Evaluation of the IMPROVE Equation for estimating aerosol light extinction

ABSTRACT

The [revised] IMPROVE Equation for estimating light extinction from aerosol chemical composition was evaluated considering new measurements at U.S. national parks. Compared with light scattering (Bsp) measured at seven IMPROVE sites with nephelometer data from 2003–2012, the [revised] IMPROVE Equation over- and underestimated Bsp in the lower and upper quintiles, respectively, of measured Bsp. Underestimation of the worst visibility cases (upper quintile) was reduced by assuming an organic mass (OM)/organic carbon (OC) ratio of 2.1 and hygroscopic growth of OM, based on results from previous field studies. This assumption, however, tended to overestimate low Bsp even more. Assuming that sulfate was present as ammonium bisulfate rather than as ammonium sulfate uniformly reduced estimated Bsp. The split-mode model of concentration- and size-dependent dry mass scattering efficiencies in the [revised] IMPROVE Equation does not eliminate systematic biases in estimated Bsp. While the new measurements of OM/OC and OM hygroscopicity should be incorporated into future iterations of the IMPROVE Equation, the problem is not well constrained due to a lack of routine measurements of sulfate neutralization and the water-soluble fraction of OM in the IMPROVE network.

Implications: Studies in U.S. national parks showed that aerosol organics contain more mass and absorb more water as a function of relative humidity than is currently assumed by the IMPROVE Equation for calculating chemical light extinction. Consideration of these results could significantly shift the apportionment of light extinction to water-soluble organic aerosols and therefore better inform pollution control strategies under the U.S. Environmental Protection Agency Regional Haze Rule.

Introduction

The U.S. Environmental Protection Agency (EPA) Regional Haze Rule (RHR) provides a basis for improving visibility in 156 federally protected U.S. national parks and wilderness areas referred to as “Class I” areas (EPA, 1999). Progress toward this goal is evaluated using protocols outlined by the EPA (EPA, 2003). The Interagency Monitoring of Protected Visual Environments (IMPROVE) network was established to monitor and evaluate trends in haze-causing aerosols and their chemical composition in these Class I areas (Malm, 2000; Hand, 2011). PM_2.5 (particles with diameters smaller than 2.5 µm) filter samples are collected every third day and analyzed for sulfate (SO₄^2-), nitrate (NO₃⁻), chloride (Cl⁻), organic and elemental carbon (OC and EC), and trace elements. Ammonium ion (NH₄⁺) is not routinely measured. Total particle light scattering (Bsp) and extinction (Bext) are measured directly at only a small subset of sites. For this reason, tracking of progress under the RHR is accomplished by estimating light extinction from the chemical composition of the aerosol. This is referred to as chemical light extinction.

Malm et al. (1994) estimated PM_2.5 Bsp from aerosol chemical concentrations as follows:

(1)

The dry scattering efficiencies were assumed to be 3 m²/g for (NH₄)₂SO₄ (AMSUL) and NH₄NO₃ (AMNIT), 4 m²/g for organic mass (OM), and 1 m²/g for fine soil. The f(RH) term accounts for the increase in Bsp caused by hygroscopic growth of AMSUL and AMNIT (RH = relative humidity) (Tang and Munkelwitz, 1994). OM was assumed to be nonhygroscopic. An OM/OC ratio (R) of 1.4 was used to convert OC to organic mass (OM) to account for unmeasured hydrogen, oxygen, and other species in organic compounds. This forms the basis of the original IMPROVE Equation for estimating Bext (Malm, 2000), which also includes terms for particle light absorption (Bap) by elemental carbon (EC), coarse particle (PM₁₀–PM_2.5) scattering, and Rayleigh scattering. Sulfates and nitrates were assumed to be completely neutralized by ammonia because ammonium is not routinely measured in the IMPROVE network.

Hand and Malm (2006) and Malm and Hand (2007) evaluated the IMPROVE Equation with a review of the literature and an analysis of IMPROVE data. The original value of OM/OC = 1.4 was based on measurements in an urban area (White and Roberts, 1977). As air masses age, organic material can oxidize to produce higher values of R. The analyses of Hand and Malm (2006) and Malm and Hand (2007) confirmed this, and a value of R between 1.7 and 1.8 was recommended for the IMPROVE network. The value of 1.8 was adopted by IMPROVE in its analysis of aerosol composition and light extinction in the IMPROVE network (DeBell, 2006). Ryan et al. (2005) evaluated the IMPROVE equation (eq 1) and concluded that in order to reconcile measured and estimated Bsp, the dry scattering efficiencies for AMSUL, AMNIT, and OM would have to increase nonlinearly with increasing concentration of these species. Hand and Malm (2006) and Malm and Hand (2007) recognized that dry mass scattering efficiencies appeared to increase as a function of concentration. This was consistent with systematic overestimation of Bsp at low measured Bsp and underestimation of Bsp at high measured Bsp using the original IMPROVE Equation (eq 1) in the IMPROVE network. Hand and Malm (2006) addressed this with a split component model for light scattering by AMSUL, AMNIT, and OM. Concentrations of these species were divided among small and large size modes with lower and higher dry mass scattering efficiencies, respectively. This reduced the systematic biases in eq 1. Hand and Malm (2006) and Malm and Hand (2007) also considered light scattering by PM_2.5 sea salt, site-specific Rayleigh scattering, and absorption by NO₂.

Pitchford et al. (2007) presented a [revised] IMPROVE Equation (IMPROVE2) as follows and evaluated its performance across the IMPROVE network:

(2)

The aerosol absorption component (Bap) is the product of EC and an assumed mass absorption efficiency of 10 m²/g. NO₂ also contributes to light absorption. The subscripts S and L denote the small and large modes for AMSUL, AMNIT, and OM (1.8 OC), with assumed lognormal size distributions with geometric mass mean diameters (Dg) and geometric standard deviations (σg) of 0.2 (2.2) and 0.5 (1.5) for the small and large size modes, respectively. Sea salt concentration is estimated as 1.8 times the chloride ion concentration. The dry scattering efficiencies and f(RH) in eq 2 were estimated theoretically based on these size distributions, species densities and refractive indices, and hygroscopic growth factors described in detail by Pitchford et al. (2007). Organic material was assumed to be non-hygroscopic, that is, f_S(RH)_OM = f_L(RH)_OM = 1. The f(RH) values for sulfates and nitrates were derived using the hygroscopic growth curve for ammonium sulfate calculated with the AIM (Aerosol Inorganic Model) thermodynamic equilibrium model (Clegg et al., 1998). Hygroscopic growth as a function of RH follows the upper branch of the ammonium sulfate hysteresis loop and thus represents the most hydrated hygroscopic state.

Lowenthal et al. (2015) examined assumptions underlying eq 2, including that (1) the OM/OC ratio is 1.8; (2) OM is not hygroscopic; (3) dry mass scattering efficiencies increase with concentration; and (4) hygroscopic growth of AMSUL and AMNIT follow the upper branch of the hysteresis loop of AMSUL. Intensive field studies were conducted at Great Smoky Mountains National Park during summer [GRSM(S)] and winter [GRSM(W)], and at Mt. Rainier (MORA) and Acadia (ACAD) national parks during summer. The results of Lowenthal et al. (2015) showed that (1) the average OM/OC ratio is at least 2.1; (2) water-soluble organic carbon (WSOC) is hygroscopic, although significantly less so than AMSUL; (3) sulfate can be acidic, that is, not completely neutralized as AMSUL; and (4) aerosols are generally in their most hydrated state, exhibiting smooth growth or falling along the upper branch of a hysteresis loop. In this paper, the results presented by Lowenthal et al. (2015) are evaluated with respect to the apportionment of estimated PM_2.5 light scattering (Bsp) among chemical species using data from 2003–2012 in the IMPROVE network and their impact on the comparison between measured and estimated Bsp.

It is not possible to generalize all of the results from four intensive studies in three national parks to the entire IMPROVE network. It is possible, however, to determine the sensitivity of the apportionment of Bsp among OM and sulfate, for example, to OM/OC ratios of 1.8 versus 2.1 and hygroscopic growth of organic aerosols using WSOC/OC ratios ranging from zero to 1.

This analysis has significant implications regarding emissions reductions necessary to attain the goals set out in the RHR.

Methods

The following analysis examines various assumptions underlying the IMPROVE equation by evaluating their effects on (1) the apportionment of Bsp to chemical components at all IMPROVE sites and (2) the comparison of measured and estimated Bsp at sites where Bsp was measured. IMPROVE aerosol data for the 10-year period from 2003 through 2012 were retrieved from the Federal Land Manager Environmental Database (FED) (http://views.cira.colostate.edu/fed/Datasets/RHRDataFiles.aspx). Only samples with valid concentrations of PM_2.5 mass, sulfate, nitrate, OC, EC, fine soil, and chloride ion (for calculating the sea salt component) were included. There were 159,843 samples with complete PM_2.5 data from 142 sites. Coarse mass (PM₁₀–PM_2.5 mass) was not available for 2775 of these samples. Since coarse-particle scattering is not an issue here, the apportionment analysis was confined to PM_2.5 chemical components and estimated PM_2.5 Bsp. EPA Regional Planning Organizations (RPOs) were established to deal with the regional nature of haze-causing pollutants (http://www.epa.gov/visibility/regional.html). Two urban sites, Washington, DC (WASH), and Phoenix, AZ (PHOE), and two noncontinental sites, Hawaii Volcanoes NP (HAVO) and Virgin Islands NP (VIIS), are omitted from the analysis. Relative humidity (RH) is needed to estimate Bsp. However, RH is not routinely measured at most IMPROVE sites. Calculation of Bsp from eq 1 or eq 2 for tracking progress under the RHR is done using monthly climatological RH for each IMPROVE site (EPA, 2003). These values were extrapolated from historical RH measurements across the United States.

Pitchford et al. (2007) compared measured and estimated Bsp at 21 IMPROVE sites where Bsp and RH were measured. Nephelometer measurements of Bsp at these sites began and ended at different times. In order to evaluate the IMPROVE Equation over a more recent time period, only sites with measured Bsp and RH data for the 10-year period of 2003–2012 were chosen: ACAD, BIBE, GRCA, GRSM, MACA, MORA, and SHEN. Hourly average Bsp and RH were also obtained from the FED website (http://views.cira.colostate.edu/fed/Datasets/RHRDataFiles.aspx).

For the nephelometer data set, there were 182,281 hours with nonmissing values for Bsp, RH, and corresponding daily chemical data, including coarse mass concentration. The nephelometer data are flagged as valid, interference, invalid, or suspect. Interference can signify fog or precipitation. There were 42,912 hours with RH ≥ 90%, but of these, only 8 cases were flagged as valid. For the following analysis, only hours flagged as valid with RH < 90% were used to estimate 24-hr average Bsp (Levin et al., 2009). Twenty-four-hour averages were accepted only when the number of hours was 18 (75%) or greater. There were 127,672 such hours, corresponding to 4184 of the 24-hr average periods.

During the review process, it was revealed that Optec NGN-2 LED nephelometers, which measure Bsp at a wavelength of 530 nm, were substituted for the original Optec NGN-2 nephelometers, which measure Bsp at a wavelength of 550 nm, at ACAD, BIBE, GRCA, GRSM, MACA, MORA, and SHEN on 11/20/11, 2/1/12, 3/27/12, 11/9/11, 11/2/11, 12/13/11, and 11/29/11, respectively (Bret Schichtel, personal communication). While Bsp measured at 530 nm should be somewhat higher than Bsp measured at 550 nm, the effect on the following analysis and results are negligible, as discussed in the following.

In the following analysis, Bsp is calculated using eq 2 with the exclusion of the last three terms, that is, light absorption by EC, Rayleigh scattering, and absorption by NO₂. The f(RH) for small- and large-mode AMSUL and AMNIT for RH ≥ 37% are taken directly from Pitchford et al. (2007). f(RH) values for water-soluble organic matter (WSOM) were derived from hygroscopic growth factors, GF, where GF is the ratio of the diameter of a hydrated particle at a specific RH to its dry diameter, determined by Lowenthal et al. (2015).

Results and discussion

OM hygroscopic growth factors and f(RH)_OM

WSOM hygroscopic growth curves (GF_OM) determined for GRSM(S), GRSM(W), MORA, and ACAD were presented by Lowenthal et al. (2015). The average of the four growth curves is plotted in Figure 1 from 30 to 90% RH. Below 30% RH, the GFs were 1. The error bars are the standard deviations of the mean growth factors from each study (i.e., the standard deviation of four data points at each RH). The data points were fitted to a third-order polynomial and the resulting function of GF_OM versus RH is shown in Figure 1. For example, predicted GF_OM are 1.01, 1.03, 1.09, and 1.15 at 30, 60, 80, and 90% RH, respectively.

Figure 1. Average hygroscopic growth factors from the four field studies at GRSM(S), GRSM(W), MORA, and ACAD. The error bars are the standard deviations of the average GFs at each RH. The solid black line is the fit to the third-order polynomial shown in the figure.

Scattering enhancement factors for OM [f(RH)_OM] were estimated as described for sulfates, nitrates, and sea salt by Pitchford et al. (2007). f(RH)_OM is defined as the ratio of Bsp at specific RH from 30 to 95% to that for RH<30%, where GF_OM = 1. The split component model assumes a bimodal OM log-normal size distribution with geometric mass mean diameters and standard deviations of 0.2 µm and 2.2 and 0.5 µm and 1.5, respectively. OM density and refractive index were assumed to be 1.4 g/cm³ and 1.55, i0, respectively, as in Pitchford et al. (2007). Water mass was added using the growth function in Figure 1 at unit RH from 30 to 95% RH. Bsp estimated using Mie theory (Mie, 1908) was integrated over the size distribution up to a diameter of 2.5 µm. The f(RH)_OM for the small [f_S(RH)_OM] and large [f_L(RH)_OM] size modes are presented in Table 1.

Table 1. Bsp enhancement factors as a function of RH [f(RH)] for small and large size distributions of OM.

RH (%)	fS(RH)OM^a	fL(RH)OM^b	RH (%)	fS(RH)OM	fL(RH)OM	RH (%)	fS(RH)OM	fL(RH)OM
0–29	1.000	1.000	52	1.377	1.306	75	1.511	1.418
30	1.321	1.267	53	1.379	1.308	76	1.524	1.428
31	1.325	1.271	54	1.382	1.309	77	1.537	1.438
32	1.329	1.274	55	1.384	1.311	78	1.551	1.449
33	1.333	1.278	56	1.387	1.313	79	1.566	1.461
34	1.337	1.280	57	1.390	1.314	80	1.582	1.473
35	1.340	1.283	58	1.393	1.316	81	1.599	1.486
36	1.343	1.286	59	1.397	1.318	82	1.617	1.500
37	1.346	1.288	60	1.400	1.320	83	1.637	1.515
38	1.349	1.290	61	1.404	1.323	84	1.657	1.531
39	1.352	1.292	62	1.409	1.325	85	1.679	1.548
40	1.354	1.294	63	1.413	1.328	86	1.703	1.566
41	1.356	1.296	64	1.419	1.331	87	1.727	1.585
42	1.358	1.297	65	1.424	1.334	88	1.754	1.605
43	1.360	1.299	66	1.430	1.338	89	1.782	1.626
44	1.362	1.300	67	1.437	1.342	90	1.812	1.648
45	1.364	1.302	68	1.444	1.346	91	1.843	1.672
46	1.366	1.303	69	1.452	1.350	92	1.877	1.696
47	1.368	1.305	70	1.460	1.355	93	1.912	1.722
48	1.369	1.306	71	1.469	1.385	94	1.950	1.750
49	1.371	1.308	72	1.478	1.393	95	1.989	1.779
50	1.373	1.309	73	1.489	1.401
51	1.375	1.311	74	1.500	1.409

Notes: ^af_S(RH)_OM for the small mode of the OM mass size distribution.

^bf_L(RH)_OM for the large mode of the OM mass size distribution.

OM and sulfate contributions to estimated Bsp

Lowenthal et al. (2015) measured average OM/OC ratios of 2.0, 2.7, 2.1, and 2.2 during the GRSM(S), GRSM(W), MORA(S), and ACAD(S) studies, respectively. The high value at GRSM(W) may be valid for that site during the study period but it is an outlier in the overall study. We therefore estimate Bsp using an OM/OC ratio of 2.1, the average for the other three studies, and compare these estimates with those calculated using OM/OC = 1.8 (Pitchford et al., 2007). Lowenthal et al. (2015) also found that the WSOC/OC ratio varied from 21% at GRSM to 93% at ACAD. The WSOC/OC ratio ranges from 0 to 1 in the following sensitivity tests. For ease of presentation, the results are averaged by RPO.

The top panel of Table 2 presents the average concentrations (dry) of PM_2.5 species by RPO. Individual sample values were averaged by site and then the site averages were averaged by RPO. The middle panel presents the corresponding chemical composition as a percent of the sum of species. This represents a base case with OM18 = 1.8 OC for calculating Bsp using eq 2. The bottom panel presents the “base case” apportionment of Bsp among the chemical species. AMSUL is the dominant component of PM_2.5 mass and Bsp in all RPOs except WRAP, where OM18 is dominant.

Table 2. Average species concentrations (dry), percent of sum of species concentrations, and percent of species light scattering by RPO.

RPO	AMSUL			AMNIT			OM18^a			SOIL			SALT			EC			N^b
	Concentration (µg/m^3)
CENRAP	2.4	±	0.8	1.16	±	0.66	2.1	±	0.7	0.79	±	0.32	0.05	±	0.02	0.23	±	0.09	18
MANEVU	2.4	±	0.7	0.45	±	0.23	1.87	±	0.36	0.26	±	0.08	0.24	±	0.28	0.23	±	0.07	13
MIDWEST	3.2	±	1.8	1.20	±	0.94	2.0	±	0.5	0.38	±	0.17	0.06	±	0.03	0.28	±	0.14	4
VISTAS	4.0	±	0.6	0.54	±	0.23	2.8	±	0.5	0.59	±	0.18	0.17	±	0.18	0.32	±	0.09	15
WRAP	0.82	±	0.33	0.35	±	0.35	1.43	±	0.56	0.67	±	0.45	0.11	±	0.29	0.14	±	0.08	89
	Percent of sum of species
CENRAP	35	±	4	15.6	±	6.1	32	±	4	12.9	±	5.7	0.95	±	0.40	3.5	±	0.4	18
MANEVU	40	±	3	8.3	±	2.3	37	±	6	5.2	±	1.4	5.2	±	5.0	4.5	±	0.9	13
MIDWEST	42	±	6	14.0	±	7.0	32	±	6	6.1	±	0.6	1.40	±	0.66	3.9	±	0.6	4
VISTAS	46	±	4	7.5	±	2.3	33	±	3	6.8	±	1.8	2.9	±	3.1	3.9	±	0.7	15
WRAP	27	±	4	9.3	±	4.5	38	±	9	18	±	7.9	3.7	±	7.1	4.0	±	1.2	89
	Percent of sum of species scattering
CENRAP	46	±	5	21	±	8	28	±	3	4.2	±	2.9	1.2	±	0.5		—		18
MANEVU	51	±	4	10.8	±	2.5	30	±	6	1.4	±	0.6	6.2	±	5.6		—		13
MIDWEST	55	±	7	18.2	±	8.8	24	±	6	1.5	±	0.2	1.7	±	0.9		—		4
VISTAS	60	±	3	9.9	±	2.8	25	±	3	1.7	±	0.6	3.5	±	3.7		—		15
WRAP	36	±	6	13.3	±	5.8	39	±	9	7.1	±	4.2	4.6	±	7.9		—		89

Note: The uncertainties (±) represent the intersite variance within each RPO.

^aBase case for OM = 1.8 × OC (OM18) and f_S(RH)_OM = f_L(RH)_OM = 1.

^bNumber of sites in each RPO.

Table 3 presents the results of varying selected parameters in eq 2. The left side of Table 3 compares the base case (Table 3, OM/OC = 1.8, OM is nonhygroscopic) apportionments of estimated Bsp to AMSUL and OM with apportionments assuming that OM/OC = 2.1, OM is hygroscopic, and the WSOC/OC ratio varying from 0 to 1. The results are averaged and presented by RPO, as in Table 2. When WSOC/OC is zero and the effect is due solely to increasing the OM/OC ratio from 1.8 to 2.1, the AMSUL contribution decreases by 2% for all RPOs and the OM21 (OM/OC = 2.1) contribution increases by 3% for all RPOs except MANEVU (4%). The relative contributions of AMSUL and OM decrease and increase further, respectively, and estimated PM_2.5 Bsp increases as the WSOC/OC ratio increases to 1. With a WSOC/OC ratio of 1, the AMSUL contribution has decreased relative to the base case by 7.4%, on average, and the OM21 contribution has increased 10.2%, on average. At the same time, increasing the contribution of OM also increases estimated PM_2.5 Bsp.

Table 3. Apportionment of estimated PM_2.5 Bsp to sulfate and OM, assuming sulfate as ammonium sulfate (AMSUL) or ammonium bisulfate (AMBSUL), OM/OC = 1.8 or 2.1, and WSOC/OC from 0 to 100%.

	Sulfate as AMSUL						Sulfate as AMBSUL
RPO^a	18–0^b	21–0	21–25	21–50	21–75	21–100	18–0	21–0	21–25	21–50	21–75	21–100
	AMSUL percent of estimated Bsp						AMBSUL percent of estimated Bsp
CENRAP	46	44	42	41	40	39	40	38	37	36	34	34
MANEVU	51	49	47	46	45	43	46	43	42	41	39	38
MIDWEST	55	53	51	50	49	47	49	47	46	45	43	42
VISTAS	60	58	56	54	53	52	55	53	51	49	48	47
WRAP	36	34	33	32	31	30	31	29	28	27	26	25
	OM percent of estimated Bsp						OM percent of estimated Bsp
CENRAP	28	31	33	35	37	38	30	33	35	37	39	40
MANEVU	30	34	36	37	39	41	33	36	38	40	42	43
MIDWEST	24	27	29	30	32	34	26	28	30	32	34	35
VISTAS	25	28	30	32	34	35	27	30	32	34	36	38
WRAP	39	42	44	46	47	49	41	44	46	48	50	51
	Estimated Bsp (Mm^−1)						Estimated Bsp (Mm^−1)
CENRAP	29	31	32	32	33	34	28	29	30	31	32	33
MANEVU	24	26	26	27	28	29	22	23	24	25	26	27
MIDWEST	38	40	41	41	42	43	36	38	39	40	40	41
VISTAS	42	44	46	47	48	50	38	40	42	43	44	46
WRAP	11.7	12.7	13.3	13.8	14.4	15.0	11.2	12.2	12.7	13.3	13.9	14.5

Notes: ^aRegional planning organizations: MANEVU (Mid-Atlantic/Northeast Visibility Union); VISTAS (Visibility Improvement State and Tribal Association of the Southeast); MIDWEST (Midwest RPO); CENRAP (Central Regional Planning Organization); WRAP (Western regional Air Partnership).

^bXX–YY, where XX is 10 OM/OC and YY is WSOC/OC, as a percentage; 18–0 is the base case.

Lowenthal et al. (2015) found that sulfate was partially neutralized, for example, as NH₄HSO₄ (AMBSUL) during two (ACAD and GRSMS) of four intensive field studies. This was also the case at GRSM during summer 1995 during the Southeastern Aerosol and Visibility Study (SEAVS) (Andrews et al., 2000; Malm et al., 2001) and at Big Bend National Park, TX (Malm et al., 2003). Since ammonium ion is not routinely measured in IMPROVE, the effect of incomplete sulfate neutralization on estimated Bsp can be evaluated by assuming that sulfate is present as AMBSUL (right side of Table 3). Scattering due to AMBSUL will decrease due in part to the difference between the molecular weights of AMSUL (132) and AMBSUL (115). AMBSUL dry scattering efficiencies for the small (D_g = 0.2 µm, σ_g = 2.2) and large (D_g = 0.5 µm, σ_g = 1.5) size modes (see earlier description) were calculated using Mie theory assuming a density of 1.78 g/cm³ and a refractive index of 1.47, i0 (Sloane, 1986). The resulting AMBSUL dry efficiencies of 1.748 (small mode) and 4.104 (large mode) m²/g are 20.6 and 14.5% lower, respectively, than the corresponding efficiencies for AMSUL, whose refractive index is 1.53, i0 (Pitchford et al., 2007). Hygroscopic growth factors for AMBSUL were calculated using the AIM model (Clegg et al., 1998). Small- and large-mode f(RH) for AMBSUL for RH ≥ 37% were calculated with Mie theory based on the assumed dry size distributions and calculated hygroscopic growth factors. The effects of lower molar mass and mass scattering efficiencies for AMBSUL compared with AMSUL are offset to some extent by its higher f(RH). At 50, 60, 70, 80, and 90% RH, the small- and large-mode AMBSUL f(RH) are 7 and 6%, 12, and 10%, 19 and 15%, 26 and 21%, and 34 and 26% higher, respectively, than the corresponding AMSUL f(RH).

The right side of Table 3 is analogous to the left side except that AMBSUL was substituted for AMSUL. For the base case where OM/OC = 1.8 and WSOC/OC = 0, the contribution of sulfates to Bsp is 5-6% lower for AMBSUL than AMSUL and the OM18 contribution is 2–3% higher. The contributions of AMBSUL and OM decrease and increase, respectively, and estimated Bsp increases as the OM/OC ratio is changed to 2.1 and the WSOC/OC ratio increases from 0 to 1. The maximum difference in the apportionments is found by comparing the first (AMSUL, 18–0) and last (AMBSUL, 21–100) columns in Table 3. Assuming sulfate is present as AMBSUL and that 100% of OM is water soluble and hygroscopic decreases the sulfate contribution to estimated Bsp by 12.4%, on average, and increases the contribution of OM to estimated Bsp by 12.2%, on average. At the same time, estimated Bsp increases by 13.5%, on average.

This analysis demonstrates that substituting measured aerosol properties in remote U.S. national parks for those assumed in IMPROVE2 can significantly affect the apportionment of estimated Bsp to sulfates and OM. This is particularly germane since estimated chemical light extinction is the only metric used to follow progress under the RHR and to devise strategies for improving visibility in Class I areas. Because measured light extinction plays no role in this process, it is important to determine how substituting these measured parameters into the IMPROVE Equation affects the comparison of estimated and measured Bsp where such measurements are available.

Comparison with measured Bsp

Measured and estimated Bsp at ACAD, BIBE, GRCA, GRSM, MACA, MORA, and SHEN are compared to evaluate the effects of changing assumptions underlying eq 2. Since the open Optec nephelometers measure total particle scattering at near-ambient RH, estimated coarse Bsp (eq 2) must be added to PM_2.5 Bsp (eq 2). Estimated Bsp was calculated using hourly RH to derive small- and large-mode f(RH) (eq 2), and these values were applied to the 24-hr IMPROVE sample concentrations. The values were then averaged to obtain a 24-hr average estimated Bsp. This was done for IMPROVE2 (base case, i.e., AMSUL, 18–0, as in Table 3), and for cases where OM/OC was assumed to be 2.1, sulfate was assumed to be present as AMBSUL, and for a range of WSOC/OC.

The comparison between measured and estimated Bsp was evaluated using the bias metric, defined as 100 (measured Bsp - estimated Bsp)/measured Bsp, as in Pitchford et al. (2007). Figure 2 compares measured and estimated Bsp for the base case. Overall, eq 2 appears to estimate measured Bsp well with an average bias of –3.1% and an R² of 0.90. However, the regression line is below the 1:1 line at high Bsp and above it at low Bsp (the two lines intersect at ~15 Mm⁻¹, near the median of measured Bsp). Estimated Bsp is larger than measured Bsp in 1936 out of 4184 cases.

Figure 2. Comparison of measured and Bsp estimated with the revised IMPROVE Equation (eq 2, Case 1, Table 4). The dashed line is the linear regression line.

Table 4 presents average bias, RH, and the number of samples (N) in quintiles of measured Bsp for all samples and for each of the seven sites. Note that estimated Bsp in the lowest (best visibility) and highest (worst visibility) quintiles of estimated Bsp are the metrics by which progress is evaluated under the RHR (EPA, 2003). Therefore, minimizing the average bias is of limited practical value. For samples from all sites combined, Case 1 (base case, eq 2) shows a bias of –21% in the lowest quintile. The bias becomes more positive until it reaches +10% in the highest quintile. This is consistent with previous evaluations of the original and revised IMPROVE Equations, which overpredicted low Bsp and underpredicted high Bsp (Pitchford et al., 2007). This trend is also apparent at BIBE, GRCA, GRSM, MACA, and SHEN. At ACAD, there is overprediction in all quintiles, although it is most severe at low Bsp. At MORA, positive bias is highest (+35%) in the highest quintile. Note that the number of samples at MORA is limited by the high frequency of RH ≥ 90% (13,659/24,388 hours).

Table 4. Distributions of bias between measured and estimated Bsp, in quintiles of measured Bsp, overall and by site.

		Case 1^a	Case 2^b	Case 3^c	Case 4^d	Case 5^e	Case 6^f	Case 7^g	RH	N	Bsp
ALL	1	–21 (–22)^h	–27	–33	–12	–18	–24	–30 (–30)	44	837	5
ALL	2	–7 (–8)	–12	–18	1	–4	–10	–15 (–16)	47	836	11
ALL	3	–3 (–4)	–8	–14	5	0	–5	–11 (–12)	51	837	18
ALL	4	5 (5)	1	–4	13	9	4	–2 (–2)	57	837	30
ALL	5	10 (8)	6	1	18	14	10	5 (4)	64	837	71
ACAD	1	–34	–42	–53	–25	–33	–44	–56	58	121	5
ACAD	2	–10	–16	–25	–2	–8	–17	–25	58	123	9
ACAD	3	–7	–13	–21	0	–6	–14	–22	61	122	12
ACAD	4	–2	–8	–16	4	–2	–10	–18	64	122	18
ACAD	5	–4	–9	–17	3	–2	–10	–17	68	122	38
BIBE	1	–24	–28	–31	–14	–19	–22	–24	31	200	6
BIBE	2	–5	–9	–12	5	1	–2	–5	36	200	11
BIBE	3	0	–4	–7	10	7	4	1	40	201	17
BIBE	4	10	7	5	21	18	15	12	44	200	25
BIBE	5	18	15	12	28	25	22	20	45	200	46
GRCA	1	–36	–42	–48	–25	–31	–37	–43	44	161	2
GRCA	2	–2	–7	–12	4	0	–5	–10	45	164	5
GRCA	3	–2	–7	–11	6	1	–3	–7	40	162	8
GRCA	4	–1	–6	–10	6	0	–4	–8	40	163	11
GRCA	5	14	8	4	19	13	9	5	43	162	20
GRSM	1	–25	–34	–43	–15	–24	–33	–42	50	123	13
GRSM	2	–6	–12	–19	4	–2	–9	–16	56	123	23
GRSM	3	0	–5	–12	10	4	–2	–9	60	124	35
GRSM	4	11	6	0	20	16	10	4	64	123	57
GRSM	5	14	10	6	26	22	18	13	69	123	113
MACA	1	–20	–27	–35	–13	–19	–27	–35	56	103	17
MACA	2	–6	–11	–18	1	–4	–11	–17	59	104	28
MACA	3	1	–4	–10	7	2	–4	–10	62	104	39
MACA	4	7	3	–2	12	8	3	–2	64	105	55
MACA	5	9	5	1	16	12	8	4	67	103	95
MORA	1	9	4	–5	12	7	–2	–11	71	28	6
MORA	2	10	4	–7	13	6	–5	–15	71	28	13
MORA	3	–1	–10	–21	2	–6	–18	–29	64	29	18
MORA	4	6	–2	–12	9	1	–9	–19	67	28	24
MORA	5	35	27	17	36	29	19	9	70	28	41
SHEN	1	–20	–25	–31	–10	–16	–22	–28	46	97	11
SHEN	2	–10	–14	–19	–1	–5	–11	–16	57	97	19
SHEN	3	3	–1	–5	9	5	1	–4	63	97	27
SHEN	4	3	–1	–5	11	7	3	–2	65	97	39
SHEN	5	11	8	4	21	18	14	10	71	97	84

Notes: ^aRevised IMPROVE Equation.

^bCase 2, as in Case 1 but with OM/OC = 2.1

^cCase 3, as in Case 1 but with OM/OC = 2.1 and hygroscopic OC with WSOC/OC = 0.5.

^dCase 4, as in Case 1 but with sulfate as ammonium bisulfate.

^eCase 5, as in Case 1 but with sulfate as ammonium bisulfate and OM/OC = 2.1.

^fCase 6, as in Case 1 but with sulfate as ammonium bisulfate, OM/OC = 2.1, and hygroscopic OC with WSOC/OC = 0.5.

^gCase 7, as in case 1 but with sulfate as ammonium bisulfate, OM/OC = 2.1, and hygroscopic OC with WSOC/OC = 1.

^hExamples of bias calculated for the first 9 years only, before the Optec NGN-2 with a wavelength of 550 nm was replaced by the Optec NGN-2 LED with a wavelength of 530 nm.

As noted earlier, Optec NGN-2 nephelometers, which measure Bsp at 550 nm, were replaced by Optec NGN-2 LED nephelometers, which measure Bsp at 530 nm. This impacted the Bsp measurements at the end of 2012 and during 2012. The effect of this change on the results presented in Figure 2 and Table 4 was evaluated using two approaches: (1) using Mie theory to calculate the wavelength dependence of the dry scattering efficiencies and f(RH); and (2) comparing the bias for all years with the bias calculated for the period (approximately 9 years) when the 550 nm nephelometers were used. Based on Mie theory, the small-mode mass scattering efficiencies are about 3, 3, and 9% higher at 530 than 550 nm for small-mode AMSUL, AMNIT, and OM, respectively. The differences are somewhat smaller for the large mode. F(RH) is the ratio of wet to dry Bsp. It should not vary significantly for the two wavelengths. This was confirmed using Mie theory to recalculate f(RH)_OM at 530 nm. OM was chosen as it exhibited the largest difference with respect to the change in wavelength. The average differences for f(RH)_OM calculated at 550 and 530 nm were 0.13 and –0.10% for the small and large OM modes, respectively.

The maximum difference between Bsp measured at 550 and 530 nm is probably smaller than the than 9% calculated for the OM mass scattering efficiency because the differences are smaller for AMSUL and AMNIT. The overall effect will be to reduce overestimation at low Bsp and increase underestimation at high Bsp. However, since the change in the nephelometer wavelength affects only 10% of the data, the error in the 10-year average bias is probably less than 1%. In Table 4, this amounts to the rounding error. Next, estimated Bsp was recalculated after eliminating data when the NGN-2 LED nephelometers were used. The remaining data set represented 3724 days. The overall bias shown in Figure 2 and for Case 1 (Table 4) changed from –3.1 to –4.0%. The average biases for the reduced data set in quintiles of measured Bsp are shown in parentheses in Table 4 for the calculations for ALL samples, Case 1 and Case 7. For both cases, the differences in bias are no more than 2%. Further, the differences in bias between Case 1 and Case 7 for the full and reduced data set are no more than 1%. Note that some of these differences result from real variation in aerosol composition and concentration in the 9-year subset compared with the entire data set. The contribution of the wavelength difference in the 10th year to any of the values in Table 4 is unlikely to be more than a fraction of a percent.

The split mode model in eq 2 was intended to address systematic over- and underprediction at low and high Bsp, respectively, but this has apparently not resolved the problem. Small- and large-mode concentration-weighted mass scattering efficiencies for AMSUL, AMNIT, and OM18 (Case 1) are presented in Table 5 Average dry mass scattering efficiencies and their standard deviations are given for all samples and in quintiles of measured Bsp. Nearly all of the values in Table 5 are lower than the efficiencies in the original IMPROVE Equation (eq 1). While the average efficiencies for the lowest quintile are nearly the same as the small-mode efficiencies (eq 2), there is still overprediction in the lowest quintiles of Bsp (Table 4). Conversely, the efficiencies in the upper quintile are not high enough to reproduce the measured Bsp.

Table 5. Average mass scattering efficiencies (m²/g) overall and by quintile of Bsp, with the small- and large-mode mass scattering efficiencies (eq 2) shown for comparison.

Quintile	AMSUL			AMNIT			OM
ALL	2.5	±	0.3	2.5	±	0.1	3.1	±	0.3
1	2.3	±	0.04	2.4	±	0.01	2.9	±	0.1
2	2.4	±	0.1	2.4	±	0.04	3.0	±	0.1
3	2.4	±	0.1	2.5	±	0.1	3.1	±	0.1
4	2.6	±	0.2	2.5	±	0.1	3.1	±	0.2
5	3.0	±	0.4	2.5	±	0.2	3.4	±	0.3
Small	2.2			2.4			2.8
Large	4.8			5.1			6.1

One way to lower estimated Bsp at low measured Bsp is to lower one or more of the dry mass scattering efficiencies in the low concentration mode. This can be done by reducing Dg in the lower concentration mode. However, this will also reduce estimated Bsp at high Bsp. The upper-mode mass scattering efficiency for AMSUL (4.8 m²/g) is at a maximum for any Dg with a σg of 1.5. For a log-normally distributed AMSUL mass size distribution, the maximum mass scattering efficiency occurs at a Dg of 0.5 µm. An efficiency larger than 4.8 m²/g can be obtained by lowering the σg below 1.5 but there is no evidence for such a narrow size distribution.

Further, the split between low- and high-mode concentrations may not be realistic. For this data set, 88 and 91% of AMSUL and OM18 concentrations, respectively, are in the low concentration mode. Pitchford et al. (2007) defined the large-mode concentration, for example, AMSUL_L, as the total PM_2.5 ammonium sulfate concentration (AMSUL) divided by 20. The use of 20 µg/m³ to split the modes appears to be too high. Only 5 (0.1%), 27 (0.6%), and 101 (2.4%) of the 4184 samples had AMSUL concentrations greater then 20, 15, and 10 µg/m³, respectively. The average concentration in the highest quintile of AMSUL was only 6.5 µg/m³. It is beyond the scope of this work to devise a more realistic approach for determining dry mass scattering efficiencies that eliminates the biases in estimating low and high Bsp. Perhaps a nonlinear split mode system, analogous to that suggested by Ryan et al. (2005), could be explored. Any generalized system will be ad hoc to some extent, since information needed for estimating realistic dry mass scattering efficiencies, for example, chemically speciated dry mass size distributions, is not likely to be measured on a routine or comprehensive basis.

The average bias and RH in each quintile of measured Bsp are plotted by site in Figure 3 There is a monotonic increase in bias with measured Bsp, which is accompanied by a similar increase in RH at BIBE, GRSM, MACA, and SHEN. The correlation between bias and RH over the 5 quintiles of Bsp was 0.98, 0.98, 0.96, and 0.98 at BIBE, GRSM, MACA, and SHEN, respectively. The direct relationship between bias and RH, at least at these sites, suggests a deficit in estimated hygroscopic growth. This is consistent with our observation that OM is hygroscopic. The correlations between bias and RH at ACAD, GRCA, and MORA were 0.62, 0.25, and 0.55, respectively, suggesting that other factors also contribute to bias at these sites.

Figure 3. Average bias in each quintile of Bsp (solid). Average fractional RH in each quintile of Bsp (dashed).

As seen earlier (Table 4), assuming an OM/OC ratio of 2.1 and that some fraction of the OM is hygroscopic will tend to increase estimated Bsp. Assuming that sulfate is partially neutralized as AMBSUL will tend to reduce estimated Bsp. This is illustrated in Cases 2–7, Table 4, where each case is defined relative to Case 1 (eq 2, base case). For Case 2 (OM/OC = 2.1), bias becomes more negative in the lowest quintile of measured Bsp but decreases from +10 to +6% in the highest quintile. Assuming that OM/OC =2.1 and that 50% of OM is water soluble (Case 3) decreases bias in the highest quintile to +1%. Assuming that sulfate is present as AMBSUL (Case 4) increases bias to +18% in the highest quintile. Assuming AMBSUL and OM/OC = 2.1 (Case 5) decreases the positive bias to +14%. Assuming AMBSUL, OM/OC = 2.1, and WSOC/OC = 0.5 (Case 6) and 1.0 (Case 7) lowers the bias in the highest quintile to +10 and +5%, respectively. Factors that reduce overprediction in the lowest quintile increase underprediction in the highest quintile.

The trends for the individual sites are similar to those seen over all samples. While low and high Bsp cannot both be estimated well with IMPROVE2 under any set of assumptions, it may be useful to focus only on the worst visibility cases, that is, the highest quintile of measured Bsp, as the primary goal of the RHR is to improve the worst visibility days. With that restriction, the best performance for ALL samples is found for Case 3, assuming AMSUL, OM/OC = 2.1, and hygroscopic OM with WSOC/OC = 0.5. This is also the case at GRCA, GRSM, MACA, and SHEN. Note that for these sites, the next best outcome was obtained for Case 7 (AMBSUL, OM/OC = 2.1, and hygroscopic OM with WSOC/OC = 1). With the worst visibility cases systematically underestimated by IMPROVE2, increasing the OM/OC ratio to 2.1 and allowing for enhanced light scattering from hygroscopic organics are reasonable revisions to the IMPROVE Equation for estimating light extinction in Class I areas.

Conclusions

The [revised] IMPROVE Equation (IMPROVE2) is used to estimate chemical light extinction to track progress under the Regional Haze Rule, the goal of which is to improve visibility in U.S. Class I areas. Our research in national parks indicates that some assumptions underlying IMPROVE2 are not realistic (Lowenthal et al., 2015). For example, the measured OM/OC ratio was closer to 2.1 than to the currently used value of 1.8. Contrary to IMPROVE2, water-soluble organic matter (OM) absorbs water as a function of RH. Sulfate was not always completely neutralized, as is assumed in IMPROVE2, but may have been present as ammonium bisulfate at Great Smoky Mountains National Park and Acadia National Park. These observations lead to a larger apportionment of light extinction to OM relative to sulfate than is implied by IMPROVE2 and should be considered in devising strategies for reducing emissions from which these chemical species are derived. Assuming that sulfate is present as ammonium sulfate, the OM/OC ratio is 2.1, and all OM is water-soluble, the apportionment of PM_2.5 light scattering (Bsp) to sulfates and OM decreases and increases relative to the current IMPROVE Equation by 7.4 and 10%, respectively, across sites in the five regional planning organizations. Under the same assumptions, except that sulfate is present as ammonium bisulfate, the apportionment of PM_2.5 Bsp to sulfates and OM decreases and increases relative to IMPROVE2 by 12.4 and 12.2%, respectively.

The reliance of the Regional Haze Rule on estimated light extinction allows for assigning light extinction to specific chemicals and their sources, but it remains problematic because the IMPROVE2 overestimates Bsp at low measured Bsp and underestimates it at high Bsp. For the data set examined here, IMPROVE2 underpredicted the upper quintile of measured Bsp, that is, cases with the worst visibility, by 10%, on average. This bias can be reduced by assuming that the OM/OC ratio is 2.1 rather than 1.8, accounting for the hygroscopic growth of water-soluble OM, or assuming that sulfate is present as ammonium bisulfate. However, increasing estimated Bsp and reducing low bias at high measured Bsp also increases overprediction at low Bsp. The split-mode model for estimating dry mass scattering efficiencies does not resolve these biases and should be reexamined. The problem of correctly estimating light extinction with the IMPROVE Equation remains unconstrained due to the lack of routine measurements of sulfate neutralization and the water-soluble fraction of OM. Given the importance of worst visibility days under the RHR and the systematic underestimation of the worst visibility cases by IMPROVE2, increasing the OM/OC ratio to 2.1 and allowing for enhanced light scattering from hygroscopic organics are reasonable revisions to the IMPROVE Equation for estimating light extinction in Class I areas until a better algorithm can be developed that works in all cases.

Funding

This research would not have been possible without the cooperation of the National Park Service and its personnel at Great Smoky Mountains National Park, Mount Rainier National Park, and Acadia National Park. The funding support was provided to Desert Research Institute from EPRI under contract 10002378.

Additional information

Funding

This research would not have been possible without the cooperation of the National Park Service and its personnel at Great Smoky Mountains National Park, Mount Rainier National Park, and Acadia National Park. The funding support was provided to Desert Research Institute from EPRI under contract 10002378.

Notes on contributors

Douglas H. Lowenthal

Douglas Lowenthal is a research professor at the Desert Research Institute, Nevada System of Higher Education, in Reno, NV.

Naresh Kumar

Naresh Kumar is a Senior Program Manager at EPRI, Palo Alto, CA.