Optical Calibration and Equivalence of a Multiwavelength Thermal/Optical Carbon Analyzer

Organic and elemental carbon (OC and EC) are operationally-defined by the measurement process, so long-term trends may be interrupted with instrumentation changes. A modification to the U.S. IMPROVE carbon analysis protocol and hardware is examined that replaces the 633 nm laser light used for OC charring adjustments with seven wavelengths ranging from 405 to 980 nm, including one at 635 nm. Reflectance (R) and Transmittance (T) values for each wavelength are made traceable to primary standards through transfer standards consisting of a range of aerosol deposits on filter media similar to that of the analyzed samples. R and T values are assigned to these filters using a UV/VIS spectrometer calibrated with these standards. Using ambient and source (e.g., diesel exhaust, flaming biomass, and smoldering biomass) samples, it is demonstrated that R and T calibration is independent of the sample type. Total carbon (TC), OC, and EC comparisons with the earlier hardware design for urbanand non-urban samples demonstrate equivalence, within precisions derived from replicate analyses, for the 633 nm and 635 nm wavelengths. Several uses of the additional multiwavelength information are identified, including: 1) ground-truthing of multi-spectral remote sensors; 2) improving estimates of the Earth’s radiation balance; 3) associating specific organic compounds with their light absorption properties; and 4) appropriating sources of black and brown carbon.


INTRODUCTION
Organic and elemental carbon (OC and EC, respectively; EC is sometimes termed black carbon [BC]) are important components of suspended particulate matter (PM), especially in the respirable fraction with aerodynamic diameters less than 2.5 micrometers (PM 2.5 ).Excessive OC and EC concentrations can adversely affect human health, visibility, climate, materials, and biotic systems (Chow and Watson, 2011;U.S.EPA, 2012;Eklund et al., 2014;Grahame et al., 2014).Brown carbon (BrC) (Andreae and Gelencser, 2006;Moosmüller et al., 2011), which absorbs light more efficiently at shorter visible wavelengths (400 to 700 nm) than BC, has become an important research topic because it is an indicator of biomass burning and secondary organic aerosol (SOA) formation and differs from BC in its effects on radiative transfer.The nature of BrC is especially important in Asian countries where source emissions differ from those of North America and Europe (Betha and Balasubramanian, 2014;Chuesaard et al., 2014;Gargava et al., 2014;Pervez et al., 2015).More information on the light absorption properties of ambient aerosols would be useful, especially if it could be acquired as part of the ongoing chemical speciation networks.
OC and EC have been measured using more than 20 variations of the evolved gas analysis (EGA) method (Watson et al., 2005).Ambient air and source emission samples are collected onto a heat-resistant filter and a portion of each filter is subjected to various temperatures and atmospheres.PM carbon in the sample volatilizes and combusts to carboncontaining gases that are oxidized to carbon dioxide (CO 2 ).The evolved CO 2 may be either directly determined by a nondispersive infrared (NDIR) detector or reduced to methane (CH 4 ) and quantified by a flame ionization detector (FID).A charring adjustment (Huntzicker et al., 1982) based on light reflected from or transmitted through the filter was introduced to separate OC and EC.Different EGA methods have been found to produce equivalent values for total carbon (TC = OC + EC) but dissimilar values for OC and EC (Watson et al., 2005 and references therein).
The Interagency Monitoring of PROtected Visual Environments (IMPROVE, 2015) thermal/optical carbon analysis protocol (Chow et al., 1993a(Chow et al., , 2007(Chow et al., , 2011) ) has been consistently applied at the non-urban IMPROVE and urban Chemical Speciation Network (CSN, U.S. EPA, 2015) sites since 1986 and 2005/2006, respectively, thereby allowing long-term trends to be assessed (Murphy et al., 2011;Chen et al., 2012).Consistent trends are needed to develop emission control strategies in urban areas and track reasonable progress toward natural visibility conditions by CY 2065 at U.S. national parks and wilderness areas, as required by the Clean Air Visibility Rule (U.S. EPA, 1999;Watson, 2002;U.S.EPA, 2005).Fig. 1 shows these trends at a representative site from the IMPROVE network, indicating that emission reduction measures are having a positive effect on lowering ambient concentrations.In addition to the IMPROVE and CSN networks, the IMPROVE protocol has been adopted for other long-term PM chemical speciation networks in the U.S. (Hansen et al., 2006), Canada (Dabek-Zlotorzynska et al., 2011), and China (Huang et al., 2012;Zhang et al., 2012), as well as for shorter-term ambient and source characterization studies in these and other parts of the world.Since OC and EC are operationally defined by the measurement protocol, changes in equipment and procedures may adversely affect the ability to track trends, as noted in Fig. 1.This is the case for many environmental measurements, and it presents a problem as newer, and possibly better, measurement technologies supplant older ones.
With recent laser diode technology, it is possible to add a multiwavelength light absorption capability to the IMPROVE thermal/optical protocol.To be quantitative, the reflected and transmitted light must be calibrated against and traceable to available standards, and the OC and EC fractions must be equivalent to those obtained with earlier instrumentation.The objectives of this study are to: 1) describe multiwavelength changes to the IMPROVE carbon analysis; 2) develop a traceable calibration procedure for quantifying multiwavelength light absorption along with OC, EC, and thermal carbon fractions; and 3) demonstrate equivalence between the multiwavelength and single wavelength systems., and a small increase in EC and a small decrease in OC may have been attributed in part to that change.However, this feature was not consistent at all of the IMPROVE sites that underwent the same change (Chen et al., 2012).This illustrates some of the difficulties in changing instrumentation and methods in assessing trends.Data were obtained from CIRA (2014).owing to charring of organic vapors adsorbed within the quartz-fiber filter (Chow et al., 2004;Watson et al., 2009, Chow et al., 2010).The two thermograms are nearly identical, except that the Model 2015 shows greater sensitivity in the R detection of OC charring, as evidenced by low point in R before O 2 is added to the analysis atmosphere.Temperaturedefined fractions are: OC1-OC4 (OC evolved from the 0.5 cm 2 circular filter punch in a pure He [> 99.999%] atmosphere at 140, 280, 480, and 580°C, respectively), and EC1-EC3 (EC evolved from the filter punch in a 98% He/2% O 2 atmosphere at 580, 740, and 840°C, respectively).The analysis temperature stays constant until each fraction is fully evolved, and total analysis times are longer for more heavily loaded samples.Carbon evolving after O 2 is added until R and T achieve their original values are termed optical pyrolyzed carbon by reflectance (OPRC) and optical pyrolyzed carbon by transmittance (OPTC), respectively.These have negative values when R or T achieve their initial values before O 2 is added.

Since
Calabasas, CA) following the IMPROVE_A thermal/optical protocol (Chow et al., 2007(Chow et al., , 2011) ) described in Fig. 2. Model 2001 replaced the earlier, and more specialized, DRI/OGC analyzers that had been used since 1986 after extensive testing and comparison to demonstrate equivalence (Chow et al., 2007) of OC and EC, as well as the thermallyevolved carbon fractions.The thermal fractions have been found useful for source apportionment (Watson et al., 1994;Maykut et al., 2003;Kim et al., 2004;Kim and Hopke, 2004a, b;Begum et al., 2005;Cao et al., 2005;Chen et al., 2007;Chow et al., 2009) and for understanding OC adsorption artifacts on quartz-fiber filters (Watson et al., 2009;Chow et al., 2010).The fractions can be summed in different ways to simulate and compare with other thermal and optical OC/EC protocols (Watson et al., 2005).
Although hundreds of thousands of samples have been reliably analyzed by the Model 2001 in more than 50 laboratories worldwide over the past decade, technology has advanced to the point that a re-design is needed.The cost of pure inert helium (He) gas has skyrocketed, and its use needs to be reduced.NDIR detectors can now quantify CO 2 at low levels, obviating the need for reduction to CH 4 for FID detection with its associated need for a hydrogen (H 2 ) source.Customized circuit boards can be replaced with off-the-shelf programmable integrated circuits, and a small computer can control the mechanical, optical, and data acquisition functions.Most importantly, small, compact, and inexpensive diode lasers with various wavelengths can replace the larger and more costly helium-neon (He-Ne) light source, thereby providing more information on PM optical properties and their changes throughout the analysis.
Fig. 3 summarizes the carbon analyzer modifications.To ensure compatibility, the sample presentation, calibration, and measurement procedures remain the same, except for the minor change in the red light wavelength from 633 to 635 nm, as illustrated in Fig. 4. The analysis procedures, calibration methods, quality control, and quality assurance measures are detailed in the supplemental information, with some enhancements over those reported by Chow et al. (2011).405, 445, 532, 635, 780, 808, and 980 nm diode lasers that are sequentially modulated to separate their signals; 2) the bifurcated fiber optic bundle is replaced with eight single fibers to convey the diode laser beams perpendicular to the sample surface; 3) the quartz-glass sample oven is modified to eliminate the methanator and to automatically replenish the manganese dioxide (MnO 2 ) oxidation catalyst, thereby minimizing its replacement; 4) manually-adjusted rotameters are replaced with more precise mass flow controllers (MFCs); 5) the nichrome oven heater is powered by 24V DC instead of 110 V AC to minimize high frequency fluctuations in glow intensity; 6) customized circuit boards are replaced with generic and programmable integrated circuits; 7) a high sensitivity non-dispersive infrared (NDIR) detector for CO 2 replaces the flame ionization detector (FID) for CH 4 , eliminating the need for ultrapure hydrogen (H 2 ) and the heated reduction catalyst to convert CO 2 to CH 4 ; and 8) software is written in LABVIEW to control analysis and acquire data.The 633 nm reflectance (R) and transmittance (T) allowed light absorption properties for each sample to be estimated by taking the ratio of the initial to the final R and T after the light absorbing carbon was removed, leaving a filter remnant that was usually white, like the unexposed filter.The R measure of filter darkening, similar to that of British Smoke (Heal and Quincey, 2012), was used to further demonstrate a consistent EC trend when the Model 2001 replaced the earlier DRI/OGC analyzers (Chen et al., 2012).To obtain the most information from multiple-wavelengths, and to ensure comparability among different analyzers, a more absolute measure of R and T is desired.This is accomplished by using transfer standards composed of increasing aerosol deposits on quartz-fiber filters identical to those used for routine sampling.Absolute R and T of the filters in percentage (%) of light reflected from or transmitted through each filter are measured for a range of wavelengths with a UV/VIS spectrometer (Lambda 35, Perkin Elmer, Waltham, MA) equipped with an integrating sphere to fully capture scattered light.For T measurements, the spectrometer output is set to 100% transmittance when the beam is unobstructed and to 0% transmittance when the incident beam is blocked; linearity of the response to intermediate transmittances is verified with neutral density filters (Travis et al., 2000;Upstone, 2002).T for each transfer standard is quantified by illuminating the deposit side of each filter located in front of the integrating-sphere and detector (see Supplemental Fig. S-3) and scanning from 200 to 1100 nm (1-nm resolution) for 300 seconds.R is standardized with a Spectralon diffusive reflectance standard (Labsphere, 2000).Spectralon is a proprietary Teflon-based material used to calibrate satellite remote sensors (Georgiev and Butler, 2007), and it is the same material that coats the spectrometer's integrating sphere.R is 100% with the white Spectralon standard inserted into the outlet of the integrating sphere and 0% when it is removed, allowing the incident light to pass through the sphere.NIST-traceable Spectralon standards of known reflectance are used to evaluate linearity of the spectrometer response.Chen et al. (2015) performed R and T optical calibrations using sections of quartz-fiber filters with different EC levels from the Fresno Supersite in central California (Watson et al., 2000;Park et al., 2006;Chow et al., 2009), assuming that all aerosols are equally useful for use as transfer standards as long as there is a range of darkening in the deposits.The basis for this assumption is that the R and T measurements by a carbon analyzer linearly correspond to absolute filter R and T, respectively, regardless of the sample type and loading.To evaluate this assumption, and to develop a more repeatable process for creating transfer standards, quartz-fiber filter samples with increased loadings were prepared from laboratory-generated aerosols using a diesel generator, flaming dry pine needles, smoldering damp pine cones, and smoldering Florida peat.Other potential transfer standards might be obtained from graphite electric arc soot (Evans et al., 2003), resuspended carbon black or graphite powders (Medalia and Richards, 1972;Jäger et al., 1999), or a lean-burning gas torch (Sheridan et al., 2005;Chow et al., 2009;Moore et al., 2014), but these are dominated by BC and do not include a major BrC component.Diesel exhaust was generated by an Onan Cummins 12.5 kW diesel generator with an accumulated 2,500 hours operating on commercial diesel fuel (< 15 ppmw sulfur content) under idling conditions.Biomass samples were combusted in a Pineridge Genuine Woodstove.Dry Sierra Nevada pine needles (recovered from the forest floor) were ignited with a butane lighter and sampled only during the flaming phase.Sierra Nevada pine cones were wetted with water and subjected to a temperature-adjustable electric heater to attain flameless smoldering prior to sampling.Florida peat was also ignited by the electric heater to a smoldering phase.
Samples were drawn from the plume of each source through a Bendix 240 PM 2.5 cyclone at a total flow rate of 113 L/min into a conical mixing chamber (Chow et al., 1993b).The cyclone inlet was placed at ~0.5-1.5 m from the emission point to allow for cooling and dilution with ambient air and to obtain a wide range of deposit opacities.Eight pre-fired 47 mm quartz-fiber filters and two Teflonmembrane filters were mounted to the 12-port manifold with Savillex fluorinated ethylene propylene (FEP) Teflon filter holders.Flow rates through the filters ranged from 1 to 32 L/min, in a logarithmic sequence, to obtain increasing deposits on each filter.A DustTrak DRX (Wang et al., 2009) monitored particle concentrations in real time to adjust sample durations, which ranged from ~5 min for diesel exhaust to 15 min for peat smoldering.This yielded aerosol deposits ranging from ~0.05-1.5 mg/filter.Filters were weighed before and after sampling to determine the amount of aerosol collected, but this is not a necessary feature of the transfer standard.Fig. 5 shows examples of the transfer standards.
The diesel and flaming standards are similar in appearance to many ambient samples, with shades changing from light gray to deep black.The smoldering standards have a yellowish-brown appearance that is distinct from the others.A set of absolute R(λ) and T(λ) were read from the UV/VIS spectrometer for each Model 2015 wavelength (i.e., 405, 445, 532, 635, 780, 808, and 980 nm) that characterize each transfer standard, as shown in Figs.S-4 to S-7.The calibration spectra in Fig. 6 show that R and T are lower at the shorter wavelengths for all of the samples.R and T also decrease as the sample loadings increase.However, the smoldering samples show small changes with loading at the longer wavelengths compared to the shorter wavelengths, making them less useful than the blacker standards for covering a large range of wavelengths and filter loadings.R has a larger range than T, especially at the shorter wavelengths, on a linear scale.The T signal is both scattered and absorbed on the aerosol deposit and within the filter, whereas the R signal is affected mostly by the surface of the filter and its deposit, thereby experiencing less attenuation.
Each transfer standard was placed into the Model 2015 sample boat and inserted into the analysis zone to measure instrument-specific R and T for one or two minutes to obtain an average photodiode detector response.The detector output is linear over several decades, and it responds to different wavelengths at both high and low R and T values, so the outputs are scaled such that they can be viewed on a similar basis.Fig. 7 illustrates the resulting calibration curves for R(λ) and T(λ) derived from this practice.These appear to be independent of the type and loading of sample, confirming the hypothesis of Chen et al. (2015).Ambient samples with different shadings may still be preferable to laboratory-generated samples as transfer standards, since their filter material is more likely to be consistent from sample to sample.Based on the calibration curves, absolute R and T in % can be inferred from carbon analyzer photodiode signals.Atmospheric absorption as a function of wavelength, b abs (λ), for the integrated filter deposit can then be estimated from attenuations (ATN) in either the absolute filter R or T, defined by: where R(λ)/T(λ) initial and R(λ)/T(λ) final are the time-averaged (typically 10 seconds) initial and final filter reflectance/ transmittances at each wavelength, respectively.The R(λ) final and T(λ) final may differ from those of a true blank filter due to refractory residues after the thermal analysis, and these rare cases are flagged when observed after analysis is complete.Nonetheless, in the first-order approximation (Hansen et al., 1984;ISO, 1993;Lindberg et al., 1999;Quincey, 2007), b abs (λ) of the thermally liberated fraction including OC and EC can be expressed as: where A is the area (cm 2 ) of the surface deposit and V is the sample volume (m 3 ).These yield b abs (λ) in the commonly used unit of inverse megameter (Mm -1 ) (Richards, 1984;Watson, 2002).The comparison of b abs (λ) by R and T will be discussed in a companion paper.Though highly correlated, they differ substantially in magnitude.More precise estimations of b abs (λ) should consider multiple scattering and loading effects of the particle-filter matrix, leading to modifications to Eqs. ( 3) and (4) (Arnott et al., 2005;Quincey et al., 2011;Heal and Quincey, 2012;Chen et al., 2015).

EQUIVALENCE OF OC, EC, AND CARBON FRACTIONS
The conversion from the DRI/OGC to Model 2001 analysis in CY2005 involved a major change in the optical configuration to accommodate simultaneous R and T measurements.The only change from the Model 2001 to multiwavelength measurements is in the laser wavelength from 633 to 635 nm, which is believed to be imperceptible for the quantification of OC and EC (see examples in Fig. 2).The small diameter of the 25 mm IMPROVE filter limits the number of comparisons that can be performed on a single sample.Collocated samples are taken at several IMPROVE sites for research purposes, and 73 of these were used in an initial comparison between the Model 2015 and Model 2001.As a contrast to non-urban IMPROVE samples, portions of 67 large (406 cm 2 ) PM 2.5 samples from the urban Fresno Supersite (Watson et al., 2000) were also tested in replicate.Results for the different carbon fractions are summarized in Table 1, with scatterplots shown in Figs.S-9 to S-12.The slopes are within three standard errors (± 3σ) of unity for the TC, OC, and EC, including OC and EC by reflectance and transmittance optical adjustment (i.e., OCR, ECR, OCT, and ECT), with high correlations.Precision bars on the scatterplots show that the 1:1 line is within the precision estimated from replicate analyses.It has been found in previous samples that much of the variability comes from inhomogeneity of the deposit rather than variability of the carbon detection.
The thermal carbon fractions are not as similar in these replicates with slopes differing from unity by more than a few standard errors.Similar variability was found by Chow et al. (2007), and this variability is typical of that among replicates from different carbon analyzers.The inter-instrumental variability was attributed to temperature calibration, which could deviate up to ±1% from the specified temperatures (e.g., ~15°C difference at the EC2 [740°C] for any two carbon analyzers, see Chow et al., 2005), as well as to trace oxygen levels in the oven (varying by a factor of two, see Chow et al., 2007).As a result, the error bars for carbon fractions are larger than those for TC, OC, and EC, with most of them still including the 1:1 line (see Figs. S-11 and S-12).In addition, OC1 is semi-volatile and is often found to be poorly reproduced even in replicates analyzed on the same instrument.

POTENTIAL USES OF MULTIWAVELENGTH R AND T DATA
Multiwavelength aethalometers (Park et al., 2006) and multi-spectral remote sensors (Hoff and Christopher, 2009) have shown the value of aerosol absorption data, especially related to radiative transfer and aerosol climate forcing.Deming (1943) regression with zero intercept and standard error (σ).This regression minimizes the perpendicular distances between the slope line and the data points, yielding the same slope when the x and y axes are interchanged.This is the correct approach when there are errors in both variables.Carbon fractions are described in Fig. 2 caption.The carbon measurement enhancement described here is complementary to, rather than a replacement for, these other methods.Since extensive thermal/optical analysis is being done for both research and compliance monitoring, multiwavelength spectral measurements as well as the charring adjustment provide added value to the data sets.Raw and calibrated R(λ) and T(λ) before and after analysis (i.e., initial and final values, respectively) will be reported routinely, while those during analysis can be retrieved from digitally-archived thermograms when needed.Additional data reporting includes OCR, ECR, OCT, and ECT for each of the seven wavelengths.Chen et al. (2015) demonstrate that based on spectral attenuation by transmittance (ATN T ), contributions of BC and BrC to light absorption as well as Angstrom Absorption Exponent (AAE) of BrC can be derived from a sample prior to thermal/optical analysis.The loading effect was addressed using diesel exhaust samples.The same can be done for attenuation by reflectance (ATN R ) to achieve consistent BC and BrC measurements.During analysis, however, charring occurs to increase ATN R and ATN T , as demonstrated in Fig. 9. Charring lowers the AAE on the filter, consistent with an optical property similar to natural BC.AAE based on ATN R approaches unity at the end of OC4 as all BrC has been removed from the filter or converted to char.AAE based on ATN T is higher than unity, however, likely due to charring within the filter.Optical adjustment based on R should not depend on wavelength if BrC does not exist, and therefore the difference in R split points between BrC sensitive and insensitive wavelengths (e.g., 445 and 808 nm, respectively) may reflect the BrC content in the sample.This can be verified by comparing with BrC derived from ATN R or ATN T prior to thermal/optical analysis.Both approaches for estimating BrC should be further investigated with standards of known BC/BrC mixtures, such as mixtures of diesel soot and aerosol humic acids.
For BC-dominated samples, such as diesel exhaust, initial tests show little variation in OC and EC fractions across wavelengths.However, for smoldering samples, such as those illustrated in Fig. 9, the OC and EC fractions, particularly OCR and ECR, differ depending on wavelength.A larger number of ambient and source samples needs to be analyzed to identify relevant patterns.Several potential data uses include: • Identifying light absorbing compounds: More specific organic compound detectors have been interfaced to thermal/optical analyzers (Grabowsky et al., 2011;Diab et al., 2015), and these can be associated with their spectral and thermal evolution properties.• Separating artifact OC from aerosol OC: Organic vapors adsorbed by the quartz-fiber filter affect the OC measurement and are the major cause of differences between ECR and ECT (Chow et al., 2004;Watson et al., 2009;Chow et al., 2010).Multiwavelenth R and T measurements offer opportunities to constrain radiative transfer models through the filter to separate charred  2) are shown along with the Angstrom Absorption Exponent (AAE) calculated from the two wavelengths.Red dots indicate the R or T split points for OC and EC.The two R split points between the two wavelengths are more diverse than the two T split points.carbonaceous material within the filter from that of the aerosol deposit (Chen et al., 2004;Petzold et al., 2005).
• Ground-truthing remotely-sensed BrC: Algorithms are being developed to estimate BrC from space-and landbased remote sensors (Wang et al., 2013), but these need to be verified with surface measurements.A growing database of spatially and temporally dispersed data at regional monitors would provide a means of verifying and validating these algorithms and determining their limitations.
• Improving radiation transfer estimates: It is widely assumed that aerosol absorption follows a λ -AAE relationship.This is a good approximation for samples dominated by BC, where AAE = ~1.However, for many aerosols, such as those from smoldering combustion, this simple relationship is not followed and AAE varies with wavelength (Moosmüller et al., 2012).Interpolations between the seven wavelengths can be used to infer more complex functions of wavelength for different sources, a fraction of which can be verified by the more detailed spectrometric analyses outlined above.These could be included as part of emissions inputs to global radiative transfer models (Chakrabarty et al., 2013).• Appropriating sources of BC and BrC: The simplest approach, outlined by Sandradewi et al. (2008a, b), extrapolates the BC absorption using an AAE = 1 assumption, then attributes the remaining absorption at shorter wavelengths to the BrC associated with smoldering Temperature combustion or SOA.Absorptions are translated to BC and BrC with mass absorption efficiencies derived from source emission measurements (i.e., part of the source profile).A more complex, and potentially accurate, source apportionment approach would look at carbon associated with each of the seven wavelengths in source samples and use these in a receptor-oriented model along with other source indicators (Watson et al., 2008).Other tests conducted at the beginning of each analysis day include a system blank check, a leak check, and a laser performance check.

S.1. Thermal/Optical Carbon Analysis Procedure
Temperature calibrations are conducted every six months, or whenever the thermocouple is replaced, using temperature-indicating materials (Templar  G, Temple Inc., South Plainfield, NJ).A Templar  G set consists of chemicals that change their appearance at a specific temperature (rated temperature).A thin layer (~25 µL) of Templar  G is uniformly applied onto a glass or quartz disk surface and covered with a sliced quartz-fiber filter punch.When the specified temperature is reached, the appearance of the sample changes, as evidenced by R and T, and by which the deviation between measured (thermocouple) and sample temperatures can be determined (Chow et al., 2005).Triplicate tests are made at 121, 184, 253, 510, 704, and 816 °C to obtain the calibration slope and intercept, which are then fed back into the temperature control program.This procedure ensures that each sample experiences a temperature plateau specified by the protocol.A significant change in calibration slope and intercept also triggers inspection for possible hardware problems.

Panteliadis et al.'s (2014) recent interlaboratory comparison demonstrates the importance
of calibrating for the sample temperature, not just the temperature at the sample location.
Seventeen different laboratories equipped with similar instrumentation followed similar calibration procedures, including those for temperature (Phuah et al., 2009).For EC, the standard deviation among laboratories was 25.5% for the NIOSH870 and 19.5% for the EUSSAR2 protocols.Part of the discrepancy was attributed to O 2 infiltration into the inert helium (He) atmosphere in these units.Oxygen (O 2 ) measurements are performed with an external gas chromatography/mass spectrometer (GC/MS) every 6 months to monitor trace O 2 levels that may trigger EC or optical pyrolysis (OP) oxidation and bias the OC/EC split.
Concentrations for each carbon fraction (c fraction ) are reported for each filter by normalizing the area of the sample punch (0.5 cm 2 ) to the exposed area of the filter, which depends on the open face of the filter holder (nominally 3.8 cm 2 for the URG 3000N 25 mm filter holder, 11.8 cm 2 for the FRM 47 mm filter holder, and 13.8 cm 2 for the Nuclepore 47 mm filter holder).The exposed area should be measured periodically to assure that accurate values are used.
Analysis precisions (Watson et al., 2001) are calculated for different concentration intervals as: where: CV = coefficient of variation for each carbon fraction N = number of samples c i = concentration of initial analysis for each carbon fraction c i,r = concentration of sample "i" replicate analysis for each carbon fraction     Major components of a carbon analysis: 1) removing a sample from the laboratory refrigerator; 2) scanning the laboratory code and initializing the analyzer; 3) removing a 0.5 cm 2 punch from the filter-three such punches are available, thereby permitting replicate analyses on a randomly-selected analyzer; 4) placing the punch into the quartz sampling boat (the thermocouple/pushrod precisely positions the sample in the analysis zone); 5) placing the analyzed punch on the thermogram (the analyzed punch may have a red or yellow hue when minerals are abundant); and 6) inspecting the results and entering analysis validation flags.Figure S-3.Procedure for calibrating quartz-fiber transfer standards for R and T: 1) sample filter holder, transfer standard, and forceps for handling filter with gloved hands in a laminar flow hood; 2) transfer standard in a custom filter cassette designed to minimize distance between the filter and the integrating sphere inlet; 3) filter cassette in standard Lambda 35 sample holder (a UV/VIS Spectrometer, PerkinElmer, Waltham, MA); 4) scan across wavelengths for 100% T with beam unblocked and Spectralon reflectance standard (Labsphere, 2000); 5) scan across wavelengths for 0% T with beam blocked; 6) insert filter with deposit toward beam (to mimic Model 2015 optical configuration) and scan across wavelengths for measurement; 7) scan across wavelengths for 100% R with beam unblocked and Spectralon reflectance standard (Labsphere, 2000); 8) scan across wavelengths for 0% R with beam blocked; and 9) insert filter with deposit toward beam (to mimic Model 2015 optical configuration) and scan across wavelengths for measurement.

Reflectance and Transmittance Calibration
CY 2005, OC and EC measurements for the IMPROVE network have been made with the DRI Model 2001 thermal/optical carbon analyzer (Atmoslytics, Fig. 1.Trends in annual average organic carbon (OC) and elemental carbon (EC) concentrations at the Shenandoah National Park IMPROVE site.The Model 2001 thermal/optical carbon analyzer replaced the DRI/OGC carbon analyzers beginning in 2005

Fig. 2 .
Fig. 2. Example of IMPROVE_A thermograms for samples from: a) Model 2001; and b) Model 2015 for the same sample from the IMPROVE network.Reflectance (R) and transmittance (T) are normalized to their initial values and carbon is expressed as the fraction of total carbon (TC) on the sample.The 633 nm or 635 nm laser beams monitor darkening of the sample by R and T, which indicates OC charring to EC during the analysis.Carbon that evolves after R returns to its initial value is classified as EC in the aerosol deposit.EC by transmittance (ECT) is usually less than EC by reflectance (ECR) owing to charring of organic vapors adsorbed within the quartz-fiber filter(Chow et al., 2004;Watson et al., 2009, Chow et al., 2010).The two thermograms are nearly identical, except that the Model 2015 shows greater sensitivity in the R detection of OC charring, as evidenced by low point in R before O 2 is added to the analysis atmosphere.Temperaturedefined fractions are: OC1-OC4 (OC evolved from the 0.5 cm 2 circular filter punch in a pure He [> 99.999%] atmosphere at 140, 280, 480, and 580°C, respectively), and EC1-EC3 (EC evolved from the filter punch in a 98% He/2% O 2 atmosphere at 580, 740, and 840°C, respectively).The analysis temperature stays constant until each fraction is fully evolved, and total analysis times are longer for more heavily loaded samples.Carbon evolving after O 2 is added until R and T achieve their original values are termed optical pyrolyzed carbon by reflectance (OPRC) and optical pyrolyzed carbon by transmittance (OPTC), respectively.These have negative values when R or T achieve their initial values before O 2 is added.

Fig. 3 .
Fig. 3. DRI Model 2015 multiwavelength carbon analyzer configuration compared to Model 2001.Major changes are: 1) the 633 nm He-Ne laser is replaced with405, 445, 532, 635, 780, 808, and 980  nm diode lasers that are sequentially modulated to separate their signals; 2) the bifurcated fiber optic bundle is replaced with eight single fibers to convey the diode laser beams perpendicular to the sample surface; 3) the quartz-glass sample oven is modified to eliminate the methanator and to automatically replenish the manganese dioxide (MnO 2 ) oxidation catalyst, thereby minimizing its replacement; 4) manually-adjusted rotameters are replaced with more precise mass flow controllers (MFCs); 5) the nichrome oven heater is powered by 24V DC instead of 110 V AC to minimize high frequency fluctuations in glow intensity; 6) customized circuit boards are replaced with generic and programmable integrated circuits; 7) a high sensitivity non-dispersive infrared (NDIR) detector for CO 2 replaces the flame ionization detector (FID) for CH 4 , eliminating the need for ultrapure hydrogen (H 2 ) and the heated reduction catalyst to convert CO 2 to CH 4 ; and 8) software is written in LABVIEW to control analysis and acquire data.

Fig. 4 .
Fig. 4. Optical configurations for: a) Model 2001; and b) Model 2015.Mechanical modulation of the 633 nm helium-neon (He-Ne) laser is replaced by sequential pulse modulation of seven diode lasers.The aerosol deposit faces the incident radiation in both cases.The single optical fiber in the Model 2015 provides more efficient light transmission than the optical fiber bundle in the Model 2001.

Fig. 5 .Fig. 6 .
Fig.5.Transfer standards with varying amounts of deposit.The leftmost samples are blanks.Fresno Supersite samples were obtained from large 406 cm 2 quartz-fiber filter deposits using a high-volume sampler equipped with a PM 2.5 inlet during 2003 and 2004.The remaining samples were generated in the laboratory from simultaneous sampling at different flow rates to obtain a range of different filter loadings.

Fig. 7 .
Fig. 7. Model 2015 R and T calibration curves using transfer standards.The laser signals are multiples of millivolt photodetector outputs normalized to a common scale for all wavelengths.Example calibrations for 633 nm in the Model 2001 (see Supplemental Fig. S-8) show similar linearity.
Fig. S-12 shows low temperature OC1 (140°C) to be the most variable, especially for the low-concentration IMPROVE samples.High temperature EC2 for Fresno samples (Fig. S-11) also showed a lower correlation and higher slope than that for IMPROVE, even though most of the values are within ± 1σ of the 1:1 line supporting a typical inter-instrumental variability.Fig. 8 contrasts replicate analyses for two Model 2001 analyzers with replicates from the Model 2015, showing comparable reproducibility.

Fig. 8 .Fig. 9 .
Fig. 8. Replicate analyses for Model 2001 vs. another Model 2001 unit (top row) and a Model 2015 vs. a Model 2001 (bottom row), showing similar comparability.Slopes use a Deming regression with zero intercept.Since only three 0.5 cm 2 punches can be obtained from the 25 mm diameter IMPROVE filter samples, comparisons among instruments for the same samples are limited.OCR and ECR are OC and EC by reflectance, respectively.

Figure S- 1
Figure S-1 shows the sequence of steps for processing of quartz-fiber filters used for carbon analysis(DRI, 2012).Chain-of-custody procedures and data management systems record the disposition of filters at each stage of analysis and storage.An analysis list is prepared prior to carbon analyses, and a small batch of samples is transferred from the archive to a laboratory refrigerator prior to analyses.Records of system calibration results, comparisons to standard solutions, traceability of calibration standards to nationally recognized primary standards, acceptance testing results, and thermograms for individual samples are retained and made available to auditors and other interested parties when needed to evaluate the measurement process and to explain potentially anomalies found by data users.. Filter inspection, pre-firing, acceptance testing, and archiving of blank filters in Steps 1 through 4 of Figure S-1 are important parts of the operation.Each filter is inspected on a light table for inhomogeneities, pinholes, or other defects, and unacceptable filters are returned to the manufacturer for credit.Acceptable filters are then baked at 900 °C for 4 hours to remove organic vapors, then replaced in their containers, sealed, and refrigerated prior to being loaded.The actual analysis in Step 8 is detailed in Figure S-2.Instrument calibration, quality control (QC), and quality assurance (QA) measures are summarized in Tables S-1 and S-2.Each analyzer is calibrated by analyzing samples of known amounts of methane (CH 4 ), carbon dioxide (CO 2 ), sucrose, and potassium hydrogen phthalate (KHP).The flame ionization detector (FID) response is normalized to a reference CH 4 bolus injected at the end of each sample analysis, and the variability of this response is kept within pre-set tolerances.Performance tests of each instrument's calibration are conducted at the beginning and end of each day's operation.Intervening samples are re-analyzed when calibration changes of more than ±10% are observed.Known amounts of American Chemical Society (ACS)-certified reagent grade crystal sucrose and KHP are analyzed to verify the OC fractions.Fifteen different standards are used for each calibration.At least 10% of the samples are analyzed in replicate to control the analyses and replicate punches are analyzed on an instrument chosen at random.
MDL = minimum detection limit (3 standard errors [3σ] of laboratory blanks) for each carbon fraction σ cfraction = precision of c fraction Data are transmitted to the master data base in any specified format, which usually includes a separate record for each sample with the sample ID, filter lot number, analysis date, validation flags, values of carbon fractions (µgC/filter), precision for each fraction, and initial, minimum, and final R and T values.These are associated with field data, such as sample volume, through the sample ID for a final output in µg/m 3 .It is at this stage that field blank values are subtracted and the errors propagated according to: carbon fraction (µg/m 3 ) c fraction = carbon fraction concentration (µg/filter) b fraction = average concentration of field blanks (µg/filter) σ cfraction = precision of c fraction (µg/filter) σ bfraction = standard deviation of field blanks V = sample volume (m 3 ) from sampling unit, actual atmospheric conditions σ V = precision of sample volume (m 3 )

Figure S- 3
Figure S-3 details the procedure for standardizing the transfer standards used to relate the mV output of the optical detector to the filter absorbance.Figures S-4 through S-7 shows R and

Figure
Figure S-2.Major components of a carbon analysis: 1) removing a sample from the laboratory refrigerator; 2) scanning the laboratory code and initializing the analyzer; 3) removing a 0.5 cm 2 punch from the filter-three such punches are available, thereby permitting replicate analyses on a randomly-selected analyzer; 4) placing the punch into the quartz sampling boat (the thermocouple/pushrod precisely positions the sample in the analysis zone); 5) placing the analyzed punch on the thermogram (the analyzed punch may have a red or yellow hue when minerals are abundant); and 6) inspecting the results and entering analysis validation flags.

"
Figure S-4.Reflectance and transmittance as a function of wavelength for diesel exhaust samples with different loadings.Vertical lines correspond with the seven wavelengths used in the Model 2015.

Figure S- 11 .
Figure S-11.Model 2015 vs. Model 2011 thermal carbon fraction scatterplots for 67 Fresno Supersite samples.EC3 was below detection limits for most of these samples and is not shown.

Table 1 .
Comparison measures for replicate measurements of carbon fractions among Model 2015 and Model 2001 units from 67 Fresno Supersite samples and 73 IMPROVE network samples.