Droplet Measurement Technology's ultra-high-sensitivity aerosol spectrometer (UHSAS) approaches this problem by using infrared (IR) light (1054 nm), keeping Mie scattering monotonic through the UHSAS maximum diameter of 1000 nm and meaning particles below 300 nm are within the Rayleigh regime, where scattering goes with the sixth power of diameter, giving a nice log-linear relationship over much of the sizing range. This has the additional benefit of suppressing the effect of shape: when the particle size is much smaller than wavelength, scattering is determined primarily by the volume of the particle rather than the cross section . It requires an intense laser to detect the smallest particles (∼ 1 kW circulating in the sample volume) and wide-ranging amplification to handle the >6 orders of magnitude scattering change over the 60 to 1000 nm nominal detection limits. To achieve this, the UHSAS uses a pair of detectors, one an avalanche photodiode and the other a low gain PIN photodiode. Each photodiode has two output channels with different gains, yielding four channels spanning the range of particle scattering.

The UHSAS sample flow rate can vary from 1 to 100 standardcm3min-1 (referenced to 298 K and 1013 hPa) and is aerodynamically focused with a nozzle of 500 µm diameter and a sheath flow typically about 700 cm3 min-3 through a 760 µm nozzle positioned within 1 mm of the laser. This jet creates a particle beam roughly 50 to 100 µm in diameter, which is much smaller than the 500 µm 1/e2 intensity diameter of the laser .

As with any optical particle sizer, the UHSAS is subject to sizing errors when the refractive index of sample particles is different than that of the calibration material. One can approach that by using calibration materials close to the refractive index of natural aerosol , use Mie scattering calculations and composition to correct calibrations for sampled aerosol , or conclude that the errors are small enough to ignore . But it appears that the UHSAS may have unique problems with black carbon. noted poor behavior of the UHSAS in a Mexican biomass burning (BB) plume and concluded that strongly absorbing particles are essentially invisible to the UHSAS.

Light-absorbing carbonaceous material (LAC) is one of the most difficult parts of aerosol to describe chemically, to measure, and to model . It is poorly defined chemically, tends to be highly aspherical, and is present in highly variable quantities. The plethora of names invented to describe it gives an indication of the complexity: soot, elemental carbon (EC), black carbon (BC), refractory black carbon (rBC), brown carbon (BrC), and tarballs all address subtly different properties. Historically, it has been practical to measure either the amount of carbon with chemical methods or the blackness with absorption measurements, but quantitatively connecting the two amidst the complexity of ambient aerosol remains a challenge .

A review of this is beyond the scope of this paper (see ), but a description of how the terms are used in this paper may be useful. BC is an umbrella term for carbon that absorbs strongly across spectrum from infrared through ultraviolet with a fairly weak (∼1/λ) wavelength dependence. The carbon atoms are largely sp2-bonded like graphite. It is refractory and insoluble in water and organic solvents . rBC refers to the LAC particles detected by a DMT single-particle soot photometer (SP2) or other laser-induced incandescence (LII) instruments, which heat IR-absorbing particles to vaporization temperatures, roughly 4000 K for graphite, where it is detected by incandescence . That clearly includes BC, but other carbonaceous species can masquerade as BC if they char as they heat . BrC is organic material that absorbs light primarily at short wavelengths and to a small and very poorly known extent in the IR e.g.,. Tarballs are a distinctively spherical variety of BrC prevalent in aged biomass burning plumes. Their optical properties are also poorly known, though it appears that they may absorb more in the IR than other BrC .

The shape of LAC particles poses another set of difficulties. Electron microscopy of particles freshly emitted from fires shows that soot particles are aggregations of tiny spherules with diameter ≤ 50 nm tacked together in branching, almost fractal structures. As those particles age, they collapse into more compact shapes, though they typically remain far from spherical. This non-sphericity affects aerodynamic drag, so mobility measurements overestimate particle size (compared to mass or volume-equivalent diameter) and changes optical properties (scattering and absorption) in ways that are not straightforward to determine (e.g., ).

ORACLES was a NASA-funded project to examine the direct, indirect, and semi-direct influence of aerosol from burning fields in Africa on the radiative balance over the southeast Atlantic Ocean . It was an aircraft-based project with field deployments in September 2016, August 2017, and October 2018. The NASA P-3B was deployed each year, with an extensive payload of aerosol, cloud droplet, radiation, and remote sensors. See https://espo.nasa.gov/oracles (last access: 11 November 2021) for a project description and links to data and other publications.

The Hawaii Group for Environmental Aerosol Research deployed a UHSAS in addition to a set of differential mobility analyzers (DMAs) to take advantage of the superior time resolution of the UHSAS (1 s or less compared with 90 s), particularly valuable during vertical profiles and when sampling in and around clouds. The UHSAS also offered detection limits of 60 nm, sufficient to detect nearly all particles likely to activate within clouds.

Aerosol in the ORACLES project was largely aged smoke from burning fields and forests in southern Africa. Plume ages were typically 2 d to 2 weeks, and the aerosol was dominated by organic material with substantial rBC. Data from the first 2 years showed that when compared to size distributions from the DMAs the UHSAS consistently sized particles too small.

During the final year, we installed tubing and valves to allow a UHSAS to sample size-selected particles from a DMA during flight. This allowed us to directly test any sizing anomalies due to the characteristics of the ambient aerosol. It was essentially an in-flight UHSAS calibration using particles representative of the plume. This kind of calibration while sampling is not new, as did essentially the same thing, but it is the first aircraft deployment we are aware of in a biomass burning plume. We also anticipated that by measuring scattering from aerosol of known size, we could determine refractive index as was done by , but on a single-particle basis.

Instead, as will be shown below, we got a dataset from the UHSAS that did not agree with scattering measurements or other sizing techniques, so in this work we explore the reasons for the poor sizing performance of the UHSAS and attempt to make the data as useful as possible by developing a correction scheme.

This test was only possible because the plume we were studying was vast ; we often spent a half hour or more in remarkably constant aerosol so there was plenty of time to try time-consuming procedures. This is normally impractical with fast-moving aircraft (∼100 ms-1) even considerably downwind of a fire.

In addition to the field deployment, we tested various materials in the laboratory, partly to familiarize ourselves with the UHSAS, but also to explore the sizing problems evident in the first 2 years. It was not practical to generate good proxies for the aged BB particles present in the field, so we tested some representative non-absorbing salts (NaCl, Na2SO4, and H2SO4) and two strongly absorbing materials containing refractory rBC : Aquadag, a suspension of graphite flakes, and fullerene soot, which is aggregations of 1–100 nm spherules which are ∼ 90 % amorphous black carbon

“Amorphous carbon” in this context is not the definition recommended by which refers to relatively small molecules in incipient or young soot; it is probably more like the turbostratic or polycrystalline graphite where graphitic regions are very small.

The aerosol sizing package aboard the NASA P-3B during ORACLES included two DMA systems, the UHSAS, a TSI 3321 aerodynamic particle sizer (APS), and a DMT SP2. In addition, a pair of TSI 3563 three-wavelength nephelometers measured aerosol light scattering, which is strongly related to particle size. The second nephelometer was in series with the first, and a 1 µm aerodynamic diameter impactor was periodically switched in between them. Sample air for all instruments was drawn through a shrouded inlet that samples aerosol particles with near 100 % efficiency to roughly 3µm .

Both DMA units used grab samplers to ensure constant size distributions through the 60 s scans. Modified versions of the software developed by controlled the systems and inverted the size distributions. One of the units was the modified TSI 3071 LDMA mentioned above. The other was a thermal tandem DMA system (TDMA) that used a nano-DMA (TSI 3085) and a radial DMA (RDMA) . It could scan with either DMA after passing sample air through unheated, 150 ∘C, or 300 ∘C thermal denuders, or it could be configured to select a mobility size with one DMA and scan the resulting particles with the other either directly or after heating to 300 ∘C . Sample air was not dried, but excess air from each DMA was desiccated and recycled as sheath air, so particles in the DMA were rapidly dried as they migrated through the sheath air, and sizing was effectively at low RH. Particles passing through the DMAs were detected with TSI 3010 condensation nuclei (CN) counters.

The UHSAS operated in slightly different configurations each year. In 2016, sample air was either unheated or passed through a 400 ∘C thermal denuder that eliminated volatile material . The valve system suffered from electrophoretic losses of small particles (due to short segments of insufficiently conductive tubing), so an empirical diameter-dependent correction was implemented by comparison with the long DMA in field data and confirmed with lab measurements with and without the denuder. A linear drop-off of passing efficiency below D0=21µm approximated the data tolerably well: E=1-3.667(D0-Dp), where Dp is the particle diameter. This correction, about a factor of 2 at 0.75 µm, has very minor effects on scattering calculations and aerosol mass but is important for cloud condensation nuclei and total particle number.

The denuder was absent in 2017 but reintroduced in 2018 with the phoretic loss problem fixed. In 2017, sample flow to the UHSAS was diluted 50:50 with desiccated zero air to ensure a low RH, reducing particle growth due to liquid water to a minimum. High RH was only rarely a problem, and only in the marine boundary layer (MBL), as the free troposphere (FT) had low water vapor content and sample air was heated to cabin temperature, so RH even before desiccation was < 52 % about 95 % of the time.

In 2018 the thermal denuder and desiccated dilution flow were used (though the volatility data are not explored here), and as mentioned above, a system was installed to divert size-selected particles from the TDMA to the UHSAS. The dilution flow system was bypassed during this mode of operation as RH was already low from the TDMA. The largest particles the RDMA could select at altitude were 180 nm; the LDMA could select 500 nm particles but was about 4m away from the UHSAS, so it was not practical to use for the in-flight calibration.

In the UHSAS, scattered light is focused on detectors on opposite sides of the scattering region. Each uses a identical set of optics consisting of a pair of Mangin mirrors that accept light scattered in an annular region between 14.8 and 57.0∘ from perpendicular to the beam (Fig. ). There must also be a direct 90∘ scattering path from the particle to the detectors that bypasses the mirrors, but that is a trivial fraction of the collected light and is ignored in these calculations.

The photodiodes are about 5 cm from the beam. If they are 1 cm in diameter (probably an overestimate) then they each subtend a solid angle of 0.031 sr, less than 1.2 % of the 2.65 sr collected by each set of mirrors.

Calculated UHSAS response to several types of particles are shown as lines in Fig. . Table  shows the corresponding refractive indices. The non-absorbing materials have slightly lower n than PSL, so they lie just underneath the almost completely hidden PSL trace. NaCl particles are somewhat oversized by the DMA since they are non-spherical; a size-dependent correction from for nebulized NaCl was used here. That correction may be counterproductive, as the non-sphericity enhances side scatter, but the corrections are relatively small. Na2SO4 also forms non-spherical particles, but we have not found corrections, so no mobility diameter corrections have been made. Another small potential issue is that sulfuric acid is intensely hygroscopic; there is inevitably some water in the particles despite the low humidity, so the refractive index is not exactly known.

In contrast to the non-absorbing salts, spheres with the refractive indices of fullerene soot and graphite have elevated n and k and scatter far more light up to about 600 nm. Coincidentally, the predicted UHSAS response to model soot particles calculated with MSTM lies almost atop the non-absorptive materials. The void space comes close to canceling the elevated refractive index.

Figure  shows the same scattering calculations recast to show how much bigger or smaller particles of different materials would appear when using a PSL calibration. This magnifies the difference between curves so one can see the complex behavior for particle diameters near the wavelength of the UHSAS laser.

The symbols in Figs.  and show the results of the lab tests. In general, the non-absorbing materials follow the expected curves fairly closely, though at diameters above the Rayleigh scattering regime, the data are a bit noisy, particularly between 0.7 and 0.9 µm. This is presumably due to a combination of non-spherical salt particles, Mie wiggles, and for the Na2SO4, a refractive index at the wrong wavelength. Overall, for this collection of non-absorbing particles, knowing the refractive index reduces sizing errors for particles < 300 nm from about 10 % to ∼ 5 %. For larger particles, the errors can exceed 20 %. It is curious that NaCl is systematically undersized despite the attempted correction, while H2SO4 is not. The most likely explanation for that is non-sphericity, since the H2SO4 particles were almost certainly spherical liquid. It may be that the nebulizer we used did not perform like the one used by or that the drying rates were different. The particle-to-particle variability of the NaCl particles was higher, yielding less sharp peaks, which might be another effect of non-spherical particles.

The two varieties of black carbon scatter far less light than the calculations suggest, from an order of magnitude for small particles to a factor of 100 for the largest particles. This results in undersizing from 30 % at 100 nm to 67 % at 900 nm. Dramatic undersizing of absorbing particles by OPCs has been noted before , but those were for particles much larger than the wavelength, and the undersizing was predicted from Mie theory. (In their case, the undersizing was not as extreme as predicted, which they attributed to non-sphericity.) It is striking that despite very different morphologies (flakes vs. spheres) the two types of rBC behaved very similarly.

One possible reason for undersizing is that particles shrink while heated by the UHSAS laser. That would not be entirely surprising, since the laser intensity is close to that of the SP2, which heats rBC to incandescence.

The biggest uncertainty in Eq. () is the accommodation coefficient α, which represents the odds that an air molecule will equilibrate with the temperature of the particle during a collision. The effect of α is illustrated in Fig. . Fully annealed graphite surfaces are very smooth and have 3αgraphite=0.28-3.23×10-5Tp+0.8e-1.53×10-3Tp×(0.175+5×10-4Ta), where Tp and Ta are particle and air temperatures, respectively . At room temperature, before the particle is heated, αgraphite is 0.25, but it drops to 0.05 at Tp=4000K. This inability to transfer energy to air molecules yields exceedingly high calculated temperatures. Even the smallest particles modeled would heat far beyond the ∼4000K vaporization temperature of graphite and would evaporate almost completely. But soot surfaces are unlikely to be fully annealed, and surface roughness is likely to be increased by crystal defects and adsorbed hydrogen, so α is likely to be higher. estimated that this increases α by 0.24, yielding α=0.29 at Tp=4000K. This drops ΔT by a factor of about 5 but still yields temperatures high enough that all soot particles should heat to boiling and shrink significantly through evaporation.

Model particles like those shown in Fig.  are sufficiently irregular that some incoming air molecules will bounce multiple times as they enter crevices, giving a high chance of equilibrating to the surface temperature. If one assumes that happens to a third of the air molecules incident on the particle, that would increase α to about 0.53, which we consider a reasonable estimate. That would mean particles above about 70 nm would heat sufficiently to begin losing mass.

Even if α=1, minimizing the temperature rise, particles as small as 120 nm would heat to their vaporization point. Larger particles, given enough time in the beam, would evaporate down to that diameter. Altogether, it appears very likely that the fullerene soot and Aquadag are undersized by the UHSAS because the particles evaporate and shrink as they enter the beam.

The actual processes involved are much more complex than the simple equilibrium calculations used here. Crucially, to exhibit reduced scattering, particles must have heated up and partially evaporated before hitting the strongest part of the beam. The peak scattering reflects the interplay between (i) a strengthening beam that increases scattering and energy delivered through absorption and (ii) a shrinking particle that absorbs and scatters less.

In the 2018 ORACLES deployment, we performed the DMA → UHSAS tests during 15 periods, using particles with (singly charged) mobility diameters between 70 and 180 nm. We usually chose four diameters and dwelled on each for 60s, and then the grab sampler would refill and another cycle would start.

A typical response looks like Fig. . The biggest peak is plainly the particles selected by the DMA as intended, with a single charge. The width of the fit reflects the transfer function of the DMA (a triangle with curves caused by diffusion), random variation in the UHSAS response, and variations in the scattering properties of the particles.

The peak at larger diameter is due to doubly charged particles. The RDMA could scan up to about 200 nm, depending on altitude; the largest-diameter well-defined peak we could get for singly charged particles was at 180 nm, and corresponding doubly charged particles were about 280 nm. (The Cunningham slip correction factor was included in the calculation, so mobility is not just the square root of diameter.) Thus we could calibrate only a small fraction of the UHSAS size range, but that included most of the particles in the accumulation mode.

To the left of the singly and doubly charged peaks, another mode, containing just a few particles, was present in all of the tests except when the DMA was selecting 80 nm or smaller particles. (That mode may have been present but at diameters below the detection limit of the UHSAS.) This mode appears to be undersized in much the same way as the Aquadag and fullerene particles in the lab calibrations. Thus, we think it most likely that these anomalously undersized particles are the ones containing rBC. This is discussed later, in Sect. .

It is obvious in Fig.  that the diameters reported by the UHSAS for both singly and doubly charged particles are considerably smaller than the mobility diameter. This pattern was consistent through all sizes of particles for each test we did. The top plot in Fig.  summarizes the results for all of the DMA → UHSAS tests in 2018. The bottom plot shows variability between tests was pretty small, varying no more than 3 nm from the average. This presumably reflects relatively constant composition through the duration of the 2018 deployment, though we did tend to perform the tests in extended level legs with substantial BB aerosol concentrations, thus perhaps biasing our samples toward constant composition.

We tested a variety of fitting functions with the z=1 and z=2 peaks (omitting the anomalous particles). The simplest function that lacked a strong pattern in the residuals was 4Dopt=a+bDmobc, where Dopt is the optical (UHSAS) diameter; Dmob is the mobility diameter from the DMA; and a, b, and c are the fitting parameters. Within the RDMA size range, 95 % confidence limits are within 3 nm, but of course the correction is highly speculative beyond that, from 300 nm to 1 µm. It is likely to be quite inaccurate for particles beyond the accumulation mode, which are usually chemically distinct from smaller particles, typically dust in the FT and sea salt in the MBL.

The undersizing of the z=1 and z=2 particles in Fig.  cannot be explained by the refractive index of the BB plume. For spherical, homogeneous, and non-absorbing particles it is straightforward to determine the refractive index required for particles of a known diameter to produce a given scattering signal. The results are shown in Fig. . The smallest particles produce scattering consistent with a refractive index of around 1.52, which is lower than usually reported but not unreasonable. But the apparent refractive index of larger particles drops rapidly to 1.35, far below any plausible aerosol material. Extrapolating to 0.5 µm gives a refractive index of 1.33, equal to that of water.

There are two obvious explanations for the undersizing. It is possible that the particles are sufficiently aspherical that their behavior is even more pronounced than in the laboratory NaCl tests. That seems unlikely for non-rBC particles since they are a product of vapor deposition to nucleation mode particles but cannot be excluded. A more likely cause of the undersizing is evaporation of volatile material from particles as they absorb heat from the UHSAS laser. That is consistent with the exaggerated undersizing for larger particles, which are likely to get hotter, as seen in Eq. () and Fig. . But what is absorbing IR light from the laser, and how warm do the particles need to get to shrink by 15 %?

The thermal denuders on the radial DMA in the TDMA showed that heating to 150 ∘C had little effect, but heating to 300 ∘C reduced particle diameters considerably. The actual mass loss was not practical to quantify, since the radial DMA could not exceed 200 nm. However, in the tandem DMA tests, when particles were selected with the radial DMA, heated to 300 ∘C, and then scanned with the nano DMA, particles typically lost 30% of their mobility diameter, though the range was pretty large (Fig. ). Since the undersizing of the UHSAS was ≤ 15 % (Fig. ), it is clear that heating to 300 ∘C would be quite sufficient to explain the results. It is curious that volatility depended very little on particle diameter, suggesting that while individual particles at a given diameter were externally mixed, with a range of volatility and hence chemical composition, particle composition differed little across the size ranges reached by the TDMA.

The only significant absorbing materials commonly present in aerosol particles are rBC, dust, and BrC. Since we think the rBC-containing particles appear as the anomalously small mode (see Sect. ), and dust was only occasionally present and is typically on larger particles, BrC is the obvious remaining possibility. We had assumed that brown carbon would be unlikely to heat this way, as BrC absorption in the IR is weak and is typically considered trivial at long wavelengths (e.g., ).

Heating calculated from Eq. () for a few reported refractive indices of BrC are shown in Fig. . The soot aerosol from Fig.  is shown for comparison, though the void volume of natural rBC is likely to be different than of the model particles. It is not clear what accommodation coefficient is most appropriate. found that α for cloud droplets was indistinguishable from 1, but that may not be the case for organic-rich BB particles at low humidity, so the ranges of calculated heating from α=0.5 to 1.0 are shown. A 200 nm particle with the reported refractive index of African biomass burning (SAFARI 2000) would reach about 300 ∘C, which is sufficient to evaporate sulfates and a large fraction of organic material and is consistent with the TDMA data. But note that few individual particles are likely to have that refractive index; the real aerosol population is likely to have some particles that contain LAC and thus have much higher absorption while particles lacking LAC absorb little.

IR absorption by BrC has been measured in a few studies. While found negligible absorption for wood tar aerosol above 550 nm, did find a small amount of absorption, k=0.002±0.005, for 1047 nm light in smoke from smoldering peat fires in a laboratory. That absorption is only sufficient to heat 200 nm particles by about 60 to 100 ∘C (Fig. ), which would not significantly shrink the particles. It is not clear what relationship the BrC in either circumstance has to the ORACLES aerosol. During ORACLES and the downwind experiment CLARIFY held in 2017, the relatively low absorption Ångström exponent of 1–1.5 between 470 and 660 nm indicates that BrC contributions to overall absorption at visible wavelengths were small but does not exclude some IR absorption.

Tarballs are another possibility. They appear to be ubiquitous in biomass burning plumes , including those in southern Africa . Their incidence increases as plumes age over a few hours, but their size does not, implying that they are primary particles undergoing photochemical aging rather than secondary material condensing . There is essentially no information available about how they age over the 2 d to 2-week period experienced by the ORACLES plume, but there were some tarballs noted in filter samples taken aboard the plane in 2018 (Michal Segal-Rozenhaimer, personal communication, 2020). There are only a couple of reported complex refractive indices for tarballs at IR wavelengths; they are shown in Table . In addition, while not offering a refractive index for diesel-derived tar, found a mass absorption coefficient 1/23 as high as rBC in 950 nm light, suggesting that tarballs absorb more effectively than BrC. Figure  shows that tarballs are likely to heat between a few hundred and several thousand degrees, sufficient to evaporate most volatile species in the aerosol and perhaps the tarballs themselves.

There are conflicting reports about the behavior of tarballs in SP2s, presumably because of chemical differences or laser strength: found that tarball-like LAC from lab and forest fires did not incandesce, indicating that they either absorb little at SP2 (and UHSAS) wavelengths or they evaporate before becoming detectable by incandescence (∼ 3000 K). In contrast, found that pyrolysis-generated tarballs charred and incandesced but that laser power had a strong effect on charring of other organics (and might for tarballs). and found that tarballs from detuned diesel engines evaporated but only a small fraction incandesced, and they concluded that vaporization temperatures were between 1000 and 3000 K. It is clear that in the UHSAS, tarballs will heat, but the extent of volatilization is uncertain.

One problem with crediting the UHSAS undersizing to tarballs is that there are only two obvious modes in the DMA → UHSAS tests: moderate undersizing and severe undersizing. But if tarballs are numerous but not dominant, there must be three different types of particles with widely varying absorption properties: tarballs, rBC-containing particles, and others. To see only two modes, one of the following must be true.

Tarballs behave like other BrC, with similar absorption in the IR, heating, and particle shrinkage. This is not consistent with reported refractive indices for tarballs or BrC, but those are not well known.

There are some other possible explanations of the modest undersizing of the majority of particles that are interesting and are likely to occur on occasion but do not appear likely to explain the presence of the two modes. There could be rBC too small for the SP2 to detect but sufficient to heat particles to a few hundred degrees. Iron has been detected in some of the ORACLES particles, and some crystal forms absorb in the IR. In particular, magnetite (Fe3O4) has a refractive index of 2.1132–0.35780i at 1054 nm , so even a small fraction of magnetite on a particle would heat considerably. In contrast, hematite (Fe2O3), a more common material, has m=2.7574–0.0110i and would not heat significantly. Another possibility is that as particles containing both rBC and non-absorbing organic material heat, mechanical stresses force the rBC to separate from the rest of the particle. The non-absorbing remainder would be smaller but would not shrink any further in the beam . This is particularly easy to imagine if the rBC is in a clump attached to the side of a host particle.

Since the DMA → UHSAS data extend only to 0.28 µm, any extension past that is speculation. When plotting volume distributions, the accumulation mode rarely extended past 0.6 µm. Above that, particles are more likely to be sea salt (in the MBL), dust (in the FT), or sulfate (in a non-dusty FT). In Fig. , we show a correction that extends the fit line to 0.6 µm and then deviates along a straight line to the expected diameter ratio for particles with n=1.54, similar to NaCl, a little above (NH4)2SO4, and below silicate minerals. This is just a crude way to suppress really exaggerated oversizing for the non-BB particles larger than the accumulation mode and should not be considered realistic.

Figure  shows the effect of applying that correction to some selected size distributions. Uncorrected and corrected UHSAS distributions are compared with LDMA distributions in the pollution layer, the MBL, and clean FT. Data from 2016 and 2017 were chosen to demonstrate that even though the correction was based on data from 2018, when BB concentrations were lowest, it is appropriate to use the correction for earlier data in the polluted FT. The top-left plot, from a period within the BB plume, shows that the correction makes the UHSAS agree quite well with the LDMA. The effect on the volume distribution, top right, is dramatic, as the uncorrected data are far below the LDMA distribution. The correction is plainly inappropriate in a clean MBL sample, where the correction overestimates peak particle diameter and accumulation mode volume and is far too high near 1 µm compared with the APS. (Without the transition to NaCl refractive index, it is far worse.) In the clean troposphere, the number peak is at sufficiently small diameters that the correction makes very little difference, so it does not matter much whether the correction is used.

When applied to accumulation mode particles in the FT plume, the overall effect of the correction is to increase the median diameter by 13 to 18 nm. In 2018, the least polluted year, the median diameter rose from 143 to 156 nm, while for 2017, median diameter went from 163 to 181 nm with the correction applied.

The danger of extrapolating the correction to diameters above 600 nm even with the arbitrary reversion to n=1.54 is obvious in the volume distributions shown on the right side of Fig. . Comparisons between UHSAS and APS are not straightforward and are highly sensitive to assumed density, but it is clear that in each of these cases the corrected UHSAS is considerably higher than the APS between 0.6 and 1 µm.

In contrast, scattering emphasizes the large particles less, so the correction fares considerably better, as can be seen in Fig. , where scattering calculated from the UHSAS distributions is compared to the TSI nephelometer. The correction increases calculated scattering by a factor of about 2.3 in each year, improving agreement considerably. The nephelometers sense scattering between 7 and 170∘ so the scattering from the corrected UHSAS size distributions was integrated over that range of angles rather than the full 0∘ to 180∘ range. This is a more direct comparison than correcting the nephelometer data for truncation errors. See Appendix  for some details. Total rather than sub-micrometer data are shown because scattering was almost always dominated by the accumulation mode, so the total and submicron nephelometers usually agreed well and over the course of the project we had more data from the total nephelometer. This means the comparison is worse at low altitude, when coarse sea salt particles can account for considerable scattering. This effect is clear, as the dark blue low-altitude points are concentrated at the bottom of the data clouds.

While encouraging, the scattering calculation is subject to some major uncertainties, so it may have compensating errors.

Since this nephelometer did not have an impactor removing sub-micrometer particles, when dust or sea salt were present they increase scattering but are missed by the UHSAS. This gives underestimates in the marine boundary layer and on the rare occasions with significant dust.

The particles sized anomalously small in Figs.  and are a bit of a mystery. The obvious explanation is that they are rBC particles and behave the same way as the lab tests of fullerene soot and Aquadag (Fig. ). But the SP2 data indicate that rBC was typically coated with a thick layer of volatile material. In one sense, that makes extensive shrinkage easier to understand – temperatures of 300 ∘C are sufficient (Fig. ) rather than the 4000 ∘C required to evaporate graphite. But in that case, these shrunken cores ought to be dramatically undersized, as the Aquadag and fullerene soot were. Instead, they look about the same. That may be due to the energy required to evaporate the coating: it may take long enough that the particle has not yet fully shrunk by the time the particle is in the center of the beam. See for a nice illustration of that process in the SP2.

It is possible that at least some of these anomalously small particles contain no BC at all but absorb enough IR that organic materials char and become rBC, which then absorbs more IR, heating the particle enough to shrink it a lot. This seems particularly likely for tarballs, as mentioned above. Such charring behavior has been identified in the lab by , who found that nigrosin particles appeared to an SP2 as rBC with approximately 40 % of the mass of the original particles. That works out to about 70 % of the original diameter, roughly consistent with our anomalous particles (Fig. ). There were also indications of this in ORACLES when we made a short-lived attempt in 2017 to explore whether volatile coatings on particles enhanced absorption. We planned to test this by passing particles through a 400 ∘C denuder on the way to one of the two PSAPs (three-wavelength particle soot absorption photometers) on board, anticipating that evaporating some of the coating would reduce absorption. However, the heat treatment increased absorption rather than decreasing it, indicating that charring was taking place.

It would be useful to send size-selected particles to the SP2 and UHSAS simultaneously in future experiments; that would allow us to directly measure the fraction of particles at each size that contain rBC and how much rBC mass was contained in those particles.

The fraction of these anomalous particles is a strong function of particle size (Fig. ). Below 100 nm, they are either absent or too small to see, but at 180 nm they are over 20 % of the z=1 peak. There were undoubtedly undersized doubly charged particles, but they are hidden by the singly charged particle peaks. This may mean the number of singly charged particles is an overestimate and hence the fraction of low scatterers is larger than evident in Fig. .

There is also general dependence on altitude, with particles from >3.5 km tending to have fewer anomalous particles than those from lower altitude, though that pattern is a bit noisy. It may be related to other chemical differences we saw with altitude reflecting plume aging , but that will require further exploration.

While we cannot prove that the anomalous particles are the ones containing rBC, it is roughly consistent with the overall fraction of rBC-bearing particles shown in Fig. . The drastically undersized particles average only 15 % to 20 % in the DMA → UHSAS tests, but that may be an underestimate, as mentioned above. In addition, the accumulation mode median particle diameter was roughly 170 nm, averaged over all 3 years, so the data shown in Fig.  only cover half of the size distribution, and the relatively high points at the right are the most representative.

Figure  also appears to shed some light on the patterns seen in the scattering closure calculations (Fig. ). Since the anomalous particles are sized roughly 30 % small, scattering calculated from them is less than half of the real value. In 2018, that affected 10 %–17 % of the particles, and the calculated scattering was ∼ 20 % low. In 2017, by contrast, roughly 30 % of the particles had rBC, and calculated scattering was 30 % low. The results for 2016 were in between.