Interactive comment on “ Characterization of coarse particulate matter in the western United States : a comparison between observation and modeling ”

This manuscript presents the first comparison of PM10-2.5 between observation and CMAQ model simulation. Previous CMAQ studies focused primarily on PM2.5 or PM10. This study however provides a very interesting result regarding the performance of CMAQ on the PM10-2.5: the CMAQ underestimated the measurements systematically. The paper provides intriguing information for the model community and for the policy maker to rethink the uncertainty on air quality modeling. Several issues needs to be resolved before it can be accepted for publication in ACP:


Introduction
Concentrations of atmospheric particulate matter (PM) are currently regulated by the US Environmental Protection Agency (EPA) with National Ambient Air Quality Standards (NAAQS) for both PM 2.5 (fine particles; particulate matter with a diameter less than 2.5 µm) and PM 10 (particulate matter with a diameter less than 10 µm) (http://www.epa.gov/air/criteria.html).In the United States, there is an annual average standard and a 24-h average standard for PM 2.5 .The 3-yr average of the annual mean PM 2.5 concentrations must not exceed 15.0 µg m −3 , and the 3-yr average of the 98th percentile of 24-h concentrations at each population-oriented monitor within an area must not exceed 35 µg m −3 .The 24-h average PM 10 concentration standard of 150 µg m −3 must not be exceeded more than once per year on average over 3 yr.The European Community also regulates atmospheric particulate matter with legal limit values (e.g.Airborne PM 10 consists of both fine particles (PM 2.5 ) and coarse particles (PM 10−2.5 ; particulate matter with a diameter between 2.5 and 10 µm).Therefore, to meet the PM 10 standards, not only PM 2.5 but PM 10−2.5 concentrations need to be controlled.Moreover, recent epidemiological and toxicological studies show that PM 10−2.5 concentrations have been linked to mortality (e.g.Malig and Ostro, 2009;Perez et al., 2008;Zanobetti and Schwartz, 2009) as well as respiratory and cardiovascular morbidity (Branis et al., 2010;Brunekreef and Forsberg, 2005;Host et al., 2008;Sandstrom and Forsberg, 2008;Zhang et al., 2002).
In addition to health impacts and legal regulations, atmospheric particles can considerably affect climate directly by influencing incoming and outgoing radiation, and indirectly by serving as cloud condensation nuclei (CCN) and ice nuclei (IN), influencing the formation and lifetimes of clouds and precipitation as well as atmospheric chemistry (DeMott et al., 2003;Koehler et al., 2009;Krueger et al., 2004;Kumar et al., 2009Kumar et al., , 2011;;Solomon et al., 2007;Wang et al., 2007;Wurzler et al., 2000).Aerosols can also affect biogeochemical cycles, which can alter carbon fluxes and further interact with climate, by influencing physical environment (e.g.diffuse radiation, precipitation and temperature) and by depositing nutrients (e.g.nitrogen, phosphorous, and iron) or toxins (e.g.copper) to ecosystems (Mahowald, 2011;Paytan et al., 2009).The indirect effects of aerosols on climate are very uncertain (Mahowald, 2011;Solomon et al., 2007).PM 10−2.5 components (e.g.sea salt and soil dust) contribute considerably to global aerosol mass, optical thickness, and surface particle concentrations (Birmili et al., 2008;Textor et al., 2006).Therefore, to better quantify the effects of atmospheric particles, the characteristics of not only fine particles but coarse particles need to be understood.
While PM 2.5 is primarily emitted from combustion processes or formed in the atmosphere through chemical reactions and gas-to-particle conversion processes, PM 10−2.5 predominantly originates from abrasive mechanical processes, with sources such as geogenic dust, sea salt, dust from construction activities, tire wear, brake wear, and organic bioaerosols such as bacteria, pollen and fungal spores (Edgerton et al., 2009;Harrison et al., 2001;Kelly et al., 2010;Malm et al., 2007;Sesartic and Dallafior, 2011;Zhu et al., 2009).Controlling variables on these sources include land use, land cover, and environmental conditions (e.g.temperature, soil moisture, snow/ice cover, wind speed).Some of these sources are a result of natural processes (e.g.windblown dust in a desert), while others are more closely tied to human activities (e.g.construction).Additionally, PM 10−2.5 has a higher deposition velocity, i.e., shorter atmospheric residence time, than PM 2.5 .These combined facts mean that PM 10−2.5 will have different spatial and temporal variability than PM 2.5 .Recent studies investigated the characteristics of PM 10−2.5 in a few US cities including Los Angeles, CA (Pakbin et al., 2010), Detroit, MI (Thornburg et al., 2009), Rochester, NY (Lagudu et al., 2011), and Denver and Greeley, CO (Clements et al., 2012).However, little research has investigated the spatial and temporal variability of PM 10−2.5 concentrations at a regional scale, or the relationships between concentrations and land use/land cover and soil moisture dependent on geographical location.
Accurate PM 10−2.5 modeling tools are needed by both the scientific community and regulatory agencies for mitigation strategy development and health effect assessments.PM 10−2.5 is simulated as part of the US EPA's Community Multiscale Air Quality (CMAQ) modeling system (Byun and Schere, 2006).However, the model performance for PM 10−2.5 has not been explicitly assessed because over the past decade both PM model and measurement studies have primarily focused on PM 2.5 .CMAQ and other chemical transport models have been primarily assessed for their performance for PM 2.5 or PM 10 (Baldasano et al., 2011;Chuang et al., 2011;Foley et al., 2010;Konovalov et al., 2011;Lonati et al., 2010;Sokhi et al., 2008;Wang et al., 2008).Yet, since fine and coarse particles have different sources as well as different chemical composition and potential health effects, they should be considered as separate classes of pollutants as suggested by Wilson and Suh (1997) and assessed individually.
Given the importance of coarse particles for air quality, climate, and human health risk assessments, improvements to our knowledge of the sources and characteristics of PM 10−2.5 are essential.In this paper, we investigate the temporal and spatial patterns of measured PM 10−2.5 concentrations in the western United States.The results of this analysis provide insights to the sources and fate of PM 10−2.5 and motivate more accurate models that describe PM 10−2.5 emissions, transport, and atmospheric concentrations.

Methods
This study was carried out using both observations and model simulations for an entire year (2005) over a domain that covers the western United States (see Fig. 1).

Measurement data
While abundant ambient PM 2.5 and PM 10 mass concentration data are available, direct measurements of PM 10−2.5 mass concentrations are very limited.Therefore, our study obtained co-located measurements of PM 10 and PM 2.5 .We obtained all available observed hourly-averaged PM 10 and PM 2.5 concentration data in the western United States (see Figs. 1 and 2) for 2005 from the Air Quality System (AQS) datamart (http://www.epa.gov/ttn/airs/aqsdatamart/) and from two state agencies.From the AQS, we obtained hourly co-located PM 10 and PM 2.5 concentration data for sites having daily data).788 789    1).To fill spatial gaps of hourly data, we obtained daily measurements (24-h filter samples) from the AQS for an additional 25 sites in the domain, shown as circles in Fig. 1 (also described in Table 1).The 24-h measurements were taken every three days at two sites (Riverside site, CA, AQS Site Number of 060830011 and Salt Lake City, Utah, AQS Site Number of 490353006), and every 6 days at the other sites.Details of all measurement sites, including associated environmental conditions output from the Pennsylvania State University/National Center for Atmospheric Research Mesoscale Meteorology Model (MM5) (i.e.hourly average temperature, wind speed, and surface soil moisture), are presented in Table 1.The concentrations of PM 10−2.5 were calculated as the difference between co-located PM 10 and PM 2.5 concentrations at all hourly and daily sites.

Model simulations
To obtain insights for regional PM 10−2.5 modeling, model simulations were carried out for the western United States.The Community Multiscale Air Quality (CMAQ) modeling system v4.7.1 (Byun and Schere, 2006;Foley et al., 2010) was used to simulate the transport and chemistry of atmospheric gases and particles.The model configuration included the AERO5 aerosol module having secondary organic aerosol treatment for fine particles (Carlton et al., 2010), ISORROPIA inorganic chemistry (Nenes et al., 1999), the Carbon-Bond 05 (CB05) gas phase chemistry mechanism (Sarwar et al., 2008;Whitten et al., 2010), aqueous phase chemistry for sulfur and organic oxidation (Carlton et al., 2008), and sea salt treatment (Kelly et al., 2010).The CMAQ aerosol module represents PM in three lognormal modes: the Aitken ("I" mode) with diameters up to about 0.1 µm, the  (Houyoux et al., 2000) were generated using MM5 version 3.7.4(http://www.mmm.ucar.edu/mm5)with the Pleim-Xiu boundary layer and land surface model (Pleim and Xiu, 2003;Xiu and Pleim, 2001), Kain-Fritsh 2 cumulus parameterization (Kain, 2004), RRTM longwave (Mlawer et al., 1997), Dudhia shortwave (Dudhia, 1989), and Reisner 2 mixed phase moisture schemes (Reisner et al., 1998).Three dimensional analysis nudging was applied only above the boundary layer for moisture and temperature and over the entire vertical atmosphere for winds.The MM5 simulations resolve the vertical atmosphere up to 100 mb with 34 layers, which were reduced to 14 layers by MCIP (Meteorology-Chemistry Interface Processor) (Otte and Pleim, 2010) for emissions and photochemical models with the thinnest layers near the surface to best resolve the diurnal boundary layer cycles.The height of the first model layer is approximately 38 m.
Simulations were performed for the year 2005 with 3 days of spin-up at the end of 2004 that were not included in the analysis.Anthropogenic emissions used to drive the modeling system were based on the 2005 National Emission Inventory (NEI) (http://www.epa.gov/ttnchie1/net/2005inventory.html).Biogenic emissions were estimated with the BEIS model using hourly temperature and solar radiation as input (Pierce et al., 1998).Emissions were processed to hourly gridded input to CMAQ with the SMOKE model version 2.5 (Houyoux et al., 2000).Over the modeling domain, annual PM 10−2.5 emissions were dominated by the non-point area sector (86 %), and their primary sources include fugitive dust from paved roads, unpaved roads, road construction, residential construction, non-residential construction, and agricultural tilling.The inventory did not include emission estimates of wind-blown (geogenic) dust.Sea salt emissions were simulated online within CMAQ following Kelly et al. (2010).

Spatial variability
Table 2 presents a summary of statistical analyses of measured and modeled PM 10−2.5 concentration data at all sites having either hourly or daily data, including mean, 5th percentile, 95th percentile and coefficient of variation (CV).CV is defined as the following: CV = Standard deviation of time series Mean of time series (1) The measured PM 10−2.5 concentrations have a distinct spatial pattern in the western United States as seen in Fig. 2, which shows observed annual mean PM 10−2.5 concentrations at all measurement sites.The highest concentrations were observed at sites in the southwestern US, where shrublands and barren/sparse vegetation dominate (Fig. 1) with generally lower surface soil moistures and higher temperatures (Table 1).The lowest concentrations were found at sites dominated by grasslands, forest, or croplands with generally higher surface soil moistures and lower temperatures (Fig. 1; Table 1).Given the dominance of shrublands and barren/sparse vegetation along with very dry soils in the southwestern US, the higher concentrations in this region are likely caused by fugitive dust emissions, which include geogenic dust.Table 2 shows that all sites having annual mean concentrations that are higher than 17.7 µg m −3 are located to the south of ∼ 36 • N, except for the Rapid City site, which has high winds (Table 1) and is significantly influenced by fugitive dust from several industrial facilities (primarily limestone quarrying and processing and cement manufacturing and processing facilities) (http://denr.sd.gov/ documents/neap.pdf).Measured PM 10−2.5 concentrations show strong spatial variations across the western US; the annual mean of measured PM 10−2.5 concentrations is more than 17 times higher at the Nogales site in Arizona than at the Wind Cave National Park site in South Dakota.Even sites in close proximity showed significant variability.For example, although the N. Phoenix (040139997) and the N. W. Phoenix (040130019) sites are located very close to each other (∼ 5 km), the annual mean of measured concentrations differed substantially, from 19.4 to 30.4 µg m −3 , respectively.In Albuquerque, NM, the annual mean measured concentration is more than two times higher at the Albuquerque South site (21.8 µg m −3 ) than at the Albuquerque East site (10.0 µg m −3 ), although they are located within the same city (∼ 13 km apart).The differences in PM 10−2.5 concentrations between the sites can be even greater at finer temporal resolutions.The daily average concentration on 8 April 2005 (during a PM 10−2.5 episode) was 3.75 times higher at the Albuquerque South site (130 µg m −3 ) than at the Albuquerque East site (34.7 µg m −3 ); the maximum hourly concentration on this day was about 6 times higher at the former site (571 µg m −3 ) than at the latter site (95.4 µg m −3 ).Observed annual average PM 10−2.5 concentrations at the Rapid City site (28.4 µg m −3 ) were more than 10 times higher than those at the Wind Cave National Park site (2.8 µg m −3 ), even though these two sites are only 61 km apart.
The spatial variability of measured PM 10−2.5 at both urban and regional scales was assessed with the correlation coefficients for measured hourly concentrations, calculated between all sites having hourly measurements.Moderate to strong correlations were observed between some sites located in close proximity to one another, including the four sites in El Paso, TX (r 2 = 0.24-0.58),two sites in Albuquerque, NM (r 2 = 0.28), three sites in the northeastern part of the domain (Crop&River, Thompson Lake, and Overlook in North Dakota; r 2 = 0.21-0.36),three sites in the northwest (Spokane, Pinehurst, and Coeur D'Alene; r 2 = 0.23-0.37)and two sites in Seattle (r 2 = 0.2).The p-values for these correlations are all less than 0.0001, so they were considered significant.No correlation was observed between any other combinations of the site pairs.Very little correlation was seen even over relatively small distances between some sites, such as two sites in New Mexico (Sandoval and Albuquerque East; r 2 = 0.05) and three sites in South Dakota (Rapid City, Badlands, and Wind Cave National Park; r 2 = 0.00-0.03),suggesting that these sites are impacted by different sources or have a different proximity to sources.These poor correlations along with high spatial variability also suggest that PM 10−2.5 concentrations are often influenced by local factors.

Variability
Figure 3 presents the time series of measured daily average PM 10−2.5 concentrations (red lines or squares) at selected representative sites having hourly (Fig. 3a-c) or 24-h (Fig. 3d-f) measurements.The green lines in Fig. 3 represent simulated daily average concentrations from the modeling study, which will be discussed subsequently.Figure 3 demonstrates that measured daily average PM 10−2.5 concentrations have strong temporal variations at each site with episodic high levels.The CV of measured PM 10−2.5 concentrations is not less than 1.0 at 22 of 25 sites, ranging from 0.7 to 2.0 (see Table 2).Figure 3a shows that measured PM 10−2.5 concentrations exceeded the level of the PM 10 NAAQS (24h average of 150 µg m −3 ) for many days in 2005 in El Paso.This result highlights the necessity to understand the behavior of coarse particles in order to develop mitigation strategies to keep the PM 10 concentrations at safe levels.

Seasonal patterns
Figure 3 also reveals seasonal patterns.The measured PM 10−2.5 concentrations show different seasonal patterns de-pendent on location.At some sites (e.g.Fargo and Fresno, two inland or valley sites influenced by agricultural sources, shown in Fig. 3c and f), the measured concentrations show a seasonal pattern with lower values in winter months.At those sites in the southwestern US and on the west coast (e.g.El Paso West, Seattle, and Riverside shown in Fig. 3a-b, d), the measured PM 10−2.5 concentrations seem to be more uniform over the year, with some episodic increased concentrations.

Weekly patterns
The red lines in Fig. 4 show one-year average weekly patterns of measured PM 10−2.5 concentrations at selected hourly sites.There are primarily three different average weekly patterns of observed PM 10−2.5 concentrations at the hourly sites.The first pattern shows that the measured PM 10−2.5 concentrations are ∼ 50 % lower for weekends than for weekdays (e.g. the Seattle site shown in Fig. 4a), reflecting significant influences of weekday versus weekend human activities on PM 10−2.5 concentrations at these sites.The second weekly pattern, on the contrary, shows that there is little difference in the observed concentrations between weekdays and weekends (e.g. the El Paso West site shown in Fig. 4c).The weekday versus weekend human activities have a negligible impact on observed PM 10−2.5 concentrations at these sites.The third pattern, which lies in between the previous two patterns, suggests that human activities have a moderate influence on the observed PM 10−2.5 concentrations, with one-year average levels being about 20 % lower during weekends than weekdays (e.g. the Santa Barbara site in Fig. 4b).The patterns are apparently dependent on the relative importance of the weekday versus weekend human activities on PM 10−2.5 concentrations compared to other sources.

Diurnal patterns
The red lines in Fig. 5a-d show one-year average diurnal patterns of measured PM 10−2.5 concentrations at selected hourly sites.The measured concentrations exhibit different diurnal patterns varying with location.Observed PM 10−2.5 concentrations at some sites (e.g.see Fig. 5a for the Denver site) show a typical diurnal pattern associated with on-road traffic.There is a rush-hour peak in the morning, followed by a decrease corresponding to a reduced volume of traffic and an increased mixing layer height in the middle of the day.Then there is a late afternoon rush-hour peak and another reduction afterwards.However, the measured patterns at other sites are more complicated with some having significantly bigger afternoon peaks (e.g.Fig. 5b for El Paso West) but with others having significantly bigger morning peaks (e.g.Fig. 5c for Seattle).The diurnal pattern at the Rapid City site, which is significantly influenced by industrial facilities (Sect.3.1), is completely different: the concentrations at night are relatively small (15 µg m −3 ), and increase steadily, reaching a maximum value of about 42 µg m −3 in the middle of the

concentrations
In addition to the measured concentrations, Table 2 also shows statistical analyses of CMAQ-predicted PM 10−2.5 concentrations.Table 2 reveals that the CMAQ model underestimated annual PM 10−2.5 concentrations at all sites except for San Jose, CA, where the agreement between modeled and measured annual average concentrations is the best among all sites.However, the good agreement at the San Jose site is only for the annual mean concentration; the model failed to reproduce the seasonal pattern at this site (see Sect. 4.3.2).The mean ratio of measured to modeled annual PM 10−2.5 concentrations, averaged across all sites, is more than 4, with the maximum ratio of 27 at the Nogales site in the southern Arizona.While CMAQ generally underestimated PM 10−2.5 concentrations at almost all sites, there are variations in model performance at different locations.
While the modeled and measured annual mean concentrations agree within a factor of two at 16 sites, 20 sites have measured annual mean concentrations that are more than four times higher than modeled values (Table 2).Among these 20 sites, 14 sites have observed annual mean concentrations being more than five times higher than simulated levels.The lower modeled concentrations are likely due to the omission or significant underestimation (or a combination of both) of important emission sources in the inventory.Further discussions on the causes for the lower modeled concentrations are in Sect. 5.   (39.9 µm m −3 ) is much higher than that at the N Phoenix site (19.4 µm m −3 ); however, the modeled value is lower at the former site (8.7 µm m −3 ) than the latter site (10.0 µm m −3 ).These sites are located within 22 km of one another.Similarly, the observed annual mean concentration is more than 2 times higher at the North Riverside site, CA (25.6 µm m −3 ) than at the Anaheim site, CA (11.4 µm m −3 ), which are 51 km apart, but the modeled value is almost two times higher at the latter site (9.5 µm m −3 ) than at the former site (5.0 µm m −3 ).Although the observed annual mean concentration is more than three times higher at the Sandoval site, NM (31.4 µm m −3 ) than at the Albuquerque East site, NM (10.0 µm m −3 ), the modeled value is much lower at the former site (4.7 µm m −3 ) than at the latter site (8.1 µm m −3 ).These results may be caused by variable emission underestimations in the inventory across the domain; another possibility is the inaccurate spatial allocation in the emission modeling system.

Temporal variability
Figure 3 and Table 2 show that the modeled daily average PM 10−2.5 concentrations are less variable than measurements at almost all sites with average ratio of measurement CV over model CV being more than 1.5.We suggest two plausible explanations.First, the modeling system, as mentioned, allocates annual emissions into hourly values using monthly, weekly and diurnal profiles.This approach does not have the representation for the strong episodic nature of PM 10−2.5 emissions such as fugitive dust from construction and agricultural tilling that can be affected by several factors including human operation and wind speed.Second, the measured concentrations were obtained at a specific location on the surface, whereas the modeled concentrations were values averaged over a box over a 12 × 12 km grid cell with a height of approximately 38 m in the first model layer.This spatially, especially vertically averaging might have lead to smoother variability of modeled PM 10−2.5 concentrations compared to the observations.Figure 3a shows that the modeling system did not capture very high concentrations in PM 10−2.5 episodes, in which the measured PM 10−2.5 concentrations alone exceeded the level of the PM 10 NAAQS in El Paso.Since this modeling system is used for air quality management and forecasts, this inability to accurately simulate the high episodic concentrations can cause serious problems in such important issues as air quality advisory issuances and health risk assessments.

Seasonal allocation
While the measured concentrations show different seasonal patterns dependent on location as mentioned in Sect. the green lines in Fig. 3 show that the modeled PM 10−2.5 concentrations exhibit the same seasonal pattern at all sites with somewhat higher concentrations in winter months.Even at the San Jose site, CA, where the best agreement was found between the modeled and measured annual average concentrations, the seasonal patterns are clearly different: the model significantly under-predicted PM 10−2.5 concentrations in warm months, whereas it overestimated concentrations in winter months.The comparisons between modeled and measured daily concentration time series suggest that the modeling system needs to be improved to better simulate seasonal patterns.

Weekly allocation
While there are three distinct weekly patterns of measured concentrations (red lines in Fig. 4), we found that the modeled weekly patterns for all sites are similar; there is little difference between modeled weekday and weekend concentrations as shown by green lines in Fig. 4a-c.Figure 4a-c further confirms that CMAQ significantly underestimates PM 10−2.5 concentrations at these sites.The comparisons between modeled and measured weekly patterns suggest that the-day-of-week allocation needs to be improved to reflect variable influences of weekday versus weekend human activities at different locations.

Diurnal allocation
Figure 5a-d show that CMAQ not only under-predicts the magnitude of PM 10−2.5 concentrations, but fails to duplicate the diurnal patterns at many locations.While the measured concentrations exhibit distinct diurnal patterns varying with location, modeled concentrations show the same diurnal pattern with two similar peaks (one in the morning and the other in the afternoon) at all sites.This means that the diurnal allocation in the modeling system is too idealized to reflect complex patterns at different locations.

Discussion
The CMAQ model not only under-predicted the magnitude and variability of PM 10−2.5 concentrations, but also failed to duplicate the spatial as well as seasonal, weekly and diurnal patterns.The causes for the underestimated concentrations in the model may differ based on different major contributing sources.For example, our results show that the measured concentrations are very high at sites in the southwestern US, such as El Paso West (Figs. 3a,4c,and 5b).Since the area is dominated by shrublands, barren or sparse vegetation land cover with very dry soils (Fig. 1) -ideal conditions for high wind-blown dust -the high concentrations may be dominantly contributed by geogenic and other fugitive dust sources.This suggests that the underestimated concentrations in this area are caused in part by the omission of wind-blown dust in the inventory, which may also contribute to the incorrect seasonality of modeled concentrations.Still, we cannot rule out the possibility of significant under-estimation of sources such as unpaved road dust and construction, which will also be important in regions with dry soils.Some sites (e.g.Seattle) have very strong anthropogenic influence indicated by large weekday-weekend difference in observed PM 10−2.5 concentrations (Sect.3.2.3);therefore, the inclusion of natural emission sources such as wind-blown dust alone could not lead to model reconciliation with measurements.Thus, efforts should be made to improve emissions associated with human activities at these sites.
The three costal sites with hourly observations (i.e.Santa Babara, South Seattle, and Seattle) are expected to have a significant marine influence.Sea salt contributes considerably to the modeled concentrations at the Santa Babara (∼ 60 %), South Seattle (∼ 20 %), and Seattle (∼ 21 %) sites, as shown in Fig. 7, which shows modeled concentration fractions of coarse particle components (i.e.sea salt, soil dust, nitrate, ammonium, sulfate, and unspecified particles) at these three coastal sites and the Denver site.Since modeled concentrations at these three coastal sites were much lower than measurements, it is possible that the sea salt concentrations might still have been underestimated by the CMAQ model.The large weekday-weekend difference in PM 10−2.5 concentrations in Seattle, mentioned earlier, suggests that some anthropogenic sources such as construction and on-road traffic might also have been under-estimated.Modeled sea salt only affects a narrow coastal zone and over the ocean (not shown).The inland sites were dominated by soil dust (e.g. as shown by Fig. 7 for the Denver site).
Recent studies show that coarse particles contain significant organics (Cheung et al., 2011;Edgerton et al., 2009); however, Fig. 7 shows that the CMAQ model does not explicitly simulate organic materials in coarse particles.Therefore the omission of organic sources, such as primary biological particles and humic-like substances from soils, is another possible cause for lower modeled concentrations compared to measurements.The organic components of coarse particles have implications for health risk assessments and atmospheric chemistry, thus improvements should be made to include organics in coarse particles in the future.Significant concentration differences and small correlations were observed between some proximate sites (Sect.3.1), suggesting that PM 10−2.5 can be largely influenced by local factors.Therefore, the exact spatial information of emission sources becomes more important for PM 10−2.5 .The current PM 10−2.5 emissions from such sources as construction, road, and agricultural tiling are provided at the county-level that are ultimately spatially allocated into user-defined grids during simulations using surrogate data.It may be necessary to further specify the detailed spatial information of coarse PM sources in future emission inventories and model developments.
While modeled PM 10−2.5 concentrations show the same seasonal, weekly, and diurnal pattern regardless of location, the observed patterns are more complex; no consistent patterns were observed at all sites, indicative of a variety of contributing sources and their relative importance.Therefore the current temporal allocation approach in the model framework needs to be improved since it is too simplified to track the real patterns at different locations.
Correlations between PM 10−2.5 concentrations and wind speed, boundary layer height, and surface moisture output from MM5 were calculated for all hourly sites (Table 4).PM 10−2.5 concentrations are positively correlated with wind speed at most sites (18 of 25 sites), with the greatest correlations occur at the Overlook site (r = 0.24), ND, and the Rapid City site (r = 0.22), SD.However, zero or negative correlations were found at a few other sites (Table 4), with r = −0.13 at the Nogales site, AZ.This result suggests that PM 10−2.5 concentrations are affected by factors other than wind speed.Although in a region where dust is expected to influence the coarse PM concentrations, the Nogales site in Arizona is unique in that it is located near a roadway and a busy border crossing between the US and Mexico.We expect that the traffic and anthropogenic activity near this site to control PM 10−2.5 emissions and concentrations.The correlation between PM 10−2.5 concentrations and the boundary layer height is −0.30 at the Nogales site, AZ, implying higher concentrations may be partly caused by lower boundary height (Table 4).PM 10−2.5 concentrations are negatively correlated with boundary layer height at only 11 of 25 sites.At the other 14 sites, positive correlations were found between the PM 10−2.5 concentrations and the boundary layer height.This result reflects the complex influences on PM 10−2.5 concentrations by the boundary layer height and other possible factors as well as their interactions.PM 10−2.5 concentrations are negatively correlated with soil moisture at all investigated sites (Table 4), indicating that high PM 10−2.5 concentrations are correlated to lower soil moisture.Since less dust can be emitted into the atmosphere in wet conditions and airborne particles can be washed out at precipitation events, higher concentrations of dust are expected under drier conditions.
Chemical and biological analyses of measured PM 10−2.5 can be employed to quantify percentage contributions from different sources at the ambient measurement sites; however, little chemical or biological speciation data exists for PM 10−2.5 .By taking an approach that combines both mass concentration observations and model simulations, this study has improved our understanding of the sources and behavior of PM 10−2.5 concentrations at a regional scale in the western United States, and has provided insights into future developments of models that simulate atmospheric PM 10−2.5 emissions, transport, and fate.
Measurement of all of criteria air pollutants is required by law.To help meet this requirement, a Federal Reference Method (FRM) and Federal Equivalency Methods (FEM) for each have been established and are documented in the Code of Federal Regulations (CFR).The FRM and FEM requirements are stringent with lengthy quality control and quality assurance protocols for each pollutant.The end result is high quality measurement data for each pollutant being reported in the AQS.AQS data have been used successfully by numerous studies (e.g.Chang et al., 2012;Drury et al., 2010;Jensen et al., 2009;Sampson et al., 2011;van Donkelaar et al., 2006;Zhang et al., 2006).FRM and FEM for PM 10−2.5 were not established until 2007 and, as such, there are no PM 10−2.5 mass concentrations in the AQS for our study year of 2005.The FRM for PM 10−2.5 involves subtraction of low volume FRM PM 2.5 mass concentration from a co-located low volume FRM PM 10 mass concentration (for more details see CFR 40 Part 50 Appendix O).Our approach here was to use the best possible substitute.We used subtraction of collocated FRM PM 2.5 mass concentration from FRM PM 10 mass concentration, without distinction of sample volume.This approach will likely increase the uncertainty associated with the resulting PM 10−2.5 mass concentrations as compared to the FRM PM 10−2.5 .Additionally, this approach may potentially introduce a small bias associated with how volatile components are assessed.However, the magnitude of the potential bias and uncertainty associated with our approach is relatively small compared to the big differences between measured and modeled PM 10−2.5 concentrations (US EPA, 2004, 2009).In other words, the uncertainties of the measurements cannot affect our conclusion that the modeling system significantly underpredicted PM 10−2.5 concentrations across the western United States.

Summary and conclusion
We investigated the characteristics of observed coarse PM in the western US, and compared CMAQ predictions to the observations.The observed concentrations showed a spatial pattern that could be explained in part with the distributions of land use and soil moistures.The highest concentrations were found in the southwestern US, where sparse vegetation, open shrublands or barren lands dominate with lower soil moistures, whereas the lowest concentrations occurred in areas dominated by grasslands, forest, or croplands with higher soil moistures.Observed concentrations show different seasonal, weekly, and diurnal patterns at different locations across the western United States, reflecting different contributing sources and their relative importance dependent on locations.CMAQ significantly under-predicted PM 10−2.5 concentrations.The under-prediction was likely due to omission of sources such as pollen, bacteria, fungal spores, and especially, geogenic dust, as well as under-estimation of other significant source types.CMAQ also failed to reproduce their spatial as well as seasonal, weekly, and diurnal patterns.Unlike observations, the modeled concentrations show similar seasonal, weekly, and diurnal pattern across the entire domain.CMAQ does not include organics in PM 10−2.5 , which recent measurements show to be a significant component.In this study we identified some important gaps for future developments of coarse PM models and emission inventories.
daily limit value of 50 µg m −3 for PM 10 ) under Directive 2008/50/EC of the European Parliament and of the Council R. Li et al.: Characterization of coarse particulate matter in the western United States of 21 May 2008 on ambient air quality and cleaner air for Europe (http://ec.europa.eu/environment/air/quality/legislation/existing leg.htm).

Fig. 1 .
Fig. 1.Map of monitoring locations and land use/land cover in the study domain (sites having hourly data are represented with black plus symbols, and black circles represent sites having daily data).

Fig. 2 .
Fig. 2. Measured annual mean PM10-2.5 concentrations at measurement sites in the western United States.

Fig. 2 .
Fig. 2. Measured annual mean PM 10−2.5 concentrations at measurement sites in the western United States.

Fig. 3 .Fig. 3 .
Fig. 3. Measured (red lines or symbols) and modeled (green lines) daily-average PM 10-2.5 794 concentrations at representative hourly (a-c) and 24-hour (d-f) sites.Note that the 24-hour 795 measurements were taken every 3 days at the Riverside site (d) but every 6 days at the 796 San Jose (e) and Fresno (f) sites.Note the differences in scale.797

Fig. 5 .
Fig. 5. One-year average diurnal patterns of measured (red line) and modeled (green line) 815 PM 10-2.5 concentrations at selected hourly sites.Note the differences in the scales.816 817

Table 1 .
Details of measurement sites.

Table 2 .
Summary of statistical analyses of measured and modeled PM 10−2.5 concentrations.

Table 4 .
Correlations between PM 10−2.5 concentrations and wind speed, boundary layer height, and surface moisture.