Interpreting the ultraviolet aerosol index observed with the OMI satellite instrument to understand absorption by organic aerosols : implications for atmospheric oxidation and direct radiative effects

Satellite observations of the ultraviolet aerosol index (UVAI) are sensitive to absorption of solar radiation by aerosols; this absorption affects photolysis frequencies and radiative forcing. We develop a global simulation of the UVAI using the 3-D chemical transport model GEOSChem coupled with the Vector Linearized Discrete Ordinate Radiative Transfer model (VLIDORT). The simulation is applied to interpret UVAI observations from the Ozone Monitoring Instrument (OMI) for the year 2007. Simulated and observed values are highly consistent in regions where mineral dust dominates the UVAI, but a large negative bias (−0.32 to−0.97) exists between simulated and observed values in biomass burning regions. We determine effective optical properties for absorbing organic aerosol, known as brown carbon (BrC), and implement them into GEOS-Chem to better represent observed UVAI values over biomass burning regions. The inclusion of absorbing BrC decreases the mean bias between simulated and OMI UVAI values from−0.57 to −0.09 over West Africa in January, from −0.32 to +0.0002 over South Asia in April, from−0.97 to−0.22 over southern Africa in July, and from−0.50 to+0.33 over South America in September. The spectral dependence of absorption after including BrC in the model is broadly consistent with reported observations for biomass burning aerosol, with absorbing Ångström exponent (AAE) values ranging from 2.9 in the ultraviolet (UV) to 1.3 across the UV–Near IR spectrum. We assess the effect of the additional UV absorption by BrC on atmospheric photochemistry by examining tropospheric hydroxyl radical (OH) concentrations in GEOS-Chem. The inclusion of BrC decreases OH by up to 30 % over South America in September, up to 20 % over southern Africa in July, and up to 15 % over other biomass burning regions. Global annual mean OH concentrations in GEOS-Chem decrease due to the presence of absorbing BrC, increasing the methyl chloroform lifetime from 5.62 to 5.68 years, thus reducing the bias against observed values. We calculate the direct radiative effect (DRE) of BrC using GEOS-Chem coupled with the radiative transfer model RRTMG (GC-RT). Treating organic aerosol as containing more strongly absorbing BrC changes the global annual mean all-sky top of atmosphere (TOA) DRE by +0.03 W m and all-sky surface DRE by −0.08 W m. Regional changes of up to +0.3 W m at TOA and down to−1.5 W m at the surface are found over major biomass burning regions. Published by Copernicus Publications on behalf of the European Geosciences Union. 2508 M. S. Hammer et al.: Interpreting the ultraviolet aerosol index observed with the OMI satellite instrument


Introduction
Absorption of solar radiation by aerosols plays a major role in radiative forcing and atmospheric photochemistry.Aerosol absorption has been estimated to be the second largest source of radiative forcing after carbon dioxide (Bond et al., 2013;IPCC, 2014;Ramanathan and Carmichael, 2008), although considerable uncertainty remains regarding the magnitude of the forcing (Stier et al., 2007;Wang et al., 2014).Absorption of ultraviolet (UV) radiation by aerosols decreases photolysis frequencies, leading to a reduction in the concentrations of atmospheric oxidants (Dickerson et al., 1997;Jacobson, 1998;Liao et al., 2003;Martin et al., 2003).Many atmospheric chemistry models tend to overestimate tropospheric hydroxyl radical (OH) concentrations compared to observations (Mao et al., 2013;Naik et al., 2013).Accurately representing aerosol absorption could help rectify the discrepancies between simulated and observed OH concentrations, and offer constraints on radiative forcing.
The ultraviolet aerosol index (UVAI) is a method of detecting aerosol absorption using satellite measurements.The UVAI is calculated by separating the spectral contrast of radiances due to aerosol effects from those due to Rayleigh scattering at two wavelengths in the near-UV region (Herman et al., 1997;Torres et al., 1998Torres et al., , 2007)).Two attributes of the UVAI method are (1) that aerosol optical properties are more readily detected over surfaces with low reflectance such as found in the UV (Torres et al., 2005), and (2) that the strong interaction between aerosol absorption and molecular scattering in the near-UV increases the sensitivity of UVradiances to aerosol absorption (Torres et al., 1998).These attributes enhance the ability of the UVAI to detect aerosol absorption that affects UV photolysis and radiative forcing.
Traditionally, black carbon (BC) is treated as the predominant absorbing carbonaceous aerosol, and all organic carbon is assumed to be primarily scattering or weakly absorbing colorless aerosol.However, a growing number of observations have found evidence of significant absorption by a subset of organic carbon, known as "brown carbon" (BrC), which is strongest in the ultraviolet and decreases into the visible and near-IR wavelength regions (Bergstrom et al., 2007;Chen and Bond, 2010;Kirchstetter et al., 2004;Martins et al., 2009;Yang et al., 2008;Zhong and Jang, 2014).The majority of BrC is emitted to the atmosphere through low-temperature, incomplete combustion of biomass and biofuel (Chen and Bond, 2010;Kirchstetter et al., 2004;Zhong and Jang, 2014).There is evidence of a possible source from residential coal burning (Bond, 2001), while the high-temperature environment associated with other fossil fuel combustion is unfavorable to BrC formation (Andreae and Gelencsér, 2006;Saleh et al., 2014).BrC has been observed to contribute significantly to the overall absorption by biomass burning aerosol, especially in the UV (Clarke et al., 2007;Corr et al., 2012;Kirchstetter and Thatcher, 2012).
The UVAI is sensitive to this absorption (Jethva and Torres, 2011).
A great deal of uncertainty exists regarding the fraction of total primary organic carbon that is brown (BrC/POC).This uncertainty arises from the variety of methods used to measure BrC absorption as well as the variable nature of organic aerosols themselves (Lack and Langridge, 2013).Filter-based measurements where the organic carbon is extracted from the total biomass burning aerosol sample with the use of solvents range from 50 to 90 % (Chen and Bond, 2010;Kirchstetter et al., 2004).A broad range of BrC/POC values have been used to simulate absorption by brown carbon.For example, Feng et al. (2013) assume that 66 % of primary organic carbon from biomass and biofuel emissions is brown, Wang et al. (2014) assume that 50 % of POC from biomass and 25 % of POC from biofuel emissions is brown, while Lin et al. (2014) assume that 100 % POC from biomass and biofuel emissions is brown.Global observations of reflected solar radiation used in the UVAI could offer a constraint on these different assumptions.
Several estimates of BrC absorption exist, but they all differ significantly.The imaginary part of the refractive index (k) for BrC derived by Kirchstetter et al. (2004) and Chen and Bond (2010) are often taken, respectively, as the upper (k ∼ 0.168 at 350 nm) and lower (k ∼ 0.074 at 350 nm) limits in modeling studies (Arola et al., 2011;Feng et al., 2013;Lin et al., 2014).Different observations may reflect different burn conditions (Saleh et al., 2014) as well as chemical loss and evaporation of BrC (Forrister et al., 2015;Zhong and Jang, 2014).Global observations are needed to infer the effective absorption across a variety of conditions.Many studies have estimated the direct radiative effect (DRE) and/or direct radiative forcing (DRF) by BrC.In Heald et al. (2014) a clear distinction is made between the DRE, which is the instantaneous imbalance of out-going longwave and incoming shortwave radiation due to the presence of an atmospheric constituent, and the DRF, which is the difference in DRE between present-day and preindustrial conditions.Prior estimates of the change in all-sky top of atmosphere (TOA) DRE from treating organic aerosol as containing BrC range from +0.04 W m −2 to +0.25 W m −2 globally (Chung et al., 2012;Feng et al., 2013), with estimates of regional seasonal DRE of organic aerosol including absorbing BrC ranging from +0.5 to 1 W m −2 (Arola et al., 2015).Most studies estimate a TOA DRF of between 0.07 and 0.57 W m −2 due to absorption by BrC (Lin et al., 2014;Park et al., 2010;Wang et al., 2014).Following submission of our paper, we learned of a submitted paper by Jo et al. (2015) that developed a global simulation of BrC and applied it to investigate atmospheric photochemistry.To our knowledge, this work and that by Jo et al. (2015) are the first two chemical transport modeling studies that have considered the effect of absorption by BrC on atmospheric photochemistry.
In this work we introduce BrC to the chemical transport model GEOS-Chem and examine its effect on at-mospheric absorption and photochemistry, in particular in known biomass burning regions.To evaluate aerosol absorption, Sect. 3 develops a simulation of the UVAI following Buchard et al. (2015) using the Vector Linearized Discrete Radiative Transfer model (VLIDORT) coupled with aerosol fields from GEOS-Chem.Section 4 compares the simulated UVAI values to observations from the Ozone Monitoring Instrument (OMI).The change in reflected solar radiation as observed by the UVAI tests the effective representation of the absorption of UV radiation by BrC.Section 5 examines the effect of the added BrC absorption on ozone photolysis frequencies and tropospheric OH concentrations in the GEOS-Chem simulation.Section 6 calculates the DRE of absorbing brown carbon.Section 7 reports the conclusions.

OMI ultraviolet aerosol index
The OMI ultraviolet aerosol index is a method of detecting absorbing aerosols from satellite measurements in the near-UV wavelength region.The UVAI was first observed from the Nimbus-7 TOMS (Total Ozone Mapping Spectrometer) (Herman et al., 1997;Torres et al., 1998) and is currently a product of the OMI Near-UV algorithm (OMAERUV) (Torres et al., 2007).OMI flies on NASA's Aura satellite and has been taking global daily measurements since 2004.The OMAERUV algorithm uses the 354 and 388 nm radiances measured by OMI to calculate the UVAI according to Torres et al. (1998Torres et al. ( , 2007)): where I meas 354 is the TOA at 354 nm as measured by OMI, and I calc 354 is the radiance at 354 nm calculated for a purely Rayleigh scattering atmosphere bounded by a Lambertian surface of reflectance R * 354 , which is known as the adjusted Lambert equivalent reflectivity (LER) (Dave, 1978).R * 354 is calculated by correcting the LER at 388 nm (R * 388 ) for the spectral dependence of the surface reflectance at 354 nm.
Positive UVAI values indicate absorbing aerosol.Negative values indicate non-absorbing aerosol.Near-zero values indicate clouds, minimal aerosol, or other non-aerosol related effects such as unaccounted for land surface albedo wavelength dependence, ocean color effects or specular ocean reflection (i.e., sun glint).These second-order effects yield UVAI values ±0.5 within the noise level (Torres et al., 2007).The OMAERUV product identifies clouds using the measured scene reflectivity and the UVAI (Torres et al., 2013).We reject cloudy conditions (quality flag of 1) to focus on cloudfree conditions (quality flag 0).
In this work we use the OMI UVAI to evaluate simulated UVAI values, as described in Sect.3.

Absorption Ångström exponent (AAE)
We use observations of the absorption Ångström exponent (AAE) for biomass burning aerosol to test our representation of the spectral dependence of absorption.The AAE is the slope of aerosol absorption optical depth (AAOD) versus wavelength (λ) in log-log space.Using the AAE, the AAOD can be related to wavelength with the power-law relationship where k is a constant.Aerosols with spectrally independent absorption display an AAE of about 1, while aerosols with spectrally dependent absorption have an AAE > 1. BC exhibits spectrally independent absorption, and is often accepted as having an AAE close to 1 (Bergstrom et al., 2002;Bond and Bergstrom, 2006).The observed AAE over the near-UV to near-IR spectral regions can indicate aerosol type, with urban pollution aerosols dominated by BC exhibiting an AAE near 1, biomass burning aerosols displaying an AAE near 2, and desert dust having an AAE > 2 (Bergstrom et al., 2007;Russell et al., 2010).
Several recent studies have attributed the spectrally dependent absorption by biomass burning aerosols to the presence of BrC (Clarke et al., 2007;Corr et al., 2012;Kirchstetter and Thatcher, 2012;Rizzo et al., 2011;Zhong and Jang, 2014).Kirchstetter et al. (2004) measured over the 300-1000 nm range an AAE of ∼ 2 for biomass burning aerosol and an AAE of ∼ 1 for motor vehicle aerosol.They found that after extracting the organic carbon from the samples using acetone, the AAE of the biomass burning aerosol decreased to around 1, while the motor vehicle aerosol AAE remained unchanged.They concluded that the spectral dependence of absorption by biomass burning aerosol was due to BrC, while the absorption by motor vehicle emissions was due to BC.
Table 1 contains a summary of measured AAE values for biomass burning aerosol.AAE values increase toward UV wavelengths as expected for BrC absorption.Variability in the AAE at visible wavelengths may reflect differences in burn conditions and fuel type.Observations in the ultraviolet, such as the UVAI, offer an exciting opportunity to exploit the large AAE of biomass burning aerosol at short wavelengths to assess the global magnitude of BrC absorption (Jethva and Torres, 2011).

Simulated ultraviolet aerosol index
We simulate the UVAI following Buchard et al. (2015) using the VLIDORT model (Spurr, 2006), which includes polarization effects and uses the discrete ordinates method to solve the radiative transfer equation.We supply VLIDORT with the OMI pixel viewing geometry to calculate the UVAI for that pixel.The UVAI values are calculated from TOA radiances computed by VLIDORT at 354 and 388 nm, the wavelengths used by the OMAERUV product. where is the TOA radiance calculated with VLI-DORT at 354 nm assuming an atmosphere containing aerosol and Rayleigh effects, and I RAY 354 is the TOA radiance calculated with VLIDORT at 354 nm assuming a purely Rayleigh scattering atmosphere bounded by a Lambertian surface of reflectance R * 354 (adjusted LER).R * 354 is calculated by correcting the LER at 388 nm (R * 388 ) for wavelength dependence: where R 388 and R 354 are surface reflectance values from a revisited TOMS-based climatology data set (Torres et al., 2013).R * 388 is calculated by relating TOA radiance to diffuse reflectivity using the equation (Buchard et al., 2015) where is the TOA radiance calculated with VLI-DORT at 388 nm assuming an atmosphere containing aerosol and Rayleigh effects, I RAY 388 is the TOA radiance at 388 nm calculated with VLIDORT for a purely Rayleigh scattering atmosphere bounded by a Lambertian surface, T RAY 388 is the simulated transmittance at 388 nm for a Rayleigh atmosphere, and S RAY b 388 is the spherical albedo of a Rayleigh atmosphere at 388 nm.
For the calculation of TOA radiances, we provide VLI-DORT with vertical profiles of aerosol extinction, singlescattering albedo, and 32 Legendre-function expansion coefficients of the scattering phase function.We assume all aerosol particles are spherical.We obtain these aerosol optical properties using daily aerosol fields at satellite overpass time from GEOS-Chem version 9-01-03 (http://geos-chem.org), a global three-dimensional chemical transport model driven by assimilated meteorological data from the Goddard Earth Observation System (GEOS-5) of the NASA Global Modeling and Assimilation Office (GMAO).Our simulation is conducted at a spatial resolution of 2 • × 2.5 • with 47 vertical levels for the year 2007.
We calculate UVAI values for two cases.The base case simulation treats the aerosol optical properties as currently implemented in GEOS-Chem in which all organic carbon aerosols are weakly absorbing and colorless as shown in Table 2.The second case adds the more strongly absorbing BrC as described in Sect.4.2 below.

Base case simulation
Figure 1 shows the monthly mean OMI UVAI observations for January, April, July, and September of 2007.Clear signals are apparent over regions dominated by mineral dust and biomass burning (Herman et al., 1997;Torres et al., 1998).Absorption over desert regions occurs for all four months, giving UVAI values between 1 and 3, particularly over the Saharan, Iranian, and Thar deserts.Aerosol absorption from biomass burning primarily follows the seasonal cycle of agricultural burning (Duncan et al., 2003).In January, absorp-  3 and 4. tion over West Africa yields UVAI values between 1 and 2.5.In April aerosol absorption is visible over South Asia with UVAI values between 0.5 and 1. UVAI values of 1-1.7 occur over southern Africa in both July and September, while UVAI values of up to 3 occur over South America in September.
Figure 2 shows the monthly mean UVAI values for our base case simulation without BrC, The simulation captures the major absorption features compared to OMI over the desert regions, giving UVAI values of 1-3, however it fails to capture the absorption by biomass burning aerosol, giving UVAI values between −2 and 0 in all biomass burning regions for the four months.These negative values indicate that the UVAI simulation is insensitive to the absorption by BC and is dominated by the scattering from organic carbon aerosol.A sensitivity test with doubled BC concentrations increased UVAI values by only ∼ 0.1.We also conducted a sensitivity test to determine whether the heights of the biomass burning plumes could explain the underestimated absorption.Raising the aerosol layer height to unrealistic altitudes (∼ 10 km above the surface) increased the UVAI by only 0.1-0.3,which is insufficient to account for the differences between the simulation and observations.
To further analyze the discrepancies between simulated and observed UVAI, we choose four regions corresponding to the seasonal biomass burning outlined in Fig. 1 between the simulated and OMI UVAI as well as the mean bias (simulated-OMI UVAI).The correlation between the OMI and simulated UVAI is low (0.09-0.48) in all regions, with large mean biases of −0.32 to −0.97.
Uncertainty in aerosol optical depth also cannot explain the UVAI bias.Table 4 shows the simulated AOD compared with AOD retrieved from the MODIS (Moderate Resolution Imaging Spectroradiometer) and MISR (Multi-angle Imaging Spectroradiometer) satellite instruments.Overall the simulated values are within the range of satellite-retrieved AOD values.The maximum difference in simulated versus satellite AOD is found with the MODIS Dark Target algorithm over South America in September.Matching the simulated AOD to the satellite AOD changed the UVAI by less than 0.1.
We attempt to reconcile the differences between the simulated and OMI UVAI in biomass burning regions by introducing absorbing BrC into GEOS-Chem, as described below.

Treatment of brown carbon
Here we apply the OMI UVAI observations to estimate the effective absorption by BrC.We exploit the fact that the TOA radiances used in the OMI UVAI contain implicit information on the BrC from actual burn conditions, on the BrC that remains after chemical loss or evaporation, on Br-SOA, and on the BrC/POC fraction.We use the term effective to denote the implicit dependence of the UVAI upon these multiple processes.Through sensitivity simulations, we derive the effective k values for BrC given the assumed BrC/POC fraction required to reproduce the observed absorption by the OMI UVAI.This is accomplished by conducting the sensitivity simulations for several cases of BrC/POC fraction, assuming the same fixed spectral dependence for each case, and adjusting the magnitude of the effective k values to match the OMI UVAI.We treat the relative spectral dependence of k, log( k)/log( λ), as 3 for wavelengths between 300 and 600 nm to represent the mean from laboratory and field measurements of 3.2 ± 0.7 (Kirchstetter et al., 2004;Zhang et al., 2013;Zhong and Jang, 2014).At wavelengths ≥ 600 nm we leave the absorption properties of POC unchanged since BrC absorption decreases significantly into the visible and near-IR (Bergstrom et al., 2007;Chen and Bond, 2010;Yang et al., 2009).
The filled circles in Fig. 3 show the effective k values of BrC derived from seven sensitivity simulations that all achieve the same simulated UVAI.Only the BrC/POC fraction varies between simulations.The choice of simulated UVAI was selected to represent the global OMI UVAI over major biomass burning regions.The imaginary part of the refractive index decreases with increasing wavelength following an exponential relationship as prescribed based on laboratory and field measurements, and decreases with increasing BrC/POC fraction as required to reproduce the OMI UVAI.The effective k values increase with decreasing BrC/POC fraction because as BrC concentration decreases, BrC must for BrC as a function of wavelength and of the assumed fraction of primary organic carbon that is brown (BrC/POC).The background spectrum represents k values calculated using Eq. ( 6).The filled circles represent the k values obtained from sensitivity simulations.An organic carbon density of 1.3 g cm −3 is assumed.
be more absorbing to match the absorption observed by OMI.Given the smoothly varying relationship between k and BrC/POC we develop the following expression to represent this relationship: • exp (−10.5λ) ; BrC/POC ≥ 0.4, 300 nm ≤ λ ≤ 600 nm, where λ is wavelength (µm), ρ is the density of organic carbon (g µm −3 ), and c is a conversion constant equal to 1.0 × 10 12 /4π µm 2 g −1 .
The background spectrum of Fig. 3 shows the k values calculated using Eq. ( 6).This expression reproduces the full radiative transfer sensitivity simulations with a root mean squared error (RMSE) of 0.004 and a coefficient of determination (r 2 ) of 0.99.Equation (6) does not apply for BrC/POC fractions less than 0.4 since they do not reproduce the absorption observed by OMI.We emphasize that multiple choices of k and BrC/POC will yield the same TOA radiance and UVAI.The effects on tropospheric OH concentrations and radiative forcing remain unaffected as BrC/POC and effective k vary together, since the distribution of scattering and absorption remains the same.
Table 5 contains our derived imaginary parts of the refractive index for BrC/POC fractions of 0.5 and 1.0.Table 5 also contains effective k values derived for an organic carbon density of 1.8 g cm −3 which has been assumed in prior studies of BrC.The range of values for k covered by varying the BrC/POC fraction encompasses the range of values for BrC found in the literature (Chen and Bond, 2010;Feng et al., 2013;Kirchstetter et al., 2004;Lin et al., 2014;Sun et al., 2007;Wang et al., 2014;Zhang et al., 2013;Zhong    The columns with effective k values for BrC/POC fraction of unity offer the convenience of representing the effective absorption by BrC without needing to assume an arbitrary BrC/POC fraction, or to introduce a separate BrC tracer.The effective k values for unity BrC/POC fraction can be thought of as the effective imaginary refractive index for an internal mixture of BrC and colorless POC. Evidence for the existence of brown secondary organic carbon (Br-SOA) also exists.The majority of Br-SOA is from anthropogenic sources (Jo et al., 2015;Liu et al., 2013;P. F. Liu et al., 2015;Zhang et al., 2013), while SOA formed from biogenic carbon is largely non-absorbing (Flores et al., 2014;P. F. Liu et al., 2015).On a global scale it is estimated that the majority of SOA is formed from biogenic carbon (Hallquist et al., 2009;Lack et al., 2004;Tsigaridis and Kanakidou, 2003).Therefore we treat SOA as non-absorbing.We tested this approach in a sensitivity study with the standard SOA mechanism in GEOS-Chem v9-01-03 (Henze and Seinfeld, 2006;Henze et al., 2008;Liao et al., 2007) by assigning 100 % anthropogenic SOA as brown, and found that the change in UVAI was negligible (less than 0.1).Alternative SOA implementations (e.g., as used in Jo et al., 2015) may have a larger effect.

Simulation including brown carbon
Figure 4 shows the monthly mean UVAI values for the simulation including BrC for the months of January, April, July, and September of 2007.The simulated absorption features including BrC are much more consistent than the base case simulation at reproducing the OMI UVAI over biomass burning regions (Fig. 2).The simulated UVAI in the four biomass burning regions now ranges from 0.5 to 3. As summarized in Table 3, the correlation coefficients between the simulated and OMI UVAI for the four biomass burning regions now range from 0.57 to 0.68, with mean biases of −0.22 to +0.33.
The simulated UVAI using global mean k values underestimates the OMI observations for the West Africa and southern Africa regions, but overestimates observations in the South American region.We tested how k would need to change to explain the regional UVAI bias if k were the only error source.We find that these regional biases could be eliminated by changing k at 350 nm by +2 % over West Africa, by +10 % over southern Africa, and by −30 % over South America.The presence of more absorbing BrC over West and southern Africa where savannah fires dominate, and less absorbing BrC over the South America region where forest fires dominate, is consistent with work by Saleh et al. (2014) who found that the absorptivity of BrC from biomass burning is greater for flaming fires associated with burning grasslands than for smouldering fires associated with burning forest.
The absorption in the West and southern Africa cases appears to be concentrated closer to the source for the simu-Table 5. Imaginary part of the refractive index (k) values for brown carbon (not total organic carbon) inferred for case 2. We include k values associated with multiple densities (ρ), and multiple fractions of brown carbon to primary organic carbon (BrC/POC).All four columns for case 2 yield the same absorption.lated values (Fig. 4) compared to the OMI values (Fig. 1), which show an even distribution of UVAI values away from the source.By contrast, the absorption in the South American region appears to be distributed farther from the source in the simulation than in the OMI observations.Evidence exists of atmospheric photochemical loss and evaporation of BrC that causes it to become less absorbing over a lifetime of less than a day (Forrister et al., 2015;Zhong and Jang, 2014).Representing these processes would improve the simulation in the South American region but degrade the simulation in the West Africa and southern Africa regions.Regional treatment of BrC loss processes may be warranted in future work.
The current implementation offers our best representation of the effective BrC absorption at the global scale.
Table 6 shows the calculated AAE values for biomass burning aerosol (i.e., black carbon + organic carbon aerosol) from our simulations for comparison with the literature values in Table 1.Large biases are apparent for the base case simulation without BrC.We evaluate the case 2 simulation including BrC in detail.For the 350-400 nm wave-length region our mean AAE value of 2.8 ± 0.22 for the four biomass burning regions is within the recommended values of 2.5-3.0 by Jethva and Torres (2011).In the 350-700 nm range our mean AAE of 2.2 ± 0.17 is close to the value of 1.9 from Kirchstetter and Thatcher (2012).The slight positive bias could arise from the fact that Kirchstetter and Thatcher (2012) took their absorption measurements from wood smoke emitted from houses in rural California during the winter, which have different conditions than the tropical open burning considered here.We obtain a mean AAE value of 1.7 ± 0.15 for the 300-1000 nm range, which falls within the literature values of 1.1-2.For the 450-550 nm wavelength region, we obtain a mean AAE value of 2.5 ± 0.14, which is biased high compared to the values from Corr et al. (2012) extracted from an examination of biomass burning plumes in North-Central Canada, where burn conditions differ from the mostly tropical regions considered in our analysis.Over the 400-700 nm region we obtain a mean AAE of 2.1 ± 0.15, falling within the range of the literature values (1.3-2.1).In the 400-1000 nm region, we obtain a mean 0.8 1.1 0.7 1.0 0.9 ± 0.18 400-700 1.0 1.2 0.9 1.1 1.1 ± 0.13 450-700 1.0 1.1 0.9 1.0 1.0 ± 0.08 300-1000 0.9 1.1 0.9 0.9 0.9 ± 0.10 400-1000 0.7 0.9 0.7 0 AAE value of 1.3 ± 0.005, which is at the low end of the literature values (1.3-1.7).The overall consistency between observed and simulated AAE provides a measure of confidence in the spectral dependence of aerosol optical properties from the UV to the IR.We now examine the implications of this absorption for OH and DRE.

Analysis of the effect of absorption by BrC on OH concentrations in GEOS-Chem
The strong absorption in the UV by BrC aerosol decreases photolysis frequencies, which has implications for ozone photolysis and OH production.Here we examine the effect of the added absorption by BrC on the O 3 → O( 1 D) photolysis frequency, J (O( 1 D)), and tropospheric OH concentrations.
Figure 5 shows the percent differences in lowertropospheric OH concentrations between the GEOS-Chem simulation including absorbing BrC versus the base case simulation.The most significant decreases correspond with the major biomass burning regions.The addition of BrC decreases OH concentrations by up to 30 % over the South American biomass burning region in September and up to 20 % over the southern Africa biomass burning region in July.OH concentrations decrease by up to 15 % over West Africa in January and southern Africa in September, with decreases of up to 5 % over North America in July, South America in July, and South Asia in January, April, and September.The spatial and seasonal pattern of J (O 1 (D)) differ-ences closely reproduces the changes in OH (r 2 = 0.85) (not shown).
Methyl chloroform observations provide a valuable constraint on global OH (Prather et al., 2012;Spivakovsky et al., 2000).The addition of BrC to the GEOS-Chem simulation reduces global mean tropospheric OH concentrations.The reduction in global mean OH concentrations increases the methyl chloroform lifetime to tropospheric OH from 5.62 to 5.68 years.This change is noteworthy given the buffered nature of OH.This change reduces the bias with observations giving a methyl chloroform lifetime of 6.0 (+0.5, −0.4) years (Prinn et al., 2005).

Radiative impact of brown carbon
We calculate the direct radiative effect (DRE) of including absorbing BrC relative to that of the base case simulation.We use GEOS-Chem coupled with the radiative transfer model RRTMG (Iacono et al., 2008), a configuration known as GC-RT, that is described in Heald et al. (2014).GC-RT calculates both the longwave (LW) and shortwave (SW) instantaneous total radiative fluxes as well as the flux differences due to a specific constituent of the atmosphere (e.g., organic aerosol).The DRE is calculated by adding the LW and SW flux differences determined through perturbation of the constituent of interest.Our GC-RT simulations use version 10.1 of GEOS-Chem with the same aerosol emissions described in Sect. 3 (e.g.,GFED3 open fire emissions).We calculate the DRE of absorption by BrC as the difference in the DRE of organic aerosol when including BrC (case 2) minus the DRE of organic aerosol in the base case.We focus on the DRE rather than the DRF to avoid ambiguity in preindustrial BrC.
Figure 6 shows all-sky DRE values for 2007.The top two panels are the DRE values for organic aerosol from the case 2 simulation including BrC.The overall DRE of organic aerosol including BrC is negative, with the largest effects over major biomass burning regions.The bottom two panels show the DRE for absorption by BrC, calculated as the difference between the DRE of organic aerosol for the case 2 simulation including BrC minus the base case simulation.At the surface BrC absorption reduces the DRE by −1.5 W m −2 over South America and southern Africa, and by −0.5 to −0.25 W m −2 over South Asia, North America, West Africa, Australia, and Europe.At TOA, BrC absorption increases the DRE by up to 0.3 W m −2 over South America and southern Africa, and by 0.05 to 0.15 W m −2 over broad regions.This overall cooling effect at the surface and warming effect on the atmosphere are consistent with previous work (e.g., Chen and Bond, 2010).
Table 7 contains LW and SW global annual mean flux differences as well as the resulting DRE values for both organic aerosol and BrC absorption.The values for organic aerosol are from the base case simulation assuming weakly absorbing organic carbon and the case 2 simulation including BrC, while the values for BrC absorption are calculated as their difference.Absorption by BrC has a mean all-sky DRE at TOA of +0.03 W m −2 and at the surface of −0.08 W m −2 .
Our findings are at the lower end of the range of values from other studies that estimate the DRE of BrC absorption.Feng et al. (2013) introduce absorption by BrC based on the optical properties from Kirchstetter et al. (2004) and Chen and Bond (2010)

Conclusions
We interpret OMI observations of the ultraviolet aerosol index (UVAI), which provides a measure of absorbing aerosols, by developing a simulation of the UVAI using the vector radiative transfer model VLIDORT coupled with GEOS-Chem aerosol fields.The base case simulation without brown carbon (BrC) well represents the observed UVAI in most of the world but significantly underestimates the absorption in biomass burning regions.We apply the OMI UVAI to estimate absorption by BrC.This approach exploits the strong absorption by BrC at ultraviolet wavelengths and its effect on top of atmosphere (TOA) radiance.We express the imaginary part of the refractive index of BrC that is required to obtain near-identical TOA radiance values as a function of the fraction of primary organic carbon that is brown.This effective refractive index of BrC provides a measure of the degree of browness needed to represent the complex processes (e.g., burn conditions, photochemical loss) affecting global BrC and in turn the UVAI.This effective refractive index of BrC eliminates the need to know the BrC/POC ratio to model www.atmos-chem-phys.net/16/2507/2016/ the radiative effects of BrC.Rather, an effective refractive index can be chosen (Eq.6) to represent an internal mixture of BrC and colorless POC.The simulation including absorbing BrC is much more consistent than the base case at reproducing the OMI UVAI over biomass burning regions.The mean bias between simulated and OMI UVAI values is reduced from −0.57 to −0.09 over West Africa in January, from −0.32 to +0.0002 over South Asia in April, from −0.97 to −0.22 over southern Africa in July, and from −0.50 to +0.33 over South America in September.The updated optical properties for BrC result in AAE values for biomass burning aerosol ranging from 2.9 in the UV to 1.3 across the UV-Near IR, which are broadly consistent with field observations.We apply this constraint on ultraviolet absorption to examine implications for the O 3 → O( 1 D) photolysis frequency.We find that the inclusion of absorbing BrC into GEOS-Chem decreases J (O( 1 D)) and lower tropospheric OH by up to 30 % over South America in September, up to 20 % over southern Africa in July, up to 15 % over West Africa in January and southern Africa in September, and up to 5 % elsewhere.The decrease in global mean OH concentration in GEOS-Chem increases the methyl chloroform lifetime to tropospheric OH from 5.62 to 5.68 years, reducing the bias with estimates from observations of 6.0 (+0.5, −0.4) years.
We calculate the direct radiative effect (DRE) of BrC using GEOS-Chem coupled with the radiative transfer model RRTMG (GC-RT).We obtain global annual mean all-sky TOA DRE values for BrC absorption of +0.03 W m −2 and values of −0.08 W m −2 at the surface.Regional changes of up to +0.3 W m −2 at TOA and down to −1.5 W m −2 at the surface are found over major biomass burning regions.Our results are within the range of prior estimates of DRE for BrC absorption.
Ample opportunities exist to further develop the simulations of BrC and more generally of the UVAI.These opportunities include explicitly accounting for the range of processes affecting BrC such as burn conditions, photochemical loss, secondary production, as well as regional treatment of BrC.An explicit simulation of BrC (e.g., Jo et al., 2015) would facilitate these developments.Interpretation of the long observational record of the UVAI from 1979 to the present should offer constraints on trends in aerosol composition, ultraviolet absorption, and radiative effects.The forthcoming TROPOMI instrument and geostationary constellation (e.g., TEMPO, Sentinel-4, and GEMS) will offer UVAI observations at 5-20 times higher spatial resolution, as well as information on diurnal variation, both of which may offer additional constraints on BrC evolution.

Figure 1 .
Figure 1.Monthly mean ultraviolet aerosol index (UVAI) observations from the OMI satellite instrument for 2007.White space indicates cloud or snow contamination.Gray boxes outline regions examined in Tables3 and 4.

Figure 2 .
Figure 2. Monthly mean UVAI values for 2007 simulated for OMI observing conditions using the vector radiative transfer model VLIDORT coupled with GEOS-Chem aerosol fields for the base case without brown carbon (BrC).White indicates cloud or snow contamination.

Figure 3 .
Figure 3. Imaginary part of the refractive index (k) values inferredfor BrC as a function of wavelength and of the assumed fraction of primary organic carbon that is brown (BrC/POC).The background spectrum represents k values calculated using Eq.(6).The filled circles represent the k values obtained from sensitivity simulations.An organic carbon density of 1.3 g cm −3 is assumed.

a r :
Pearson correlation coefficient.b Mean bias = simulated UVAI -observed UVAI.c n = number of GEOS-Chem grid boxes in region.d Regions are defined in Fig. 1.

Figure 4 .
Figure 4. Monthly mean UVAI values for 2007 simulated for OMI observing conditions using the vector radiative transfer model VLIDORT coupled with GEOS-Chem aerosol fields for case 2, which assumes the presence of absorbing BrC aerosol.White indicates clouds or snow contamination.

Figure 5 .
Figure 5. Percent difference between OH concentrations in the lower troposphere from a GEOS-Chem simulation including absorbing BrC minus a simulation without BrC.The values are monthly means for January, April, July, and September 2007.
into a global model, and calculate an allsky TOA DRE for BrC absorption of +0.04 to +0.11 W m −2 , and an all-sky surface DRE for BrC absorption of −0.06 to −0.14 W m −2 .Chung et al. (2012) estimate BrC absorption by subtracting the absorption by BC and desert dust from total aerosol AAE values from AERONET to calculate an allsky TOA DRE for organic aerosol when including BrC between −0.15 and +0.12 W m −2 and an all-sky surface DRE between −1.50 and −0.75 W m −2 .Arola et al. (2015) use AERONET retrieved imaginary parts of the refractive index for BrC at 440, 670, 870, and 1020 nm to estimate over the Indo-Gangetic plain monthly all-sky TOA DRE values for organic aerosol including BrC absorption up to +0.5 W m −2 in spring and as low as −1 W m −2 in the winter.

Figure 6 .
Figure 6.Annual mean all-sky DRE values for 2007 (W m −2 ).The top two panels are the DRE values for organic aerosol from the case 2 simulation including BrC at (a) the surface and (b) TOA.The bottom two panels are the change in DRE values for absorption by BrC calculated as the difference between the DRE values for organic aerosol from the case 2 simulation and the base case simulation (without BrC) at (c) the surface and (d) TOA.

Table 1 .
Absorption Ångström exponent (AAE) values for biomass burning regions from the literature.

Table 2 .
Imaginary part of the refractive index (k) values for the base case with weakly absorbing "colorless" primary organic carbon.

Table 3 .
Comparison of the simulated versus observed (OMI) UVAI values for the biomass burning regions in the months examined.The base case corresponds to a simulation without BrC, while case 2 corresponds to a simulation including absorbing BrC.

Table 4 .
The mean AOD values for each region from the GEOS-Chem (GC) base case simulation, the MISR instrument, and the MODIS Terra satellite instrument.The MODIS values are included for both the collection 6 Deep Blue and Dark Target algorithms.
* Regions are defined in Fig.1.and Jang, 2014).The four columns with BrC yield identical wavelength-dependent global distributions of scattering and absorption that in turn yield the same UVAI, OH, and DRE.

Table 6 .
The absorption Ångström exponent (AAE) values for major biomass burning regions and seasons obtained from the base case simulation without BrC and the case 2 simulation including absorbing BrC.