Long-term PM_2.5 source apportionment at 12 sites in the western United States from 2000–2020

Abstract

Long-term source attribution studies help evaluate the effectiveness of PM_2.5 reduction strategies, but inconsistencies in chemical speciation methods complicate source attribution. This study addresses these challenges by integrating Positive Matrix Factorization and Chemical Mass Balance to assess source contributions to PM_2.5 over a 20-year period (2000–2020) at 12 monitoring sites in five western U.S. states: Utah (Bountiful, Hawthorne, Lindon), California (Fresno, Bakersfield, Modesto, Visalia, Sacramento), Nevada (Reno, Las Vegas), Idaho (Boise), and Colorado (Commerce City). At each site, this study identified five to eight source factors including secondary ammonium nitrate (AN), secondary ammonium sulfate (AS), dust, chloride, organic carbon (OC) rich, elemental carbon (EC) rich, EC Cu rich, Cl Zn rich, and aged sea salt. Over two decades, PM_2.5 concentrations significantly declined (−0.38 to −0.037 μg/m³ per year) at all sites except Boise and Commerce City. Declines in AN, AS, EC rich, and EC Cu rich concentrations suggest the role of stringent national regulations on mobile, point, and area sources, resulting in reductions in NO_x, SO₂, and direct PM_2.5 emissions. Despite increasing vehicle miles traveled (VMT), EC rich (vehicle) concentrations decreased, indicating the importance of lower per-vehicle emissions. Winter OC rich concentrations, linked to biomass burning, saw the largest seasonal decline due to national and local efforts to curb residential wood combustion emissions. In contrast, dust concentrations generally increased, likely driven by rising regional aridity and VMT. These findings underscore the long-term effectiveness of air-quality policies in reducing PM_2.5 concentrations.

1. Introduction

In response to growing evidence of the health risks associated with PM_2.5 exposure (particulate matter with a diameter of 2.5 microns or smaller), the United States (US) began monitoring PM_2.5 levels in 1999 to set ambient standards required by the Clean Air Act (National Ambient Air Quality Standards (NAAQS) (Esworthy, 2015)). These standards have evolved in response to new scientific and public health findings, with long-term PM_2.5 measurements supporting stricter standards. Regulatory actions based on these evolving standards have contributed to reductions in PM_2.5 and its precursor emissions over time. PM_2.5 is influenced not only by direct, or primary, emissions but also by emissions of PM_2.5 precursors, or secondary emissions (CARB, 2025). Secondary PM_2.5 forms in the atmosphere from the oxidation of gases like sulfur dioxide (SO₂), nitrogen oxides (NO_x), and volatile organic compounds (VOCs) (CARB, 2025).

States that do not meet the NAAQS are required to develop State Implementation Plans (SIPs) to develop policies to reduce the emissions from the sources contributing to PM_2.5. Evaluating whether these policies successfully reduce primary and secondary PM_2.5 from those sources, long-term studies of PM_2.5 sources are essential. These long-term studies capture shifts in emission patterns over time and have provided insight into how changes in energy use, technological advancements, industrial activities, and policies have impacted air quality (Kotchenruther, 2016, 2020; Masiol et al., 2019). Source attribution, the process of identifying and quantifying the contributions of different emission sources to ambient pollutant concentrations, is a crucial tool for understanding long-term trends in PM_2.5 contributions. Kotchenruther (2020) used source attribution to evaluate changes in major contributors to winter PM_2.5 over 11 years at nine sites in the Northwestern U.S. He observed significant reductions in ammonium nitrate (AN), ammonium sulfate (AS), and wood smoke, but reported an increase of 6% to 226% in gasoline engine contributions between two three-year winter periods (2007–2009 vs. 2015–2017). He associated the increased gasoline engine contribution to a shift in the atmospheric chemistry of particulate nitrate production, with less AN production and more particulate nitrate association with the gasoline-engine factor. Similarly, Masiol et al. (2019) analyzed long-term trends at eight sites in New York. They found decreases in secondary sulfate and secondary nitrate at urban sites between 2005 to 2016 and attributed to the reductions in emissions resulting from a combination of regulatory controls and economic drivers. Additionally, long-term measurements can provide insights into how meteorological factors such as temperature, wind patterns, and humidity can influence source concentrations and their seasonal and annual variability (Do et al., 2023; Gao et al., 2022, 2023; Quevedo et al., 2024). In the context of climate change, understanding the interplay between meteorology and emissions is critical for developing more effective air quality management and adaptation strategies.

Although source attribution and specifically positive matrix factorization (PMF)-based studies are common (Amato & Hopke, 2012; Bae et al., 2019; Brown et al., 2015; Clements et al., 2014; Hasheminassab et al., 2014; Hwang & Hopke, 2007; Karanasiou et al., 2009; Kelly et al., 2013; Kim & Hopke, 2006; Lee et al., 2008; Park et al., 2020; Reff et al., 2007; Reizer et al., 2021; Shi et al., 2018; Wang et al., 2008; Zhang et al., 2009), fewer studies consider long time periods (i.e., 10 to 20 years) (Farahani et al., 2021; Kotchenruther, 2013, 2016, 2020; Masiol et al., 2019; E. H. Park et al., 2020; Squizzato et al., 2018; H. Wang et al., 2008). Long-term source-attribution studies of PM_2.5 rely on long-term chemical speciation measurements. However, the chemical speciation sampling, instrumentation, and analysis methods have changed over time (Gorham et al., 2021; Hopke et al., 2023; Kaur et al., 2024; Malm et al., 2011; Young, 2021), resulting in discontinuities in the chemical speciation measurements. For example, these changes in the Chemical Speciation Network (CSN) caused concentration discontinuities in the time series measurements for PM_2.5 ions and elemental carbon (EC) (Kaur et al., 2024). Consequently, long-term studies require strategies to address these discontinuities.

Many areas in the Western US experience prolonged episodes of elevated PM_2.5 levels due to natural meteorological patterns, such as persistent cold air pools (PCAPs), combined with geographical features like valleys surrounded by mountains, that create barriers that hinder the dispersion of air pollution (UDEQ, 2024). Additionally, climate-change induced drought is increasing the frequency and severity of wildfires in the Western US, which complicates efforts to meet PM_2.5 standards, as wildfire smoke significantly contributes to ambient PM_2.5 concentrations (Burke et al., 2023). Further, climate-induced droughts increased the frequency and intensity (increased average mass concentration per event) of regional and long-range transport of dust (African and Asian dust) (Hwang & Hopke, 2007; Kim & Hopke, 2006; E. H. Park et al., 2020; Perry et al., 1997). Population growth and urbanization further exacerbate air quality challenges by increasing vehicle traffic, industrial emissions, and energy consumption. Long-term studies of PM_2.5 in such regions are crucial for deeper insights into source contributions, for example, identifying decreasing contributions from residential wood burning in winter and increasing wildfire smoke contributions in summer and fall (Kotchenruther, 2016). The long-term studies also allow policy makers to understand the effectiveness of regulations and control measures, which are particularly important for areas where PM_2.5 regularly exceeds the NAAQS (non-attainment areas) like the Salt Lake Valley (SLV) and the San Joaquin Valley (SJV) (San Joaquin Council of Governments, 2024; UDEQ, 2007). Long-term studies of sites in the Western U.S. are limited, with the exception of the Kotchenruther (2020) 11-year study focusing on Pacific Northwest and intermountain west.

This study estimates sources of PM_2.5 over 20 years for 12 different monitoring sites in 5 different western states, which are affected by both wintertime PCAPs and summertime wildfires. This study can achieve long-term source apportionment of PM_2.5 by addressing methodological inconsistencies in long-term speciation measurements. It also analyzes the long-term trends in the source contributions, potential policy changes affecting the trends, and how the trends correlate with meteorological factors and emission inventories.

2. Methods

2.1. Data sources

This study used observations from CSN, which included monitoring sites funded by the US Environmental Protection Agency (EPA) and operated by state and local agencies. This study focused on 12 western US cities that have EPA CSN monitors and experience poor air quality due to PCAPs. The CSN monitors collect the 24-hour integrated PM_2.5 mass on filters, which are then analyzed by EPA-contracted labs for total PM_2.5 mass, elemental composition by energy dispersive X-ray fluorescence, organic and elemental carbon (OC and EC, respectively) by thermal evolution in 8 temperature fractions, and anions and cations by ion chromatography. Gorham et al. (2021) provided details about CSN methods and individual species analysis. The CSN data for the years between 2000 – 2020 was accessed through the Air Quality System database (EPA, 2025a).

Table 1 provides the list of the cities, their CSN site code, measurement frequency, and time range. The relocation of monitoring stations in Fresno, Las Vegas, Bountiful, Reno, and Commerce City resulted in multiple CSN site codes for these locations. Ten of the cities in Table 1 are on the American Lung Association (ALA) 2024 State of the Air report of the 25 most polluted cities in the US for short-term PM_2.5 (24-hour averaged) (ALA, 2024). Since 1999, all of these cities, except Boise, have, at some point, violated the PM_2.5 NAAQS, and all the California cities are still in non-attainment for PM_2.5.

Table 1:

Site names, locations, and the date range used in this study. The number of measurement days corresponds to the days used in the analysis.

City	County	CSN Site Code	Abbreviation (Site code)	ALA ranking short-term PM2.5	Start Date	Sampling Frequency	End Date	# measurement days
Hawthorne, Salt Lake City, UT	Salt Lake	49-035-3006	HW	19	2/9/2000	3rd day	12/26/2020	2224
Lindon, UT	Utah	49-049-4001	LND	19	6/18/2001	6th day	3/28/2020	1040
Bountiful, UT	Davis	49-011-0004; 49-011-0001	BT	19	12/15/2001	6th day	12/29/2020	1031
Reno, NV	Washoe	32-031-0016; 32-031-0031	RNO	6	4/7/2001	3rd day	12/29/2020	2212
Las Vegas, NV	Clark	32-003-0540; 32-003-0560; 32-003-0561; 32-003-0020	LV	23	7/1/2002	6th day increased to the 3rd day	12/29/2020	1324
Commerce City, CO	Adams	08-001-0006; 08-001-0008;	COM		2/24/2001	6th day	10/30/2020	1373
Boise, ID	Ada	16-001-0010	BOS	17	7/1/2006	3rd day	12/29/2020	1629
Fresno	Fresno	06-019-0011: 06-019-0008	FRN	2	2/9/2000	3rd day	12/29/2020	2181
Bakersfield, CA*	Kern	06-029-0014	BFD	1	5/25/2001	3rd day	12/23/2020	1589
Sacramento, CA	Sacramento	06-067-0006	SAC	9	2/15/2000	3rd day	12/23/2020	2315
Modesto, CA	Stanislaus	06-099-0005	MOD		1/14/2002	6th day	3/16/2020	968
Visalia, CA	Tulare	06-107-2002	VIS	5	1/14/2002	6th day	3/22/2020	1013

^*Bakersfield data unavailable between 9/7/2013 - 2/23/2015.

^**LV sites: 0560 and 0561 code between 07/02/2002 until 03/07/2007; 0020 code between 03/13/2007 - 04/26/2010; 0540 code from 05/02/2010 onward.

National Emission Inventory (NEI) county-level emissions data came from the EPA’s NEI website (EPA, 2024c). Annual vehicle miles traveled (VMT) came from each state’s department of transportation for the county hosting the site (Table 1). The state of Idaho does not report county-level VMT data; therefore, this information was not available for the BOS site. The meteorological measurements (daily average temperature, daily precipitation, daily snowfall, and daily snow depth) came from the National Center for Environmental Information, National Oceanic and Atmospheric Administration (NOAA, 2025). Wildfire area burned per state came from the National Interagency Coordination Center (NICC, 2025). The state-wide, power-plant emissions of NO_x and SO₂ for each fuel type came from EPA’s Egrid database (EPA, 2025c).

2.2. Changes in the CSN analysis methods and corresponding correction strategy

Over the last twenty years, CSN has updated its sampler and analysis methodologies. The contractors performing the analyses have also changed, with different contractors using different instruments. The details of these changes can be found in Malm et al. (2011) and CSN advisories (Young, 2021). These changes often lead to discontinuities in the species concentrations, potentially influencing source-attribution results. This study accounted for three main changes:

1. 1. Between 2007–2009, EPA switched the carbon sampler from MetOne to URG3000, to be consistent with the IMPROVE (Interagency Monitoring of Protected Visual Environment) network. During this same time period, the carbon analysis protocol also switched from the thermal optical transmittance to the IMPROVE_A thermal optical reflectance method. This change affected the OC and EC concentrations, and details of these changes as well as proposed correction factors, can be found in Malm et al. (2011) and below:

$$\begin{gathered} O{C}^{CSNadj}=\frac{\left(O{C}^{CSN}-0.3\mathrm{*}E{C}^{CSN}-a_i\right)}{\left(1+b_{OC}\right)} \\ E{C}^{CSNadj}=1.3\mathrm{*}E{C}^{CSN} \end{gathered}$$

Where, the $O{C}^{CSNadj}$ are the adjusted values, $O{C}^{CSN}$ and $E{C}^{CSN}$ are before the sampler change OC and EC concentrations. The $a_i$ is the additive corrective factor that varies with the month and its values can be found in Malm et al. (2011). The $1+b_{OC}$ is the multiplicative factor and is 1.2. This study applies the Malm et al. (2011) correction factors for OC and EC concentrations prior to the sampler changes, with the application date depending on the date of the sampler change.

1. 2. The EC measurements for the period after Oct 2018 and the ions (NO₃, SO₄, and NH₄) for the period between 20 Nov 2015 – 30 September 2018 were corrected using a spline fit, discussed in Kaur et al. (2024).

2. 3. The days with possible chromium contamination (<2%) were assessed using the method provided by the CSN advisory (Raffuse, 2020) and were removed if chromium contamination was likely.

2.3. Blank correction and MDL

Starting on 1 Jan 2016, CSN blank corrected the OC and EC measurements. Before 2016, the OC and EC were not blank corrected. In this study, blank corrections were applied to OC and EC measurements collected from 2007/2009 (site-dependent, corresponding to the timing of the transition from MetOne to URG3000 carbon samplers) through January 1, 2016, using CSN-reported blanks. Measurements prior to 2007/2009 were not blank corrected, as they had already undergone corrections (discussed in Section 2.2) to account for both negative and positive biases. To address gaps in the blank data, the median of the previous three blank values was substituted for days without blank measurements, ensuring continuity in the time series despite inconsistencies in blank data availability. Kotchenruther (2016, 2020) applied this method in previous studies. CSN did not blank correct ion measurements before 1 Jan 2016, but they performed field-blank corrections for the remaining period (Young, 2021). Our study did not apply blank corrections for ions for the period before Jan 2016 because the blanks available before Jan 2016 were filter-lot background blanks and not field blanks. Also, the CSN advisory (Young, 2021) reported minimal effect of field blank correction on the NH₄, NO₃, and SO₄, i.e., less than a 5% difference between the raw and blank-corrected concentrations. Although, the same advisory reported a substantial effect of blank correction on the chloride and sodium ion concentrations. Elements were corrected for filter-lot-specific background prior to November 20, 2015. Between November 20, 2015, and January 31, 2017, the element blank correction method shifted to using the laboratory filter blank median areal density. From February 2017 onward, elements were reported as blank-corrected using the median of field blanks (Young, 2021). This study used the CSN element data without any additional modification.

This study used the CSN-reported MDLs, regardless of the different methods of MDL calculation (Young, 2021), which changed beginning 20 Nov 2015 (Section S2, supplementary material), because there is no good way to correct these MDLs.

2.4. Receptor modeling design

This study used positive matrix factorization (EPA PMF 5.0) and Chemical Mass Balance (EPA CMB 8.0) for source apportionment at the 12 sites (Table 1). PMF run individually for each site with measurements from the whole time period returned either un-converged solutions or unmapped bootstrap (BS), possibly due to the changes in contractors, instrumentation, and analysis protocols (some of which could not be corrected, for example, Na⁺ (as discussed under section 3.1.5) starting 20 Nov 2015). Therefore, this study performed PMF for each site individually between the starting date (see Table 1) and until 19 Nov 2015. The resulting PMF factor profiles and CSN measurements (prepared for PMF) were input into the CMB to obtain source contributions for the whole sampling duration (from the start date in Table 1 through 2021).

2.4.1. PMF

PMF is a popular multivariate factor analysis model used in air quality studies (Amato & Hopke, 2012; Bae et al., 2019; Brown et al., 2015; Clements et al., 2014; Hasheminassab et al., 2014; Hwang & Hopke, 2007; Kelly et al., 2013; Kim & Hopke, 2006; Lee et al., 2008; Park et al., 2020; Reizer et al., 2021; Shi et al., 2018; Zhang et al., 2009). In PMF, each measuremenťs uncertainty is taken into account to appropriately weight the data points, and a non-negativity constraint is imposed during optimization for the resulting source matrices. The model decomposes a measurement data matrix (𝑋) into two matrices, a factor contribution (𝐺) and a factor profile (𝐹) matrix, which need to be estimated as in Eq. (1) (Paatero & Tapper, 1994):

$$X=G\cdot F+E$$ (1)

Where $X$ is a $n$ by $m$ matrix, $G$ is a $n$ by $p$ matrix, and $F$ is a $p$ by $m$ matrix with $n$ being the number of samples, $m$ is the number of species, and $p$ is the number of sources. $E$ is the residual matrix. PMF minimizes the objective function $\left(Q\right)$ to solve the mass balance equation, using known uncertainty for the measurement data $\left(\sigma \right)$ stated as follows (Paatero, 1997).

$$Q={‖\frac{(X-G\cdot F)}{\sigma }‖}_F^2$$ (2)

Further details about the PMF method can be found in previous studies (Reff et al., 2007). In this study, data were prepared by following the guidelines found in Polissar et al. (1998). Briefly, when measurements were less than the detection limit, the values were set to half of the detection limit, and the uncertainties were set to 5/6^th of the detection limit or the reported uncertainties, whichever was larger. The uncertainties for measurement values greater than the detection limit were set as the sum of reported analytical uncertainties and 1/3^rd of MDL (Reff et al., 2007). Any negative measurement value was treated as below the detection limit. Missing uncertainties were supplemented using a formula provided by Kim et al. (2005). Days were removed if either all elements, all ions, or all carbon and fractions were missing. If a single species was missing, this missing measurement was replaced with the mean of all the previous measurements, and the uncertainty was set to four times the mean (Reff et al., 2007).

We used the signal-to-noise ratio criteria from Paatero and Hopke (2003) to select species to be included in the PMF model. Species with S/N less than 0.2 were not included, whereas species with S/N greater than 0.2 and less than 1 were set as “weak” variables and down-weighted by a factor of 3. Species with S/N>1 were set as strong species and used as reported. Regarding duplicate species, SO₄, sodium ion, and elemental potassium, which generally had a higher S/N ratio, were selected, and sulfur, elemental sodium, and potassium ions were excluded from the model. An extra modeling uncertainty up to 10% was assigned in the PMF when needed. To remove the effect of fireworks on the model output, days corresponding to New Years (31 Dec and 1 Jan) and the Fourth of July (in and around 3^rd, 5^th, and 6^th July) as well were removed. For sites in Utah, Pioneer Day and adjacent days (23rd, 24th, and 25th July) were removed (Kelly et al., 2013).

PMF for each site was performed individually between the starting date (see Table 1) and until 19 Nov 2015. The PMF was performed individually for each site, with a total of 40 base runs under robust mode (a model that minimizes the influence of outliers in the data by down-weighting extreme values during the factorization process) to ensure the solution represented the global minimum. We determined the optimum number of factors (starting with four factors) using the following criteria: the factor profiles represented realistic factors that could be present at that site, the quality of least square fit (Qrobust and Qtrue analysis), and the result of three error estimation methods, i.e., BS, Displacement (DISP), and BS-DISP. A total of 100 BS runs were performed for each solution and the major species of each factor (for example, Si for dust, NO₃ for secondary nitrate, SO₄ for secondary and primary sulfate, etc., more details in Supplementary Material) were used in BS-DISP. An acceptable solution and number of factors was defined as the BS mapping of a minimum of 80%, 0 swaps in DISP, and 0 swaps in BS and DISP at minimum dQ and a decrease in dQ of less than 0.5% in BS-DISP. The factor number was considered optimum when increasing the factor number further led to an unacceptable solution. The scaled residuals for the final model solutions were normally distributed, generally falling within the recommended range of −3 to 3.

2.4.2. CMB

The CMB receptor model provides a least-square solution to a set of mass balance equations that express each receptor chemical concentration as a linear sum of the product of factor profile abundances and factor profile contributions (Begum et al., 2007; Chelani et al., 2008; Chow & Watson, 2002; Ivey et al., 2019; Watson et al., 2001; Xu et al., 2021; Zhai et al., 2017). Factor profile abundances are the mass fraction of species in the emissions from each source type that might contribute. CMB requires predetermined factor profiles, and the choices of factor profiles affect the uncertainties in the CMB solution, as the number of sources and composition of sources may change with region and season (Ivey et al., 2017). More details about the CMB receptor model can be found in Watson et al. (1984).

The CMB model input parameters include ambient species concentrations and corresponding uncertainties, and factor profiles and associated uncertainties. For ambient species concentrations, we used the same data preparation as for PMF, i.e., the PMF-prepared dataset, including uncertainties, was used in CMB. The ambient concentrations prepared for PMF were rearranged for use in CMB. The PMF-driven profiles were utilized as factor profiles. The uncertainties for the factor profiles were obtained from the 100 BS runs of PMF, as described previously (Brown et al., 2015). The PMF-derived profiles based on the starting date (Table 1) −19 Nov 2015 period were used in CMB for the entire time period (starting date – end date (Table 1)) for each individual site. Again, this was done to avoid running PMF for the entire time range, which returned unconverged results, and to eliminate the impact of post-2015 concentrations (due to contractor changes) on source profiles.

2.5. Trend Analysis

This study analyzed trends in long-term source contributions, and seasonal, weekday, and monthly patterns using R 4.4. We excluded year and site combinations with fewer than 20 days of measurements. This resulted in the starting year being 2002 for BT and LND, and the year 2020 being excluded for MOD and VIS. The long-term trends in source contributions were based on annual concentrations and normalized source contributions (normalized by the federal reference method (FRM) PM_2.5 mass concentration). These were analyzed using Theil-Sen (which is a robust, non-parametric method that is less sensitive to outliers, particularly effective for analyzing data with irregularities or extreme fluctuations, both of which are common in long-term environmental data) from the OpenAir package (in R) to estimate the linear trends for the seasonal and annual averages. The long-term seasonal trends were based on the following seasons–fall: September, October, and November; winter: December, January, and February; spring: March, April, May; summer: June, July, and August.

We compared species concentrations from weekdays to weekends using a two-sample t-test (conducted in R). The t-test assumes that the data are approximately normally distributed. We also assessed the strength and direction of the linear relationship through Pearson correlation between the following: 1) source vs. source; 2) source concentration vs. NEI emissions (for NO_x and SO₂); 3) source concentration vs. VMT; 4) source concentration vs. wildfire area burned; 5) source concentration vs. meteorological parameters (daily average temperature, snow depth, and precipitation). A significance level of 0.05 was used throughout to determine statistical significance.

3. Results and Discussion

3.1. PMF factor profiles

Depending on the site, PMF resolved 5 to 8 source factors (Figure S1). These factors included secondary nitrate (AN), secondary sulfate (AS), crustal, chloride, OC rich, EC rich, OC EC rich, EC Cu rich, Cl Zn rich, and aged sea salt. The AN and AS factors occurred at all sites. AN was dominated by NO₃ and NH₄, while AS contained SO₄, NH₄, and OC, with some mixing of secondary organics at select sites. The chloride factor appeared at most sites except BOS, COM, and RNO, explaining over 80.0% of Cl mass. Utah sites (HW, LND, BT) and some California sites (BFD, FRN, VIS) had high ammonium, whereas LV, MOD, and SAC showed greater Na⁺ contributions, suggesting mixed chloride sources. The aged salt factor, identified at BFD, BOS, FRN, MOD, RNO, and SAC, explained over 85.0% of Na⁺ variation, with notable contributions from NO₃, SO₄, and OC, indicating atmospheric processing. Dust was present at all sites, and this factor was dominated by crustal elements and minor anthropogenic metals, often mixed with NO₃, SO₄, OC, and EC. Two EC dominant factors, EC rich and EC Cu rich, representing vehicle emissions, emerged. EC rich was linked to OC, NO₃, SO₄, and trace metals at all sites except COM, FRN, SAC, and VIS, while EC Cu rich occurred only at FRN. The OC-rich factor, apportioned at all sites except, COM and VIS was primarily composed of OC with contributions from EC, K, NO₃, and SO₄, indicating biomass burning. For three sites (COM, VIS, SAC), an OC-EC-rich factor represented a mixture of both EC rich and OC rich emissions. A Cu-rich factor, identified at BFD, VIS, and SAC, showed high explained variation for Cu (90.0%–95.6%), with other components like OC, EC, and NH₄, suggesting vehicle and industrial emissions. The Cl-Zn factor, observed only at COM, was enriched in Zn (11.8%) and Cl (14.5%), with EC and OC also contributing, pointing to tire wear and road salting as potential sources.

Subsequent sections discuss each factor, temporal trends, and prevalence at each site. Table S1 lists the sites and the factors observed at these sites. Figure S1 reports the factor profiles for each site. Table S2 provides a summary of the BS, DISP, and BS-DISP results.

3.1.1. PM_2.5

The CMB-calculated PM_2.5 was less than the gravimetric PM_2.5 concentration, with the slopes between the measured and calculated PM_2.5 ranging from 0.692 to 0.824 (Figure 1). Slopes less than 1 are expected, as the PMF-derived profiles used in CMB did not account for all of the measured PM_2.5. Approximately 62 – 83% of the total PM_2.5 was included in the CMB model, with the remaining portion attributable to unmodeled sources or factors not captured by the PMF profiles. The R² between the measured and calculated PM_2.5 concentration was >0.706 for all sites, with values >0.9 for seven sites (BFD, BOS, FRN, HW, LND, MOD, and VIS).

Figure 1:

Comparison of PM_2.5 concentration (μg/m³): measured (by FRM) vs. calculated by CMB using the PMF-derived factor profiles. The line represents the linear regression fit with S, I, r² representing slope, intercept, and coefficient of determination.

The PM_2.5 seasonal mean concentrations across all sites were highest during winter (9.88 – 27.3 μg/m³) and fall (7.53 – 21.1 μg/m³) and lower for summer (7.22 – 12.2 μg/m³) and spring (3.76 – 10.7 μg/m³) (Figure 2). Higher wintertime PM_2.5 concentrations were expected because all sites experienced wintertime PCAPs. The annual PM_2.5 concentrations decreased over time, i.e., negative Thiel-Sen slopes (−0.38 to −0.037 μg/m³ per year) with significantly negative slopes for BFD, FRN, LND, BT, HW, and MOD (Figure 4 and Figure S2). The slopes were negative but not significant for LV, RNO, SAC, BOS, COM, and VIS. PM_2.5 declined significantly during spring, fall, and winter, at most sites, and the summer trends were inconsistent (Figure 5). The mean annual PM_2.5 concentration showed different trends for different sites (Figure 4). For example, BFD, FRN, LV, RNO, and SAC showed a decreasing PM_2.5 concentration until 2015, when it began increasing. For COM, PM_2.5 concentrations decreased between 2007 and 2015, followed by relatively stable concentrations (Figure 4). The somewhat different trends in the PM_2.5 concentrations could be associated with different temporal trends of the contributing sources, and these are discussed in the following subsections.

Figure 2:

Seasonal concentrations (μg/m³) of sources and PM_2.5 for each site.

3.1.2. Ammonium Nitrate (AN)

This factor was dominated by the presence of NO₃ and NH₄, and it occurred at all 12 sites (Figure S1). Previous studies focusing on the western US have identified this factor and associated it with secondary AN (Grover et al., 2006; Hallar et al., 2020; Hansen et al., 2010; Hasheminassab et al., 2014; Hwang & Hopke, 2007; Ivey et al., 2016, 2019; Kelly et al., 2013; Kim & Hopke, 2006; Kotchenruther, 2016, 2020, 2021; Malek et al., 2006; Motallebi, 1999; Stanimirova et al., 2024; Sun et al., 2021; Whiteman et al., 2014). Nitrate and ammonium contributed on average 67.6 ± 6.54% and 19.6 ± 3.67%, respectively to the AN profile. The ratio of nitrate to ammonium is 3.44, which is near the 3.5 ratio, expected when the nitrate gets fully neutralized by ammonium (Kotchenruther, 2016). Figure S1 details the contribution of NO₃ and NH₄ for individual sites.

AN was predominantly observed during winter and fall, with little contribution during summer (Figure 2). The average concentration of AN during winter over 20 years ranged between 2.07 – 12.7 μg/m³ (Figure 2 and Figure S3). During winter, AN concentration reached levels as high as 96.3 μg/m³ (at VIS during 2013) (Figure 2) and contributions reached ~90% of the total PM_2.5 concentration (Figure S3). All the sites in this study experience wintertime PCAPs, which facilitates the formation of secondary nitrate. Furthermore, the PCAP conditions limit dispersion or vertical mixing, resulting in elevated concentrations of pollutants, primarily AN (Chen et al., 2012). The low summer contributions of AN (<5% at all sites except VIS and COM) are likely due to the thermodynamic instability of AN and its decomposition at elevated temperatures (>20°C, Figure S4) (Tian et al., 2023).

The key precursor for AN formation is NO_x. In the presence of water vapor and cold temperatures (like those during PCAPs) NO_x gets converted into HNO₃ (nitric acid), which then reacts with NH₄ to form AN (Stockwell et al., 2000). This study’s sites, except LV, are NH₄ rich (SLV, (Kelly et al., 2013); SJV, (Kleeman et al., 2005 and Lurmann et al., 2006;); Boise (DEQ, 2013)), making NO_x the limiting reagent in AN formation. For LV, the arid environment limits agricultural activities and associated ammonia emissions, making NH₄ likely the limiting reactant (Brown, 2014). The extent of NO_x to AN conversion is influenced by factors including the NO_x/VOC ratio. For example, the high ratio of NO_x to VOCs in the SLV alters the traditional O₃ photochemical cycle, leading to the formation of particulate nitrate (Kleinman, 1994; Lin et al., 1988; Womack et al., 2019). Other factors include meteorological conditions (temperature, relative humidity, snow cover, and solar intensity (Kleeman et al., 2005; Mwaniki et al., 2014; Nguyen & Dabdub, 2002)), topographical factors (Kleeman et al., 2005; Pusede et al., 2016), and the presence of oxidants like halogens (Womack et al., 2023). For example, in the SLV, PM_2.5 exceedances, primarily due to AN, are observed four times more frequently on snow-covered days compared to those without snow cover (Whiteman et al., 2014). Green et al. (2015) also reported significantly higher AN concentrations for days with snow-cover compared to snow-free days at Salt Lake City, Reno, and Boise. We observed a moderately positive correlation between snow cover and AN concentration (r of 0.521, 0.237, 0.330, 0.334, 0.415, and 0.378 for RNO, LND, HW, COM, BT, and BOS) (Figure S5). Womack et al. (2023) also highlighted the importance of halogens (chlorine and bromine from US Magnesium) on nitrate formation in the SLV; releases of Br and Cl increase the production of oxidized VOCs and nitric acid (D. S. Wang & Hildebrandt Ruiz, 2018; Womack et al., 2023).

The predominant sources of NO_x include anthropogenic sources, such as light- and heavy-duty vehicles, fuel combustion, and biomass burning (Zhu et al., 2023). Kelly et al. (2013) associated NO_x in SLV with vehicle and industrial emissions. In the SLV during PCAPs, Ivey et al. (2019) associated AN with on-road gasoline vehicles (25%), natural gas combustion (21%), and other sources including refining operations, smelting, industrial activities, and stationary diesel combustions (29%). For RNO, a previous study attributed NO_x emissions to mobile emissions (especially diesel emissions), residential wood burning, and natural gas combustion in boilers and furnaces (Chen et al., 2012). For the SJV, the sources of NO_x include mobile emissions (with HDVs responsible for more than 70% of NO_x emissions from on-road mobile sources (CARB, 2019b)), residential wood burning, industrial emissions (boilers, flares, furnaces), commercial charbroiling, and open burning (agricultural and prescribed fires) (CARB, 2024). Soil NO_x emissions, released as a byproduct of microbial nitrification and denitrification (Andreae M.O. & Schimel D.S., 1991), also contribute to NO_x in the SJV (Almaraz et al., 2018; Guo et al., 2020; Sha et al., 2021). VOCs and ammonia (NH₃) emissions are also important to AN formation and are highly dependent on agricultural and industrial sectors in each valley (Hallar et al., 2020; Kelly et al., 2013; Simon Wang et al., 2015). Wallace et al. (2012) identified roadway sources (53.7% in 2008) and fuel combustion for building heating (30.7% in 2008) as the major contributors to NO_x emissions in BOS. Major sources of NO_x in COM include energy generating units, especially the Cherokee power plant, vehicle emissions, and Suncor refinery operations (Flocke et al., 2020). Mobile sources, including on-road and non-road vehicles are the major sources of NO_x in LV (Clark County Department of Environment and Sustainability, 2024).

Only FRN exhibited weekday dependency (Figure 3), with weekend AN concentrations being significantly lower than weekday concentrations (p < 0.05). FRN is situated near busy roads, and this weekday dependence suggests the influence of vehicle emissions at this site.

Figure 3:

Weekday mean source-specific concentrations (μg/m³). The 1 on x-axis represent Sunday, and the 7 represents Saturday.

AN concentrations declined between 2000 and 2020 for all sites, with Thiel-Sen slopes ranging from −0.321 to −0.0379 μg/m³ per year (Figure 4 and Figure S2). RNO (−0.0379 μg/m³ per year) and BOS (−0.0499 μg/m³ per year) had the smallest decrease (Figure S2). BOS site had limited data availability prior to 2006, which might affect the annual Thie-Sen slope. Green et al. (2015) also reported a nonsignificant decline in AN for RNO and BOS for period between 2000 – 2012. For HW and LND, this study found a wintertime (winter season, 2007 – 2009 vs. 2015 – 2017) AN reduction of 46% and 52%, respectively. These values closely align with those reported by Kotchenruther (2020) for the same time period, at 42% and 52%, respectively.

Figure 4:

Long-term trends in source concentrations (2000 – 2020). The red solid line represents the Theil-Sen fit, with shaded red lines representing the lower and upper limits of this fit (95% CI).

Long-term trends in source concentrations 2000–2020. The dots represent the annual average. The red solid line represents the Theil-Sen fit, with shaded red lines representing the lower and upper limits of this fit (95% CI).

The seasonal trends reveal varying patterns at the different monitoring sites. For example, AN concentration declined significantly: during winter for all sites except LV, VIS, and BOS; during fall for all sites except for BOS and COM; and during spring for all sites (Figure 5). The small change in AN over time during summer (−0.176 to −0.000902 μg/m³ per year) was not surprising because the summer concentrations of AN were very low at all sites. Similar decreases were also observed in the AN contribution to the total PM_2.5 over time (Figure S6). The slopes were negative (−1.18% to −0.0781% per year) and significantly negative for all California sites (p<0.001), BOS, BT, and LND (p<0.05). The decline in AN’s contribution to the total PM_2.5 over time was not significant for HW and RNO.

Figure 5:

Season-specific Theil-Sen slope and limits (upper and lower, 95% CI) for the sources. The dot represents the slope, and the shaded bar represents the range between the lower and upper limits on the slope.

Previous studies linked decreasing AN concentrations with reduced NO_x emissions. Kotchenruther (2020) noted a 24% decline in NO_x emissions across five states (WA, OR, MT, UT, ID) between two three-year periods (2007–2009 and 2015–2017). Several regulations targeting mobile, point, and area emissions significantly contributed to this decrease. Enhanced emission controls, cleaner fuels, and improved engine designs have all lowered vehicle NO_x emissions. Tier 2 vehicle emission standards, implemented between 2004 to 2009, limited NO_x emissions from passenger cars and light trucks (Bin 5) to 0.07 grams per mile (EPA, 2017). The subsequent Tier 3 standards, phased in from 2017 to 2025, further decreased NMOG (non-methane organic gases) + NO_x emissions to 0.02 grams per mile for passenger cars (Bin 20), while also promoting the adoption of zero-emission vehicles (EPA, 2025b). Additionally, EPA NO_x limits for heavy-duty diesel vehicles (HDDV) were established in 2007 at 0.2 grams per brake horsepower, requiring the upgrade or replacement of older, higher polluting engines, further reduced NO_x emissions (EPA, 2016b). In the SJV, the Truck and Bus rule for this region resulted in a 41.8% reduction in NO_x emissions between 2013 and 2019, translating to a decrease of 53.8 tons per day of NO_x emissions (CARB, 2019a). Stricter regulations on the sulfur content in gasoline (reduced from 30 ppm in Tier II to 10 ppm in Tier III) and diesel (15 ppm beginning in June 2006) have also enhanced the effectiveness of catalytic converters (EPA, 2016a, 2016c). Furthermore, various incentive and compliance programs, such as the federally funded Diesel Emission Reduction Act, which began in FY2008, encouraged reductions in diesel emissions (EPA, 2024a). More stringent regulations on non-road diesel engines (9.2 g/kW-hr in Tier 1 to 0.4 g/kW-hr in Tier 4, phased between 2008–2015) also reduced NO_x emissions (EPA, 2016d). Vehicle inspection and maintenance programs, along with on-board diagnostics, have supported the management of vehicle emissions. Other federal regulations, including the Clean Air Interstate Rule (2005), the Cross-State Air Pollution Rule (2010), the Mercury and Air Toxics Standards (2011), and the New Source Review program, have led to lower emissions (NO_x and SO₂) from energy-generating units (Mei et al., 2024). Additionally, regions containing our study sites have developed SIPs, for PM_2.5, ozone, or PM₁₀, that mandated reductions in NO_x emissions from both point and area sources.

The NEI data shows decreasing NO_x emissions for all sites, except RNO (Figure S7). For the Utah sites prior to 2014, the concentrations of NO_x did not decrease consistently, but after 2014 they showed a steady decrease in annual NOx emissions (Figure S7). NO_x emissions are affected by economic conditions; for example, the recession between 2007 – 2011 reduced vehicle emissions and VMT (Stanimirova et al., 2024). The variability in fuel prices can also affect NO_x emissions. Elevated natural gas prices lead to the usage of alternative fuels (like coal, with higher NO_x emissions) for power production (Mei et al., 2024). NO_x emissions for RNO increased starting in 2014 (Figure S7), and this increase appears to be associated with the NEI tier 1 category “Fuel Comb. Industrial” (Figure S8). For all counties, NO_x emissions from the “Highway Vehicles” and “Off-Highway” tier 1 categories decreased, as expected, due to regulations on vehicle emissions (Figure S8).

Comparing annual NO_x emissions with AN concentrations yielded moderate (0.411 – 0.555 for BFD, HW, LND) to strong Pearson correlations (r >0.7 for VIS, SAC, MOD, FRN, COM, BT, BOS) for all sites, except RNO and LV (Figure S9). LV showed a weak correlation of 0.269, and RNO showed a negative correlation with annual NO_x emissions (−0.267) (Figure S9). Removing the NEI NO_x contributions from “Natural resources” and “Miscellaneous” (which includes prescribed burning and wildfires) from total NO_x emissions (labeled as “ParTotal” in Figure S9), the correlation improved for VIS, (0.789 to 0.905), FRN (0.767 to 0.815), and MOD (0.805 to 0.910) (Figure S9). For most sites (except LV), moderate correlations were observed for AN vs. tier 1 category “Highway Vehicles” (>0.5 for all sites except RNO, LND and BFD) and strong correlations were observed for AN vs tier 1 category “Off-Highway” (>0.7). Together, these correlations indicate the importance of regulation on decreasing vehicle NO_x emissions and consequently AN concentrations.

A previous study on long-term PM_2.5 contributions in New York reported a weak to moderate correlation between AN and NEI NO_x emissions and discussed how this could be due to the local sources affecting AN (Masiol et al., 2019). High inventory NO_x emissions may not correlate with high AN concentrations because of atmospheric conditions. Winter conditions are ideal for AN formation, and this study showed that the correlation between NO_x and AN increased with an increasing number of PCAP days per year (Figure S10). For example, years with more than 30 PCAP days showed a Pearson correlation of 0.502 between NO_x emissions and AN concentration. This increased to 0.889 for years with more than 60 PCAP days.

Jiang et al. (2018) reported a slowdown in the decrease of US NO_x emissions between 2011 and 2015 (7% annually from 2005 to 2009, compared to 1.7% annually from 2011 to 2015), based on satellite and surface measurements, and reported that this slowdown contradicts EPA’s continuous decreasing trend (from bottom-up inventories). Hallar et al. (2020) discussed the uncertainties in EPA’s NEI including: poorly quantified NO_x emissions from older and off-road vehicles, the frequency of cold starts and idling during winter, and increased NH₃ emissions from selective catalytic reduction systems during winter. The uncertainties in agricultural emissions, which are related to emissions of both NH₃ (Hempel et al., 2016) and NO_x could have a large impact on the California sites (especially, FRN, VIS, and BFD), where crop fertilization occurs in the winter months. NH₃ from heated animal housing could lead to an under-estimation of process-based NH₃ during winter (Moravek et al., 2019). Also, even when solid-fuel (wood, wood pellet, coal) burning is prohibited in a region due to air quality conditions, the primary and secondary aerosols originating from residential wood burning can persist from days prior to the implementation of the prohibition (Hallar et al., 2020; Kotchenruther, 2020).

3.1.3. Ammonium Sulfate (AS)

This factor, secondary AS, emerged at all sites (Figure S1) and was dominated by NH₄ and SO₄, along with OC at HW, RNO, LV, COM, FRN, BFD, SAC, MOD, and VIS. Sulfate, ammonium ion, and OC contributed 55.9 ±7.9%, 21.3 ± 3.95%, and 13.4±9.61%, respectively, to the average AS profile (average contribution of each species to the factor across all sites). At the BFD and MOD sites, NO₃ contributed 9.62 ± 3.85%, on average, to the AS profile. The presence of OC, NO₃, and other species in the profile indicated a mixture with other sources, like OC rich. The presence of OC in this factor could imply that secondary organic aerosol formation coincides with AS formation (Kim et al., 2005), while the presence of EC has been associated with AS formation in stagnant conditions, for example during PCAPs (Lee et al., 2008).

Hasheminassab et al. (2014) also reported an AS factor at the FRN site, mixed with road dust tracers, such as OC, EC, Fe, Ca, Mn, Si and Ti. Kotchenruther (2013) reported AS factors mixed with other sources. The ratio of sulfate to ammonium was 0.375, which is near the 0.32 ratio, expected when the sulfate becomes fully neutralized by ammonium (Kotchenruther, 2016).

The formation of AS involves the oxidation of SO₂ to H₂SO₄ (sulfuric acid), which is then neutralized by NH₄ to produce either ammonium sulfate (NH₄)₂SO₄) or ammonium bisulfate (NH₄HSO₄). The oxidation of SO₂ can occur in the gas phase through reactions with the OH radical, followed by condensation onto particles. Additionally, SO₂ oxidation can take place in the aqueous phase on particle surfaces through reactions with O₂ and H₂O₂ (Saxena & Seigneur, 1987). Due to the low vapor pressure of H₂SO₄, sulfate predominantly exists in the particle phase (Pinder et al., 2007).

Important sources of SO₂ include coal and oil combustion; industrial processes; and smelting of metal ores (IDEM, 2024). For the LND site, Hansen et al. (2010) reported that the sulfate is probably dominated by emissions from mobile sources and wood smoke. Off-shore shipping emissions in locations, such as California, can impact sulfate concentrations across the western United States (Xu et al., 2006). For BOS, the major SO₂ sources include Amalgamated Sugar (located approximately 17 km from the BOS site) with SO₂ emissions of 819–1660 tons per year between 2011– 2014 (Floyd, 2016). In COM, the major source of SO₂ until 2018 were the Cherokee Power plant (CDPHE, 2024) and the Suncor refinery, which had the largest number of tail gas incidents in the US between 2016 – 2020 (Piantanida, 2023). The dominant source for sulfur dioxide (SO₂) emissions in Nevada is fossil-fuel combustion at power plants and other industrial facilities (NDEP, 2013). The Reid Gardner Generating Station near Lag Vegas was the dominant source until 2017, when it shut down (NDEP, 2013). Smaller sources of SO₂ emissions include industrial processes, such as ore processing, as well as the combustion of high sulfur fuels by locomotives and nonroad equipment (NDEP, 2013).

AS concentrations were highest (average of 0.844 – 3.42 μg/m³, 12.1 – 30.2% of total PM_2.5; Figure S3) during the summer months at all sites except BOS (Figure 2) and were also elevated during winter (average of 0.309 – 1.86 μg/m³, 3.25 – 11.1% to total PM_2.5) at all sites (Figure 2). Although similar in magnitude in terms of concentrations, the percentage contribution of AS to the total PM_2.5 was lower during winter than summer, as winter conditions are better suited for AN generation, as discussed in the Section on AN.

Higher sulfate concentrations are expected in warmer months due to higher solar radiation, increased humidity, and increased concentration of atmospheric oxidants (predominantly hydroxyl radicals). These factors lead to a higher conversion of SO₂ to sulfate (Masiol et al., 2019; Park et al., 2020; Wang et al., 2008). During wintertime PCAPs, equilibrium partitioning occurs between the gas and particle phases when ammonia, nitric acid, sulfuric acid, and chloride are all present in an airshed. Kelly et al. (2013) stated that under ammonia-poor conditions, when there is insufficient ammonia to neutralize all acid compounds, equilibrium partitions will favor the formation of ammonium sulfate over other ammonium salts (AN and ammonium chloride (AC)). The AS factor did not exhibit much weekday dependence (Figure 3). Hwang & Hopke (2007) also reported no dependence of AS on weekends for CSN sites in Oregon.

Over the years, the AS concentrations decreased significantly at all sites, with Thiel-Sen annual slopes ranging from −0.105 to −0.0399 μg/m³ per year (Figure 4 and Figure S2), and for all seasons and at all sites (Figure 5), except for wintertime Thiel-Sen slopes for SAC and VIS. While this study found a decrease in AS of 15% to 49% (winter 2007 – 2009 vs. 2015 – 2017), Kotchenruther (2020) reported a decline of 27% to 79% at Utah sites between these two winter periods. This study corrected the ion concentrations between November 2015 and October 2018, which may be responsible for this study’s smaller AS decline. The reduction in AS is primarily due to the reduction in the SO₂ emissions (Kotchenruther, 2020; Masiol et al., 2019; Stanimirova et al., 2024). Since the 1970s, the US has implemented policies to control the emissions of sulfur oxides (SO_x), with a significant focus on reducing the sulfur content in fuels. These policies include the US EPA’s Acid Rain Program, which reduced SO₂ emissions from power plants; the Clean Air Nonroad Diesel Rule (2004); the Tier 2 Vehicle and Gasoline Sulfur Program (2006); the ultra-low sulfur diesel standard (2007); and the Tier 3 standards that required gasoline to contain no more than 10 ppm sulfur (annual average) down from 30 ppm (Lattanzio & Mccarthy, 2014). The Clean Air Interstate Rule in 2005 and its replacement, the Cross-State Air Pollution Rule in 2010, focused on SO₂ and NO_x emissions from power plants. The Mercury and Air Toxics Standards were issued in 2011 to regulate hazardous air pollutants, including SO₂. In 2010, the US EPA revised the SO₂ NAAQS to 75 ppb (1-hour, primary) from 140 ppb (over 24 hours, primary) (Wisconsin Department of Natural Resources, 2022). The cumulative effect of these policies has resulted in dramatic reductions in ambient SO₂ concentrations, from an annual national ambient average SO₂ concentration of 12.1 ppb in 1980 to 0.7 ppb in 2022 (94% reduction) (EPA, 2024b).

The NEI for county-level SO₂ emissions decreased at all sites between 2000 – 2020 (Figure S11), similar to the AS factor decrease in this study. The NEI reported SO₂ emissions remained somewhat consistent between 2011 and 2020 for BFD, FRN, SAC, VIS, BOS, and RNO (Figure S11); similarly, AS contributions (Figure 4) either remained constant or increased after 2013 for these sites. Interestingly, BOS exhibited an increase in AS between 2010 – 2015, which coincided with increased SO₂ emissions from coal and natural gas usage (Figure S12, US EPA eGrid).

Total county-level SO₂ emissions and annual AS concentrations exhibited a strong correlation (>0.85) at BT, COM, LND, and HW, while moderate to strong correlations (0.5 – 0.85) were seen at BFD, BOS, RNO, and SAC. Poor to moderate correlations (<0.5) were found at FRN, LV, MOD, and VIS (Figure S13).

Figure S13 shows the correlation between the annual SO₂ emissions from tier 1 categories and annual AS concentrations. The following tier1 NEI categories are significant contributors to SO₂ emissions: “Fuel Comb. Elec. Util.”, “Fuel Comb Industrial”, “Fuel Comb Other”, “Highway Vehicle”, “Off-Highway”, and “Miscellaneous” (Figure S15). SO₂ emissions from “Highway Vehicle” and “Off-Highway” categories show positive correlations with AS at most sites, with Utah sites and COM having very strong correlations (>0.83, Figure S13), indicating that improving engine design and decreasing fuel sulfur content have led to a substantial decrease in the SO₂ emissions and subsequently AS emissions. The “Fuel Comb. Industrial” and “Fuel Comb. Other” category also show moderate (0.4 – 0.7) to strong positive correlations (>0.7 at various sites (Figure S13), with stronger correlations at RNO, LND, HW, COM, BFD, and BT. “Fuel Comb. Elec. Util.” category exhibited strong positive correlations at FRN, COM, BT, and BFD.

3.1.4. Chloride

PMF identified this factor at all sites except BOS, COM, and RNO (Figure S1). This factor explained > 80% of the Cl concentration. Kelly et al. (2013) reported this factor at the three Utah sites, and Hasheminassab et al. (2014) reported this factor at the California sites (BFD, FRN, and SAC). Other species contributing to this factor varied by site (Figure S1). For the Utah sites, FRN, VIS, and BFD, this factor was associated with NH₄, with NH₄ contributing 9.91 – 30.9% to the profile (Figure S1). These same six sites (BT, LND, HW, BFD, FRN, and VIS) tended to have low contributions of Na⁺ (0.182 – 4.48%) to this factor, which suggest a limited contribution from road or sea salt. LV, MOD, and SAC sites exhibited substantial contributions from Na⁺ (12.2% to 24.9%) to the factor, suggesting a contribution from sea salt (Figure S1). The other contributors to this factor include NO₃, SO₄, OC, and EC.

The chloride factor was most prominent during winter at the three Utah sites (HW, BT, and LND) (Figure 2), with mean winter concentration ranging from 0.576 – 1.25 μg/m³ (2.44 – 4.77% of total PM_2.5) and a strong association with NH₄. The spring, summer, and fall mean concentrations for these three sites were low (0.0168 – 0.116 μg/m³) (Figure 2), and with mean contributions of less than 1.49% to PM_2.5 (Figure S3). Previous studies identified chlorine in association with NH₄ as AC (Kelly et al., 2013), primarily formed as secondary aerosol during winter season. The most likely pathway for the atmospheric formation of AC is the reaction of NH₄ with HCl (Silva et al., 2007). The source of HCl could be direct emissions of HCl or precursor gases such as chlorinated hydrocarbons (Silva et al., 2007). Hydrochloric acid can also be produced as a product of the nitric acid reaction with salt, which is used to salt roads during winter (Silva et al., 2007). Wintertime AC formation is highly dependent on the thermodynamics between ammonium, nitrate, sulfate, and chloride. The thermodynamic partitioning among inorganic acids (SO₄, NO₃, hydrochloric acid or HCl) and cations (NH₃, Na, and calcium from dust) is highly complex and depends on the relative amounts of each compound and environmental conditions such as temperature and humidity (Emery et al., 2020). Kelly et al. (2013) highlighted the prevalence of AC at temperatures less than 10°C. This study also observed this association for HW, BT, and LND sites (Figure S16) as well as a negative correlation between the chloride factor and average daily temperature (Pearson r of −0.238, −0.219, −0.279 for BT, HW, and LND, respectively) (Figure S17).

The chloride factor contributions at LV, MOD, and SAC were highest during summer (2.01% – 7.49%) (Figure S3). MOD and SAC exhibited low winter-season chloride factor contributions (< 1.37%), and LV exhibited similar chloride factor contributions during spring, fall, and winter (3.86 – 4.14%) (Figure S3). Previous studies identified a chlorine factor associated with Na⁺ as fresh sea salt (Hasheminassab et al., 2014; Hwang & Hopke, 2007), with higher contributions during summertime. SAC and LV, with high summer and spring contributions, showed slightly positive Pearson correlations with average T (0.103 and 0.202 for SAC and LV, respectively, Figure S17). These seasonal patterns are expected, as elevated temperature causes more evaporation, and wind speeds are higher during summer, which enables the transport of sea salt from the Pacific Ocean (Figure S18).

The BFD, VIS, and FRN sites exhibited elevated winter and summer chloride concentrations (mean concentration of 0.269 – 0.947 μg/m³ for BFD, VIS, and FRN, respectively, in winter and 0.04 – 0.11 μg/m³ for BFD and FRN, respectively in summer, Figure 2). At these sites, chloride contributes 0.928 – 4.03% during winter and 0.438 – 1.16% during summer (Figure S3). Elevated chloride factor concentrations in both winter and summer could be due to AC formation and sea salt. Biomass burning could also affect winter chloride concentrations (Wang et al., 2021). A Pearson correlation of 0.36 and 0.46 was also observed between the chloride and the OC-rich factor (representing biomass burning) at BFD and FRN, respectively. These sites showed a negative Pearson correlation (−0.264, −0.368, and −0.395 for BFD, FRN, and VIS, respectively) with average daily temperature (Figure S17). Hasheminassab et al. (2014) also reported a different seasonal prevalence of chloride sources at FRN, BFD, and SAC, sites, with high concentrations during fall and winter for FRN and BFD, and summer dominance for SAC.

Major possible sources of chlorine include industrial emissions, coal combustion (Liu et al., 2018; Reizer et al., 2021), wintertime road salting, sea salt (Reizer et al., 2021), agriculture (use of chlorine-based pesticides and fertilizer), waste incinerators (especially incineration of PVC-based waste), biomass burning (Chang et al., 2024), and biological activities linked to saltwater lakes and marshes (Rhew et al., 2014). The microbial activity in salt marshes and the ocean can release methyl halides (including methyl chloride), which can serve as carriers of halogens into the atmosphere (Hirata et al., 2017; Rhew et al., 2014; Rhew et al., 2000). In the presence of sunlight, methyl chloride (CH₃Cl) can undergo photolysis to produce Cl radical.

The Utah sites may be affected by the high chlorine and HCl emissions from US Magnesium (a magnesium production facility that lies approximately 70 km due west of Salt Lake City, with annual emissions of 255 – 5438 tons of Cl and 263 – 1498 tons of HCl. Womack et al. (2023) found a gradient in HCl concentration between the west (high) and east part of the SLV and indicated that US Magnesium was a possible reason for this concentration gradient. Womack et al. (2023) suggested that this single source, US Magnesium, may be responsible for regional PM_2.5 increases of 10% – 25% compared to a baseline without this source. The other unique source in the SLV includes emissions from the Great Salt Lake. Stutz et al. (2002) suggested that reactive halogens (Cl and Br) get mobilized from salt on the flats around the Great Salt Lake. Road salt is another possible source of Cl, although Kelly et al. (2013) rejected the road salting and sodium chloride-nitric acid interactions as the sources due to NH₄-rich conditions in the valley (all nitric acid neutralized by NH₄) and the absence of Na in the AC profile.

The chloride factor concentrations at Utah sites were elevated between 2011 – 2015 (Figure 4, Figure S19). Although the emissions of Cl+HCl were high between 2011 – 2015 (2000–3000 tons/yr of Cl and 500 – 1498 tons/yr of HCl), no correlation (Pearson r between −0.004 – 0.004) existed between the Cl+HCl emissions and total annual chloride concentrations. However, a moderate to strong correlation was observed between annual chloride factor concentrations and the product of NH₄/(NO₃+SO₄) and number of PCAPs (Figure S20, Pearson r of 0.744, 0.608, and 0.418 for BT, HW, and LND, respectively). This indicates that the formation of AC depends on the availability of NH₄ ions and halogens, and the occurrence of PCAPs, particularly when the temperatures were less than 10°C, ideal for AC formation. A more complete understanding of the availability of gas-phase chloride, as represented by HCl, is required for the development of more reliable models for the management of Wasatch Front air quality (Martin et al., 2019).

Over the 20-year study period, the concentration of this factor decreased significantly during winter and fall at BFD (−0.0183 and −0.00403 μg/m³ per year, respectively), during fall at FRN (−0.00454 μg/m³ per year), and during summer at MOD (−0.0130 μg/m³ per year) (Figure 4 and Figure S2). The factor concentrations increased during summer and spring for SAC (0.00359 μg/m³ per year for spring and 0.00904 μg/m³ per year for summer) and HW (0.00178 μg/m³ per year) (Figure 4 and Figure S2). For all other sites and seasons, the trends were unclear (slopes not significant).

Chloride concentrations decreased during winter at all sites (Figure 5). Since chloride is formed via a complex mechanism, this decrease could be due to decreased emissions that affect one of the chloride formation pathways. For example, policies on winter-time biomass burning, industrial emissions, and vehicle emissions and the corresponding decrease in NO_x and SO₂ concentrations could affect AC formation. For example, decreasing NO_x concentrations could limit AC formation by decreasing HCl produced by the reaction between nitric acid and salt. Decreasing SO₂ concentrations could reduce AC formation via the reaction of chlorine with sulfuric acid to release HCl. The sites with winter-dominant AC factors exhibited a strong correlation with the AN factor (Pearson r of 0.386 – 0.597 for VIS, LND, HW, FRN, BT, and BFD), the OC-EC-rich factor (0.292 for VIS), the EC factor (0.447 for LND and 0.119 for BT), and the OC-rich factor (0.188 – 0.464 for HW, FRN, and BFD) (Figure S21). These correlations suggest that the effectiveness of policies reducing AN, OC-rich, and EC-rich factors are also helping to reduce the wintertime AC.

Sites with increasing chloride summer concentrations (SAC and MOD) could be caused by increasing temperatures leading to increased evaporation and more sea salt in the atmosphere (Figure 5, Figure S22). SAC and MOD exhibited moderate correlations between the aged salt (discussed in the next section) factor and chloride factor (Pearson r of 0.480 and 0.318 for SAC and MOD, respectively; Figure S21); LV showed a slight positive correlation (0.147) with the crustal factor (Figure S21), indicating possible relationship between the road salt/dust and chloride factor.

This factor generally did not show a clear weekday dependency (Figure 3), although the Utah sites showed higher weekday contributions (significant for LND; p<0.05). This dependence suggests the role of industrial emissions (if operations are reduced on weekends) or vehicle emissions on chloride factor concentrations.

3.1.5. Aged Salt

PMF identified an aged salt factor at BFD, BOS, FRN, MOD, RNO, and SAC, and this factor explained >85% of the variation in Na⁺ at all sites. NO₃ (24.9 – 40.9%), SO₄ (17.9 – 32.2%), and OC (20.2 – 34.7%) also contributed significantly to this profile (Figure S1). Mg contributions to the profile were small (<1%) at BFD, FRN, MOD, and SAC. MOD and RNO exhibited small chlorine contributions (<1%, Figure S1).

Previous studies identified high concentrations of Na⁺, SO₄, and NO₃ as aged salt (Kim et al., 2010; Kotchenruther, 2016; Squizzato et al., 2018). Squizzato et al. (2018) apportioned this factor at sites in New York and found >80% of the explained variation of Na⁺ was associated with this factor. The lack of chlorine in this profile has been explained by the Cl displacement by acidic gases (H₂SO₄ and HNO₃)(Song & Carmichael, 1999; Squizzato et al., 2018). For California sites, the marine aerosol can react with acidic gases, especially nitric acid, resulting in depletion of halogens from the aerosol and formation of alkaline salts (Motallebi, 1999; Song & Carmichael, 1999).

For the California sites, this factor’s concentration was highest for summer (Figure 2, mean concentration between 0.998 – 1.66 μg/m³, and lowest for winter (0.415 – 0.605, μg/m³). The aged salt factor also had the largest contribution to the total PM_2.5 during spring (12.2 – 21.8%) and summer (9.74 – 21.6%) and the lowest contributions during winter (<6.07%, Figure S3). For BOS, the concentrations were highest during winter (0.701 μg/m³) and lowest during spring (0.505 μg/m³, Figure 4). The contribution to total PM_2.5, at BOS, was similar during summer and fall (~9.87%; Figure S3). RNO showed similar mean concentrations (0.323 – 0.378 μg/m³; Figure 2) and contributions (5.93 – 9.35%; Figure S3) for all seasons.

Summertime-dominant aged salt has been associated with marine aerosol aging, promoted by the seasons’ elevated onshore winds, solar radiation, humidity, and concentration of atmospheric oxidants (predominantly hydroxyl radicals) (Hasheminassab et al., 2014). Hasheminassab et al. (2014) reported the highest concentrations of aged sea salt in the SJV during summer and the lowest during winter. The California sites showed a positive correlation (Pearson r between 0.102 – 0.374) between average daily temperature and aged salt contributions. Wintertime aged salt has been linked to road salt (Chen et al., 2012; Masiol et al., 2019; Reizer et al., 2021), which subsequently reacts with nitric acid or sulfuric acid to form aged salt. No strong weekday dependence was observed for this factor (Figure 3).

Over the 20-year study period, the aged salt concentrations decreased at all sites except BOS (Figure 4, Figure S2). This decrease was significant for BFD, FRN, and SAC (−0.0233 to −0.00586 μg/m³ per year, Figure S2). For the BFD, FRN, RNO, and SAC sites, the aged salt concentration dropped substantially between 2000 and 2005 (Figure 4), followed by a slower decline. BOS exhibited no clear trends. The BOS measurements began in 2006, collected every sixth day, which may contribute to the lack of a clear trend (since the major decrease occurred between 2000 – 2005 at the other sites) (Figure 4).

Except for BOS, all sites exhibited a significant decrease in winter-time aged salt (Figure 5). For the California sites (FRN, BFD, SAC, and MOD), which typically experience no snow fall, the aged salt concentration depends on marine salt and the acid gas concentrations. For sites with snow, like RNO and BOS, the aged salt concentration could depend on the road salt and acid-gas concentrations. Because the sites in this study have NH₄-rich conditions during winter, sulfuric and nitric acids are mostly neutralized by NH₄, limiting the availability of these acids for the formation of aged salt. The decreasing emissions of NO_x and SO₂ (Figure S7 and S11) limit the availability of nitric acid and sulfuric acid for aged salt formation and may be one cause of the observed decline in aged salt concentrations, especially between 2000 – 2005 (when more stringent NO_x and SO₂ regulation were implemented).

The other seasonal trends were less clear, FRN, RNO, and SAC exhibited significant decreases in aged salt in spring (Figure 5). FRN and BFD showed significant declines in summer, and BFD and SAC showed significant declines in fall. BOS did not show significant trends for any seasons (Figure 5).

Methodological changes associated with the CSN network could affect the concentration of sodium (Na⁺) and aged salt factor concentrations. Starting in January 2016, CSN began using field blanks to correct ion concentrations, and Na⁺ experienced substantial blank corrections. Additionally, the contractor responsible for ion analysis changed between November 2015 and October 2018 (Young, 2021). This transition could have further affected Na⁺ concentrations, but this study was unable to assess the extent of this impact. Shifts in CSN-reported Na⁺ concentrations resulting from these changes complicate the interpretation of Na⁺ and aged salt concentration trends.

3.1.6. Dust

PMF identified a dust factor at all sites, and this factor profile was rich in crustal soil and mineral elements including Si (22.8% – 42.2%), Al (6.68 – 13.9%), Ca (4.74 –18.9%), Fe (6.35% – 15.3%), Ti (<2%), Mg (<2%), and K (<2%) (Figure S1). This profile explained >50% of the variation in Al, Fe, Ca, Mg, Si, and Ti. SO₄, NO₃, OC, and EC were also present at most sites in varying quantities (Figure S1). Crustal elements, such as Al, Fe, Ca, Mg, Si, and Ti, are markers of mineral and soil dust (Frie et al., 2017; Reizer et al., 2021; Shi et al., 2018; Taghvaee et al., 2018; Wang et al., 2008). This study identified trace amounts of Na⁺, Mn, P, Sr, Br, Cu, Ni, and Pb, indicating the presence of road dust in this factor (Amato et al., 2010; Squizzato et al., 2018; Thorpe & Harrison, 2008). Previous studies have also reported significant amounts of OC, EC, SO₄, and NO₃ in road dust factors (Hasheminassab et al., 2014; Kelly et al., 2013; Lee et al., 2008; Squizzato et al., 2018). Ti is sometimes used as a tracer for soil and road dust (Kotchenruther, 2016; Reizer et al., 2021; Watson et al., 1994). The dust profiles in this study show contributions from both natural (soil materials) and anthropogenic (traffic emissions, industrial processes) sources, as observed in previous studies (Brown et al., 2007; Hasheminassab et al., 2014; Lee et al., 2008; Perrone et al., 2019; Squizzato et al., 2018; H. Wang et al., 2008).

Fugitive dust emissions originate from unpaved roads, disturbed soil, construction sites, and agricultural activities (Kelly et al., 2013; Reff et al., 2009; Watson et al., 2008). Chen et al. (2007) reported that farmlands, pasture lands, and unpaved roads are major sources of windblown dust in the SJV. Road dust can arise from brake and tire wear, road surface abrasion, and resuspension (Amato et al., 2010; Squizzato et al., 2018; Thorpe & Harrison, 2008). The shrinking Great Salt Lake and exposed playas also contribute to the wind-blown dust in the SLV (Brahney et al., 2024). Flowerday et al. (2025) associated the dust in the SLV with winds from the south and west, which pass over playas including Milford Flats, Sevier Dry Lake, Tule Dry Lake, Great Salt Lake Desert, Dugway Proving Grounds, and the West Desert of Utah (Flowerday et al., 2025). The sources of dust can also differ with the season (Kelly et al., 2013). Non-regional dust sources, such as intercontinental dust storms (for example, Asian dust episodes) also affect the dust concentrations in the western US (Hwang & Hopke, 2007; Kim & Hopke, 2006).

All of the locations exhibited the lowest dust concentrations and contributions during winter (Figure 2), which is expected because these sites experienced increased moisture (Figure S23) in the winter and some locations (BOS, BT, HW, LV, RNO, and COM) experienced snow cover (Figure S24 and Figure S25). Snow cover reduces fugitive dust (Chen et al., 2012; Kelly et al., 2013). All sites (excluding California) showed negative correlations between the dust factor and snow depth (r < −0.32) (Figure S5). Kelly et al. (2013) found that high wind speeds during winter months did not increase the fugitive dust factor at HW, BT, and LND sites.

In other seasons, some differences exist in seasonal concentrations. The Utah sites and BOS exhibited the highest concentrations in summer (peaked in June and July), with average mean summer concentrations of 0.594 – 1.00 μg/m³ (Figure 2). LV exhibited the highest concentrations in summer and spring (0.743 and 0.764 μg/m³, respectively). RNO and COM had similar mean concentrations (0.273 – 0.382 μg/m³ in RNO and 0.788 – 0.797 μg/m³ in COM) during fall, spring, and summer. BFD, FRN, MOD, SAC, and VIS exhibited the highest concentrations in fall with mean concentration of 1.06 μg/m³ (Figure 2).

The dust contributions to total PM_2.5 varied by season and location. In spring, the dust factor contributed 7.83% – 11.82 % to PM_2.5 in BT, HW, LV, COM, and RNO (Figure S3). The higher wind speeds during spring (Figure S18) can suspend crustal material (Khan et al., 2010; Shi et al., 2018; Tian et al., 2013). For LND, BFD, BOS, and VIS, the dust contribution peaked in summer (8.72% – 13.35%). Squizzato et al. (2018) linked elevated concentrations of dust at a New York site in spring and summer months to the combined effects of increased traffic, drier conditions, and the absence of snow cover. MOD (5.64% – 6.78%) and SAC (4.15% – 4.62%) had similar dust contributions in fall, spring, and summer. For FRN, dust contributions were highest during fall and summer (4.81% – 5.35%) (Figure S3). FRN, BFD, VIS, and MOD are located in major agricultural regions of California, and these rural agricultural areas in California experience higher fugitive dust during drier, non-winter periods (Chow et al., 2007). Among all the California sites, SAC exhibited the lowest dust concentrations (Figure 2), likely because of the higher RH and precipitation at this site (Figure S23 and Figure S25) (Harrison et al., 2001; Hasheminassab et al., 2014) and the residential location of this site

Weekday dependence was observed at BFD, COM, FRN, MOD, RNO, SAC, and VIS (Figure 3, p<0.05). Brown et al. (2007) reported decreased concentration of soil components (e.g., Si, Ca, Fe, and Ti) on weekends, highlighting the impact of road dust during the week. Hwang & Hopke (2007) reported a lower average weekend contribution (0.24 μg/m³) of airborne soil sources compared to the weekday contribution (0.29 μg/m³) for sites in southwestern Oregon. The proximity of these sites to the nearby highways (Site description, supplementary material section S1) could be related to the higher weekday concentrations of this factor. Elevated dust concentrations were observed on Sundays for FRN, HW, and LND sites (Figure 3).

The contribution of the dust factor to total PM_2.5 increased at all sites over the study period, except for VIS and RNO (Figure S6). This increase, ranging from 0.0875% to 0.336% per year, was significant at BFD, BOS, BT, COM, FRN, and HW. The annual averaged Theil-Sen fit (Figure 4 and Figure S2) indicated rising dust concentrations (0.000672 to 0.0240 μg/m³ per year) at most sites, with significant trends at BFD, COM, and FRN. However, RNO (−0.00387 μg/m³ per year, p<0.05) and VIS (−0.00186 μg/m³ per year) exhibited declining concentrations (Figure 2 and Figure S2). The seasonal trends in dust concentrations exhibited significant variability across different monitoring sites. Notably, BFD, FRN, and COM demonstrated positive and statistically significant increases in dust concentrations during fall and summer, whereas RNO showed a statistically significant decreasing trend in spring (Figure 5). Utah sites showed positive trends for spring dust concentrations. This variation underscores the seasonal and site-specific nature of dust concentration dynamics.

Previous studies have associated increasing dust concentrations with increasing aridity, decreasing precipitation, decreases in relative humidity, and soil moisture (Brey et al., 2020; Hand et al., 2016; Pu & Ginoux, 2018). Pu & Ginoux (2018) found that increasing fine dust concentrations in the southwestern US in spring are associated with a significantly decreased precipitation. Other factors contributing to increased dust concentrations include intensified land use and disturbed lands, wildfires (Hahnenberger & Nicoll, 2014; Miller et al., 2012), increased population and economic activity, and impacts of climate change (Hallar et al., 2020). Wildfires can contribute to increased crustal PM by creating erodible surfaces and enhancing dust emissions (Menut et al., 2022). This study identified a positive Pearson correlation between the total annual dust concentrations at each site and state-level land area burned for HW (r = 0.108), FRN (0.762), BT (0.356), BFD (0.501), SAC (0.530), and LV (0.244) (Figure S26). Schlosser et al. (2017) discussed how wildfire plumes can significantly impact downwind states, leading to elevated levels of airborne soil at the downwind states (for example, Arizona for a California fire). In recent years, the long-range transport of dust (African and Asian dust) has also increased in frequency and intensity (Hwang & Hopke, 2007; Kim & Hopke, 2006; Park et al., 2020; Perry et al., 1997).

VMT is another potential factor in increasing dust levels. More vehicles on the road cause more resuspended road dust. Farahani et al. (2021) linked high resuspension of road dust in the Los Angeles Basin to increased traffic activities. The increasing number of electric vehicles (EVs) has also been associated with higher resuspended dust (Farahani et al., 2021; Timmers & Achten, 2016) because EVs are heavier than internal combustion engine vehicles, resulting in increased tire and road surface wear. In this study, VMT has increased substantially in the last 20 years at all the sites (Figure S27). We observed a positive Pearson correlation between county-specific annual VMT and yearly dust concentrations for BFD (0.393), BT (0.457), FRN (0.326), HW (0.239), LV (0.705), and VIS (0.415), respectively (Figure S28). These positive correlations suggest the association between increasing VMT and resuspended dust.

3.1.7. EC rich (vehicle emissions)

PMF identified this EC-rich factor at all sites except COM, FRN, SAC, and VIS (Figure S1). The factor profile explained over 50% of the variation in EC. EC’s contribution to the total factor varied between 30.0 – 45.9%, depending on the site. This factor contained other components, such as OC (19.9 % – 37.3%), NO₃ (0.24 – 32.5%), SO₄ (0.132 – 9.82%), and trace elements including Fe (2.00 – 6.33%), Ca (0.372 – 12.9%), Na⁺ (0 – 1.41%), Mn (0.0308 – 0.285%), Cr (0.0366 – 0.0871%), Cu (0.00600 – 0.316 %), Pb (0.0463 – 0.133%), Zn (0.205 – 1.32%), and Ni (0.0249 – 0.0356%), although their quantities varied by site (Figure S1). While these trace elements are not major contributors to the factor profile, the majority of their explained variation was associated with this factor. V, Sr, and Ce were present in the LV site, and Sn, Se, and P were also present in MOD. The percentage contribution and explained variation of each species can be found in the supplementary material (Figure S1).

The dominance of EC along with substantial OC and trace elements in this factor has been linked to mobile emissions, with diesel emissions being a major source of EC (Kelly et al., 2013; Kotchenruther, 2016; Park et al., 2020). Gasoline emissions are richer in OC than in EC (Park et al., 2020). Major sources of EC in the U.S. include internal combustion engines, wild and controlled burning of vegetation, and in some cases, wood stoves and biofuels (Bond et al., 2004; Murphy et al., 2011). Chen et al. (2012) identified diesel engine exhaust as the primary source of EC at the RNO site.

The EC/OC ratio is commonly used to differentiate between gasoline and diesel emissions. Heo et al. (2009) reported EC/OC values of 0.4 for gasoline and 1.4 for diesel, while Hasheminasaab et al. (2014) reported EC /OC ratios of 0.4 for their gasoline source profile and 2.2 for their diesel vehicles source profile. Chen et al. (2007) extracted an EC/OC ratio of 0.8 for mobile vehicles (representing a combination of diesel and gasoline emissions) in the SJV. Mobile sources generally have EC/OC ratios of around 1 (Hand et al., 2013), while rural and biomass-burning aerosols have much lower ratios (~0.1). In this study, the EC/OC ratio was slightly above 1 (1.03 – 2.13) for the EC-rich factor, suggesting that this factor represents mobile emissions, with both gasoline and diesel characteristics.

The concentrations of this factor (Figure 2) were highest in winter (mean: 1.08 – 2.35 μg/m³) and fall (0.855 – 2.05 μg/m³), and lowest for the summer (0.357 – 0.853 μg/m³). The contribution of this factor to PM_2.5 varied between 8.78 – 21.8%, 10.4 – 24.7%, 10.7 – 20.8%, and 5.46 – 12.8% for winter, fall, spring, and summer, respectively (Figure S3). All the sites in this study experience wintertime PCAPs, that increase pollution accumulation (Kelly et al., 2013; Kotchenruther, 2016), explaining the high wintertime factor contributions. Hand et al. (2013) linked elevated winter EC in the western U.S. to urban combustion sources, including residential heating and mobile sources, that are trapped by poor ventilation. Kelly et al. (2013) found that the impacts of their mobile factor peaked in winter months in Utah, likely due to the high frequency of wintertime PCAPs, rather than actual increase in mobile emissions. Additionally, cold-start emissions, which can comprise a substantial portion of on-road vehicle emissions, are higher in winter because temperatures are lower (Reiter & Kockelman, 2016). Watson et al. (2007) attributed 38–49% of the PM_2.5 to mobile sources in southern Nevada, including LV.

This factor exhibited a weekday dependence (Figure 3), with significantly lower concentrations of the EC-rich factor on weekends at all sites (p<0.001). Murphy et al. (2008) reported a strong Sunday-Monday minimum in EC concentrations, even at remote IMPROVE sites, and attributed this to lower use of diesel trucks and off-road equipment on weekends. Kelly et al. (2013) also reported significantly lower contribution of their diesel factor at HW, BT, and LND on weekends.

The Thiel-Sen slopes showed significantly decreasing concentrations over time for BFD, BT, HW, LV, and RNO, with slopes ranging between −0.0995 to −0.0284 μg/m³ per year (Figure 4 and Figure S2). The slopes were positive for BOS (0.00472 μg/m³ per year) and MOD (0.00219 μg/m³ per year) but not statistically significant. Green et al. (2015) also reported a positive, although insignificant, Thiel regression slope for BOS for EC concentrations for the period between 2000 – 2012. The Theil-Sen slopes varied by season and site (Figure 5). For example, for BFD, the slopes were positive for winter (0.0101 μg/m³ per year), but significantly negative for all other seasons (Figure 5). BT exhibited a significant decrease for all seasons except for spring. For the Utah sites, this study’s results differ from those of Kotchenruther (2020), who reported an increased contribution from gasoline and diesel emissions at BT, HW, and LND between two 3-year winter periods (2007–2009 vs. 2015–2017). The Kotchenruther (2020) study applied PMF using organic carbon (OC) and EC subfractions, as well as organic matter (OM, defined as 1.4 × OC) and was able to differentiate between diesel and gasoline sources. The Kotchenruther (2020) study identified EC2 (a subfraction of EC) as a dominant marker of diesel emissions, with EC2 concentrations increasing from 2% to 7% at HW and BT across the two periods. This study used only total OC and EC in the PMF analysis, without subfraction data, and showed decreasing EC rich contributions. However, these methodological differences make a direct comparison difficult.

Over the years, EPA regulations on mobile emissions have become increasingly stringent, driving advancements in engine technology and reductions in PM and PM precursor emissions over the past two decades. On-road diesel particulate standards were introduced at 0.6 g/bhp-hr in 1988 and subsequently reduced to 0.25 g/bhp-hr in 1991, 0.1 g/bhp-hr in 1994, and 0.01 g/bhp-hr in 2007 (Murphy et al., 2011). The U.S. EPA's nonroad diesel engine emission standards progressively tightened PM limits, starting with Tier 1 (1996–2000) at 0.4–0.6 g/bhp-hr, Tier 2 (2001–2006) maintaining similar PM limits, Tier 3 (2006–2008) focusing on NO_x reductions without significant PM changes, and Tier 4 (2008–2015) requiring advanced after-treatment technologies to reduce PM emissions to 0.02 g/bhp-hr for most engines (EPA, 2004). Regulations for light-duty vehicles also became more stringent over time. Under Tier 2 (2004–2009) emission standards set a particulate matter (PM) limit of 0.01 g/mile for Bins 1–6, while Tier 3 (2017–2025) further tightened the PM standard to 0.003 g/mile for most light-duty vehicles (EPA, 2025b). California has implemented stricter vehicle emissions standards than federal regulations. Starting in 2010, with a rolling compliance period beginning in 2015, all on-road HDDVs operating in California were required to either retrofit existing engines with particle filters or replace engines older than the 2007 model year (Schwarzman et al., 2021). Although HDDVs remain California’s largest source of diesel particulate matter emissions, regulatory actions reduced HDDV emissions by 85% between 1990 and 2014 (Schwarzman et al., 2021).

Federal initiatives, such as the Diesel Emission Reduction Act, have provided funding to retrofit, repower, and replace diesel engines with newer, cleaner technologies (EPA, 2024a). Other incentives include tax credits for purchasing EVs and plug-in hybrid electric vehicles (PHEVs), access to high-occupancy vehicle lanes for EVs and alternative fuel vehicles, and improved charging infrastructure (U.S. Department of Energy, 2024). The fraction of registered EVs and PHEVs has increased from 0.060 – 0.841% to 0.215 – 2.01% in the five states (California, Colorado, Nevada, Utah, and Idaho) between 2016 and 2020 (Figure S29). EVs also eliminate emissions associated with cold starts, which can account a substantial fraction of on-road vehicle emissions (Reiter & Kockelman, 2016). In addition, renewable fuel standards have mandated the blending of renewable fuels, like ethanol and biodiesel with gasoline and diesel, and these fuels tend to emit less PM_2.5 and SO₂ (Yacobucci, 2013).

The decline in EC-rich factor concentrations (Thiel-Sen slope: −0.0995 to 0.0219 μg/m³ per year) has been less pronounced compared to reductions in AN (−0.321 to −0.0378 μg/m³ per year) and sulfate (−0.105 to −0.0398 μg/m³ per year) concentrations (Figure 2 and Figure S3). While stringent tailpipe emission regulations have reduced emissions per vehicle, the number of registered vehicles has increased (diesel vehicles by 10.1 – 26.9% and gasoline vehicles by 1.98 – 13.2% between 2016 and 2020) (Figure S29) corresponding to greater VMT, which has been steadily rising since 2011 (Figure S27). The Pearson correlation between this factor and county-level annual VMT data shows negative correlations for RNO (−0.216), LV (−0.638), LND (−0.436), HW (−0.378), and BT (−0.216), suggesting that reductions in per vehicle emissions have helped lower tailpipe emissions despite rising VMT (Figure S28). Because total emissions depend on both emission per vehicle and VMT, the combination of increasing VMT and decreasing emission rates has resulted in a slower decline in vehicle emissions than nitrate and sulfate emissions. Murphy et al. (2011) and Kotchenruther (2020) also attributed the muted decrease in vehicle emissions to rising VMT. While the number of zero-emission vehicles is growing, their share remains small, making up less than 2% of the total fleet. MOD, however, showed a slightly positive Pearson correlation with VMT (0.241), and BFD exhibited a near-zero correlation (0.0447). MOD also displayed positive Theil-Sen slopes for the EC-rich factor, suggesting that increasing VMT may be contributing to a modest rise in emissions there.

3.1.8. EC Cu rich (vehicle emissions)

PMF only identified this factor in FRN (Figure S1). This factor could be a mixture of Cu-rich and EC-rich factors. The factor profile had a high fraction of EC (28.6%), nitrate (41.4%), OC (16.7%), SO₄ (7.22%) Fe (3.61%), Cu (0.367%), Ca (0.445%), Zn (0.203%), Ti (0.117%), and Cr (0.097%). Although Fe, Cu, Cr, Pb, Mn, Ni, Ti, and Zn did not contribute much to this overall factor, 34.1%, 82.3%, 53.9%, 17.1%, 21.2%, 35.9%, 20.5%, and 20.6% of the explained variation of respective elements was associated with this factor (Figure S1). Park et al. (2020) found that elevated Cu and Zn concentrations were linked with diesel emissions. The high EC/OC ratio (1.72) indicates diesel contributions mixed with other combustion sources. Significantly higher weekday concentrations (Figure 3) were observed (p<0.001), indicating the possible association with vehicle or industrial emissions. This factor showed elevated mean average concentrations for winter (1.51 μg/m³) and fall (1.30 μg/m³) (Figure 2), which could be associated with wintertime PCAPs, and low concentrations for spring (0.636 μg/m³) and summer (0.532 μg/m³). This factor contributed 7 – 8% to the total PM_2.5 concentration, depending on the season (Figure S3).

Overall, the Theil-Sen fit showed a negative slope (−0.0172 μg/m³ per year, p>0.05) (Figure 4 and Figure S2). The slopes were significantly negative for Spring (−0.0131 μg/m³ per year; p<0.05) and Summer (−0.0467 μg/m³ per year; p<0.001), negative for fall (−0.0247 μg/m³ per year) and positive slopes for winter (0.00997 μg/m³ per year, p>0.05) (Figure 5). The annual average concentrations increased between 2000 – 2006, followed by a decrease until 2010, a small increase until 2012, consistent concentrations until 2016, followed by a sharp decrease until 2020 (Figure 4). These trends generally follow VMT (Figure S27). The sharp decrease after 2016 may be due to the on-road HDDV and non-road diesel engine regulations, which began in 2015. Starting in 2015, all on-road HDDVs operating in California were required to either retrofit existing engines with particle filters or replace 2007 and older engines (Schwarzman et al., 2021).

This factor exhibited a positive correlation (r = 0.491) with county-level annual VMT (Figure S2), suggesting that the increasing VMT affected the contribution of this factor, in spite of the increasingly stringent vehicle emission standards (discussed in the previous section). California has struggled to reduce diesel emissions from the agriculture sector, which accounted for up to 18% of the state’s total diesel particulate emissions from mobile sources in 2014 (Schwarzman et al., 2021). Fresno is the top agricultural producer in California and is expected to have substantial diesel particulate emissions.

3.1.9. OC rich (biomass burning)

PMF identified this factor at all sites except COM and VIS (Figure S1). The dominant contributor to this factor profile was OC, with contributions ranging from 67.9% – 89.7%, depending on the site. The other major contributors included EC (8.07 – 20.7%), K (1.27 – 2.53%), NO₃ (0 – 5.83%), and SO₄ (0 – 5.35%) (Figure S1). This profile explained >60% of the variation in OC and >40% of the variation in K (Figure S1). High concentrations of OC and K, and moderate concentrations of EC indicated biomass burning (Jeong et al., 2011; Park et al., 2020; Parworth et al., 2017). The EC/OC ratio was in the range of 0.0900 – 0.305, and EC/OC ratios (~0.1) represent biomass burning (Hand et al., 2013; Kotchenruther, 2016). The K to OC ratio was in the range 0.0149 – 0.0309. Kotchenruther (2016) has associated K/OC ratio of 0.012 with aged wood smoke and 0.021 to primary wood smoke. The K/OC range observed in this study suggests a mixed contribution from fresh and aged smoke.

This factor exhibited elevated concentrations during winter (1.03– 7.25 μg/m³) and fall (1.36 – 5.16 μg/m³) (Figure 2). Summer peaks were also observed at all sites (1.18 – 2.53 μg/m³). The concentrations of this factor were lowest during the spring months (0.629 – 1.77 μg/m³) at all sites. For BOS the average summer and fall concentrations were higher than winter concentrations. RNO showed a similar mean concentration for fall, summer, and winter (2.27 – 2.53 μg/m³). This factor’s mean contribution to total PM_2.5 was 11.6 – 35.0% in winter, 18.0 – 27.8% in Fall, 14.7 – 30.5% in summer, and 13.6– 20.5% in spring (Figure S3).

Elevated winter OC rich source concentrations have been associated with residential heating (Hand et al., 2013; Jeong et al., 2011; Park et al., 2020; Xue et al., 2019). Xue et al. (2019) associated wintertime elevated concentrations of OC in wintertime at FRN (32−47%) with biomass burning, which is a significant heating source in this region. Chen et al. (2007) and Gorin et al. (2006) also associated increased wintertime atmospheric OC in SJV in the winter with residential wood combustion. Wood smoke from residential burning is largest source of wintertime PM_2.5 emissions in Sacramento County (Brown et al., 2020). For the northwest and intermountain west US, Kotchenruther (2016) associated wintertime wood smoke with residential wood combustion, since there are typically no prescribed fires or wildfire activity during winter. Chen et al. (2012) attributed the high OC levels in Reno to residential wood combustion. For LV, residential biomass burning can contribute more than 26% to PM_2.5 between 17:00 and 05:00 LST (Brown et al., 2016).

This factor’s summertime contribution was also substantial at all sites (Figure 2). Kotchenruther (2016) associated high summer impacts of biomass burning with summertime wildfire activity and noted contributions from primary wood smoke, aged wood smoke, and secondary OC. Due to the use of OC and EC instead of the OC and EC thermal splits, in this study, such distinction between the primary, aged, and secondary OC could not be made. We believe all these three sources (fresh, aged, and secondary OC) are pooled in this one factor. Summertime impacts of this factor are due to regional (rather than state-specific) impacts of the woodfire smoke. For example, a low number of acres burned in Utah and Nevada (i.e., 2014 and 2015; Figure S14) do not necessarily correspond to low OC-rich factor concentrations, because the Utah and Nevada sites are affected by California, Oregon, and Washington fires. Kotchenruther (2016) associated high summer concentrations of an OC-rich factor at LND to aged smoke and secondary OC formation, even during the low wildfire activity years in Utah.

No strong dependence on the days of the week was observed at any sites, except the Utah sites and LV (Figure 3). Utah sites and LV exhibited higher OC rich concentrations during the weekends (significant for p<0.05), potentially due to residential wood combustion. A slightly higher magnitude was observed for Saturday at all other sites.

Thiel-Sen analysis yielded negative slopes for all sites (−0.0918 – −0.00706 μg/m³ per year) except LV (0.0217 μg/m³ per year) (Figure 4 and Figure S2). The most pronounced and statistically significant negative trends were observed at HW, MOD, and LND. Seasonally, winter exhibited the most consistent and significant decline across all sites (−0.238 to −0.0440 μg/m³ per year), except for LV (Figure 5). Spring also showed negative trends at all sites, with significant decline at FRN and LND, while summer had positive but insignificant slopes at BFD, BOS, FRN, MOD, RNO, and SAC. In LV, seasonal trends were mixed, although none were significant. The interpretation of LV trends is complicated by multiple site relocations.

The observed decline in wintertime OC levels is consistent with previous findings. Kotchenruther (2020) reported a decrease in total wood smoke in BOS, BT, HW, and LND due to EPA’s National Performance Standards for new residential wood heaters (revised in 2015. Several locations offer incentives to replace older, high-emitting devices with cleaner options. These include tax deductions in Idaho to replace old wood stoves (Idaho Department of Environmental Quality, 2024), funding to replace wood combustion with natural gas or electric heating (BT, HW, LND), and wood-stove change-out programs (SJV) (San Joaquin Valley Air Pollution Control District, 2025).

Wood-burning bans at many of these sites have contributed to decreased wood-burning emissions (Yap & Garcia, 2015). For example, Northern Ada County (BOS) issues voluntary burn bans when PM₁₀ levels are expected to reach 64 μg/m³ and issues bans when PM₁₀ reaches 100 μg/m³ (Department of Environmental Quality, 2013). California has no-burn days that have become increasingly strict beginning in 1993 (San Joaquin Valley Air Pollution Control District, 2021). In 2019, Fresno and Kern counties lowered the “no burning unless registered” threshold from 20 μg/m³ to 12 μg/m³ and the “no burning for all” threshold from 65 μg/m³ to 35 μg/m³ (San Joaquin Valley Air Pollution Control District, 2021). Colorado limits burning to approved devices between October 31 to March 31 when air quality is unhealthy or conditions are expected to worsen (CDPHE, 2025). Utah banned solid-fuel burning when PM_2.5 levels reached 35 μg/m³ in response to EPA’s updated NAAQS in 2008. By 2013, it began declaring no-burn days based on air quality forecasts in non-attainment counties (HW, BOS, and LND). Nevada, including Washoe County (RNO), discourages burning when AQI is yellow and prohibits burning if the AQI code is red (NNPH, 2024).

The positive Thiel-Sens slopes during summer (BFD, BOS, FRN, RNO, and SAC; Figure 5) are associated with the increasing area burned by wildfires (Figure S30), especially in California (Figure S14). This factor summer concentrations correlates moderately to strongly with the state total area burned (acres) (0.437 – 0.824 for BFD, BOS, FRN, LV, MOD, and SAC, Figure S31). RNO showed no correlation between NV-specific area burned and OC-rich factor concentrations (Figure S31), but it showed an increasing trend, possibly due to regional fires from California and Washington. Wildfire plumes can be transported long distances (Nisantzi et al., 2014).

3.1.10. OC EC rich

PMF identified this factor at COM, VIS, and SAC (Figure S1) sites where either OC rich or EC rich or both OC rich and EC rich were not resolved. Therefore, this factor represents a mixture of OC-rich and EC-rich factors. The proportion of OC and EC varied with each site (Figure S2). For SAC, the contribution of OC and EC to the total factor profile was 28.3% and 24.8%, respectively, and the explained variation of OC was 6.76% and EC was 34.1%, respectively (Figure S1). The OC-rich factor was already apportioned at SAC. The explained variation of either OC or EC were not high enough to assign this factor to “OC rich” or “EC rich”. The contribution of other species to the factor included NO₃ (36.1%), SO₄ (6.57%), and Fe (2.83%), and trace amounts of Cr, Mn, Ni, Ba, Zn, and Br. The explained variation of Fe, Cr, Ni, Mn, Zn, and Ba was greater than 30% (Figure S1). Due to the presence of EC, the EC/OC ratio of 0.879, and elements, including the absence of K, this factor represented vehicle emissions.

For COM, the contributions of the OC and EC were 51.0% and 35.3%, respectively (Figure S1). The explained variation of OC and EC was 73.1 and 85.7%, respectively. The other species present included NO₃ (6.33%), Fe (2.06%), Ca (0.788%), Na⁺ (0.767%), K (0.611%), Si (1.06%) and trace amounts of Cr, Cu, Pb, Mn, Ni, Ba, Br, and P. The explained variation in Ba, Br, Cr, Cu, Fe, Pb, Mn, Ni, P, Na⁺, and Sr was greater than 30% (Figure S1). The high explained variation of trace elements and the presence of K indicates that this factor may represent a mixture of E- rich and OC-rich factors, with a higher contribution from the OC-rich factor.

For VIS, OC and EC contributions to the total factor were 52.1% and 10.6%, respectively (Figure S1). The other significant contributors included NO₃ (27.8%), K (1.50%), NH₄ (6.03%), and Fe (0.512%). The explained variation of OC and EC were 58.8 % and 59.1 %, respectively (Figure S1). The explained variations of K, Zn, Pb, and Br were 52.2%, 62.3%, 50.0%, and 31.1%, respectively. The presence of K suggests that the OC-EC-rich factor was also associated with biomass burning; the presence of Zn and Pb suggest vehicle emissions, with Zn as a marker for tire and road wear (Klöckner et al., 2019). This factor represented a mixture of OC-rich and EC-rich factors, with high contributions from the OC-rich factor.

For COM and VIS, the factor contributions were highest in winter (3.51 – 5.89 μg/m³) and fall (3.16 – 5.29 μg/m³), along with substantial summer (2.45 – 3.59 μg/m³) impacts (Figure 2). For SAC, the contributions were higher during fall (1.21 μg/m³) and winter (1.65 μg/m³) and remained low during summer (0.447 μg/m³) and spring (0.658 μg/m³). For COM and VIS, the net contribution of this factor to total PM_2.5 was 32.6–34.3% during fall and winter, and 23.4–29.4% during spring and summer (Figure S3). For SAC, the contribution to total PM_2.5 was less than 12.1% during all seasons (Figure S3).

The factor showed weekend low concentrations for the COM and SAC (p<0.001) (Figure 3), which indicated the effect of vehicle emissions, especially diesel, which are low on weekends (Murphy et al., 2008).

Over the study period, this factor exhibited a decreasing trend at COM with a negative Thiel-Sen slope of −0.0576 μg/m³ per year (0.05<p<0.1) (Figure 4 and Figure S2) and a significant negative slope for winter (−0.114 μg/m³ per year; p<0.05) (Figure 5). On an annual basis, this factor showed little change over time in SAC although it showed a decrease for summer (−0.0189 μg/m³ per year; p<0.05) (Figure 5). This factor exhibited a minimum around 2010 at the VIS site (Figure 4).

For SAC, this factor most closely represented vehicle emissions, and it showed a strong positive correlation (r=0.719) with VMT (Figure S28). The increasing VMT (Figure S27) over time might explain this factor’s positive slope (an increasing trend with a Thiel-Sen slope of 0.00525 μg/m³ per year, not significant). For VIS, this factor most closely represented biomass burning. It showed a positive slope for summer (Figure 5) and a strong correlation with California’s total acres burned (r = 0.765; Figure S31), indicating the summer concentration increase could be linked to wildfire smoke. For COM, the factor was a mixture of OC rich and EC rich, and it showed a decrease in concentration (Figure 4), significantly for winter (Figure 5). The increasingly stringent vehicle emission regulations (as discussed under EC-rich factor) and wood burning restrictions (discussed under OC-rich factor) are likely related to this decrease.

3.1.11. Cu rich

PMF identified this factor at BFD, VIS, and SAC. This factor was named as “Cu rich” due to the high explained variation of Cu (90.0%, 85.2%, and 95.6% for BFD, SAC, and VIS, respectively) (Figure S1). For BFD, this factor is composed of Cu (5.02%), OC (41.3%), EC (8.87%), NO₃ (11.6%), SO₄ (10.6%), Na⁺ (3.39%), Si (4.03%), Fe (2.74%), Ca (1.43%), NH₄ (9.19%) (Figure S1). For SAC this factor is composed of Cu (5.55%), OC (37.3%), EC (46.5%), Fe (3.42%), NH₄ (4.48%), and traces of Al, NO₃, Si, and K (Figure S1). For VIS this factor had Cu (3.43%), OC (49.1%), EC (11.3%), NO₃ (26.9%), SO₄ (1.42%), K (0.764%), NH₄ (5.37%) and traces of Fe, Si, and Na⁺ (Figure S1). Cu is a marker of vehicle emissions or industrial emissions, like Cu smelters. Hasheminassab et al. (2014) reported a Cu-smelter factor at the Bakersfield site but not at the Sacramento site. This factor showed a positive correlation (r = 0.516; Figure S28) with VMT for SAC, indicating the potential association of this factor with vehicle emissions.

The concentrations of this factor were low with mean concentration for all seasons between 0.0605 – 0.212 μg/m³ for BFD, 0.0619 – 0.154 μg/m³ for SAC, and 0.189 – 0.406 μg/m³ for VIS (Figure 2). For BFD, this factor had higher summer-fall concentrations (0.208 μg/m³) (Figure 2). For SAC, higher winter (0.152 μg/m³) and fall concentrations (0.154 μg/m³) were observed, and for VIS higher summer, fall, and winter (0.368 – 0.406 μg/m³) contributions were seen. This factor exhibited a weekday dependence at BFD (Figure 3), with significantly (p<0.01) lower concentrations during weekends, possibly indicative of industrial emissions (Cu smelters). SAC and VIS did not exhibit weekday dependence for this factor (Figure 3).

The Thiel-Sen analysis on the mean annual concentration yielded a negative slope for VIS (−0.0132 μg/m³ per year) but a positive slope for SAC (0.00162 μg/m³ per year) and BFD (−0.000635 μg/m³ per year) (Figure 4 and Figure S2). Seasonally, SAC had a positive slope for fall and winter, but negative slope for spring and summer (Figure 5). VIS had negative slopes for all seasons. BFD showed a positive slope for fall, but negative slopes for all other seasons (Figure 5).

3.1.12. Cl Zn factor

This factor was rich in Cl and Zn and was identified only at the COM site (Figure S1). Its major contributors were Zn (11.8%), Cl (14.5%), EC (39.3%), OC (7.14%), SO₄ (6.60%), and NO₃ (3.19%) (Figure S1). The factor had a high explained variation for Cl (78.9%) and Zn (76.1%). Zn is a marker for tire and road wear (Klöckner et al., 2019). Chlorine could come from road salting during winter and transportation corridors, including highways (~0.6 km to I76) and railways (~0.8 km) could contribute to this factor.

This factor showed strong winter (mean concentration of 0.365 μg/m³) dominance (Figure 2), which could be associated with air stagnation during this season. The summer concentrations were low (0.0635 μg/m³). Winter dominance also supports the possible contribution of chlorine from road salting. This factor concentrations were weekday dominant (Figure 3), supporting its possible relationship with vehicle and industrial emissions.

The Thiel-Sen analysis of the mean annual concentration yielded a negative slope (−0.00340 μg/m³ per year). For all seasons, the slope were negative, with statistically significant slopes for summer (−0.00289 μg/m³ per year) (Figure 5).

4. Conclusions

This study assessed the long-term trends in PM_2.5 source contributions at 12 monitoring sites across five western US states using PMF and CMB. It identified between five and eight source factors at the sites, and these factors included AN, AS, dust, chloride, OC rich, EC rich, EC Cu rich, Cl Zn rich, and aged sea salt. AN, AS, and dust occurred at all sites. Over the course of this 20-year study, PM_2.5 concentrations declined (−0.379 to −0.0373 μg/m³ per year) at all sites, significantly at all sites except BOS and COM. AN concentrations declined (−0.321 to −0.0378 μg/m³ per year) at all sites, with more pronounced reductions in winter, fall, and spring. AS concentrations also showed significant declines across all sites (−0.105 to −0.0399 μg/m³ per year) and seasons (−0.122 to −0.0222 μg/m³ per year), except for winter at SAC and VIS. The declining AN and AS concentrations were likely driven by reductions in NO_x and SO₂ emissions, respectively, resulting from regulatory policies. The chloride concentrations decreased in winter at all sites, likely due to reduced emissions affecting chloride formation pathways, while summer increases at SAC and MOD may be linked to higher temperatures and sea salt evaporation. Aged sea salt concentrations declined significantly at BFD, FRN, and SAC, with the largest reductions occurring between 2000 and 2005, coinciding with stricter NO_x and SO₂ regulations. However, methodological changes in CSN sodium measurements after 2016 introduced uncertainties in trend interpretation for this factor. EC-rich and EC-Cu-rich factor concentrations, representing vehicle emissions, decreased at several sites (−0.099 to −0.014 μg/m³ per year), although the reductions were less pronounced than those for AN and AS. This suggests that rising VMT partially offset improvements in per-vehicle emissions.

In contrast to the other source factors, dust concentrations increased at most sites (0.000672 to 0.0240 μg/m³ per year), except for VIS and RNO. Factors contributing to this increase include decreasing precipitation, increased aridity, land disturbances, VMT, and wildfire activity. The OC-rich (biomass combustion) factor concentrations decreased significantly during the wintertime, likely due to wood-burning policies. However, summer increases in this source factor, particularly in California, were linked to increased wildfire activity. This study highlights the effectiveness of emission-control policies in reducing key PM_2.5 contributors, while also identifying emerging challenges, such as increasing dust and wildfire-related emissions. Addressing these evolving sources will be critical for future air quality management efforts in the western U.S.

Highlights.

1. PM_2.5 concentrations declined significantly at most sites over the past two decades.

2. Regulations on mobile, point, and area sources reduced AN, AS, EC rich, and EC Cu rich concentrations.

3. Winter OC rich concentrations decreased due to biomass burning policies.

4. Dust levels increased, likely due to increased regional aridity and VMT.

5. Data Availability

All raw data were sourced from federally and state-maintained web databases. Chemical speciation data were obtained from the Air Quality System database, while annual vehicle miles travel were collected from each state’s Department of Transportation. Meteorological measurements were retrieved from the National Centers for Environmental Information, part of the National Oceanic and Atmospheric Administration. Data on wildfire areas burned per state were acquired from the National Interagency Coordination Center. Statewide power plant emissions, categorized by fuel type, were obtained from the EPA’s eGRID database.