Spatial and Temporal Characterization of PM_2.5 Mass Concentrations in California, 1980–2007

ABSTRACT

Systematic measurement of fine particulate matter (aerodynamic diameter less than 2.5 μm [PM_2.5]) mass concentrations began nationally with implementation of the Federal Reference Method (FRM) network in 1998 and 1999. In California, additional monitoring of fine particulate matter (PM) occurred via a dichotomous sampler network and several special studies carried out between 1982 and 2002. The authors evaluate the comparability of FRM and non-FRM measurements of PM_2.5 mass concentrations and establish conversion factors to standardize fine mass measurements from different methods to FRM-equivalent concentrations. The authors also identify measurements of PM_2.5 mass concentrations that do not agree with FRM or other independent PM_2.5 mass measurements. The authors show that PM_2.5 mass can be reconstructed to a high degree of accuracy (r ² > 0.9; mean absolute error ∼2 μg m⁻³) from PM with an aerodynamic diameter ≤10 μm (PM₁₀) mass and species concentrations when site-specific and season-specific conversion factors are used and a statewide record of fine PM mass concentrations by combining the FRM PM_2.5 measurements, non-FRM PM_2.5 measurements, and reconstructions of PM_2.5 mass concentrations. Trends and spatial variations are evaluated using the integrated data. The rates of change of annual fine PM mass were negative (downward trends) at all 22 urban and 6 nonurban (Interagency Monitoring of Protected Visual Environments [IMPROVE]) monitoring locations having at least 15 yr of data during the period 1980–2007. The trends at the IMPROVE sites ranged from -0.05 to -0.25 μg m⁻³ yr⁻¹ (median -0.11 μg m⁻³ yr⁻¹), whereas urban-site trends ranged from -0.13 to -1.29 μg m⁻³ yr⁻¹ (median -0.59 μg m⁻³ yr⁻¹). The urban concentrations declined by a factor of 2 over the period of record, and these decreases were qualitatively consistent with changes in emissions of primary PM_2.5 and gas-phase precursors of secondary PM. Mean PM_2.5 mass concentrations ranged from 3.3 to 7.4 μg m⁻³ at IMPROVE sites and from 9.3 to 37.1 μg m⁻³ at urban sites.

IMPLICATIONS

Mean measured and reconstructed fine particulate matter mass concentrations declined by about a factor of 2 in California over the period 1980 to 2007 and varied by about a factor of 4 among air basins. The integrated data record is of interest for epidemiological studies and for assessments of emission control programs.

INTRODUCTION

Comparison of emission trends with changes in ambient concentrations of air pollutants provides information on source-receptor relationships and is often used to evaluate the effectiveness of emission control programs. The detectability of emission reductions depends on their magnitude relative to weather-driven variations in ambient pollutant concentrations and on the quality and length of record of the monitoring data. A long record with greater spatial coverage is of value both for detecting trends and for assessing temporal and spatial variations in exposures to air pollutants, a crucial step in developing a quantitative understanding of the effects of specific pollutants on particular health endpoints.

The U.S. Environmental Protection Agency's (EPA) recently released final integrated science assessment for airborne particulate matter (PM)1 and earlier reports2–4 have associated PM with increased rates of hospital admissions and mortality resulting from respiratory and cardiovascular diseases. The evidence linking fine particles with morbidity and mortality led the U.S. EPA to establish an ambient air quality standard for fine PM (aerodynamic diameter less than 2.5 μm [PM_2.5]) in 19975 and to fund a national network of PM_2.5 Federal Reference Method (FRM) samplers for determining compliance.

Individual exposures depend on many factors, including in particular the amount of time spent indoors, the time spent in areas having high emissions of air pollutants (e.g., in highway traffic,6 adjacent to heavily traveled roads, or in workplaces with occupational exposures), a person's daily movements, and the indoor and outdoor ambient concentrations occurring each day. Ambient concentrations typically constitute a major fraction of the particles to which people are exposed, especially for PM_2.5.7 However, indoor concentrations may be higher or lower than outdoor concentrations; for example, ammonium nitrate may be present at high concentrations outdoors during winter but at lower concentrations in the indoor environment due to dissociation and deposition.8 Models have been developed to estimate personal exposures from ambient concentration data.9–12 Because of the complexity of quantifying actual personal exposures, epidemiological studies of air pollution may instead rely on ambient concentration data obtained at one or more monitoring sites as a measure of exposure.13,14 Monitoring data coupled with land use information have been used for spatial interpolation.15,16 Different procedures for optimizing network design exist,17,18 and various approaches to combining monitoring data with modeling and additional information have been described.19–24

Although FRM measurements of PM_2.5 mass concentrations (fine mass) are available (beginning in 1998 or 1999), studies of fine PM exposure may require longer monitoring records, denser spatial coverage, or better understanding of actual exposures than is provided by the FRM network. For times and locations having no fine PM data, investigators have reconstructed PM_2.5 mass concentrations from visibility, light-scattering, or other measurements. For example, previous studies have reconstructed PM_2.5 mass concentrations from airport visibility measurements in California for the years 1966–198625 and for 12 cities nationwide for the years 1978–1981.26 All reconstructions are subject to uncertainty; even actual PM_2.5 mass measurements may exhibit either systematic or random differences from FRM measurements or both.27,28 Consequently, there is a need for long-term data that have been reviewed for compatibility and adjusted to a common reference. Here, the authors develop a unified California PM_2.5 data set for the period 1980–2007 using FRM measurements, other measurements of PM_2.5 mass, and reconstructions based on PM with an aerodynamic diameter ≤10 μm (PM₁₀) mass and related data. The authors also evaluate the comparability of the different measurement methods, document data quality, consolidate the data into a fine PM database, and assess temporal changes in fine PM concentrations in California. Although the spatial coverage of the database is limited to California, the approaches described here are of potential interest for reconstruction of historical PM_2.5 data sets in other areas.

DATA

The EPA has identified FRM methods29 and documented FRM data quality.30,31 The FRM measurement program commenced in 1998 at six sites in the Sacramento Valley and in 1999 at additional sites throughout California. FRM data are available from 94 California monitoring locations, after combination of data from sites that were relocated during the period 1999 through 2007 (the distances between the initial and replacement sites ranged from 64 m to 2.9 km).

Before implementation of the FRM network, fine PM (PM_2.5) mass concentration measurements were made in California by the California Air Resources Board (CARB) with a routine monitoring program (dichotomous [dichot]) and by a variety of special studies (Table 1). The dichot sampler uses a low-volume PM₁₀ inlet followed by a virtual impactor that separates the particles into the fine (PM_2.5) and coarse (PM_10–2.5) fractions. The dichot network was operated in accordance with CARB's Air Monitoring Quality Assurance Manual. The limit of detection for PM_2.5 dichot samplers in the California network was 2 μg m⁻³. PM_2.5 concentrations obtained from the dichot sampler are reported as corrected to standard temperature and pressure (STP). PM_2.5 measurements from the CARB dichot network commenced in 1981 and continued until 2001, but systematic sampling, data validation, and electronic archiving of data began in 1988. Dichot data from 1981 to 1987 have not previously been reported and are used here with the understanding that measurement uncertainties are greater than for the 1988–2001 data. Up to 19 dichot sites operated each year, with the largest number of sites (7–10) in the San Joaquin Valley. Thirty-one sites reported dichot data from 1 or more yr between 1988 and 2001; measurements were made at 6–10 sites per year (14 sites total) in 1981–1987. PM_2.5 measurements from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network are available for 24 sites in California. Nine of these sites have PM_2.5 data for 10 yr or more. The sampling network of the California Acid Deposition Monitoring Program (CADMP) was designed to determine spatial and temporal patterns of acidic pollutant concentrations. Originally, the CADMP sampler had separate units for collection of two 12-hr samples of PM_2.5, PM₁₀, and acidic gases. In September 1995, to better complement the routine PM monitoring network, the sample collection was changed to one 24-hr sample beginning at midnight, and the sampling was reduced to the PM_2.5 size fraction only. The Two-Week Sampler (TWS) network was deployed to provide information for an on-going study of the chronic respiratory effects in children from long-term exposure to air pollution in southern California; the TWS sampler provides 2-week integrated measurements of vapor-phase acids and PM_2.5 mass and inorganic ions. Operational characteristics of different sampling devices have been described and compared.27

Table 1. Summary of PM_2.5 mass concentration measurements made in California

Network	Spatial Coverage	Time Period	Sample Frequency ^a	No. Sites	References
CalTech	SoCAB	1982; 1986; 1993	1–6	5–11	32
IMPROVE	Statewide; national parks	1988–2007b	2 per week	24	33, 34
VAQS	San Joaquin Valley	1988–1989	1–6	6	35
CADMP	Statewide	1988–1995	1–6	10	36, 37
		1995–1998	1–6	5
TWS	SoCAB	1994–2002	2 per month	15	38, 39
IMS95	Central California	12/95–1/96	1–1	10	40, 41
PTEP	SoCAB	1995–1996	1–1; 1–3; 1–6	6	42, 43
Dichotomous	Statewide	1981–1987	1–6	14 ^c	CARB data
		1988–2001	1–6	31	27, 28
FRM	Statewide	1999–2007 ^b	1–1; 1–3; 1–6	94	27, 28
Notes: SoCAB = South Coast Air Basin; IMPROVE = Interagency Monitoring of Protected Visual Environments; VAQS = Valley Air Quality Study; CADMP = California Acid Deposition Monitoring Program; TWS = Two-Week Sampler; IMS95 = Integrated Monitoring Study of 1995; PTEP = PM Enhancement Program;
^a1–1 = every day; 1–3 = every 3rd day; 1–6 = every 6th day;
^bOngoing; data available at the time of this study;
^c6–10 sites per year, including sites with partial annual records.

In addition to the measurements listed above, continuous measurements of PM_2.5 mass concentrations were made during recent years (1999 or later) at 29 California monitoring sites using either tapered element oscillating microbalance (TEOM) or beta attenuation meter (BAM) instrumentation. EPA's published guidance for relating continuous measurements to FRM data indicates that particle measurements from non-FRM monitors may be used for the purpose of reporting a real-time air quality index if a linear relationship between the continuous and FRM measurements can be established by statistical linear regression, where the squared correlation coefficient (r ², coefficient of determination) is defined as the parameter of interest for determining whether the model relating FRM with continuous PM_2.5 measurements is acceptable.44 EPA guidance further classifies uncorrected or statistically corrected hourly continuous data that are seasonally or yearly within 10% bias and have a correlation (r) of at least 0.9 (squared correlation of at least 0.81) with the FRM as “acceptable PM_2.5 air quality index and speciation mass” (parameter code 88502).45 Otherwise, continuous data are reported as “PM_2.5 raw data” (parametercode 88501).45 BAM data correlated well (r = 0.98) with FRM data at one California location over a 2-yr period,27 but no other BAM sites had colocated FRM measurements, precluding direct comparison and conversion of BAM data to FRM-equivalent concentrations at California locations. The TEOM and FRM data were not highly correlated (r ² < 0.3) at the four sites with colocated measurements, and because colocated data were lacking at other sites, no continuous data were used in the equivalency analysis. TEOM instruments may lose semivolatile material, such as ammonium nitrate, and fail to replicate gravimetric measurements.46 High correlations (r ² > 0.9) between FRM measurements and either BAM or filter dynamics measurement system TEOM instruments have been observed at some locations,47,48 but colocated BAM and TEOM devices at the Fresno, California, supersite yielded values differing by a factor of 2 at times.49 Despite these differences, the continuous data are useful for describing diurnal PM variations.

Related data are available from long-term, routine monitoring operations in California. CARB collected filter-based 24-hr PM₁₀ mass samples once every 6 days at up to 85 sites from 1984 to 2002.27 Concentrations of major ions, primarily sulfate and nitrate, were measured on a subset of samples; smaller numbers of measurements of ammonium, chloride, and total carbon concentrations were also made. Measurements of coefficient of haze (CoH) were made at 88 sites in California during part or all of the period from 1980 to 2007. CoH is a measure of light absorption, which largely depends on the amounts of black carbon (BC) in the air. Measurements of light scattering, which may exhibit a functional relationship with fine PM levels,50 were made at 25 sites in California during parts of the period from 1980 to 2007.

METHODS

An integrated data set was developed using FRM PM_2.5 mass measurements for the period 1999–2007, supplemented by other measurements of PM_2.5 mass made from 1982 to 2002. Conversion factors were developed to adjust the non-FRM measurements. Reconstructions of PM_2.5 mass were then developed from PM₁₀ mass and species concentrations for sites and time periods lacking PM_2.5 mass measurements. The conversion and reconstruction methods are described in the following sections.

Fine Mass Concentrations

In this section, the authors evaluate the comparability of PM_2.5 measurement methods, identify biases, and standardize all data to a common reference point using linear regression to convert fine mass measurements from other methods to an equivalent FRM value. Standardized measurements are referred to as “Federal Reference Method equivalent values,” but this terminology does not imply that the EPA has designated the measurement devices as FRM-equivalent methods. Fine mass concentration measurements from non-FRM samplers may exhibit either additive or multiplicative biases relative to FRM data, but FRM fine mass concentrations are predictable from non-FRM measurements if the two sets of measurements are correlated.27,28 The coefficients of determination (r ², the squared correlation coefficients) of PM_2.5 mass measuredby colocated samplers from different networks ranged from 0.82 to 0.98. The highest coefficients of determination were those between FRM and dichot data (r ² = 0.96) and between FRM and IMPROVE data (r ² = 0.98). The regression of FRM measurements against all colocated 1998–2001 dichotomous sampler data was

(1)

As indicated by eq 1, the dichot fine mass concentrations averaged ∼15% lower than colocated FRM measurements. Equation 1 fit the data from all sites (as shown by r ² =0.96); site-specific FRM-dichot conversion factors varied somewhat (n = 10 sites with 35–210 days per site; intercepts =−1.3–0.8; slopes = 0.9–1.3; r ² = 0.88–0.99). For the 10 sites having more than 30 days of colocated FRM-dichot data, site-specific conversion factors improve the agreement between FRM and dichotomous sampler measurements and also correct for the difference between FRM and dichot measurement reporting conventions (ambient temperature and pressure for FRM and STP for dichot). Without daily temperature and pressure data, it was not possible to convert PM concentrations from STP to ambient conditions. However, systematic differences between values reported at STP and those reported at ambient conditions are accommodated within the site-specific regression coefficients. For 21 sites lacking a site-specific conversion factor (because the FRM and dichot records did not overlap), the authors accounted for elevation when converting dichot data to FRM-equivalent concentrations with use of the following equation:

(2)

where z is height above mean sea level, P _z is atmospheric pressure at height z (obtained from site elevation and the hydrostatic condition), and P ₀ is mean atmospheric pressure at sea level.

The PM_2.5 mass concentration measurements from the CalTech, Valley Air Quality Study (VAQS), CADMP, and PM Enhancement Program (PTEP) networks (Table 1) can be compared with each other and with data from dichotomous samplers, but they have no overlap with FRM data. PM_2.5 from each of these four special studies exceeded the dichot values by ∼10–30% on average, but the correlations were high (r ² = 0.84–0.94) (eq 3). These internetwork differences are qualitatively consistent with the dichot-FRM comparisons (eqs 1 and 2), which indicated that dichot values were lower than FRM concentrations. The CADMP data at one location (Bakersfield) failed to replicate colocated dichotomous sampler measurements, so the Bakersfield CADMP data were excluded from further analyses.

(3)

The correlations between the 1981–1987 dichot data and the CalTech data (r ² = 0.55, 0.75, and 0.85 at three sites with colocated instruments) were lower than those listed in eq 3 (which are based on 1993 data), possibly reflecting less rigorous quality assurance for the 1981–1987 dichot data compared with the 1988–2001 dichot samples. Therefore, the 1993 colocated measurements were used as the basis for CalTech-dichot conversion factors.

To convert special-study data to FRM-equivalent values, the authors first converted to the dichot equivalent (eq 3), followed by conversion from dichot equivalent to FRM equivalent (eq 2). The no-intercept regressions in eq 3 were used because they fit the data (root mean square errors were 0.3–8% greater for the no-intercept regressions than for regressions with intercepts), they permitted a more straightforward propagation of uncertainties, the PTEP and VAQS intercepts were not statistically significant, and graphical analyses indicated that the positive intercept terms did not represent systematic offsets. Because measurements from special studies were reported at STP, the authors completed the conversion of the special-study data to their FRM-equivalent values using eq 2.

The data from the TWS obviously could not be paired with daily FRM measurements. Instead, the authors obtained monthly averages of the 2-week sampler data from CARB and compared those with monthly averages of the FRM data. The FRM fine mass concentrations exceeded the TWS values as discussed previously,51 but the correlations were high (FRM = 2.36 + 1.05 · TWS, r ² = 0.87, or FRM = 1.18 · TWS).

Reconstruction of Fine Mass Concentrations

Reported mean ratios of PM_2.5 mass to PM₁₀ mass vary from ∼0.2 to 0.8, depending on continent, aridity, and proximity to marine influence.52 The average ratio of PM_2.5 mass to PM₁₀ mass in the data varied among sites and seasons. The mean PM_2.5/PM₁₀ ratio tended to be greatest during the months of November through January and lowest during the months of May through September. This seasonal variation results from higher concentrations of coarse PM (PM_10–2.5) during the drier months and from higher concentrations of ammonium nitrate during cooler months. Considerable intersite variation occurred in the mean PM_2.5/PM₁₀ ratio. FRM PM_2.5 mass concentrations were regressed against PM₁₀ mass by site and month, and the regression coefficients were then used to calculate an estimated PM_2.5 mass from PM₁₀ mass. The overall correlation was high (r ² = 0.8), and 80% of the estimates were within ±5 μg m⁻³ (94% within ±10 μg m⁻³) of the actual measurements. However, some predictions exhibited larger deviations from the measured PM_2.5 mass, and the predictions tended to overestimate low concentrations and underestimate high values. Therefore, the authors restricted use of PM₁₀ mass as a predictor of PM_2.5 mass to sites where the coefficients of determination were r ² ≥ 0.9 (for comparison, see eqs 1 and 3 for the coefficients of determination of PM_2.5 mass measured by different networks). A total of 85 sites had PM₁₀ mass measurements available, but only 9 met the r ² ≥ 0.9 criterion. Predictions of PM_2.5 mass from total suspended particulate (TSP) (a measure of PM with no size selection) mass were weaker (lower r ²) than predictions from PM₁₀ mass, so no reconstructions were attempted from TSP mass alone.

The principal constituents of PM_2.5 mass in California are organic and black (elemental) carbon, sulfate, ammonium, and nitrate.53 Typically, these PM components are present at greater concentrations in PM_2.5 than in PM_10–2.5,35,36,41–43 potentially providing a better method for reconstructing fine mass concentrations using sulfate, nitrate, and carbon from PM₁₀ samples in addition to PM₁₀ mass, as described in the remainder of this section.

The PM₁₀ species measurements that were reported most often are sulfate and nitrate. In some cases, measurements of PM₁₀ ammonium or total carbon were reported (PM₁₀ total carbon is the mass of carbon with no distinction between organic and elemental carbon [EC] and no adjustment for the mass of elements such as hydrogen and oxygen that are normally associated with organic carbon). Because the PM₁₀ measurements of total carbon are limited to a few sites and years, whereas the CoH database is extensive, correlations between the PM₁₀ carbon data and CoH were established. A linear relationship was previously reported between monthly average CoH and black (elemental) carbon at Fresno, California, described by BC = 5.13 · CoH + 0.57, r ² = 0.96.54 For daily data, PM carbon as a nonlinear function of CoH based on consideration of the CoH measurement methodology was estimated as

(4)

The first factor (3.4) is an average ratio of total carbon to EC in central California PM samples that is consistent with reported literature values: for sites in the San Joaquin Valley, ratios of fine PM mean total carbon to mean EC range from 2.3:1 to 4:135 and from 3.1 to 3.7.41 For urban Los Angeles sites, reported ratios range from 2.5:1 to 2.7: 1 42 and from 6:1 to 8:132 (differences between the latter two studies are due in part to differences in measurement techniques). Because CoH is a measurement of light absorption, the second factor (1/10) is needed to convert from absorption to concentration units; reported values of BC light ∼5 to 20 m² g⁻¹, with most researchers using 10 m² g⁻¹.33 The factor 1/10 is the inverse of 10 m² g⁻¹. The remaining terms convert from the reported CoH units (soiling index) to inverse megameters, according to CARB guidance. Calculated daily carbon concentrations were set to “missing” if the calculated values were physically impossible, as indicated by occurrence of the following condition:

(5)

If ammonium (NH₄ ⁺) was not measured, it was conservatively estimated that ammonium concentrations would be at least half the levels expected if sulfate and nitrate were present as ammonium sulfate and ammonium nitrate, respectively:

(6)

Because the purpose of eqs 5 and 6 was to identify physically unrealistic, gross outliers, these equations led to exclusion of carbon estimates from CoH only when such estimates were obviously and objectively erroneous. Comparisons of measured daily PM₁₀ total carbon with total carbon calculated according to eq 4 showed good agreement at two locations with colocated CoH and PM₁₀ total carbon (PM₁₀ C =-0.54 + 0.85 CoH C, r ² = 0.73 at Bakersfield; PM₁₀ C =-0.70 + 0.94 CoH C, r ² = 0.88 at Stockton). Equation 4 yields estimates of EC that are ∼30% greater than the linear formula for Fresno data54 for CoH values in the range of 0.2 to 0.8 and closer agreement for CoH less than 0.2 or greater than 0.8. Uncertainties in the carbon estimates are not critical for use of the data in bounding calculations.

PM₁₀ mass was used as an upper bound for PM_2.5 mass. A small number (∼0.5%) of the PM₁₀ mass concentrations were less than the PM_2.5 mass concentrations, because of limitations in measurement precision and accuracy. The available data do not include a measurement that is a theoretically rigorous lower bound for PM_2.5 mass. However, a practical lower bound for fine mass concentration is the sum of sulfate, nitrate, and carbon from PM₁₀ samples, because those species do not represent all the PM_2.5 mass and because the majority of the mass of sulfate, nitrate, and carbon exists in the fine size range as explained previously. The monthly averages of these sums were always lower than the monthly average PM_2.5 mass (this was usually but not always the case for individual days). Two other potential lower bounds were examined: (1) the sum of PM₁₀ sulfate and nitrate (excluding carbon) and (2) TSP sulfate, TSP nitrate, and carbon from CoH. The use of TSP mass and species is discussed later.

The daily-average PM_2.5 mass concentrations (X) were estimated from the measured upper bound (U) and lower bound (L) concentrations as

(7)

where f is between 0 and 1. The value of f is typically not 0.5; that is, splitting the difference between the upper and lower bounds does not give a correct answer, because f varies from site to site and from month to month. f was estimated from the daily-average measurements by site and month using a second equation that is algebraically identical to eq 7:

(8)

Equation 8 is a no-intercept regression of the quantity U − X (which is PM₁₀ mass minus PM_2.5 mass) against U − L (which is PM₁₀ mass minus the sum of the PM₁₀ species concentrations). The slope is f. The authors obtained values of f for each site and month (13 sites, ∼20 days per site-month), calculated estimates of PM_2.5 mass concentrations, and compared them with the FRM measurements (Figure 1). The predicted and measured values are highly correlated (r ² = 0.94) and are close to the 1:1 line.

Figure 1. Measured PM_2.5 mass versus predicted PM_2.5 mass, by season. The data are daily-average concentrations from 12 monitoring sites, 1999–2006. The predictions were computed as described in the text using PM₁₀ mass concentration as an upper bound and the sum of major species concentrations as a lower bound.

It was not possible to directly compare measured FRM PM_2.5 and reconstructions from TSP measurements because their periods of record do not overlap. Instead, the authors compared TSP with non-FRM measurements of PM_2.5 mass (dichot or CADMP measurements) when overlapping data were available (1988–1991). The TSP mass concentrations were used as an upper bound. TSP sulfate and nitrate correlated with PM₁₀ sulfate and nitrate concentrations at r ² = 0.77 and 0.84, respectively, but the TSP sulfate and nitrate concentrations were higher than the PM₁₀ sulfate and nitrate concentrations by ∼2 and ∼4 μg m⁻³, respectively. Therefore, the lower bound was adjusted downward to be one-half the sum of TSP species concentrations. Month-specific estimates of f (eq 8) were obtained by combining data from five urban sites (Stockton, Fremont, San Jose, Bakersfield, and Fresno), because the number of overlapping TSP and PM_2.5 measurements was too small to obtain month-specific values for the individual sites. The predicted PM_2.5 concentrations from TSP bounding compared well with colocated PM_2.5 mass measurements (predicted = 3.59 + 0.876  × measured, r ² =0.89 for combined sites; r ² = 0.85–0.90 for individual sites). However, when the Fresno and Bakersfield predictions were compared with colocated dichot data from 1981 to 1987 (93 and 161 pairs, respectively), ∼5–10 paired comparisons at each site seemed to be suspect, excluding apparent outliers, r ² = 0.70 at Bakersfield and 0.87 at Fresno with the estimates from TSP mass and species averaging ∼8 and 3% higher than the dichot concentrations (all concentrations converted to FRM equivalent). The TSP bounding estimates for a given day could be biased if the actual value of f (eq 8) on that day is substantially different from the month-specific mean. Because validation information for the 1981–1987 dichot data is lacking, the conclusion is that individual daily reconstructions from TSP and daily dichot data from 1981 to 1987 are both subject to greater uncertainty than are more recent data, but the 1981–1987 monthly and annual mean concentrations are robust enough to be useful. The reconstructions from TSP were limited to the five urban sites identified.

Fine Mass Concentration and Light Scattering

Light scattering measurements provide another potential predictor of fine PM concentrations,50 and an extensive set of CARB measurements of light-scattering is available, especially for locations in the Sacramento Valley. Light scattering is measured by nephelometers, which are not size selective, and coarse particles also contribute to light scattering, albeit much less than fine particles. An additional confounding factor is the contribution of fog or cloud droplets, which, when present, tend to cause very high light scattering. The CARB nephelometers are heated, which is expected to minimize the fog contribution. Light scattering due to particles (b _sp) was calculated as

(9)

Equation 9 converts the light-scattering data in the CARB database from 10⁻⁴ to 10⁻⁶ m (inverse megameters [Mm⁻¹]), subtracting 10 to approximately remove Rayleigh scattering (light scattering by molecules).55 Many very large values (b _sp > 1000) occurred in the data, indicating that fog or cloud droplets probably were affecting the nephelometer values despite sample heating. Samples having b _sp > 800 Mm⁻¹ were excluded. For comparison, open (unheated) nephelometer measurements made in the San Joaquin Valley during a special study were less than 500 Mm⁻¹ whenever the relative humidity (RH) was less than 90% and greater than 500 Mm⁻¹ on nearly all occasions with RH exceeding 90%.56 Extinction efficiency (light extinction per unit concentration of a chemical component) varies with RH and chemical composition but is generally in the range of 3 to 20 m2 g⁻¹ for sulfate, nitrate, and organic carbon,56 so that 800 Mm⁻¹ corresponds to ∼100–150 μg m⁻³ fine mass concentration for RH<90%. For comparison, the maximum recorded 24-hr FRM fine mass concentrations in the database were 87.8 μg m⁻³ at Los Angeles–North Main, 101 μg m⁻³ at Stockton, and 154 μg m⁻³ at Bakersfield.

Measured fine mass concentrations were regressed against b _sp (for b _sp<800 Mm⁻¹) and the regression coefficients were used to predict fine mass from b _sp. Then outliers, which were defined as points for which predictions of fine mass exceeded measured levels of PM₁₀ or TSP mass concentrations, were identified. The suspect outliers were excluded and regressions of fine mass against the b _sp measurements were repeated.

The nephelometer data correlated well (r ² > 0.8) with the fine PM measurements from the dichotomous samplers for the years before 1995 but were poorly correlated with both dichot and FRM fine PM mass concentrations from 1995 to 2002. The authors were unable to determine the cause of the difference, but CARB staff indicated that the nephelometers were not routinely calibrated for compliance purposes; they were instead operated to provide qualitative information relevant to permitting or prohibiting agricultural burn activities on a daily basis. Although site-specific regressions in the later years exhibited better agreement than did regressions with all sites included, one location showed an unexplained shift in the regression line. The authors attempted to refine the predictions of fine mass from b _sp by incorporating measurements of maximum and daily-average RH and temperature, as well as precipitation. No obvious improvements were obtained in the coefficients of determination. All routine CARB nephelometer data were excluded from further use because of the unresolved discrepancies.

RESULTS

Long-Term Monitoring Record

The authors developed a data set with the best available daily-average PM_2.5 mass concentrations for each site and day. If multiple observations existed for a site and day, the selection priority was (1) FRM, (2) dichot, (3) other PM_2.5 mass measurements, and (4) reconstructions of PM_2.5. For data from 1980 to 1987, the second and third priorities were reversed because quality assurance documentation for the dichot data before 1988 were lacking. The daily data set comprises 135,852 site-days, of which 127,408 site-days (93.8%) are PM_2.5 mass measurements and 8444 site-days are PM_2.5 reconstructions. The reconstructions include 1802 site-day estimates based on TSP mass and species concentrations made at five monitoring sites (Bakersfield, Fremont, Fresno, Stockton, and San Jose) during the years 1980 through 1987. For southern California sites, all values before 1988 are actual PM_2.5 mass measurements.

Of the 94 FRM sites, 38 had data from years before 1999 (either measurements of PM_2.5 mass or measurements that were used to reconstruct PM_2.5 mass concentrations according to the methods discussed above) and had at least 10 full years of measurements. In addition, 9 IMPROVE sites had PM_2.5 mass measurements covering at least 10 yr. These 47 sites are depicted in Figure 2. Figure 3 shows the number of nonduplicative observations of FRM data, non-FRM measurements of PM_2.5 mass, and reconstructions of PM_2.5 mass for a subset of 28 sites having at least 15 yr of data each.

Figure 2. Monitoring sites having at least 10 years of data, including PM_2.5 mass measurements and reconstructions.

Figure 3. Number of nonduplicative daily-average PM_2.5 concentrations at 28 long-term monitoring sites by measurement type.

For each site, regardless of the length of monitoring record, a monthly average was determined from all days in a month having valid data. Most of the samples collected before the implementation of the FRM network were obtained once every 6 days, typically generating ∼5 samples per month. The authors do not report a monthly average for site-months having fewer than 3 samples or for a sampling period that spanned fewer than 12 days in a month. The monthly-average TWS concentrations were used for site-months lacking daily data and are included in the final data set as an alternate value otherwise.

Annual averages were computed from (1) monthly averages (requiring at least 9 months per years) and (2) daily concentrations (requiring at least 45 sampling days and a sampling span of at least 300 days). The completeness criteria were intended to ensure representation of seasonal variations within annual averages. The two approaches yielded comparable annual averages, but the second invalidated more data. The 300-day criterion was more effective than construction of annual averages from four quarters, each having a minimum number of days (e.g., 11), because the data record included nonstandard sampling schedules (e.g., quarters with 14 daily samples occurring entirely within a single 2-week period). All data are available from CARB.57

Uncertainties

An uncertainty estimate was obtained for each monthly average concentration. The principal sources of uncertainty are (1) sampling, which includes accuracy, precision, variability, and number of days sampled (e.g., during a month) and (2) conversion uncertainty, determined from the regressions of measured FRM fine mass against other measurements. The conversion uncertainty was quantified for each predicted daily PM concentration by the prediction S ², which is the uncertainty of a predicted daily-average FRM mass (y, not measured) given knowledge of the value of another daily measurement (x, the predictor).58

(10)

where other terms are defined as

(11)

Equation 10 is the standard statistical formula for the uncertainty of an individual predicted value. If the term 1 were excluded from eq 10, the result would be the standard formula for the confidence envelope of a line, which expresses the uncertainty of the mean rather than the uncertainty of an individual data point.

For each monthly average, the estimated uncertainty is the sum of the variances associated with sampling and prediction:

(12)

Here, SD refers to the standard deviation of the daily-average PM values, and n is the number of daily averages in a month, which gives the conventional standard error of the mean.58 The standard error of the mean does not account for prediction uncertainty, so the second variance term must be included also. The prediction variances of the individual days in a month (eq 10) are averaged (mean prediction S ²) to give the estimated prediction variance of the monthly average. The reasonableness of the uncertainty estimates for the monthly averages was evaluated by comparing them with prediction errors when possible. For the period 1999 through 2007, it was possible to calculate the differences between monthly averages determined from FRM monitors and averages determined from other measurements or reconstructions. If the uncertainties calculated from eq 12 represent true uncertainty reasonably well, ∼95% of the prediction errors should be within the two-sigma uncertainty limits. This result held. The authors also examined the prediction errors with respect to season and year to check for systematic biases. The median prediction errors were ∼1–3 μg m⁻³ during all months, indicating that no seasonal bias exists. However, the range of errors was larger during November through February than in other months, consistent with a greater range of fine PM concentrations during winter months than at other times.

Temporal Trends in Fine Mass Concentrations

The rates of change of annual-average fine PM mass were computed using linear regressions for the 28 sites presented in Figure 3 that had at least 15 yr of data during the period 1980–2007. Regression slopes were negative (downward trends) at all these sites (Table 2). Downward trends were greater at the 22 urban locations (decreasing at 0.13–1.29 μg m⁻³ yr⁻¹, median −0.59 μg m⁻³ yr⁻¹) than at the 6 IMPROVE sites (decreasing at 0.05–0.25 μg m⁻³ yr⁻¹, median −0.11 μg m⁻³ yr⁻¹), possibly a reflection on the higher concentrations at the urban sites or on differences in source influence (e.g., proportions of primary and secondary material). The rates of PM decrease tended to be similar at sites within each air basin and to differ more among basins. Similar results were obtained when trends were analyzed using monthly average concentrations. On a statewide basis, mean urban PM_2.5 mass concentrations trended downward by a factor of 2 (Figure 4) despite increasing population and vehicle miles traveled.

Table 2. Trends in mean annual PM_2.5 mass concentrations

Basin	EPA Code No. ^a	Name	Network	No. Yr	Mean PM2.5 (μg m^−3)	Slope (μg m^−3 yr^−1)	SE ^b
Mojave Desert	60719010	San Gorgonio	IMPROVE	17	7.4	−0.25	0.04
Mountain Counties	60430003	Yosemite NP	IMPROVE	17	4.9	−0.06	0.04
North Central Coast	60690003	Pinnacles	IMPROVE	16	5.6	−0.13	0.02
North Coast	60150002	Redwood NP	IMPROVE	17	4.3	−0.12	0.03
Sacramento Valley	60070002	Chico-Manzanita Avenue	CARB	16	15.5	−0.43	0.07
Sacramento Valley	60610006	Roseville-N Sunrise Blvd	CARB	15	12.2	−0.41	0.07
Sacramento Valley	60670006	Sacramento-Del Paso Manor	CARB	17	15.7	−0.32	0.17
Sacramento Valley	60670010	Sacramento-T Street	CARB	18	14.8	−0.46	0.08
Sacramento Valley	60893003	Lassen Volcanic NP	IMPROVE	17	3.3	−0.05	0.03
Salton Sea	60250005	Calexico-Grant/Ethel	CARB	15	15.6	−0.52	0.08
Salton Sea	60251003	El Centro-9th Street	CARB	17	11.9	−0.42	0.06
San Francisco Bay	60011001	Fremont-Chapel Way	CARB	26	16.0	−0.47	0.06
San Francisco Bay	60130002	Concord-2975 Treat Blvd	CARB	19	14.6	−0.58	0.09
San Francisco Bay	60410002	Point Reyes	IMPROVE	18	6.9	−0.10	0.03
San Francisco Bay	60750005	San Francisco-Arkansas St	CARB	19	15.1	−0.60	0.07
San Francisco Bay	60811001	Redwood City	CARB	19	13.5	−0.59	0.07
San Francisco Bay	60850005	San Jose-4th/Jackson	CARB	26	19.3	−0.68	0.11
San Joaquin Valley	60190008	Fresno-1st/Olive	CARB	23	24.5	−0.63	0.08
San Joaquin Valley	60290014	Bakersfield-5558 California	CARB	28	28.2	−0.93	0.12
San Joaquin Valley	60310004	Corcoran-Van Dorsten/Patterson	CARB	16	20.2	−0.13	0.18
San Joaquin Valley	60771002	Stockton-Hazelton Street	CARB	25	19.4	−0.61	0.08
San Joaquin Valley	60990005	Modesto-Oakdale/I/14th	CARB	19	19.0	−0.50	0.11
San Joaquin Valley	61072002	Visalia-N Church Street	CARB	21	23.8	−0.66	0.14
South Central Coast	60792006	San Luis Obispo-Marsh/Higuera	CARB	19	9.3	−0.36	0.04
South Coast	60370002	Azusa	CARB	21	24.7	−0.92	0.13
South Coast	60371103	Los Angeles-North Main St	CARB	19	24.1	−0.74	0.08
South Coast	60374002	North Long Beach	CARB	23	24.1	−0.91	0.10
South Coast	60658001	Riverside-Rubidoux	CARB	25	37.1	−1.29	0.10
Notes:
^aAerometric Information Retrieval System code;
^bStandard error of the slope (one sigma).

Figure 4. Statewide mean annual PM_2.5 mass concentrations and number of trend sites included in each annual average. The means were determined from 22 urban sites having long-term monitoring records (Table 2) using monthly site averages and weighting inversely by uncertainties (eq 12). The error bars are two-sigma uncertainties and are larger in the 1980s because there were fewer monitoring sites, sampling frequencies were limited to once every 6 days, and some of the data are reconstructions of PM_2.5 mass (see text).

Downward trends in basin average PM_2.5 mass con centrations were qualitatively consistent with emission trends (Figure 5). Trends in PM_2.5 composition were not analyzed. However, downward trends were observed in PM₁₀ sulfate and nitrate at many locations, and inorganic species comprise approximately half the PM_2.5 mass on average at California sites,53 suggesting that the downward trends in PM_2.5 mass concentrations were linked at least in part to decreasing concentrations of secondary species including sulfate and nitrate. Because BC concentrations in the San Francisco Bay Area declined by ∼30% from the late 1960s to the early 2000s,54 the trends in PM_2.5 mass concentrations throughout California may also be due in part to trends in primary PM_2.5 emissions. Both BC and sulfate declined from 1982 to 1993 in the South Coast Air Basin (SoCAB).32 The ambient PM_2.5 trend in the SoCAB parallels trends in emissions of CO, oxides of nitrogen (NO_x), and reactive organic gases, but also seems to be related to trends in SO₂ and primary (directly emitted as PM) PM_2.5 emissions (Figure 5). Further work is needed to apportion the ambient PM_2.5 trends to changes in the emissions of specific primary species and PM precursors.

Figure 5. Trends in emissions and basin mean PM_2.5 mass concentrations in the SoCAB. Emission trends are from the 2009 California Almanac.63 Basin annual averages of ambient PM_2.5 mass concentrations were aggregated into 5-year means centered on the plotted years; the value for 1980 was based on 2 years of available data (1981 and 1982). Two-sigma error bars are shown for the ambient data. Basin mean PM_2.5 annual averages were determined from the annual averages of all monitoring sites. The location of the SoCAB is shown in Figure 6.

Figure 6. Mean basin PM_2.5 mass concentrations by decade (1980s, 1990s, and 2000s). Averages consist of 8–10 years data except Lake Tahoe 1990s (7 years), Mojave Desert 1980s (5 years), and Sacramento Valley 1980s (4 years). Population data are for year 2000.63

Spatial Variations of Fine Mass Concentrations

Mean fine PM mass concentrations were greater at sites in southern California and the San Joaquin Valley than in the San Francisco Bay area or the Sacramento Valley (Figure 6). The majority of site mean annual concentrations ranged from ∼15 to 35 μg m⁻³ in the SoCAB and from ∼8 to 20 μg m⁻³ in the San Francisco Bay area. Sites at Burbank and in San Bernardino and Riverside counties showed higher mean PM_2.5 values than did other SoCAB sites. In the San Joaquin Valley, urban sites (e.g., Fresno and Bakersfield) exhibited means of ∼20 to 30 μg m⁻³, whereas less urban locations had means of ∼10–20 μg m⁻³.

Although the data from the 47 long-term sites are useful for characterizing broad-scale spatial variations over time, they lack sufficient resolution to support fine-scale (e.g., 10 km or less) spatial interpolations over a geographical area as large and complex as the state of California, because some counties are unrepresented (Figure 2) and spatial variation exists within and between air basins (Figure 6). With this caveat, the data nonetheless show decreasing areas of high (>15 μg m⁻³) mean PM_2.5 concentrations (Figure 7).

Figure 7. Mean annual PM_2.5 mass concentrations in (a) 1989, (b) 1999, and (c) 2007. Monitoring data were interpolated using weights inversely proportional to the square of the distance between monitors and grid points.

The data set could be expanded by using less rigorous criteria than the authors did for inclusion of PM_2.5 data and reconstructions from PM₁₀ mass, TSP mass, and nephelometer data. In addition, county-level and gridded emission inventories are available; spatial interpolations that use such supplementary information potentially provide added resolution. Air quality model predictions23 and, for recent years, satellite data59–62 could also be combined with the monitoring data to increase the spatial resolution of PM_2.5 concentration fields.

CONCLUSION

Gravimetric PM_2.5 measurements from dichotomous and IMPROVE samplers and from special studies were calibrated against FRM data to provide standardized PM_2.5mass concentrations for years before FRM monitoring. Nongravimetric measurements of PM_2.5 mass concentrations were not used, because they did not agree with colocated FRM or other independent PM_2.5 mass measurements. PM_2.5 mass can be reconstructed to a high degree of accuracy (r ² > 0.9; mean absolute error ∼2 μg m⁻³) from PM₁₀ mass and species concentrations when determined on a site-specific and month-specific basis. By using non-FRM PM_2.5 measurements and reconstructions, the 10-yr PM_2.5 FRM monitoring record was extended to earlier years for 47 sites. Trends and spatial variations were evaluated using data from the 47 long-term sites and from a subset of 28 sites having 15 or more years of data. PM_2.5 trends were downward at all 28 trend sites and were in the range -0.13 to [1.29 μg m⁻³ yr⁻¹, median -0.59 μg m⁻³ yr⁻¹ at the 22 urban sites. Although the data from the 47 long-term sites are also useful for characterizing broad-scale spatial variations over time, additional efforts are needed to incorporate gridded emission estimates, modeling predictions, and other supplementary information to permit higher spatial resolution of fine PM concentrations.

ACKNOWLEDGMENTS

Many people collected samples, made measurements, created databases, and made their results available. The authors thank all who contributed data and especially thank L. Guo and C. Wang for preparing spatial interpolations of PM_2.5 mass concentrations, L. Dolislager and B. Croes for reviewing earlier drafts, and anonymous reviewers for their suggestions. The statements and conclusions in this report are those of the authors and not necessarily those of the CARB. The mention of commercial products, their source, or their use in connection with material reported herein is not to be construed as actual or implied endorsement of such products.