Interactive comment on “ Multi-wavelength optical measurement to enhance thermal / optical analysis for carbonaceous aerosol ” by L .-W . A

We thanks for the reviewers’ comments. A table of abbreviation will be added to the supplemental information, along with pictures of the multi-wavelength module and transfer standards for retrofit. Figure 3 is expanded to include comparisons of OCT and ECT, and LRλ and LTλ are specified clearly in the new Figure S1. The reason why EC2 is used in developing the τa,λ−ATNλ relationship is that for diesel soot, carbon evolved during EC2 is exclusively EC. Therefore, dτa,λ/dt can be derived from carbon measurement (Eqs. [4]-[6]). It should also be noted that EC1 and EC3 are very minor for the diesel exhaust samples tested (e.g., Figure S1[d]). For other samples, EC1 may not be entirely EC due to the existence of POC with an unknown mass


Introduction
Thermal/optical analysis (TOA) quantifies particulate-matter (PM) organic carbon (OC) and elemental carbon (EC) collected on quartz-fiber filters (Watson et al., 2005;Cao et al., 2007;Bougiatioti et al., 2013).TOA based on the IM-PROVE_A protocol (Chow et al., 2007a(Chow et al., , 2011) has determined OC and EC concentrations in tens of thousands of samples each year from long-term chemical speciation networks operated in the US (IMPROVE, 2014; U.S. EPA, 2014), Canada (Dabek-Zlotorzynska et al., 2011), and China (Zhang et al., 2012).IMPROVE_A specifies stepped heating up to 580 • C in an inert helium (He) atmosphere (> 99.99 % purity), where most organic compounds are either evaporated or decomposed (Chow et al., 1993), followed by a second stage of stepped heating to 840 • C in 98 % He / 2 % O 2 to remove EC on the filter.Since some of the OC is converted to EC through pyrolysis in pure He, as evidenced by darkening of the filter, IMPROVE_A also specifies a reflectance pyrolysis adjustment.Reflected light at wavelength λ = 633 nm is monitored throughout the heating (Huntzicker et al., 1982).OC and EC are defined as carbon evolved before and after the filter reflectance (R) returns to its initial level, respectively.
In addition to reflectance, some TOA protocols use transmitted light (T) to monitor the pyrolysis (Birch and Cary, 1996;Turpin et al., 1990).ECs based on R or T splits are referred to as ECR and ECT, respectively.In addition to particle deposits, adsorbed organic vapors within the quartz-fiber filter (Chow et al., 2010;Watson et al., 2009) can pyrolyze during the analysis (Yang and Yu, 2002;Chow et al., 2004).ECR differs from ECT since the R signal is dominated by pyrolyzed OC (POC) on the filter surface, while the T sig-Published by Copernicus Publications on behalf of the European Geosciences Union.
L.-W. A. Chen et al.: Multi-wavelength optical measurement nal is influenced by POC both on and within the filter (Chen et al., 2004;Chow et al., 2004).Unlike IMPROVE_A, which reports both ECR and ECT, other TOA protocols employ different temperature steps, often reporting only the ECT results (Birch and Cary, 1996;Cavalli et al., 2010;NIOSH, 1999;Schauer et al., 2003;Peterson and Richards, 2002).
The R and/or T measurements as part of TOA can infer the light absorption coefficient (b abs ), analogous to the principle of optical absorption monitors such as the aethalometer (Hansen et al., 1984), particle-soot absorption photometer (PSAP; Bond et al., 1999), and the multi-angle absorption photometer (MAAP; Petzold and Schönlinner, 2004).Both the aethalometer and PSAP apply T attenuation, while the MAAP incorporates both R and T attenuations in the calculation of b abs .In any case, it is necessary to compensate for multiple-scattering and loading effects of the particlefilter matrix that cause deviations from the simple Beer's law (Chen et al., 2004;Arnott et al., 2005b;Virkkula et al., 2005).Black-carbon (BC) concentrations can be derived from b abs by applying a mass-and wavelength-specific absorption efficiency (MAE λ , typically in m 2 g −1 ).Many collocated measurements show high correlations but different slopes in BC/EC comparisons (Ahmed et al., 2009;Quincey et al., 2009;Reisinger et al., 2008;Snyder and Schauer, 2007;Chow et al., 2009).This confirms the overlapping concept of EC and BC but also signifies the complex nature of carbonaceous material and uncertainties in such measurements (Andreae and Gelencséer, 2006;Moosmüller et al., 2009;Petzold et al., 2013;Lack et al., 2014).
While BC (or EC) absorbs light strongly across the solar spectrum (300-1000 nm), some organic compounds that evolve in the OC step can also absorb light, especially at shorter wavelengths (< 600 nm).These compounds have been termed "brown carbon" (BrC) and are associated with the smoldering phase of biomass burning and some end products of secondary aerosol formation (Andreae and Gelencsér, 2006;Clarke et al., 2007;Zhang et al., 2011).The spectral dependence of b abs is often described by α: where α(λ) is the absorption Ångström exponent.For BC (or EC) with graphitic-like structure and a constant refractive index, α = 1 and b abs is proportional to λ −1 .For BrC and mineral dust, α varies with λ and is mostly > 1 (Moosmüller et al., 2009;Chen et al., 2015), causing b abs to increase more rapidly towards shorter wavelengths (blue and ultraviolet) than is the case for BC.The aerosol deposit thus appears to be brown -or sometimes yellow, red, or chartreuse -as the longer wavelengths of illuminating light are reflected and the shorter wavelengths are absorbed.The multi-wavelength aethalometer has revealed different spectral patterns that are indicative of BC, BrC, dust, and their mixtures (Sandradewi et al., 2008;Favez et al., 2009;Yang et al., 2009).
Given the large number of samples per year analyzed by TOA worldwide, the optical data acquired as part of the analysis could be used, in addition to OC and EC, for studies relevant to source apportionment, human health, visibility, and climate.Described and characterized here is a retrofit of a TOA carbon analyzer that expands the singlewavelength R and T monitoring to seven wavelengths for the IMPROVE_A analysis, hereafter designated thermal spectral analysis (TSA).Equivalence of the OC and EC fractions from TOA and TSA is demonstrated for several source and receptor samples, and the wavelength dependence of the OC-EC split is investigated.An approach to report spectral b abs (as absorption optical depth) and α for decoupling the BC and non-BC components is also introduced.As an aid to readers, all abbreviations used in this paper are listed in the Supplement (Table S1).

Instrument design and calibration
The 633 nm He/neon (Ne) laser in the DRI Model 2001 carbon analyzer (Chow et al., 2011;Chen et al., 2012) is replaced with a package of seven diode lasers with wavelengths (λ) of 405, 450, 532, 635, 780, 808, and 980 nm.The use of diode lasers substantially reduces the cost of the optical module.It also provides stronger signals than using a lightemitting diode (LED) in other designs (e.g., Hadley et al., 2008), despite being limited by the wavelengths commercially available.While the 635 nm approximates the He/Ne laser, other wavelengths were selected to cover the visible and near-infrared regions.The lasers are alternately pulsed (2 consecutive pulses per laser, 14 pulses per cycle) and lockin amplified at a frequency of 30 Hz, resulting in two cycles and ∼ four pulses for each wavelength every second.The bifurcated fiber optic for delivering the He / Ne laser to the reflectance light pipe is replaced with an eight-furcated opticalfiber bundle, one for each of the lasers and the last for transferring the light reflected from the filter punch (0.5 cm 2 ) to a photodiode (Fig. 1).Another light pipe on the opposite side of the filter directs the transmitted light toward a separate photodiode detector.
Photodiode signals are acquired with an NI6216 data acquisition system (National Instruments, Austin, TX) at a rate of up to 100 000 data points per second.The system integrates the product of photodiode and reference (30 Hz square wave) signals every second to suppress noise (e.g., from 60 Hz power supply and oven glow, random noise, and baseline drift).The resulting integrals are reported as the spectral laser reflectance and transmittance (LR λ and LT λ , respectively).Example thermograms with LR λ and LT λ are illustrated in Fig. S1 of the Supplement.
LR λ and LT λ are relative terms depending on not only the optical properties of the sample but also laser intensity, the geometry of the laser/detector setup, and the response of the photodiode.They are related to absolute fil- ter reflectance and transmittance (FR λ and FT λ , i.e., fraction of light reflected from or transmitted through the filter, respectively), using a set of transfer standards consisting of eight PM 2.5 quartz-fiber filters acquired using high-volume samplers from the Fresno Supersite (Watson et al., 2000;Chow et al., 2009), with EC loadings ranging from 0.9 to 15.8 µg cm −2 (Fig. S2).The high-volume filters (406 cm 2 ) represent well-characterized urban aerosol, while providing sufficient sample for extensive testing.FR λ and FT λ of the transfer standards were predetermined using an integrating-sphere spectrometer (Lambda 35, Perkin Elmer, Massachusetts, USA; see, e.g., Zhong and Jang, 2011;Chen et al., 2015) traceable to NIST (National Institute of Standard and Technology) standards for wavelengths between 300 and 1000 nm.
Figure 2 compares LR λ and LT λ with FR λ and FT λ , respectively, for the transfer standards.Within the range of FR λ (0.1-0.8) and FT λ (0.0001-0.1), highly significant linear relationships (r 2 > 0.9) demonstrate the stability of the LR λ and LT λ measurements and the feasibility of converting them to FR λ and FT λ through calibration with a standardized spectrometer.Figure S1 illustrates the changes in FR λ and FT λ , retrieved from LR λ and LT λ , throughout a typical TSA.The uncertainty of the FR λ and FT λ retrievals may be evaluated by the coefficient of variance of the root mean square residual (CV-RMSR) in the FR λ / LR λ and FT λ / LT λ regression, respectively: where Y obs is the FR λ or FT λ measured by the integratingsphere spectrometer, Y cal is the FR λ or FT λ calculated from LR λ or LT λ , respectively, and n is the number of samples.FR 633 nm and FT 633 nm based on the He / Ne laser of a typical Model 2001 analyzer exhibit a ∼ 3 % uncertainty (Fig. 2).For this retrofit, FR λ uncertainties range from 3 to 11 % and FT λ uncertainties range from 5 to 18 %, with the best precision shown at 450 and 808 nm.The divergence in uncertainty is attributed to the quality of the laser and the sensitivity of the photodiode detector for different wavelengths.Ongoing efforts to improve the signal-to-noise ratio of R and T measurements by using different lasers and averaging algorithms will be documented in subsequent papers.

Consistency of the OC-EC split
Several source and ambient PM 2.5 samples (Table 1) are used for testing the system.Samples from the Fresno Supersite (2 April 2003-28 December 2003) contain a mixture of carbonaceous materials from fresh engine exhaust, biomass burning, and cooking (Chen et al., 2007;Chow et al., 2007b;Gorin et al., 2006;Schauer and Cass, 2000).Reno ambient samples were acquired during the period of Rim Fire (17 August 2013-24 October 2013) and are dominated by an aged biomass burning plume containing mixed flaming and smoldering emissions.Source testing as part of the Lake Tahoe Prescribed Burning Study (Malamakal et al., 2013) and Gasoline/Diesel Split Study (Fujita et al., 2007) provided pure biomass burning and diesel exhaust samples, respec- tively.All these samples were analyzed by both TSA (using the retrofit) and TOA (using conventional Model 2001 analyzers) following the IMPROVE_A protocol.Table 1 compares total carbon (TC), OC, and EC by reflectance (i.e., OCR and ECR) and transmittance (i.e., OCT and ECT) between TSA with the 635 nm and TOA with the normal 633 nm OC-EC split.As expected, TC is equivalent, with the averages agreeing within ± 5 % and regression slopes (m) ranging from 0.91 and 1.06 for each of the four sample types.The relative difference (RD), defined as the ratio of the difference divided by the average of two measurements (i.e., TSA and TOA) on the same sample, does not differ from 0 at the 5 % significance level (p > 0.05).The standard deviations of RD, a measure of random error, are higher for source (14-15 %) than for ambient (4-9 %) samples, indicative of greater deposit inhomogeneity for these samples, possibly due to variable sampling conditions over short sampling durations.
TSA and TOA also yield statistically equivalent OC and EC results, either by R or T (Table 1).Figure 3 visualizes the comparisons.With respect to the standard deviation of RD, OCR and OCT are similar to TC, while ECR and ECT are higher (up to 33 %) due to a lower fraction of EC in TC.By category average, ECR and ECT account for 10-50 and 5-46 % of TC, respectively.In general, ECR > ECT, as reported in previous studies (Khan et al., 2012;Han et al., 2013 et al., 2001;Schmid et al., 2001), consistent with organic vapors pyrolyzed within the filter leaving the sample after native EC and POC in the surface deposit have evolved (Chen et al., 2004).POC was least apparent for the diesel exhaust samples where optical adjustments were negligible.The basic assumptions for optical adjustment include the following: (1) OC does not absorb light and (2) POC has the same apparent MAE as EC.Within-the-filter POC shows no apparent MAE by R (i.e., cannot detect it for any wavelengths) but high MAEs by T (higher than EC due to a multiple-scattering effect (Chow et al., 2004;Subramanian et al., 2006)).This leads to divergent ECR and ECT results.Since the multiple-scattering effect is larger for shorter wavelengths, ECT is expected to decrease with the wavelength at which the split is made.On the other hand, Chow et al. (1993) observed that the operational definition for EC by any TOA protocol might contain some light-absorbing OC.BrC, if present, would lower the baselines of R and T (prior to thermal analysis) for shorter wavelengths, resulting in earlier split points and larger EC reported than that from longer wavelengths.
Table 2 compares OC-EC splits for 635 nm with splits derived from shorter (450 nm) and longer (808 nm) wave- lengths that have the lowest LR or LT uncertainties.ECRs based on the 635 nm split agree well with those based on the 808 nm split (i.e., slopes within the standard error from unity and p(RD) > 0.02) for all the sample types.This demonstrates that the 633 nm R split commonly used for optical adjustment since Huntzicker et al. (1982) does not significantly respond to BrC absorption compared with longer visible or infrared wavelengths.For ECT, the 880 nm split yielded higher values than the 635 nm split, especially for biomass-burning-dominated samples.This is consistent with ECT being influenced by POC within the filter and decreasing ECT with decreasing wavelength.
The 450 nm split appears to be sensitive to BrC content, as evidenced by larger ECR than those determined with 635 nm (except for diesel exhaust samples, where BrC contents are low).The difference is largest for the aged Reno wildfire samples, which may also contain some secondary organic aerosol formed during transport over the 200 km distance between Yosemite National Park and Reno, Nevada.As demonstrated in Fig. S1, shorter wavelengths produce earlier splits, resulting in higher ECR concentrations.The increase of ECT due to a 450 nm optical correction is not evident, which may be due to the opposing effects of BrC and POC at the shorter wavelength.

Multi-wavelength absorption retrieval
Light absorption by particles on the filter is often estimated by transmittance attenuation (ATN): where i and f indicate FT λ measured before and after thermal analysis, respectively.FT λ,f approximates a blank filter since all of the carbon has been removed.ATN λ includes scattering and absorption within the substrate.On the other hand, the absorption optical depth (τ a,λ ) measures only the light absorption.For diesel soot samples with negligible BrC and POC, where [EC] is the areal concentration of EC on filter in µg cm −2 .If there were no filter effects, ATN and τ a would be identical for a given λ as described by Beer's law.With the knowledge of sampling volume (V ) and filter area (A), ambient b abs can be calculated from τ a × A/V .The relationship of [EC] and ATN 635 nm throughout the EC2 step (740 • C in a 98 % He / 2 % O 2 atmosphere) of the IMPROVE_A analysis for a diesel exhaust sample is shown in Fig. 4. Since in this case carbon evolved during EC2 is exclusively EC, the temporal variation in [EC], i.e., d[EC] / dt, can be determined from carbon released and detected by the carbon analyzer.Arnott et al. (2005b) proposed a quadratic relationship between [EC] and ATN, derived from and thus where M λ and β λ account for the wavelength-specific multiple-scattering and loading effects, respectively.Eq. ( 6) fits the relationship in Fig. 4 well (r 2 > 0.99), thereby allowing M λ and β λ to be estimated.EC (quantified by IM-PROVE_A ECR or ECT) of all the diesel exhaust samples exhibits a consistent dependence on the initial sample ATN 635 nm (prior to TSA) (Fig. 4).This supports the use of ATN 635 nm as a surrogate for EC and light absorption.Simi-   740• C in a 98 % He / 2 % O 2 atmosphere) of IMPROVE_A analysis for a diesel exhaust sample (CIFQ074 from the Gas/Diesel Split Study (Fujita et al., 2007)).Also shown are paired [EC]-ATN 635 nm of 11 diesel exhaust samples, where EC is determined from IMPROVE_A and ATN 635 nm from initial filter transmittance.Circles and error bars indicate the average and spread, respectively, of EC by transmittance and reflectance (i.e., ECT and ECR, respectively).τ a,635 nm was further calculated from a MAE of 7.4 m 2 g −1 × [EC], and the regression result between τ a,635 nm and ATN 635 nm is shown in green.lar relationships hold between [EC] and ATN for other wavelengths (Fig. S3).The largest scatter observed for 532 nm corresponds to the highest uncertainty in FT 532 nm retrievals; it results from the slow laser response time to modulation and the low sensitivity of the transmittance photodiode detector at this wavelength, which will be addressed in a subsequent design of the retrofit optics.
To relate τ a,635 nm to ATN 635 nm , a MAE (EC) of 7.4 m 2 g −1 at 635 nm was used.The MAE was derived from concurrent b abs , by a photoacoustic sensor at 1047 nm, and EC, by IMPROVE_A, measurements during the Gasoline/Diesel Split Study (Arnott et al., 2005a) and assuming an α of 1 in Eq. ( 1).The quadratic relationship holds for all of the available wavelengths, as seen in Fig. 5. Light absorption is enhanced by the filter since any τ a,λ < 1 corresponds to a larger ATN λ , with more amplification towards shorter wavelengths.These τ a,λ -ATN λ relationships would apply to any samples, regardless of the nature of light-absorbing material (e.g., EC, BC, BrC, or mineral dust).Particle penetration depth and, to a lesser degree, single-scattering albedo and the asymmetry g factor can influence the τ a,λ -ATN λ dependence, though for a similar sampling configuration and filter material/thickness with typical loading, the perturbation is expected to be small (Chen et al., 2004;Arnott et al., 2005b).Retrieval algorithms employing both R and T such as those in Petzold and Schönlinner (2004) should be developed in the future to utilize all the information available.Using the relationships in Fig. 5, τ a,λ was calculated for all samples from the initially measured ATN λ values, with the average τ a,λ × λ by sample type compared in Fig. 6.The nearly constant τ a,λ × λ for diesel exhaust samples, i.e., τ a,λ ∝ λ −1 , is consistent with the exclusive contribution of EC to light absorption.BC derived from the τ a,λ would be equivalent to diesel EC.Averaged τ a,λ × λ increases by factors of 1.5, 3.0, and 3.9 from 980 to 405 nm for the Fresno ambient, Reno wildfire, and prescribed-burning samples, respectively.This reflects different levels of non-EC contribution.

Separation of BC and BrC contributions
A simplified two-component model consisting of BC and BrC, each with explicit absorption Ångström exponents (α BC and α BrC ), is used to explain the spectral dependence of τ a,λ in the samples: where q BC and q BrC are fitting coefficients.This is analogous to the approach of Sandradewi et al. (2008), who considered the two components to be traffic and wood-burning particles and Hadley et al. (2008), who modeled two components of BC and char.Assuming an α BC of 1 the same as diesel EC, then τ a,λ × λ = q BC + q BrC × λ −(α BrC −1) .
Fitting coefficients in Eq. ( 8) were obtained for α BrC values between 2 and 8 by least-square linear regression, and the α BrC that led to the overall best fit in terms of r 2 was selected as the effective absorption Ångström exponent of BrC with which τ a,λ,BC and τ a,λ,BrC can be calculated from the first and second terms of Eq. ( 7).This fitting takes advantage of all six wavelengths.For each of the 44 samples, Fig. S4 shows that fitted τ a,λ are within ±5 % of the measured values for τ a,λ > 0.01.Examples of the τ a,λ decomposition as a function of wavelength are shown in Fig. S5.
Table 3 summarizes the apportionment of τ a,635 nm into BC and BrC fractions along with average α BrC , "diesel-ECequivalent" BC (termed BC d hereafter), ECR, and ECT by sample type.Consistent with Fig. 5, BrC contributions to τ a,635 nm are much higher in prescribed burning than in diesel exhaust samples (averaging 46 % versus 1 %) while somewhere in between (10-26 %) for Fresno and Reno wildfire samples.Effective α BrC compares well among Fresno, Reno wildfire, and prescribed-burning samples (4.2-4.8) and is consistent with BrC of a similar nature from biomass burning (Bahadur et al., 2012;Kirchstetter and Thatcher, 2012).Even in the infrared region BrC accounts for 3, 6, and 24 % of τ a,980 nm for the Fresno, Reno, and prescribed-burning samples, respectively, on average.α BrC in diesel exhaust, detectable in 5 of 11 samples, appears to be significantly lower (2.3 ± 0.1) than in other sample types.
Table 3 shows that BC d as determined from τ a,635 nm,BC / MAE 635 nm (7.4 m 2 g −1 ) are lower than ECR 635 nm .The differences are especially significant (i.e., p value of RD < 0.01) for Reno wildfire and Tahoe prescribed-burning samples with relatively high BrC contributions.The comparisons do not change with ECR 808 nm (with low BrC influence) replacing ECR 635 nm .A continuum of light-absorbing carbon from biomass burning -ranging from BrC and char to soot -as suggested by Pöschl (2003) and Masiello (2004) may explain the phenomenon.As char and soot resulting from pyrolysis and high-temperature graphitization, respectively, are both quantified as EC by TOA (Han et al., 2009), they may have distinct optical properties.BC d that was calibrated against diesel EC would represent just the soot fraction because there is little char material in diesel exhaust.ECT 635 nm is substantially lower than ECR 635 nm due to the aforementioned POC effect and is much closer to BC d for the Reno and Tahoe biomass burning samples but not the Fresno and diesel exhaust samples.

Conclusions
Thermal/optical analysis that combines thermal separation and optical monitoring is potentially a powerful tool for analyzing carbonaceous aerosol on filters.Spatiotemporal variations and long-term trends in aerosol loading, chemical composition, sources, and effects have been inferred from OC and EC measurements (e.g., Chen et al., 2012;Hand et al., 2012;Malm et al., 1994;Murphy et al., 2011;Park et al., 2006).As many archived samples may be retrieved for reanalysis and ∼ 40 000 new samples are collected per year in the US long-term networks alone, an enhanced multiwavelength thermal/optical analyzer would benefit the scientific community that uses the data.
The seven-wavelength (visible to near-infrared regions) TSA with both R and T sensors allows the determination of the OC-EC split at different wavelengths and light absorption measurements to be made with wavelength-specific loading corrections.In the selected ambient and source PM 2.5 samples, contributions of BC and BrC to light absorption were decoupled, assuming an absorption Ångström exponent of unity for BC and much higher values for BrC.Thus, BC concentrations optically equivalent to diesel exhaust EC, i.e., BC d , can be calculated.BrC with an average absorption Ångström exponent of 4.2-4.8 is found to be enriched in samples influenced by biomass burning.
Despite the modifications in light source and detection technique, it is shown that the TSA measures OC 635 nm and EC 635 nm equivalent to OC 633 nm and EC 633 nm from conventional TOA following the same IMPROVE_A protocol with either R or T pyrolysis adjustment.ECR 635 nm is also consistent with those determined with longer wavelengths (e.g., 808 nm), though OC-EC splits with shorter wavelengths (e.g., 450 nm) increase ECR appreciably, showing the effect of BrC.For ECT, the BrC effect is somewhat canceled by an opposite POC effect.The optically derived BC d underestimates ECR 635 nm or ECR 808 nm in biomass-burningdominated samples with relatively high BrC content though the agreements are good for other samples.This discrepancy calls for further studies on the optical properties of EC, including soot and char, from biomass burning in contrast to those of diesel soot particles.
The Supplement related to this article is available online at doi:10.5194/amt-8-451-2015-supplement.

Figure 1 .
Figure 1.Configuration of optical sensing (left) modified from the DRI Model 2001 analyzer (Chen et al., 2012).The seven-laser module (i.e., L1-L7) represents seven diode lasers with wavelengths of 405, 450, 532, 635, 780, 808, and 980 nm.Reference (top right) and laser (middle right) signals are modulated at 30 Hz for a lock-in amplification of the reflectance or transmittance signals (bottom right).Only 30 Hz signals with the same phase as the reference are amplified at gains proportional to the reference signal voltage and number of data points integrated.

Figure 3 .
Figure3.Comparison of organic and elemental carbon by reflectance optical correction (i.e., OCR and ECR, respectively) between the conventional thermal/optical analysis (TOA based on 633 nm optical split) and seven-wavelength thermal/spectral analysis (TSA based on 635 nm optical split) following the IMPROVE_A protocol.Note that x and y axes are on log scales.

Figure 4 .
Figure 4. Relationship of EC concentration, [EC], versus ATN 635 nm from all 1 s measurements during the EC2 step (740 • C in a 98 % He / 2 % O 2 atmosphere) of IMPROVE_A analysis for a diesel exhaust sample (CIFQ074 from the Gas/Diesel Split Study (Fujita et al., 2007)).Also shown are paired [EC]-ATN 635 nm of 11 diesel exhaust samples, where EC is determined from IMPROVE_A and ATN 635 nm from initial filter transmittance.Circles and error bars indicate the average and spread, respectively, of EC by transmittance and reflectance (i.e., ECT and ECR, respectively).τ a,635 nm was further calculated from a MAE of 7.4 m 2 g −1 × [EC], and the regression result between τ a,635 nm and ATN 635 nm is shown in green.

Figure 5 .
Figure 5. τ a,λ -ATN λ relationships for 405-980 nm wavelengths, based on a diesel exhaust reference sample.Results for 532 nm are not shown owing to lack of sufficient detector signal-to-noise ratio at this wavelength.

Figure 6 .
Figure 6.Product of absorption optical depth (τ a,λ ) and wavelength (λ) as a function of λ by sample type.τ a,λ shown represents averages over each of the sample types.

Table 1 .
Comparison of TC, OCR, OCT, ECR, and ECT between the 633 nm (TOA) and 635 nm (TSA) optical splits following the IM-PROVE_A protocol.a (Chow et al., 2007a)sion: y = m × x + b RD: 2(y − x)/(y + x) a TC: total carbon; OCR: organic carbon by reflectance; OCT: organic carbon by transmittance; ECR: elemental carbon by reflectance; and ECT: elemental carbon by transmittance following the IMPROVE_A thermal/optical carbon analysis protocol(Chow et al., 2007a).TOA: thermal/optical analyses; TSA: thermal/spectral analyses.b x is by retrofitted seven-wavelength carbon analyzer; y is by conventional single-wavelength (633 nm) DRI Model 2001 thermal/optical carbon analyzer.c Concentration in µg cm −2 ; m is the slope; b is the intercept in µg cm −2 .d σ : standard deviation.e Student's t test p values.

Table 2 .
Comparison of TSA-determined 635 nm ECR and ECT with 880 nm and 450 nm following the IMPROVE_A protocol.
a x and y are both by modified seven-wavelength carbon analyzer but with optical pyrolysis adjustment at different wavelengths.b Concentration in µg cm −2 ; m is the slope; b is the intercept in µg cm −2 .c σ : standard deviation.e Student's t test p values.

Table 3 .
Average τ a,635 nm for four sample types and their respective BC and BrC fractions, BrC absorption Ångström exponent, and diesel-EC-equivalent BC (BC d ) concentration.IMPROVE_A ECR and ECT determined by the 635 nm optical adjustment are compared to BC d in terms of average and p value of the relative difference (RD).type τ a,635 nm τ a,635 nm,BC τ a,635 nm,BrC α a ) (µg cm −2 ) (µg cm −2 ) (RD BC,ECR ) (RD BC,ECT ) Calculated from τ a,635 nm,BC /(7.4 m 2 g −1 ).
a Pre-assumed values.b