Anthropogenic iron oxide aerosols enhance atmospheric heating

Combustion-induced carbonaceous aerosols, particularly black carbon (BC) and brown carbon (BrC), have been largely considered as the only significant anthropogenic contributors to shortwave atmospheric heating. Natural iron oxide (FeOx) has been recognized as an important contributor, but the potential contribution of anthropogenic FeOx is unknown. In this study, we quantify the abundance of FeOx over East Asia through aircraft measurements using a modified single-particle soot photometer. The majority of airborne FeOx particles in the continental outflows are of anthropogenic origin in the form of aggregated magnetite nanoparticles. The shortwave absorbing powers (Pabs) attributable to FeOx and to BC are calculated on the basis of their size-resolved mass concentrations and the mean Pabs(FeOx)/Pabs(BC) ratio in the continental outflows is estimated to be at least 4–7%. We demonstrate that in addition to carbonaceous aerosols the aggregate of magnetite nanoparticles is a significant anthropogenic contributor to shortwave atmospheric heating.

D ark-coloured aerosols such as combustion-induced carbonaceous particles and wind-blown mineral dust absorb solar radiation and perturb the climate system by heating the atmosphere and reducing the snow albedo 1,2 . Shortwave atmospheric heating by such aerosols can alter the cloud fraction and hydrological cycle on both regional and global scales [3][4][5][6] . The global mean increase in precipitation per degree of increase in global-mean surface temperature, attributable to human-induced global warming, strongly depends on the emission of black carbon (BC), a proxy for anthropogenic shortwave absorbers 6 . To evaluate the effects of dark-coloured aerosols on climate using numerical models, observational datasets are needed to constrain the sources, atmospheric abundance, and detailed microphysical properties (for example, size distribution and morphology) of individual light-absorbing aerosols 1,7 . The long-term data on absorption aerosol optical depth measured by ground-based remote sensing 8 have been the main observational constraints for numerical models used to evaluate the effects of absorbing aerosols on climate change 1,7 . Inverting the remote sensing data to the column abundance of absorbing aerosols requires a priori assumption on the optical properties of each type of aerosols that contributes to absorption aerosol optical depth 9 . Therefore, in situ observations of all the important contributors to atmospheric shortwave absorption form the basis for the quantitative investigation of the complex effects of absorbing aerosols on climate.
Until now, two types of carbonaceous aerosols, BC and brown carbon (BrC), and wind-blown mineral dust (DU) have generally been considered as the only three aerosol species that contribute significantly to shortwave absorption in the atmosphere and snowpack 1,2,8,10,11 . BC and BrC are mostly co-emitted during the burning of fossil fuels, biofuels and biomass 7 . Light absorption by DU is mostly due to iron oxide nanoparticles attached on the surfaces of host mineral materials 12 . DU absorption dominates the atmospheric shortwave absorption over the desert and dust outflow regions 13 . A number of attempts have recently been made to quantify BrC absorption separately from BC absorption 11,[14][15][16] . These studies suggest that BC absorption almost invariably dominates BrC absorption, except at ultraviolet wavelengths. In Beijing, an indirect estimate using ground-based remote-sensing technique 14 indicated that BrC absorption is B10% of BC absorption at the mid-visible wavelength in the winter and spring seasons. Another estimate of BrC absorption in Beijing using in situ optical and chemical measurements in March provided a similar result 11 . Using a global chemical transport model, Feng et al. estimated that BrC accounts for 7-19% of aerosol absorption (global average) 17 .
Recent observational studies using electron microscopy methods revealed that anthropogenic iron oxide particles in the form of aggregated FeO x nanoparticles are ubiquitous in urban atmospheres [18][19][20] and roadside environments 21 . They are emitted from, for example, the blast furnaces of iron manufacturing facilities 22 and from the engines and brakes of motor vehicles 23,24 . The major iron oxide phases of aggregated FeO x nanoparticles are magnetite (Fe 3 O 4 ) 18,20,21,23,24 , maghemite (g-Fe 2 O 3 ) 21,24 and hematite (a-Fe 2 O 3 ) 18,21,24 . However, their abundances in the troposphere and radiative effects have not yet been evaluated.
In this study, in situ aircraft measurements using a modified single-particle soot photometer (SP2) 25 and electron microscopy are performed to show that anthropogenic FeO x particles, particularly aggregated magnetite nanoparticles, are ubiquitous in the continental outflows from East Asia. We then evaluate their contribution to atmospheric shortwave absorption on the basis of the observed size-resolved mass concentrations and particle morphologies. Our results indicate that the absorption by anthropogenic FeO x is at least 4-7% of the BC absorption over East Asia.

Results
Observation. We used observational data from the Aerosol Radiative Forcing in East Asia (A-FORCE) 2013W aircraft campaign 26 over Yellow Sea and East China Sea in February and March, 2013. Our modified SP2 can measure individual BC and FeO x particles in the mass-equivalent diameter (D m ) domains of 70 nmrD m r850 nm and 170 nmrD m r2,100 nm, respectively 25 . The aerosol-sampling system, which consisted of a forward-facing shrouded solid diffuser inlet, transport tubes and aerosol measuring instruments, was designed to observe submicron-sized particles. The theoretical transmission efficiency curves Tr(D m ) of FeO x particles began to drop at D m ¼ B600 nm and decreased to B0.5 at D m ¼ 2,100 nm ( Supplementary Fig. 1). These theoretical Tr(D m ) curves suggest that our measurement system underestimates the ambient FeO x concentrations at D m 4B600 nm.
Characterization of anthropogenic FeO x . In this section, we characterize the fundamental properties of airborne FeO x particles in the East Asian continental outflows. For this purpose, we focus on the air parcel passed through the planetary boundary layer over eastern China and was directly transported to the sampling point on the flight track below 2 km altitude without experiencing wet removal of aerosols. The aircraft observation data of this air parcel is called dry PBL air 26 . The detailed criteria for selecting the dry PBL air according to the backward trajectory analysis is described in Kondo et al. 26 .
First, we classified individual Fe-bearing particles in dry PBL air depending on the composition and morphology, based on the electron-microscopy analyses of 1,460 particles collected by an onboard aerosol-impactor sampler. A transmission electron microscope (TEM) and a scanning transmission electron microscope equipped with energy-dispersive X-ray spectrometry (STEM-EDS) 19 were used for these analyses. Table 1 lists the number of Fe-bearing particles for each of the three morphological types measured by the STEM-EDS analyses. The most abundant type of Fe-bearing particle was aggregated FeO x nanoparticles; the TEM images and associated elemental mappings of these  Fig. 3). The second most abundant type of Fe-bearing particles in dry PBL air was fly ash, which is a complex internal mixture of combustion-induced refractory materials commonly includes Si as a major component. The Fe-bearing fly ash particles in the dry PBL air typically contained only several per cent of Fe by mass. A third type of Fe-bearing particle, Fe-bearing mineral dust, was rare in the dry PBL air; the fraction of mineral dust particles among all the Fe-bearing particles was only 0.035 (Table 1). Although the mineral dust was observed to be minor in the dry PBL air, it would be the dominant type of Fe-bearing particles over the East Asia during an Aeolian dust (that is, Kosa) outflow event. Next, we present results provided by the light-scattering signals of individual FeO x particles measured by the SP2. Figure 2 shows a scatterplot of the scattering cross-section at the onset of the incandescence (C s-oi ) 28 and the mass-equivalent diameter D m for both laboratory and ambient FeO x -containing particles. Figure 2 also includes experimental results for pure magnetite particles and mineral dust particles (Icelandic dust and Taklamakan dust) 25 . Compared with pure magnetite, the mineral dust samples exhibited large C s-oi values beyond the detectable limit of B4 Â 10 À 14 m 2 , because the host minerals internally mixed with the incandescing FeO x also contribute to the C s-oi value. On the basis of these results, we introduced a criterion for classifying the mixing state of detected  Table 1). The particles magnified in a and b were collected on the same substrate with an inter-particle distance of B4 mm, as shown in c. The elemental distributions of Fe and O for particles a and b are shown in d and e, respectively, and indicate that these particles contain iron oxide.   Figure 2 shows that a majority of the FeO x -containing particles in the dry PBL air were similar to the pure magnetite particles in terms of their C s-oi value, whereas a minority had substantially larger C s-oi values. In the dry PBL air, the fraction of dust-like FeO x particles among the detected FeO x -containing particles was B0.1, as discussed later.
In addition to the direct microscopic observations using the TEM-EELS, the optical signals measured using the SP2 also provided indirect information on the material properties of the airborne FeO x . The timing of the onset of incandescence (t oi ) of FeO x -containing particles has been experimentally shown to be a qualitative indicator of colour darkness (that is, light-absorbing efficiency) in FeO x materials 25 . The distribution of the t oi values of the FeO x -containing particles in dry PBL air was similar to that for black-coloured magnetite but dissimilar to that of red-coloured hematite (see the Methods section and Supplementary Fig. 4). This implies that the major component of incandescing FeO x particles in dry PBL air is magnetite, which is consistent with the TEM-EELS results ( Supplementary Fig. 3).
We found that the FeO x number concentration in the dry PBL air was highly correlated with the CO mixing ratio (r 2 ¼ 0.87) and BC number concentration (r 2 ¼ 0.69; Fig. 3a,b, respectively). These results suggest that the spatial distribution of the FeO x emission flux over the East Asian continent is similar to those of CO and BC. On the other hand, the FeO x /BC number concentration ratio in the dry PBL air-correlated neither with the relative abundance of the supermicron-sized aerosols (Fig. 3c) nor the number fraction of the dust-like FeO x -containing particles (Fig. 3d). The scatterplots imply that the majority of the detected FeO x particles in the dry PBL air are not associated with DU, which is consistent with the results of the direct STEM-EDS analyses of the aerosol-impactor samples ( Table 1).
The two most important conclusions drawn from these observations are as follows. First, the FeO x particles detected using the modified SP2, with the exception of the dust-like FeO x , are primarily aggregated magnetite nanoparticles of anthropogenic origin. Second, the anthropogenic magnetite is a major type of Fe-bearing aerosol in East Asian continental outflow.
Altitude profiles. Figure 4 shows the size-resolved number and mass concentrations of FeO x at each altitude along with those in the dry PBL air. Within the observed size domain of 170 nmrD m r2,100 nm, the size-resolved number concentration of FeO x was approximated using a power function with an offset dN dlogD m where y 0 , a and p are numerical parameters listed in Table 2. The size-resolved mass concentration of FeO x was approximated as where r FeOx is the assumed bulk density of FeO x (5.17 g cm À 3 ). In contrast, the size-resolved mass concentration of BC within the observed size domain of 70 nmrD m r850 nm (see Figs 6 and 14 of Kondo et al. 26 ) was approximated with a two-modal lognormal function as where A i , D i and s gi (i ¼ 0,1) are numerical parameters listed in Table 2. Figure 4 shows that the mode D m values of the number size distribution were smaller than 170 nm, whereas the mode D m values of the mass size distribution may have been larger than 2,100 nm, that is, the detectable D m domain of 170 nmrD m r2,100 nm was insufficient to reveal the entire shape of the size-distribution function of ambient FeO x . As mentioned earlier, our aerosol-sampling system likely underestimated the ambient FeO x concentrations for D m 4B600 nm. In contrast, our BC measurements in the domain of 70 nmrD m r850 nm were likely sufficient to determine the total BC mass concentrations, as expected from the Figure 6 and 14 of Kondo et al. 26 and the parameters listed in Table 2.
The BC size distributions tended to shift towards smaller sizes at higher altitudes 26 (see parameter D 0 in Table 2). This trend has been explained by the preferential wet removal of larger BC particles during vertical transport 26,29 . In contrast, the FeO x size distributions shifted towards larger sizes with increasing altitude (Fig. 4). The decrease of the parameter p with altitude reflects this trend ( Table 2). As shown later, the number fraction of dust-like FeO x was observed to increase with altitude. We hypothesize that the mineral dust particles lifted from the deserts in central China and transported by the westerlies cause the larger shift of the FeO x size distributions over the Yellow Sea and East China Sea. Figure 5a,b show the vertical profiles of the FeO x concentrations and the FeO x /BC concentration ratios, respectively. The mean FeO x mass concentration was 100-400 ng m À 3 at standard temperature and pressure (STP) in the boundary layer (altitu-deoB2 km), and monotonically decreased with altitude to 30-90 ng m À 3 STP in the free troposphere (altitude4B3 km). The mean FeO x /BC number concentration ratio was 0.004 in the boundary layer and decreased to 0.002 in the free troposphere. The mean FeO x /BC mass concentration ratio was 0.3-0.5 in all altitudes below 6 km, and increased to B0.8 at higher altitudes (6-8 km). The observed FeO x /BC number (mass) ratio would be substantially larger if we were able to detect smaller (larger) FeO x particles with D m o170 nm (D m 42,100 nm). The number fraction of dust-like FeO x showed a monotonic increase with altitude from B0.07 at 0-1 km to B0.5 at 6-8 km (Fig. 5c). In an aerosol-impactor sample collected during a time period of the highest FeO x number concentration observed at 6-8 km altitude (B0.4 cm À 3 STP), we actually found mineral dust particles including Fe (Supplementary Fig. 5). However, in the free tropospheric samples, the number of Fe-bearing particles collected on each TEM grid was too low to evaluate the relative abundance of each of the three morphological types listed in Table 1.
Mass-absorption cross-sections. An original discrete-dipole approximation code was used to calculate the mass-absorption cross-sections of the BC-and FeO x -containing particles  by assuming the particle shape, mixing state and refractive index of each material. Before discussing the results, we briefly explain the nontrivial assumptions used. For both BC and FeO x , the particle shape was assumed to be a fractal-like aggregate of spherical monomers. The mixing state of the aggregate with other non-absorbing materials was assumed to be either bare or coated: the bare state denotes a pure aggregate, whereas the coated state denotes an aggregate coated by a large amount of non-absorbing material. In the coated state, the volume of the coating material was prescribed such that the theoretical log(C s-be /C s-oi ) value of the model BC-containing particles was greater than the measured log(C s-be /C s-oi ) values of the real BC-containing particles, where the C s-be denotes the scattering cross-section of a BC-containing particle in the SP2 laser beam before the onset of evaporation 28 (see the Methods section). The mean þ 1s value of the observed log(C s-be /C s-oi ) value for BC-containing particles with D m B200 nm was B0.9 at all altitudes below 8 km (Fig. 5d). On the basis of this observation, the coating/aggregate volume ratio in the coated state was chosen to be approximately 3-4 so that the theoretical value of log(C s-be /C s-oi ) for the coated BC was B1.0 at D m B200 nm (Supplementary Fig. 6). The mass-absorption cross-sections (s a ) calculated for the BC-and FeO x -containing particles as functions of wavelength (l) and mass-equivalent diameter D m are shown in Fig. 6. In general, the s a value at a particular (l, D m ) is approximately three times larger for BC than for FeO x , primarily because the assumed bulk density of FeO x (5.17 g cm À 3 ) was 2.87 times greater than that of BC (1.8 g cm À 3 ). Compared with the bare state, s a was enhanced by a factor of approximately two in the coated state because of the so-called lensing effects 30 . The s a (l, D m ) of FeO x did not change appreciably with l in the D m domain largely contributing the total ambient FeO x mass concentration. Thus, we expect that the atmospheric absorption coefficient (b abs ) attributable to anthropogenic FeO x depends little on the wavelength. In contrast, the b abs attributable to carbonaceous aerosols (BC þ BrC) is known to decrease sharply with wavelength 14,15,31 . This large difference in the wavelength dependence of b abs between anthropogenic FeO x and carbonaceous aerosols will be useful to classify them in remote sensing observations. Shortwave atmospheric heating. Here we quantify the instantaneous shortwave atmospheric heating attributable to FeO x and BC using their observed size-resolved mass concentrations and theoretical mass-absorption cross-sections. The contribution to shortwave atmospheric heating from particles of a particular size is quantified by the size-resolved absorbing power dP abs /dlogD m (W m À 3 ), which is defined as where F ac (l) is the spectral actinic flux (W m À 2 nm À 1 ), s a (l, D m ) is the mass-absorption cross-section (m 2 g À 1 ) and dM/dlogD m (g m À 3 ) is given by equations (2)   angle (see the Methods section). Figure 7 shows the size-resolved absorbing powers of BC and FeO x in (a) dry PBL air and (b) at the highest altitude (6-8 km). The bell-shaped dP abs /dlogD m distributions for BC suggest that the BC-containing particles within the detectable D m domain (70 nmrD m r850 nm) predominantly contribute to the total BC absorption. On the other hand, the broad dP abs /dlogD m distributions of FeO x suggest that the unobserved FeO x -containing particles outside the detectable D m domain (170 nmrD m r2,100 nm) also contribute significantly to the total FeO x absorption. Figure 7 also shows the theoretical transmission efficiency curves (Tr(D m )) of our aerosol-sampling system for BC and FeO x particles. The steeply decreasing Tr(D m ) curve for FeO x suggests that we underestimated the dP abs /dlogD m value for FeO x at D m 4B600 nm.
The total shortwave absorbing power P abs (W m À 3 ) was computed by integrating equation (4) over D m : where D min and D max denote the lower and upper limits of the detectable D m domain, respectively. Figure 8a shows the altitude profiles of P abs (BC) and P abs (FeO x ). For both BC and FeO x , P abs values decreased monotonically with altitude, following their mass concentration profiles. For the dry PBL air, the mean P abs (BC) and P abs (FeO x ) values in the bare (coated) state were 4.0 (10.0) and 0.24 (0.44) mW m À 3 , respectively. These P abs values are equivalent to atmospheric heating rates of 0.29 (0.72) and 0.017 (0.032) K per day, respectively. Figure 8b shows the altitude profiles of the mean P abs (FeO x )/P abs (BC) ratio. In the boundary layer wherein anthropogenic FeO x dominated the detected FeO x particles (Fig. 5c), the mean P abs (FeO x )/P abs (BC) ratio was estimated to be 0.04-0.07. The altitude dependence of P abs (FeO x )/P abs (BC) was similar to that of the FeO x /BC mass concentration ratio (Fig. 5b). In the highest altitude (6-8 km), wherein the dust-like FeO x particles contributed 50% of the detected FeO x particles (Fig. 5c), the P abs attributable to anthropogenic FeO x may be B50% of the calculated P abs value. It should be noted that the P abs (FeO x )/P abs (BC) ratios reported herein are lower than the real values because our measurements underestimated the ambient FeO x mass concentration in the D m domain of B600 nmrD m r2,100 nm and totally ignored  the FeO x particles outside the detectable D m domain. Considering these facts, the real P abs (FeO x )/P abs (BC) ratios in the boundary layer would be as large as 0.1.

Discussion
On the basis of the previous observations of BrC near Beijing in March 2013 (refs 11,14), we roughly estimate the typical P abs (BrC)/P abs (BC) value in the boundary layer to be B0.1. Thus, we expect that P abs (FeO x ) is as large as P abs (BrC) in the East Asian continental outflows. In future studies, it is highly desirable to use an aerosol-sampling system with a higher transmission efficiency of large FeO x particles. In addition, some technical improvements in the SP2 are necessary to perform FeO x measurements beyond the current detectable D m domain (170 nmrD m r2,100 nm). Despite the room for such improvements, an essential conclusion can be drawn from our results, namely, in addition to BC and BrC, airborne FeO x in the form of aggregated magnetite nanoparticles should also be recognized as a significant anthropogenic contributor to shortwave atmospheric heating.
In addition to clear-sky shortwave absorption, we briefly discuss some other potential climate effects of anthropogenic FeO x particles. Low stratiform clouds are of frequent climatological occurrences over the mid-to high-latitude ocean and southeast China 32 . Under modest maximum supersaturation in such clouds, particle's mass fraction activated to cloud droplets will be substantially larger for FeO x than BC, because the critical supersaturations of FeO x -containing aerosols will be lower than those of BC due to the larger D m . Under these situations, P abs (FeO x )/P abs (BC) is enhanced because the lensing effect in droplets 33 is larger for FeO x than for BC. We expect that the number of co-emitted cloud condensation nuclei and their precursor gases in anthropogenic FeO x -rich sources, which remain uncertain, are substantially different from those in anthropogenic BC-and BrC-rich sources such as residential coal, industrial coal and biomass fuels 7 . Thus, without comprehensive investigations, it is not clear whether the net positive climate forcing of anthropogenic FeO x -rich sources is negligible or comparable with those of BC-and BrC-rich sources. Finally, it should be mentioned that the anthropogenic FeO x particles may also play a role in the biogeochemical cycles 34 .

Methods
Modified single-particle soot photometer. A modified single-particle soot photometer (SP2), which detects light-absorbing refractory aerosols on the basis of intra-cavity laser-induced incandescence, was used to measure BC and iron oxide FeO x particles 25 . The BC and FeO x are discriminated from each other on the basis of the colour ratio, which is an indicator of the boiling temperature 25 , and the peak amplitude of the blue-band incandescence signal, which is an indicator of an incandescing particle's size 25 . Supplementary Figure 7 shows a scatterplot of the peak amplitudes and colour ratios of all incandescing particles detected during the A-FORCE 2013 W campaign. The boundary lines for discriminating FeO x from BC were determined on the basis of experimental results 25 and are also shown in Supplementary Fig. 7.
The masses (m) of the individual FeO x and BC particles were determined from the peak heights of the incandescence signals using the experimentally determined mass-to-peak height relationships 25 . The mass-equivalent diameters (D m ) for BC and FeO x were calculated from the observed masses m assuming the void-free densities of 1.8 g cm À 3 and 5.17 g cm À 3 (density of magnetite), respectively. In this study, the detectable size domains of the BC and FeO x particles were 70 nmrD m r850 nm and 170 nmrD m r2100 nm, respectively.
The detailed morphological properties of the individual particles containing a particular mass of each incandescing material were evaluated on the basis of time-resolved scattering cross-sections C s (integrated over the solid angle of light collection) in a laser beam derived from the scattering signal [35][36][37] . We evaluated whether the incandescing material (BC or FeO x ) was attached to the surface of another particle on the basis of the magnitude of C s at the onset of the incandescence signal (C s-oi ) 28 . In our observations, the fraction of the attachedtype 28 BC-containing particles with D m ¼ 200 nm was o3% at any altitudes below 8 km. Thus, we assumed the C s-oi of BC-containing particles to be equivalent to the C s of BC core. For BC-containing particles, the ratio of C s before the onset of particle evaporation (C s-be ) to C s-oi is used as an indicator of the amount of non-refractory material (for example, sulfate) coating the BC 28 .
The timing of the onset of the incandescence signal (t oi ) is an indicator of the heating rate of an absorbing particle in the laser beam of the SP2; t oi tends to be lower (earlier) for more efficient light absorbers. Our experiments 25 showed that the t oi of magnetite (Fe 3 O 4 ) particles was markedly earlier than that of hematite (a-Fe 2 O 3 ), reflecting the greater absorption efficiency of magnetite. It should be noted that comparisons of t oi values derived from the laboratory experiments and field data are meaningful only under similar SP2 conditions (that is, the same laser power and width of the Gaussian beam) because the t oi of a particular composition depends on these parameters. Supplementary Fig. 4 shows the t oi distributions for the ambient BC and FeO x particles in dry PBL air. The SP2 laser power during the A-FORCE 2013W campaign was similar to that in our laboratory experiments 25 , as expected from the comparison of t oi À t cen (0.2 ms) for BC between ambient data ( À 55ot oi À t cen o À 30) and our laboratory results ( À 50ot oi À t cen o À 35; refer to Fig. 2c of Yoshida et al. 25 ). The t oi distributions of the FeO x particles for two different mass ranges (10 fgrmr88 fg and m 4 530 fg) were similar, indicating that the primary FeO x material does not change appreciably with FeO x particle mass ( Supplementary Fig. 4). Our experiment illustrated that the incandescing probability of pure hematite particles was zero for moB100 fg because the absorption efficiency of hematite particles in this size range is insufficient to heat the particles to the incandescing temperature 25 . Thus, the incandescing FeO x particles in dry PBL air had greater absorption efficiency than hematite. The t oi À t cen (0.2 ms) distributions of FeO x with 10 fgrmr88 fg and m4530 fg largely overlapped with the t oi À t cen distributions of pure magnetite with m oB100 fg ( À 40ot oi À t cen o À 20) 25 and m4530 fg ( À 40ot oi À t cen o À 30) 25 , respectively. On the basis of these results and the EELS spectra shown in Supplementary Fig. 3, the major constituent of the detected FeO x particles in dry PBL air is likely magnetite (Fe 3 O 4 ).
Electron microscopy analysis. An aerosol impactor-sampler 19 onboard the aircraft was used to collect aerosol samples on the Cu TEM grids with collodion substrate at 12-min intervals during each flight. A 120-kV transmission electron microscope (JEM-1400, JEOL) equipped with an energy-dispersive X-ray spectrometer (Oxford Instruments) was used for the STEM-EDS analysis. A 200-kV transmission electron microscope (ARM 200, JEOL) was used for the EELS analysis.
Transmission efficiency of aerosols. Special care is required to measure the number concentrations of supermicron-sized aerosols in ambient air as the transmission efficiency (Tr) through the entire tubing apparatus connecting the aerosol inlet to the particle detection volume unavoidably decreased because of the inertial and gravitational depositions. Since a substantially large fraction of the total FeO x mass was expected to be in the supermicron size range (Fig. 4), our conclusion is strongly dependent on the degree of sampling loss of large FeO x particles. According to the theoretical formulae compiled by Pramod et al. 38 , we estimated the Tr(D m ) curves of the BC and FeO x particles through our aerosol-sampling system. In our A-FORCE 2013W aircraft campaign, the sample air was aspirated through a forward-facing shrouded inlet (DMT Inc., Boulder, CO, USA) installed on the top of the aircraft fuselage. This aerosol inlet is a replica of 'University of Hawaii shrouded solid diffuser inlet' described and evaluated by McNaughton et al. 39 . An isokinetic aspiration was performed to maintain the aspiration efficiency at B1.0, independent of the particle size. The geometric specifications and flow rates of the tubing apparatuses comprising the sampling system are listed in Supplementary Table 1. The void-free densities of BC and FeO x were assumed to be 1.8 and 5.17 g cm À 3 , respectively. The dynamic shape factors of the BC and FeO x particles were assumed to be 1.5. Our experiment using pure magnetite particles confirmed the reasonable agreement between the measured and theoretical Tr(D m ) curves for a 1 4 -inch tube (length ¼ 0-3 m) assuming the same density and dynamic shape factor. orientations. When applying block-DDA to the scattering problems for fractal-like aggregates of BC or FeO x monomers, the monomer dipole formulation was assumed to avoid the discretization shape error in each monomer 46 . The coupled electric and magnetic dipoles formulation 46,47 was used to mitigate the multipoletruncation error associated with the monomer-dipole assumption 46 . To apply efficient fast Fourier transform-based algorithms in DDA 48 , we adjusted the center position of each monomer dipole to the nearest site of the computational cubic lattice (CL) 46 . In this study, the lattice spacing of the CL was set to one-half of the monomer diameter. The geometry of the coating material surrounding a fractal-like aggregate of spheres was computed by the following procedure. First, the CL sites in the proximity of each monomer (at a distance of less than T monomer radii) were selected as candidates for coating volume elements, where the coating thickness was controlled by the parameter T. In this study, T was selected to be 3. Next, the CL sites overlapping with monomer volume, if present, were removed from the candidates. Finally, the coating volume was assigned to each candidate.
We used the complex refractive indices of BC and FeO x listed in Supplementary Table 2. Ackerman and Toon 49 also used this refractive index data set to estimate the radiative effects of atmospheric aerosols containing BC and magnetite. The refractive index of the coating material was assumed to 1.5 þ 0.0i, independent of the wavelength.
Radiative transfer calculations. The spectral actinic fluxes F ac (l) under the A-FORCE 2013W condition were calculated using the radiative transfer software package libRadtran (version 1.6 beta) 50 . At each altitude, the absorption coefficients b abs (l) (m À 1 ) of the BC and FeO x particles were calculated on the basis of observational D m /dlogD m data and theoretical s a (l, D m ) values. The extinction coefficients b ext (l) were evaluated as b abs (l)/(1 À o), where the single-scattering albedo o was assumed to be 0.85, independent of l. The Henyey-Greenstein function with asymmetry parameter 0.7 was used for the scattering phase function. Shortwave absorption by gases (for example, water vapour and ozone) was calculated using the LOWTRAN/SBDART parameterization assuming the default atmospheric profiles (US-standard atmosphere). Effective solar zenith angle averaged over the daylight hours (local noon ± 6 h) was assumed for the actinic flux calculation. All the input parameters used in the radiative transfer calculations are listed in Supplementary Table 3.
Cloud and aerosol spectrometer probe. A cloud and aerosol spectrometer probe (CAS; Droplet Measurement Technologies, Inc.) installed under a major wing of the aircraft was used for measuring size-resolved number concentration of total aerosol in the light-scattering diameter (D p ) domain of 0.5 mmrD p r50 mm. The CAS instrument used herein was also described in Koike et al. 51 .
Code availability. The aggregate_gen code for computing the fractal-like aggregates of spheres and the block-DDA code for electromagnetic scattering were developed by the corresponding author (N.M.) and are available in the GitHub repository at: https://github.com/nmoteki.