissions for 2010. The distribution of model-simulated mean AOD for data points selected in this manner are shown in Fig. S1a in the Supplement. For comparison purposes, the ground-based AOD measurements were obtained from the Aerosol Robotic Network (AERONET) version 3 level 2.0 dataset (Holben et al., 1998). To compare with model-simulated AOD data at 550 nm, the AOD measurements at 500 and 675 nm were used to compute the Ångström exponent (Ångström, 1964) and calculate the corresponding AOD values at 550 nm (Kumar et al., 2013). Figure 6a compares the monthly mean fire-dominated 550 nm AOD observations in 2010 in AERONET with corresponding monthly mean model results for selected months and regions. The correlation coefficient r is high at 0.88. We performed principal-component (PC) regression analysis of observed and simulated data. The low regression slope (0.56) indicates that the observed AOD data were underestimated, implying a low bias in fire emissions.

Figure 8Comparison between observed and simulated vertical profiles of BrC absorption at 365 nm, the ratio between BrC absorption at 365 nm and BC (BrC∕BC), and concentrations of BC and CO for the DC3 (left column) and SEAC4RS (right column) missions. Black lines and shaded areas show the means and standard deviations of the observations binned in 1 km intervals, respectively. The colored vertical lines and horizontal bars show the means and standard deviations of corresponding model results, respectively. Model sensitivity simulations of BrC are listed in Table 2. The difference among simulated BC and CO vertical profiles is negligible, and the ICB simulation results are shown.

Download

We also compared AOD with the measurements from the Moderate Resolution Imaging Spectroradiometer (MODIS) on Terra (EOS AM-1) satellite for the months and regions in model simulations for which >50 % of monthly mean AOD data are from fire emissions for 2010. We used collection 6 of the MODIS level 3 Deep Blue–Dark Target merged product with a resolution of $\mathrm{1}{}^{\circ }\times \mathrm{1}{}^{\circ }$ (Platnick et al., 2017). Figure 6b shows the comparison. Both the correlation coefficient r (0.67) and the PC regression slope k (0.49) are lower than the comparison with AERONET observations. One reason is that the high AOD data in the outflow region of the tropical Atlantic from fire emissions over Africa were significantly underestimated (Fig. S1b); similar low biases were also found in the outflow region of fire emissions in South America. Additionally, CAM5 underestimates AOD at high latitudes (Liu et al., 2012). The general low bias of fire aerosol emissions was also found by Ward et al. (2012). For these data, the effect of BrC absorption on AOD is small; we estimate that BrC absorption contributes 0.37 % of the total AOD.

In addition, we compared the model simulations to the AAOD data from the AERONET version 3 level 2.0 inversion dataset (Holben et al., 2006). Since the AAOD estimation is highly uncertain in the low AOD conditions (Dubovik et al., 2000), we used only AAOD measurements for AOD at 440 nm≥0.4 (Holben et al., 2006). Monthly mean AAOD data were computed for AERONET sites with more than 10 d of daily averaged observed AOD at 440 nm>0.4 in a month. Because of the model underestimation, the corresponding model threshold of AOD at 440 nm is 0.315 based on the PC regression between AERONET observations and model simulation results (Fig. 6a). Daily model results with AOD at 440 nm>0.315 were used to compute simulated monthly means for the grid cells corresponding to the AERONET sites. To show the performance of the model simulation of aerosol absorption, here we compare the observed and simulated AAOD values at 550 nm, which is near the peak wavelength of solar intensity. Because of the strong wavelength dependence of BrC absorption, the enhancement of AAOD by BrC absorption at wavelengths lower than 550 nm is more significant. Figure 6c compares the monthly mean 2005–2014 AERONET AAOD data over fire-dominated regions and months with the corresponding monthly mean model results. The observations showed significant interannual variability, which was not included in the model results for the climatological year 2010. With BrC absorption, the simulated higher AAOD data are in better agreement with AERONET observations with a PC regression slope of 0.59 compared to a slope of 0.43 for the simulation without BrC absorption. For these observations, the model underestimated the AERONET AAOD observations by 39 % without BrC absorption. Including BrC absorption reduced the low bias to 17 %, which is well within the large variability of the observations. Globally, the AAOD absorption at 550 nm is higher by 8.5 % on average when BrC is considered in the model simulations.

Figure 9Simulated zonal averaged annual mean BrC absorption at 365 nm (Mm⁻¹) for (a) all sources, (b) biomass burning emissions, (c) secondary BrC formation and (d) biofuel BrC emissions. The unit is Mm⁻¹. The colored bar is in log scale.

Download

5.1 Model simulations of BrC for the DC3 and SEAC4RS missions

We evaluated BrC model simulations using the measured BrC absorption data from the airborne measurements of Studies of Emissions, Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (Toon et al., 2016) and Deep Convective Clouds and Chemistry Project field experiments (Barth et al., 2015). The SEAC4RS campaign was conducted from 6 August to 23 September 2013 over the central and southeast US, and the DC3 campaign was conducted from 18 May to 22 June 2012 over a similar region. Flight tracks for these experiments are shown in Fig. 7. Fresh fire plume data, diagnosed by plumes with a coefficient of determination between CO and CH₃CN >0.5 during the period of enhanced CO, were not included in the model evaluation as in previous studies (De Gouw et al., 2004; Liu et al., 2014).

We described in Sect. 3.1 the rationale for sensitivity simulations to evaluate the effects of BrC photobleaching and convective wet scavenging. The model sensitivity simulations are listed in Table 2. In the NCNB (base) model, neither effect was included. In the NCB model, the photobleaching effect is included. In the ICNB model, the wet scavenging efficiency of convective transported BrC was decreased from 75 % simulated in the base model to 30 % such that ∼70 % of BrC was transported through convection to the free troposphere as suggested by Zhang et al. (2017). In the ICB model, both photobleaching and reduced convective scavenging effects were included. The ICBB model is similar to ICB model, but photobleaching of all BrC was included; in the other models including the photobleaching effect only non-convectively transported BrC was affected (Zhang et al., 2017).

Figure 10Annual averaged global distributions of (a) BC DRE, (b) BrC DRE and (c) the ratio of BrC∕BC DRE for 2010. The unit is W m⁻². The global averaged DRE is shown in the upper right corner. In (c), BrC∕BC DRE ratios larger than 1.0 are specified by a differently colored bar.

Figure 8 shows the observed vertical profiles of BrC absorption, the BrC-to-BC absorption ratio (BrC∕BC ratio), and concentrations of BC and CO during the DC3 and SEAC4RS experiments in comparison to the corresponding model simulation results. The difference between BC and CO vertical profiles is negligible among the sensitivity simulations. Simulated mean BC concentrations are within the uncertainties of the measurements. The underestimation at 2–5 km during SEAC4RS likely reflects underestimated fire emissions since the coefficient of determination (R²) is 0.6 for HCN and BC at 2–5 km and 0.5 for HCN and BrC, reflecting the effects of biomass burning emissions on BC and BrC. The higher CO concentrations in the model than the observations, particularly near the surface, suggest that the model overestimates surface CO emissions.

Table 2 lists all sensitivity simulations. For BrC and BrC∕BC simulations, Fig. 8 shows that the NCNB model clearly overestimated BrC compared to the observations at 0–8 km (the overestimate is not as apparent in the BrC∕BC comparison because it is a logarithmic scale). The overestimation reflected the importance of photobleaching (Forrister et al., 2015; Sareen et al., 2013; Lee et al., 2014; Wang et al., 2016; Wong et al., 2017, 2019; Zhong and Jang, 2011). The overestimation in the lower troposphere in NCNB led to a reasonable simulation of BrC in the upper troposphere, although the underestimation at 12 km was obviously relative to the ICB simulation during the DC3 experiment. Similarly, considering enhanced convective transport, but not photobleaching, the ICNB simulation clearly overestimated BrC absorption relative to the observations. Including photobleaching, but not enhanced convective transport of BrC, the NCB simulation clearly underestimated BrC and the BrC∕BC ratio in comparison to the observations. We also included a simulation of ICBB, in which enhanced convective transport of BrC was included with photobleaching. Compared to the observations, upper-tropospheric BrC and the BrC∕BC ratio in the ICBB simulation were clearly underestimated. At 12 km, the observed BrC∕BC ratio is ∼10 and ∼20 times higher than BrC∕BC near the surface during DC-3 and SEAC4RS, respectively. This increase in the BrC∕BC ratio in the upper troposphere was captured by the ICB simulation. On the basis of our current understanding of BrC processes (Forrister et al., 2015; Sareen et al., 2013; Lee et al., 2014; Wang et al., 2016; Wong et al., 2017, 2019; Zhang et al., 2017; Zhong and Jang, 2011) and the model evaluation with the observations, we chose the ICB simulation to investigate the effects of global BrC radiative forcing.

Figure 11Annual averaged global distributions of BrC DRE for all sensitivity simulations (Table 2). The unit is W m⁻². The global averaged DRE is shown in the upper right corner.

Figure 12Same as Fig. 10 but for seasonal global DRE distributions of BrC for (a) December–January–February (DJF), (b) March–April–May (MAM), (c) June–July–August (JJA) and (d)  September–October–November (SON) in the ICB simulation.

During the SEAC4RS experiment, Fig. 8 shows that the models overestimated both BC and BrC in the upper troposphere (except ICBB and NCB, which underestimated BrC in both experiments), that all model simulations except ICNB underestimated the BrC∕BC ratio in the upper troposphere, and that all model simulations except NCB were overestimated the BrC∕BC ratio in the middle and lower troposphere. The simulation bias is mostly due to biases in the BC simulation in that BC was overestimated in the upper troposphere and underestimated in the middle and lower troposphere (Fig. 8).

In the ICB simulation, wet scavenging of BrC was reduced relative to BC in order to simulate the observed BrC∕BC ratios in DC3 and SEAC4RS. The mechanisms are not yet clear due to a lack of laboratory and field observations. Hydrophobic OC, such as humic-like substances (HULISs), is more likely to have high light absorption compared to hydrophilic OC (Hoffer et al., 2006). BrC with a high molecular weight dominates the aged biomass burning plume (Wong et al., 2017, 2019). Since higher molecular weight compounds have lower hygroscopicity (Dinar et al., 2007) and it is harder to activate hydrophobic OC in clouds, less BrC is removed in deep convection. Another possible mechanism is production of BrC through in-cloud heterogeneous processing of fire plumes (Zhang et al., 2017). However, there is no observation data to implement such a mechanism in a model.

Figure 13Global zonal mean distributions of (a) BC mass concentrations (µg std m⁻³; mass concentration under standard conditions) and (b) BrC absorption at 365 nm (Mm⁻¹) for 2010.

Download

5.2 Simulated global zonal mean distribution of BrC

We performed diagnostic model simulations to investigate the contributions of BrC absorption from biomass burning emissions, biofuel emissions and secondary formation, respectively. Figure 9 shows the results. Secondary BrC production has a relatively small contribution because the photobleaching of secondary BrC is 100 %, while a small fraction of BrC is left after photobleaching of biomass burning and biofuel BrC (Forrister et al., 2015). Both biofuel and secondary production are largest at northern mid latitudes since they are due to anthropogenic emissions.

Biomass burning BrC shows drastically different distributions from biofuel BrC. The latitudinal maximum is in the tropics and subtropics, with a secondary peak at 60^∘ N due to fires over Canada and Siberia. The vertical extent of biomass burning BrC is much higher than biofuel BrC due to fire plume rise (Ke, 2020) and the higher vertical extent of tropical convection than at mid latitudes. While the effect of biofuel BrC is primarily in the lower troposphere, the radiative forcing of biomass burning BrC is much more substantial in the free troposphere and therefore more strongly affects the atmosphere since solar heating of the atmosphere is generally weak.

Figure 14Global averaged vertical profile of the BrC-to-BC heating rate ratio for 2010. The black and red lines are the average profiles for regions without and with deep-convection events, respectively. Standard deviations are indicated by the horizontal bars.

Download

Figure 15Global zonal mean distributions of the heating rate of (a) BrC and (b) BC for 5 years of present-day simulations. The dashed line denotes the tropopause.

Download

5.3 Global directive radiative effect of BrC

Aerosol DRE represents the instantaneous radiative effect of aerosols, which is sometimes confused with DRF (Heald et al., 2014; Ghan, 2013). We applied the Rapid Radiative Transfer Method for GCMs to BrC and BC radiative forcing. We parameterized the imaginary part of the BrC refractive index as an external input of RRTMG. As discussed in Sect. 3.2, the imaginary refractive index is specified at 0.045 at 550 nm and 0.043 at 365 nm for primary and secondary BrC, respectively. The RRTMG shortwave and longwave wavelength boundaries are listed in Tables S1a and S1b in the Supplement (Neale et al., 2010; Iacono et al., 2008; Mlawer et al., 1997). We calculated the imaginary refractive index at a different wavelength by introducing wavelength dependence w (Saleh et al., 2014).

where k_BrC,λ denotes the imaginary refractive index of BrC and w is the wavelength-dependent AAE value. The calculation and parameterization of MAE and the imaginary refraction index of BrC were discussed in Sect. 3.1.

In our estimation of BrC DRE, we only considered the absorption of BrC, and the effect of scattering is not considered. We computed the clear-sky net solar flux at the top of atmosphere in two simulations, one with BrC-tracer-absorbing light and the other without. The difference between the two simulations is BrC DRE. The same method was used to calculate BC DRE.

The ICB model calculated global DRE distributions of BrC, BC, and the DRE ratio of BrC to BC; the DRE distributions are shown in Fig. 10. For 2010, we estimated the global averaged DRE of BrC absorption at 0.10 W m⁻² in comparison to 0.39 W m⁻² by BC. While the global DRE by BrC is less than BC, regional BrC DRE can be as large as that of BC due to the large difference in emission distributions. BC DRE is large at northern mid latitudes due to anthropogenic emissions from China and India. BrC emissions are relatively low in these regions (Fig. 1), and consequently BC DRE dominates. Over regions with large fire emissions, both BC and BrC are important. Over most regions of the remote tropical ocean, BrC DRE is larger than BC, despite a model high bias of simulated BC in the middle and upper tropical troposphere (Fig. 5), suggesting significant broad regional effects by BrC radiative forcing in the tropics. This simulated feature is due to two factors. In the ICB simulation, wet scavenging removes much more BC than BrC. Therefore, BrC is enriched relative to BC in the free troposphere (e.g., Zhang et al., 2017). In the tropics, the easterly trade winds in the boundary layer become westerlies in the middle and upper troposphere. The regions of boundary layer BC transport and free-tropospheric BrC transport are in opposite directions. As a result, the DRE ratio of BrC to BC is low to the east of the fire emission regions, and it is high to the west of the fire emission regions.

To discuss our simulation results in the context of previous modeling studies, which did not use the ICB assumptions, we show the annual mean DRE distributions for all model simulations (Table 2) in Fig. 11. Comparing the global mean DRE relative change of ICB (0.10 W m⁻²) to NCB (0.013 W m⁻²) with that of NCNB (0.077 W m⁻²) to NCB shows that the global effect of a convective scavenging efficiency decrease is larger than photobleaching. A similar conclusion can be obtained by comparing the global mean DRE relative change of ICNB (0.26 W m⁻²) to NCNB with that of ICNB to ICB. The DRE relative change from ICB to ICBB (0.030 W m⁻²) indicates that the photobleaching effect of convectively transported BrC is larger than the enhancement of BrC convective transport. The 0.013 W m⁻² DRE in the NCB simulation is lower than previous model studies considering the photobleaching effect (Wang et al., 2018; Brown et al., 2018). In the NCB simulation, remote BrC concentrations are mostly affected by the threshold for photobleaching, which is 6 % in this study (Forrister et al. 2015) in comparison to 25 % in Wang et al. (2018) and Brown et al. (2018), causing the difference in the global DRE estimates with photobleaching between this work and previous studies. The 0.077 W m⁻² DRE in the NCNB simulation is comparable to previous studies (Feng et al., 2013; Jo et al., 2016; Wang et al., 2014).

Only ICB simulation results are discussed hereafter. Figure 12 shows the seasonally BrC DRE distributions. The seasonal variation is due primarily to biomass burning. Figure 2 shows that the largest fire emissions in August, September, July and June. While fire emissions are mostly in the tropics in SON, burning at northern mid and high latitudes is much more pronounced in June–July–August (JJA) in addition to tropical burning.

5.4 Global effects of BrC absorption on the atmosphere

As found by Zhang et al. (2017), the importance of radiative heating by BrC relative to BC increases with altitude due to convective transport. We compared the difference of average vertical profiles of the BrC-to-BC heating rate ratio in Fig. 14. Over regions directly affected by convection, where the convective mass flux is $>{\mathrm{10}}^{-\mathrm{5}}$ kg m⁻² s⁻¹, the simulated result is in agreement with Zhang et al. (2017). Globally, the average BrC∕BC heating rate ratio is 15 % below 500 hPa and 44 % above 500 hPa. In deep-convection regions, the average BrC∕BC heating rate ratio is 60 % below 500 hPa and 118 % above 500 hPa, indicating that in deep-convection regions, atmospheric heating of BrC is stronger than that of BC. In comparison, over the regions not directly affected by deep convection, the globally averaged BrC∕BC heating rate ratio increases from 9 % at surface level to 53 % in the upper troposphere. Geographically, Fig. 15 shows that the difference between BrC and BC is particularly large over the tropics.

The heating rate from aerosols, especially its vertical profile, has significant implications on cloud dynamics (e.g., Bond et al., 2013) and can also induce feedback from regional circulation and the planetary boundary layer dynamics (e.g., Ramanathan and Carmichael, 2008). We conducted three 5-year free-running model simulations for the present day (the year 2010). In the control simulation, both BrC and BC heat were included. We then conducted sensitivity simulations in which only BrC or BC heating was included. By comparing the sensitivity simulation results to the control simulation, we diagnosed the climate response to differential BrC and BC heating.

The latitudinal and vertical difference between BrC and BC heating (Figs. 14 and 15) implies that BrC heating is more dominant in the tropics and tends to decrease the vertical gradient of temperature. Allen et al. (2012) suggested that light-absorbing aerosols may contribute to tropical expansion. They did not consider the effect of BrC, the atmospheric heating effect of which is more concentrated in the tropics than BC heating (Fig. 15). Using the latitude where the mean meridional circulation (MMC) at 500 hPa becomes zero on the poleward side of the subtropical maximum to diagnose the boundary of the tropics (Zhou et al., 2011), we estimated a $\mathrm{1.0}\pm \mathrm{0.9}{}^{\circ }$ latitude of tropical expansion due to BrC heating in comparison to a $\mathrm{1.2}\pm \mathrm{2.9}{}^{\circ }$ expansion due to BC heating. The large uncertainty comes from 5 years of free-running simulations and the relatively low model spatial resolution. Another effect from BrC heating is the decrease of deep-convective mass flux over the upper troposphere (Feingold et al., 2005; Yoshimori and Broccoli, 2008). We estimated a decrease of deep-convective mass flux by $\mathrm{4.41}\times {\mathrm{10}}^{-\mathrm{5}}$ kg m⁻² s⁻¹ or 4.1 % over the tropics, which is about one third of the corresponding BC heating effect ($\mathrm{1.52}\times {\mathrm{10}}^{-\mathrm{4}}$ kg m⁻² s⁻¹ or 12.9 %).

Hodnebrog et al. (2016) suggested that biomass burning aerosols suppress precipitation regionally due primarily to aerosol–cloud interactions. On a global basis, BrC heating reduces precipitation by 0.9 %±7.0 %, which is about 60 % of the precipitation reduction by BC simulated in the model. Over the tropical region with high-intensity convection and precipitation, BrC heating decreased precipitation by 3.9 %±17.8 %, similar to that from BC heating (4.0 %±17.1 %). The effect of BrC heating on tropical precipitation (∼100 %) is larger than on convective mass flux (about one third) relative to BC heating because of the stronger BrC than BC heating in the upper than lower troposphere (Figs. 14 and 15). BrC heating decreased precipitation by 0.3 %±10.7 % in the northern mid and high latitudes, which is much lower than the effect of BC heating ($-\mathrm{4.8}\mathit{\%}\pm \mathrm{13.5}$ %).