Characterization of Atmospheric PM_2.5 Inorganic Aerosols Using the Semi-Continuous PPWD-PILS-IC System and the ISORROPIA-II

Abstract

A semi-continuous monitoring system, a parallel plate wet denuder and particle into liquid sampler coupled with ion chromatography (PPWD-PILS-IC), was used to measure the hourly precursor gases and water-soluble inorganic ions in ambient particles smaller than 2.5 µm in diameter (PM_2.5) for investigating the thermodynamic equilibrium of aerosols using the ISORROPIA-II thermodynamic equilibrium model. The 24-h average PPWD-PILS-IC data showed very good agreement with the daily data of the manual 5 L/min porous-metal denuder sampler with R² ranging from 0.88 to 0.98 for inorganic ions (NH₄⁺, Na⁺, K⁺, NO₃⁻, SO₄²⁻, and Cl⁻) and 0.89 to 0.98 for precursor gases (NH₃, HNO₃, HONO, and SO₂) and slopes ranging from 0.94 to 1.17 for ions and 0.87 to 0.95 for gases, respectively. In addition, the predicted ISORROPIA-II results were in good agreement with the hourly observed data of the PPWD-PILS-IC system for SO₄²⁻ (R² = 0.99 and slope = 1.0) and NH₃ (R² = 0.97 and slope = 1.02). The correlation of the predicted results and observed data was further improved for NH₄⁺ and NO₃⁻ with the slope increasing from 0.90 to 0.96 and 0.95 to 1.09, respectively when the HNO₂ and NO₂⁻ were included in the total nitrate concentration (TN = [NO₃⁻] + [HNO₃] + [HONO] + [NO₂⁻]). The predicted HNO₃ data were comparable to the sum of the observed [HNO₃] and [HONO] indicating that HONO played an important role in the thermodynamic equilibrium of ambient PM_2.5 aerosols but has not been considered in the ISORROPIA-II thermodynamic equilibrium model.

1. Introduction

Particles smaller than 2.5 µm in diameter (PM_2.5) exposure has been confirmed to be associated with total mortality, cardiovascular and respiratory mortalities, lung cancer, influenza, etc. [1] due to its chronic adverse human health effects [2]. Some previous studies [3,4] have revealed that PM_2.5 constituents such as sulfate, nitrate, and ammonium were responsible for these effects. Sulfate (SO₄²⁻), nitrate (NO₃⁻), and ammonium (NH₄⁺) are the major water-soluble inorganic (WSI) species, which represent very large fractions in PM_2.5 depending on source emissions, human activities, chemical reactions, and meteorological conditions [5]. For example, SO₄²⁻, NO₃⁻, and NH₄⁺ ions accounted for 39% to 63% on the episode days and from 14.5% to 39% on the non-episode days in Taiwan [6]. These WSI species influence the hygroscopic nature, acidity properties, optical properties, and lifetime of PM_2.5 [7,8] in which SO₄²⁻, NO₃⁻, and NH₄⁺ ions are the secondary species mainly formed from the chemical reactions of precursor gases (NH₃, HNO₃, and H₂SO₄) and nucleation and condensation processes [9]. For instance, up to 80% of NO₃⁻ and NH₄⁺ in PM_2.5 can be formed from precursor gases in the South Coast Air Basin in the USA [10].

Generally, NH₃ prefers to react homogeneously with H₂SO₄ to form (NH₄)₂SO₄ first, and then the excess NH₃ reacts with HNO₃ to form NH₄NO₃ when the molar ratio of NH₄⁺/SO₄²⁻ is greater than two [11,12]. Therefore, the total molar concentration of SO₄²⁻ and NO₃⁻ shows a good correlation (R² > 0.9) with that of NH₄⁺ [12]. The formation of sulfate and nitrate particles is also affected by environmental factors including temperature (T), relative humidity (RH) [12], wind speed [13], and pressure [11]. As compared with (NH₄)₂SO₄, NH₄NO₃ is more unstable at high temperature and low relative humidity conditions [5] due to the reversible phase equilibrium with HNO₃ and NH₃. The NO₃⁻ concentration can also be increased by the uptake of HNO₃ on particles at high RH and low T conditions [13,14]. In addition, NO₃⁻ _(p) along with HONO can also be generated by the hydrolysis of NO₂ on the particle surface with the NH₃ promotion [15,16]. The additional HONO source also contributes to the formation of NO₃⁻_(p) due to the enhanced oxidation of hydroxyl radical (OH) with NO₂ to produce more HNO₃ [17]. In addition, HNO_3(g)/NO₃⁻_(p) can be photolyzed to form HONO under moderate to low NO₂ conditions [18,19].

To determine the chemical composition of PM_2.5 aerosols by measuring the mass concentrations of precursor gases and particles simultaneously, a system consisting of a manual filter sampler and a denuder (i.e., annular denuder, coiled denuder, and honeycomb denuder) is usually used [20] since manual filter samplers show good measurement accuracy [21]. The samples are, then, extracted and mass concentrations are quantified using ion chromatography (IC). However, manual filter samplers have negative artifacts due to the semi-volatile nature of NH₄⁺ and NO₃⁻ [22] resulting in underestimation [23,24]. A porous-metal denuder sampler (PDS) [25], which uses acidic/basic coated porous metal discs to capture the basic/acidic precursor gases combined with a Teflon filter to collect particles and back-up filters to collect the semi-volatile materials evaporated from collected particles, can be used for determining the accurate mass concentrations of gases and particles simultaneously without evaporation loss [26]. Nevertheless, this method can be time-consuming and labor-intensive and is not fast enough to characterize the rapid evolution of atmospheric particles.

Currently, there are some hourly monitoring systems for water-soluble ions in PM_2.5 including the wet-annular denuder and steam-jet aerosol collector (WAD/SJAC) [27], the monitor for aerosols and gases in ambient air (MARGA) [28], the gas particle ion chromatography system (GPIC) [29], the ambient ion monitor ion chromatograph (AIM-IC) [30], and the parallel plate wet denuder and particle into liquid sampler) (PPWD/PILS) [31]. Some previous studies have shown that the WAD/SJAC measured NH₄⁺ and SO₄²⁻ ion concentration incorrectly at high mass loading conditions [27], the MARGA biased low HNO₃ and NH₃ gas concentrations as compared with those measured by the denuder/filter pack method [32], the GPIC overestimated NH₃ and SO₂ mass concentrations and underestimated HNO₃ concentration as compared with the manual denuder sampler [33], and the AIM-IC overestimated SO₄²⁻ ion concentration [34] and underestimated 11% NH_3(g), 19% SO_2(g) and 12% HNO_3(g) [35]. In comparison, the PPWD/PILS consisting of PPWD [36,37] and PILS [38,39] and coupled with the IC (i.e., PPWD-PILS-IC) developed by our group [31] showed good agreement with the PDS for precursor gases (slope 0.92–.04) and PM_2.5 ion species as well (slope 0.75–0.97 and R²: 0.77–0.94). In addition, the PPWD-PILS-IC system has low MDLs (method detection limit) which are less than 0.01 µg/m³ for Na⁺, Ca²⁺, NO₂⁻, and SO₄²⁻ and 0.02 µg/m³ for NH₄⁺, NO₃⁻, and Cl⁻, respectively. However, the performance of the PPWD-PILS-IC needs to be studied further to improve the linear regression parameters for gases (R² 0.76–0.83) and NH₄⁺ ion (slope 0.84 and R² 0.89).

The equilibrium partitioning of the WSI species in different phases (particle (p), aqueous (aq), and gas (g) phases) at different T and RH conditions is complex and nonlinear, which needs the support of some thermodynamic equilibrium models to simulate the physical and thermodynamic processes. To meet these needs, many thermodynamic equilibrium models have been developed including AIM2 [40], SCAPE2 [41], EQUISOLV II [42], ISORROPIA/ISORROPIA-II [43,44], GFEMN [45], EQSAM2 [46], HETV [47], MESA [48], and UHAERO [49]. Among these methods, the ISORROPIA-II was used extensively since it includes the crustal elements (K⁺, Ca²⁺, Mg²⁺, and Na⁺), can simulate the thermodynamic equilibrium of particles in the stable state or metastable state, and can be used to solve the forward problem or the reverse problem in which particle-phase species can be used to determine the partial pressure of the gas phase species under the equilibrium condition [47]. Additionally, the ISORROPIA-II thermodynamic equilibrium model has been proven to be an order of magnitude faster than the SCAPE2 [44]. Previous studies have shown that the ISORROPIA-II thermodynamic equilibrium model predicted accurately for NH₃ but had biases for NH₄⁻, NO₃⁻, and HNO₃ as compared with the observations due to the uptake of HNO₃ on coarse particles [44,50,51] or the evaporation properties of NH₄NO₃ salt [52]. Therefore, this model should be studied further to improve its partitioning capability.

In this study, the hourly concentrations of precursor gases (NH₃, HNO₃, SO₂, and HONO) and WSI ions (NH₄⁺, Na⁺, K⁺, SO₄²⁻, NO₃⁻, and Cl⁻) in PM_2.5 were measured by the PPWD-PILS-IC monitoring system and the 24-h average data were averaged from the hourly concentrations and compared to the daily average data of the manual 5 L/min PDS (daily data no. (N) = 30). Then, the hourly data (N = 720) were input to the ISORROPIA-II thermodynamic equilibrium model to study the thermodynamic equilibrium of ambient PM_2.5 WSI aerosols.

2. Experimental Method

2.1. Five Liters Per Minute PDS Design and Laboratory Test

In this study, the 5 L/min PDS was scaled up from the original 2 L/min PDS developed by our group [25,26] to increase the collected mass. The new thinner and larger porous metal discs could be cleaned easily to minimize the background influence. As shown in Figure 1, the 5 L/min PDS, henceforth referred to as PDS, consisted of a 2.5 µm cutpoint cyclone to remove particles larger than 2.5 μm, two coated porous metal discs with a diameter of 37 mm, a thickness of 2.5 mm, and a pore diameter of 100 μm (P/N1000, Mott Inc., Farmington, CT) to capture precursor gases, 47 mm Teflon filter to collect non-volatile species in PM_2.5, and two back-up filters (47 mm nylon filter and 47 mm acid-coated quartz filter) to collect acidic and basic semi-volatile species, respectively. The first and second porous metal discs were coated with sodium carbonate/glycerin (1% w/v) and citric acid (1% w/v) to collect acid gases (HNO₂, HNO₃, and H₂SO₄) and basic gas (NH₃), respectively. The inner wall of the PDS was coated with Teflon material to avoid gas adsorption and minimize the loss of acidic/basic gases. The mass concentration of each WSI species is the sum of those on Teflon filter and back-up filters.

The preparation of the porous metal discs and filters is shown in detail in Section S1 in the Supplementary Information (SI) and described briefly here. The porous metal discs were cleaned 4–5 times with deionized (DI) water under the vacuum condition and ultrasonicated for 15–30 min before every coating and sampling process to eliminate the background effect. After cleaning, the background concentrations of the cleaned porous metal discs were 0.002, 0.008, 0.035, and 0.005 ppbv for NH₃, HCl, HNO₂, HNO₃, and SO₂⁻, respectively. After sampling, the porous metals were extracted with H₂O₂ solution (15 mL and 5 mM) under the vacuum condition and ultrasonicated for 30 min, and the filters were extracted with DI water for 60 min before the extracted samples were analyzed by IC to determine mass concentrations.

The PDS was tested for the particle penetration curve of the cyclone and the particle loss in the laboratory first, before the field test. Two porous metal discs and three filter holders were removed for testing the penetration curve of the cyclone, and then the cyclone and the three filter holders were removed for testing the particle loss of the porous metal discs. Figure 2 shows the experimental setup of the cyclone penetration curve and the PDS particle loss tests with the method shown in the previous studies [53,54]. A vibrating orifice aerosol generator (VOAG, TSI Model 3450, USA) was used to generate monodisperse NaCl particles ranging from 1.20 to 5.05 μm and an aerodynamic particle sizer (APS, model 3321, TSI Incorporated, St. Paul, MN, USA) was used to measure the upstream and downstream particle concentrations. The collection efficiency (%) and the particle loss (%) of each particle size are calculated as the difference of the upstream and downstream concentrations divided by the upstream concentration times 100%.

A spreadsheet was used to perform the numerical integration of the cyclone penetration curve and the particle loss curve with the three idealized ambient distributions to estimate the collected PM_2.5 and particle loss mass concentrations [53]. The percentage of the total particle loss is the ratio of the estimated mass concentration of particle loss and collected PM_2.5 times 100%.

2.2. Field Comparison Test

The field comparison test of the PPWD-PILS-IC system and the PDS was conducted at the 6th floor of the building of the Institute of Environmental Engineering, National Chiao Tung University (NCTU), Hsinchu City, Taiwan from 1 October 2018 to 15 April 2019. The sampling site is about 1 km away from a heavy-traffic road which is the major particle source at this site [55]. Thirty daily samples were collected by the PDS and 720 hourly monitoring data were obtained by the PPWD-PILS-IC system, respectively. The hourly data of the monitoring system was converted to the 24-h average data for comparison with the PDS data. The PPWD-PILS-IC monitoring system was presented in detail in Li et al. [31] and is described briefly below. A 16.7 L/min EPA PM₁₀ inlet [54] and a PM_2.5 VSCC inlet (very sharp cut cyclone) were used to remove particles larger than 2.5 μm in aerodynamic diameter. Sample air was drawn through the PPWD at 12.3 L/min to capture water-soluble precursor gases [36,37] first, before being introduced into the PILS to collected WSI ions [38]. The samples from the PPWD and the PILS were stored temporarily in syringes, and then injected in the IC in sequence for determining hourly mass concentrations of water-soluble gases and ions. The PDS sampling system was set up side-by-side with the PPWD-PILS-IC system, in which the flow rate was controlled at 5 L/min by a mass flow controller. The cyclone is covered by a rain cap and its inner wall is coated with silicone grease to eliminate the particle bounce effect.

To examine the ion balance measured by the monitoring system and the sampler, the anion and cation equivalents are calculated from the mass concentrations of WSI ions (i.e., [ion]) as:

Anion equivalent (A) = [F⁻]/19 + [Cl⁻]/35.5 + [NO₃⁻]/62 + [NO₂⁻]/46 + [SO₄²⁻]/48

(1)

Cation equivalent (C) = [Na⁺]/23 + [NH₄⁺]/18 + [K⁺]/39.1 + [Mg₂⁺]/12 + [Ca₂⁺]/20

(2)

To evaluate the performance of the monitoring system as compared to the PDS, the normalized mean bias (NMB, %) and the normalized mean difference (NMD, µg/m³) for the PPWD-PILS-IC data are calculated as:

$$\mathrm{NMB}(\%)=\frac{{{\displaystyle \sum }}_{i=1}^N\frac{C_1-C_2}{C_2}}{N}\times 100\%$$

(3)

$$\mathrm{NMD}\left(\raisebox{1ex}{$\mathsf{\mu }\mathrm{g}$}\!\left/ \!\raisebox{-1ex}{${\mathrm{m}}^3$}\right.\right)=\frac{{{\displaystyle \sum }}_{i=1}^N{C}_1-C_2}{N}$$

(4)

where C₁ (µg/m³) and C₂ (µg/m³) are the mass concentrations of the PPWD-PILS-IC and the PDS for different WSI ions and gases, respectively. N is the total number of daily samples (N = 30).

2.3. ISORROPIA-II Thermodynamic Equilibrium Model

To predict the equilibrium partitioning of WSI gases and species in PM_2.5, the ISORROPIA-II thermodynamic equilibrium model (NH₄⁺–Na⁺–K⁺–Mg²⁺–Ca²⁺–Cl⁻–NO₃⁻–SO₄²⁻–H₂O system) was used [44]. The forward mode (or closed system) and stable state were applied, in which the total concentrations of nitrate (TN = [HNO₃] + [NO₃⁻]), ammonium (TA = [NH₃] + [NH₄⁺]), other ion concentrations, temperature, and relative humidity were used as the input parameters. In this study, the TN input parameter was also revised as TN = [HNO₃] + [NO₃⁻] + [HONO] + [NO₂⁻] to evaluate the effect of HONO on the nitrate (HNO₃/NO₃^-) system and ammonium (NH₃/NH₄⁺) system. To evaluate the performance of the ISORROPIA-II thermodynamic equilibrium model, the NMB and the NMD are calculated using Equations (3) and (4), in which C₁ and C₂ are the mass concentrations of the ISORROPIA-II thermodynamic equilibrium model and the PPWD-PILS-IC for different WSI ions and gases, respectively with N = 720.

3. Results and Discussion

3.1. Laboratory Test Results of the 5 L/min PDS

Figure 3 shows the particle penetration curve of the cyclone (Figure 3a) and the particle loss curve of the PDS (Figure 3b). The cutpoint diameter (D_pa50) of the cyclone is 2.45 µm, which meets the U.S. EPA criteria for PM_2.5 inlets (D_pa50 = 2.5 ± 0.2 µm) and is similar to that of the VSCC (D_pa50 = 2.52 ± 0.02 µm) [56]. The sharpness of the particle penetration curve is 1.27, which is less sharp than that of the VSCC (sharpness = 1.16) but still meets the requirements of 1.2 ± 0.1 regulated in China and European countries [57]. It indicates that the PDS cyclone is applicable as a PM_2.5 classifier. In addition, the particle loss of the PDS casing is very small, which is less than 2% for particles smaller than 4.0 µm, whereas the particle loss of the two porous metal discs is from 1.9% to 67.5% with particles ranging from 1.2 to 5.0 µm, that is, the porous metal discs can cause a large particle loss for coarse particles while the loss for fine particles is smaller than 14%. Moreover, the total PM_2.5 loss is 2.8, 2.2%, and 3.7% based on the three idealized fine, typical, and coarse ambient particle size distributions, respectively with the average particle loss of only 2.9 ± 0.8% [58]. It means that the PM_2.5 loss of the porous metal disc is acceptable and would not affect the sampling accuracy of the PDS.

3.2. Field Comparison Test Results of PPWD-PILS-IC and PDS

Table 1. Mass concentration and percent mass of the particles smaller than 2.5 µm in diameter (PM_2.5) water-soluble inorganic species measured by the PDS.

NCTU	Na+	NH4+	K+	Mg2+	Ca2+	F−	Cl−	NO2−	NO3−	SO42−	Total	A/C
Mass conc. (μg/m3)	0.35 ± 0.19	3.22 ± 1.38	0.20 ± 0.07	0.07 ± 0.02	0.04 ± 0.02	0.31 ± 0.15	0.72 ± 0.33	0.22 ± 0.33	2.58 ± 1.60	4.02 ± 1.87	12.63 ± 1.71	0.92 ± 0.2
Percentage (%)	2.74	25.52	1.57	0.58	0.32	2.47	5.71	1.75	20.41	38.94	100

shows the average mass concentration and the mass percentage of each WSI species over the total mass concentration of the WSI species measured by the PDS. Among these WSI species, the mass concentrations of NH₄⁺, NO₃⁻, and SO₄²⁻ account for 25.52%, 20.41%, and 38.94%, respectively with the mass concentrations ranging from 0.61 to 5.87 μg/m³, from 0.95 to 6.64 μg/m³, and from 1.08 to 8.27 μg/m³, respectively. The other WSI species’ mass concentrations represent just less than 6.0% in PM_2.5 with the average mass concentrations lower than 1.0 μg/m³. It implies that SO₄²⁻ is the most dominant ion in PM_2.5 followed by NH₄⁺ and NO₃, while Na⁺, K⁺, Ca²⁺, Mg², and Cl⁻ are the minor contributors to PM_2.5, at the NCTU site. The equivalent ratio of anions and cations (A/C ratio) of PM_2.5 WSI species collected by the PDS is 0.92 ± 0.20, which falls within the range of acceptable ion balance of 0.85–1.15 [59] indicated that all the WSI ions in PM_2.5 were measured accurately.

The field comparison results of the PPWD-PILS-IC with the PDS for some major WSI species (Na⁺, K⁺, NH₄⁺, Cl⁻, NO₃⁻, and SO₄²⁻) are shown in Figure 4. It is found that the linear regression slopes and coefficient (R²) of the determination range from 0.99 to 1.17 and from 0.88 to 0.98, respectively, which are better than those (slope 0.75–0.97 and R² 0.77–0.94) in Li et al. [31]. For K⁺, Na⁺, and NH₄⁺, the improvement is even more obvious with higher R² of 0.89, 0.92, and 0.93, respectively (R² = 0.77, 0.77, and 0.89, respectively in the previous study) implying good stability of the monitoring system and the manual sampler. The slopes of all WSI species are close to 1.0 with the NMBs (NMDs) of the PPWD-PILS-IC less than ±10% (<±0.2 µg/m³) except for NH₄⁺ (NMB = −13.95% and NMD = −0.36 µg/m³). This implies that the evaporation loss of NH₄⁺ is similar to that found in previous studies [31,60].

The field comparison results of the PPWD-PILS-IC with the PDS for the precursor gases (NH₃, SO₂, HNO₃, and HNO₂) are shown in Figure 5. The average volume concentrations of NH₃, SO₂, HNO₃, and HNO₂ are 4.37 ± 2.40, 0.79 ± 0.38, 0.27 ± 0.18, and 0.73 ± 0.26 ppbv, respectively. The results show very good agreement between the PPWD-PILS-IC data and the PDS data with the slope and R² varying from 0.87 to 0.95 and from 0.89 to 0.98, respectively and the NMBs (MNDs) less than ±10% (<0.1 ppbv). It is also found that the agreement between two methods for these precursor gases is improved as compared with that in the previous study. In this study, the PDS with a higher flow rate (5 L/min) and easy-to-clean discs are the key to the improvement. Now, the discs can be cleaned thoroughly with low blank values (<0.008 ppbv). In summary, the PPWD-PILS-IC system is validated with the current 5 L/min PDS for measuring hourly precursor gases and WSI species in PM_2.5 with great confidence.

3.3. ISORROPIA-II Thermodynamic Equilibrium Model Predictions Versus PPWD-PILS-IC Observations

The hourly data of the PPWD-PILS-IC (N = 720) was used in the ISORROPIA-II thermodynamic equilibrium model to study the gas-particle partitioning in PM_2.5. As shown in Figure 6, the observed hourly data have a good ion balance with an A/C ratio of 1.10 ± 0.18 and R² of 0.94 and a strong relationship between 2[SO₄²⁻] + [NO₃⁻] and [NH₄⁺] with the slope of 0.87 and R² of 0.92. It indicates that the NH₄NO₃ and (NH₄)₂SO₄ salts are dominant at this sampling site and the observed anion and cation data are balanced.

Figure 7 shows the linear correlation of the ISORROPIA-II results (i.e., predicted results) and the PPWD-PILS-IC data (i.e., observed data) for NH₄⁺_(p), SO₄²⁻_(p), NO₃⁻_(p), Cl⁻_(p), NH_3(g), and HNO_3(g) with the input parameters of T, RH, [SO₄²⁻]_(p), TN = [HNO₃]_(g) + [NO₃⁻]_(p), TA = [NH₃]_(g) + [NH₄⁺]_(p), TC = [HCl]_(g) + [Cl⁻]_(p), [Na⁺]_(p), [Mg²⁺]_(p), [K⁺]_(p), and [Ca²⁺]_(p). The results show that the predicted SO₄²⁻ and NH₃ correlate very well with the observed data with the slopes close to 1.0 and R² higher than 0.97, whereas the predicted NH₄⁺ and NO₃⁻ are also in good agreement with the observed data (R² > 0.9). However, the ISORROPIA-II thermodynamic equilibrium model underestimates NH₄⁺ and NO₃⁻ with the NMBs (NMDs) of −27.57% (−0.56 µg/m³) and −47.60% (−1.06 µg/m³), respectively. We observed that lower concentrations and more scattered data for NO₃⁻ resulted in a greater negative NMB value, which was also found in the previous study [61]. Assuming that 2[SO₄²⁻] fully balances with [NH₄⁺] in the model, since NH₃ is much preferred to react with H₂SO₄ to form SO₄²⁻ salts and 2[SO₄²⁻] + [NO₃⁻] shows the good correlation with [NH₄⁺], the slight under-prediction of NH₄⁺ should be due to the under-prediction of NO₃⁻. In comparison, the predicted values of Cl⁻ are comparable with the observations (slope = 0.74) but data are scattered at low concentrations (648/720 data < 1 µg/m³) resulting in a moderate correlation with R² of 0.49. The distributions of the WSI ions at this sampling site were tested for five days from July 2016 to January 2018 using the NCTU micro-orifice cascade impactor (NMCI) [62,63]. The results, shown in Figure S1 in SI, indicated that Cl⁻ was mainly formed in coarse particles (PM_10–2.5) which were not measured in this study, resulting in the underestimation of the model for Cl⁻ and the overestimation for HCl (Figure S2 in SI). Since the sampling site is in the urban area and far from the coastal line (>12 km), Cl⁻ existing in coarse particles could be formed from the heterogeneous reactions of HCl with the magnesium- and calcium-containing coarse particles [64] from the road dust source, resulting in low observed HCl and over-predicted HCl concentrations.

The poorest correlation between the predictions and observations is of HNO₃, as shown in Figure 7f, which is also found in many previous studies [50,51,61]. It is revealed that the predicted HNO₃ is very sensitive to the error of NO₃⁻ prediction as predicted [HNO₃] = TN − predicted [NO₃⁻] [44]. As compared with other species, the NO₃⁻ variability is found to be sensitive to the change of the T and RH, since the deliquescence RH of the NO₃⁻ salts varies with the T variation. Figure S3 shows that the NO₃⁻ concentration increases with the decreasing T and the increasing RH depending on the total concentration of the WSI species. In addition, unlike SO₄²⁻ and NH₄⁺ which are only dominant in fine particles (PM_2.5) and fewer in coarse particles (Figure S2), NO₃⁻ is present in both coarse and fine particles, as shown in Figure S2, which contributes to the error prediction for NO₃⁻ and HNO₃. A similar result was also found in the previous study [50]. This is because SO₄²⁻ and NH₄⁺ are mainly formed from the homogeneous reactions of precursor gases, whereas NO₃⁻ formation in fine particles is from the homogeneous reactions of HNO₃, and NO₃⁻ formation in coarse particles is from the heterogeneous reactions of HNO₃ with crustal coarse particles (i.e., Ca²⁺ and Mg²⁺) [12,61]. The coarse particle NO₃⁻ formation process results in low observed HNO₃ and over-predicted HNO₃ concentrations because the concentrations of Na⁺, Mg²⁺, Ca²⁺, and NO₃⁻ in the coarse mode are not included in the model. It can be seen that Mg²⁺, Ca²⁺, and Na⁺ are dominant in coarse particles since the source of coarse particles at this sampling site is road dust. However, the K⁺ distribution is complicated since it exists in both fine and coarse particles. The source of K⁺ in the fine and coarse modes could be associated with the burning activities from the nearby temple and the road dust from the heavy-traffic road, respectively [65].

As found in many previous studies, the main formation pathway of HONO is related closely to the formation of NO₃⁻ in fine particles [66], which is not included in the model resulting in the underestimation of the predicted NO₃⁻. In the humid and NH₃-rich environment, HONO and HNO₃ can be generated from the heterogeneous reaction on particle surfaces, and then HNO₃ can be converted to NO₃⁻ in humid conditions (Equation (5)) [15,16] as:

$$2{\mathrm{NO}}_2+{\mathrm{H}}_2\mathrm{O}\stackrel{ surface }{\to }\mathrm{HONO}+{{\mathrm{NO}}_3}^{-}{}_{\left(\mathrm{aq}\right)}+{\mathrm{H}}^{+}$$

(5)

Therefore, the formation of NO₃⁻_(p) is not only from the homogeneous reactions of HNO₃ with NH₃ and the uptake of HNO₃ on the particle surfaces but also the formation of the HONO by reactions shown above. As shown in Figure 8, the observed NO₃⁻ shows a better correlation with the observed HONO during the nighttime (from 6 p.m. to 6 a.m.) with R² of 0.41, while a poorer correlation is found during the daytime (from 7 a.m. to 5 p.m.) in the NH₃-rich environment (3.90 ± 3.02 μg/m³). This result is because HONO is easily consumed in the daytime (up to 80% reduction) by photolysis (Equation (6)) [66,67] as:

HONO + hv→NO + OH

(6)

Additionally, the heterogeneous formation of HONO on the wet particle surfaces can also be associated with the formation of NH₄⁺ with the presence of NH₃ and SO₂ as follows [68]:

SO₂ + 2NO₂ + 2NH₃ + 2H₂O→2NH₄⁺_(aq) + SO₄²⁻_(aq) + 2HONO

(7)

The relationship of the predicted/observed HNO₃, the observed HONO, the predicted/observed NO₃⁻, the T, and the RH, during 24 h, is shown in Figure 9 which presents the average data of each hour of HNO₃, HONO, NO₃⁻, T, and RH obtained in this study and NO₂, NO, and O₃ obtained from the nearest Taiwan Environmental Protection Administration (TW EPA) station [69] during the sampling period. The predicted HNO₃ is much higher than the observed HNO₃, especially during the daytime when the RH is low and the T is high. In contrast, the predicted/observed NO₃⁻ is high during the nighttime and low during the daytime, which is similar to the HONO variation. It indicates that the predicted HNO₃ determined in the model is just based on the thermodynamic equilibrium with NO₃⁻ at the given T and RH. It is observed that the HONO concentration is high during the nighttime due to the hydrolysis formation at high NO₂ concentration and high RH and low during the daytime due to the photolysis reaction represented by the peak of O₃. It is also seen that the concentration of HONO is quite close to that of the predicted HNO₃ during the nighttime, suggesting that there could be the heterogeneous conversion of HNO₃ on particle/ground surfaces under high NO concentrations as shown in Equation (8) below [66,70] as:

$$\mathrm{NO}+{\mathrm{HNO}}_3\stackrel{ surface }{\to }\mathrm{HONO}+{\mathrm{NO}}_2$$

(8)

It is noted that HNO₃ could also be converted to HONO during the daytime due to the photolysis of surface-adsorbed HNO₃/NO₃, as shown in Equation (9) below [18,19]. Therefore, the difference between the prediction and the observation for HNO₃ could be due to these conversion mechanisms which are not included in the ISORROPIA-II thermodynamic equilibrium model.

HNO₃ + hv→HONO + O(³P)

(9)

As shown in Figure 9, the diurnal trends of NO, NO₃⁻, and HONO are found to be related to the variation of the particle source. NO, NO₃⁻, and HONO show the peak concentrations at 6–8 a.m., the decrease in the early afternoon and the increase at 5–7 p.m., which is associated with the variation of the traffic flow rate (i.e., the total number of vehicle per hour) [55,71]. During the rush hours in the morning (6–9 a.m.), a high NO concentration is released from vehicles with a high flow rate leading to the formation of NO₃⁻ due to the homogeneous/heterogeneous reactions of HNO₃ resulting in the peak of NO and NO₃⁻ at 6 a.m. and HONO is formed from the reactions shown in Equations (8) and (9) resulting in a peak concentration at 8 a.m. During the noontime with a low traffic flow rate and a low relative humidity condition, NO, NO₃, and HONO are decreased. During the rush hours in the evening (5–8 p.m.) with a high traffic flow rate, NO is increased again leading to an increase in NO₃⁻ and HONO.

Figure 10 shows the comparison of the ISORROPIA-II predictions with the observation data for NH₄⁺_(p), NO₃⁻_(p), NH_3(g), and HNO_3(g), when TN input parameter was changed to TN = [HNO₂]_(g) + [HNO₃]_(g) + [NO₂⁻]_(p) + [NO₃⁻]_(p). It is found that the predicted NH₃ still correlates well with the observation, whereas correlation slopes for NH₄⁺ and NO₃⁻ are increased to 0.96 and 1.09, respectively, indicating that the prediction by the ISORROPIA-II equilibrium model for NH₄⁺ and NO₃⁻ is improved. In addition, the NMBs (NMDs) of the NH₃, NH₄⁺, and NO₃⁻ are decreased from 26.37% (+0.52 µg/m³) to 18.49% (+0.31 µg/m³), from −27.57% (−0.56 µg/m³) to −16.87% (−0.33 µg/m³), and from −47.60% (−1.06 µg/m³) to −14.94% (−0.27 µg/m³), respectively. It suggests that the hydrolysis formation of HONO contributes greatly to the formation of NH₄NO₃ salt and the equilibrium partitioning of the ammonium and nitrate aerosols. It can be seen that the sum of [NO₃⁻] and [NO₂⁻] is very close to [NO₃⁻] which implies that no salts are formed from NO₂⁻ ions. The underestimation and variabilities of the predicted NO₃^- at low concentrations (<5 µg/m³) were found, which led to the overestimation and variabilities of the predicted HNO₃ (Figure 10d), similar to what was found in previous results (Figure 7f) with the NMB of +579% and the NMD of +1.09 µg/m³. The comparable correlations between the predicted [HNO₃] and the sum of observed [HNO₃] and [HONO] are found with the NMB and the NMD of +129% and +0.89 µg/m³, respectively. Since HNO₃ is the terminal species in the ambient atmosphere, its measured concentrations could be very low (<1.0 μg/m³) because of the photolysis of surface-adsorbed HNO₃/NO₃⁻ to form HONO [18,19] or heterogeneous conversion of HNO₃ on particle/ground surfaces under high NO concentrations [66,70]. Therefore, the overestimation for the predicted HNO₃ by the ISORROPIA-II thermodynamic equilibrium model as compared with the observed HNO₃ and the better correlation between the predicted HNO₃ with the sum of the observed HNO₃ and HONO effectively explains the important conversion process of the HNO₃ to HONO in the ambient atmosphere. In the future, a quantitative study of these dynamic processes is needed to improve the model prediction capabilities of the ISORROPIA-II thermodynamic equilibrium model.

4. Conclusions

This study conducted field comparison tests of the PPWD-PILS-IC monitoring system and the manual 5 L/min porous-metal denuder sampler for precursor gases (NH₃, HNO₃, HNO₂, and SO₂) and water-soluble inorganic ions (Mg²⁺, Ca²⁺, Na⁺, K⁺, NH₄⁺, NO₃⁻, SO₄⁻, Cl⁻, NO₂⁻, and F⁻) in PM_2.5 to evaluate the sampling performance of the PPWD-PILS-IC system. Among these WSI species, SO₄²⁻, NH₄⁺, and NO₃⁻ are the major species accounting for 38.94%, 25.52%, and 20.41% of the total mass concentration of the WSI species, respectively. The 24-h average PPWD-PILS-IC data agreed well with the daily PDS data, with the slope and R² ranging from 0.99 to 1.17 and from 0.88 to 0.98, respectively, and the NMBs were less than 10%, except for NH₄⁺ (NMB = −13.95%), due to the evaporation loss effect of the PILS. Moreover, the correlation of the monitoring system data and the manual sampler data was improved, especially for precursor gases and K⁺, Na⁺, and NH₄⁺ ions as compared with those in the previous study which used 2 L/min PDS as the reference. The 5 L/min PDS showed good performance with small particle loss in the porous metal discs and low background interferences. In addition, the ISORROPIA-II thermodynamic equilibrium model was used to study the partitioning of the WSI species in PM_2.5 aerosols. The predicted results correlated well with the observed data of the PPWD-PILS-IC system for SO₄²⁻ (slope = 1.0 and R² = 0.99) and NH₃ (slope = 1.02 and R² = 0.97), when TN = [HNO₃] + [NO₃⁻]. When TN was the sum of [HNO₃], [NO₃⁻], [HONO], and [NO₂⁻], the correlation for NH₄⁺ and NO₃⁻ was improved with less under-prediction (slope = 0.96 for NH₄⁺ and 1.09 for NO₃⁻). This implied that the hydrolysis formation of HONO on particle surfaces contributed to the formation of NH₄NO₃ salt. The predicted HNO₃ is comparable to the sum of the observed [HNO₃] and [HONO] indicating that the HNO₃ could be converted to HONO due to the heterogeneous conversion and the photolysis reaction.