Interactive comment on “ Sources , trends and regional impacts of fine particulate matter in southern Mississippi Valley : significance of emissions from sources in the Gulf of Mexico coast ”

My main criticism is related to the choice of parameters (chemical markers or compounds) used for PMF analysis. Although the use of different form of the same elements (or compounds) sometime could give a best source fingerprint, in the paper some parameters are redundant in the PMF analysis. In particular, I understand the use of both K and K+ in the PMF analysis because K (total content) is more related to crustal source, conversely K+ is more specific for biomass burning source (and primary marine). I’m less convinced about the use of both S and SO4, actually S comprise also methanesulfonate (MS) from atmospheric oxidation of biogenic dimethylsulfide, but at this site MS have to be negligible because the SO4/S ratio is higher than 3 (line 13 page 836) likely due to underestimation of S and absence of MS. I disagree to the use of OC1, OC2, OC3 and OC4 together with total OC. The use of more specific fraction of OC makes unnecessary insert total OC in the PMF. Besides, in quantification procedure by PMF all these parameters (K K+, S-SO4, OC1, OC2, OC3, OC4 –OC), representing more or less the same amount, give an overestimation of the contribution of the source. May be the authors exclude the double parameters from source quantification, but this is not reported in the text; please the authors better explain their approach.


Introduction
Atmospheric aerosol modifies Earth's energy balance by scattering sunlight back to space, absorbing solar and infrared radiation and changing the microphysical and thermodynamic properties of clouds (Ostrom and Koone, 2000;Lohmann and Feichter, 2001;Quinn and Bates, 2005).Overall, atmospheric aerosols have a negative radiative forcing from −0.2 ± 0.2 W m −2 to −0.8 ± 0.2 W m −2 , both directly and indirectly, through the cloud albedo effect (IPCC, 2007).Exposures to particulate matter are also associated with acute and chronic health problems and lead to increased mortality rates from respiratory, cardiac and circulatory diseases, increased emergency care visits and hospital admissions for bronchitis and asthma (Brunekreef and Holgate, 2002;Peng et al., 2005;Brunekreef and Forsberg, 2005).In the US, PM 2.5 (particles with aerodynamic diameter less than 2.5 µm) mass concentrations decreased by 27 % between 2001 and 2010 (US EPA, 2012a).These trends were attributed to the significant reductions of gaseous sulfur and nitrogen oxides from coal-fired power plants and mobile sources.These gaseous pollutants are precursors of particulate sulfate and nitrate aerosol, the dominant species of PM 2.5 aerosol in the Midwest and eastern US.The slower decline on particle mass levels for "cool" months (October to April) as compared to "warm" months was explained by the elevated emissions from residential wood burning and the formation of temperature inversion layers that trigger the accumulation of particles near the ground.
Shipping emissions are recognized as an important source of particulate matter and its precursors around the world (Wang et al., 2003;Deniz et al., 2008;Minguillon et al., 2008).The diesel engines in ships operate on fuel that can have extremely high sulfur content and porphyrin-content rich in V and Ni (termed as bunker oil).They are subjected to modest emission controls in the US and around the world, with the exceptions of passenger ships in the Baltic Sea, North Sea and English Channel.It is predicted that in the absence of emission controls, SO 2 emissions to the total emissions in the US would increase from 21 % today to 81 %, NO x emissions would increase to 28 % of total mobile NO x emissions in the US, and PM 2.5 emissions would almost triple to 170 000 tons yr −1 by 2030 (Corbett et al., 2003;Dalsoren et al., 2009;Eyring et al., 2010).Diesel particles from ships and secondary aerosol from the oxidation of NO 2 and SO 2 are shown to be related to 50-500 cancer cases, 750 asthma attacks and 29 premature deaths per million people within 15 miles of the port and influence the air quality in receptor sites far away from the coast (Corbett et al., 2007;Linder et al., 2008).
The Little Rock/North Little Rock Metropolitan Statistical Area (MSA) is a region on the western edge of the southern Mississippi Valley with a population of approximately 700,000.24 h PM 2.5 levels varied from 1 to 54 µg m −3 over the past 10 years, with annual PM 2.5 from 11.4 to 12.6 µg m −3 in 2010.According to the 2008 Environmental Protection Agency National Emission Inventory, the highest contributing source to air pollution was prescribed fires (1.076 tons yr −1 ), followed by road dust (617 tons yr −1 ); industrial and mobile sources emitted 100 and 300 tons yr −1 , respectively (US EPA, 2012b).Winds were typically from the northwest and south/southeast of the Gulf of Mexico, enabling the transport of particles and its precursors from several regions with diverse characteristics.The objectives were: (i) to apportion the contributions of sources to fine particulate matter in central Arkansas using positive matrix factorization; (ii) to assess the annual trends of PM 2.5 and its sources; and (iii) to identify and quantify the impacts of regions to fine particle mass and its sources using the trajectory residence time regression analysis.The latter was previously applied to assess the influences of source regions to sulfate concentrations in continental background sites (Gerbhart et al., 2001;2006;Xu et al., 2006) and black carbon in Arctic (Huang et al., 2010).Here, we applied this approach on physico-chemically active mixtures of fine particles from different types of sources (local and regional) in an urban area.

Sampling site and measurements
The concentrations of PM 2.5 mass and chemical species measured at the NCore site in North Little Rock (EPA AIRS ID: 051190007; Lat.: N 34.756072;Long.: W 92.281139) for the 2002-2010 period were retrieved from US Environmental Protection Agency's Air Quality System (AQS).The NCore network (previously known as PM 2.5 Chemical Speciaton Network) is comprised of 63 urban sites and 18 rural sites across the US.In each site, 1 h and 24 h PM 2.5 mass, 24-h PM 10−2.5 mass, PM 2.5 chemical speciation, NO x , NO y , O 3 , SO 2 , CO were measured.Filter sampling was done once every three days, using a four channel speciation sampler (Demokritou et al., 2001).Elements (from Na to U) were measured by X-ray fluorescence spectroscopy, watersoluble ions (sulfate, nitrate, chloride, ammonium, sodium, potassium, calcium and magnesium) by ion chromatography, atomic absorption and colorimetry.Elemental, four fractions of organic carbon (evolved from ambient to 140 • C for OC1 (volatile), from 140 • C to 280 • C for OC2 (semivolatile), from 280 • C to 480 • C for OC3 (nonvolatile) and from 480 • C to 580 • C for OC4 (non volatile)) and carbonate carbon were measured by the thermal optical reflectance (TOR) method using EPA's laboratory standard operating protocols (http://www.epa.gov/ttn/amtic/specsop.html).
In Little Rock, the NCore site is located in a park adjacent to the intersection of Pike Avenue and River Road in North Little Rock, AR by the Arkansas River.The annual average daily traffic (AADT) on these two roads is very limited (no estimates are provided by Arkansas State Highway and Transportation Department).The streets with the highest AADT are W. Riverfront Dr and W. Broadway St with 5000 and 12 000 vehicles/day, respectively (ASHTD, 2011).The site is classified as commercial/neighborhood by EPA.

Positive matrix factorization
In PMF (positive matrix factorization), the concentrations of m-aerosol species for n-sampling days are described by the sum of the product of the source contribution (G(nxp)) and the source profile matrix (F (pxm)) where p is the number of sources and the residual component (E(nxm)) (Paatero and Tapper, 1994;Paatero, 1997) (Eq. 1) through a solution that minimizes the objective function (Q in Eq. 2) based on measurements uncertainties: where x ij and σ ij are the concentration and associated uncertainty of j -species in i-sample, g ik is the contribution of the k-factor to particle mass in i-sample, and f kj is the mass  fraction of j -species on k-factor.The PMF2 algorithm applies a least-squares approach by considering that sources profiles and contributions are not negative during the optimization analysis.The F peak value, a user-defined non-zero rotational parameter, controls the subtraction of the profiles from each factor to eliminate the remaining rotational ambiguity by forcing the addition of one G vector to another, and subtracting the corresponding F factors from each other and thereby yielding realistic solutions.The optimum number of factors (sources) and the rotation (controlled by the F peak value) lies with the mathematical solution in which Q remains relatively constant, the highest individual column mean (IM) and standard deviation (IS) from the scaled residual matrix drop significantly and the highest element in rotmat increases (Paatero et al., 2002;2005): where r ij and r j are the individual and mean residuals, respectively.The rotmat matrix evaluates the rotational freedom of the solution, with the maximum value of the matrix being indicative of the case with the largest rotational freedom.
In this effort, we run the PMF2 model in the robust method with an α =4.0 using the error model "−12" (that uses observed values).Missing concentration data are replaced by the geometric mean of the measured concentrations while missing uncertainties are substituted by four times the geometric mean of measured uncertainties.The α value, a userspecified variable, defines the distance of outliers (ασ ij ) from the fitted value in order to include them into the analysis.
The error model determined the s ij values as follows: where t ij and v ij are the uncertainties and relative errors of x ij .In our case, a seven-factor model with a rotation with F peak = +0.2 was selected.It was done on trial and error analysis of the solutions and by comparison of the source profiles with previous studies.The agreement between the calculated and estimated mass concentrations was examined by the percent root mean square error (% RMSE).

Residence time regression analysis
Five-(5) day back trajectories were generated every 1 h using the NOAA Hybrid Single Particle Lagrangian Trajectory (HYSPLIT) model (Draxler, 2007) with the hemispheric Global Data Assimilation System (GDAS) meteorological data as inputs beginning at 00:00 UTC (a total of 2880 trajectory points per day -24 trajectories per day × 5 days backward per trajectory × 24 h per trajectory day.The trajectory starting height was defined at 500 meters based on climatological mean boundary layer heights in the United States, showing that a 500 m starting height would usually be in the boundary layer (Seidel et al., 2012).Trajectories calculated at lower altitude are subject to interference from topography, while trajectories at higher altitude would have been above the mixed layer at times and not representative of the air mass in the mixed layer (Kavouras et al., 2013).The residence time for each 0.5 • × 0.5 • cells was equal to the sum of the number of trajectories points (Ashbaugh et al., 1985;Poirot and Wishinski, 1986).The geographical domain for the trajectory regression analysis covered all cells with estimated residence time at least 72 h over the entire monitoring period (∼1 ‰).
The source regions included four regions (5 • × 5 • ) around the monitoring site to describe local/state contributions and 16 source regions considering that the geographic size of the source regions should increase with distance from the urban area of each source region (Fig. 1).Because of the definition of the local regions, the impact of sources within the Little Rock/North Little Rock metropolitan area cannot be separated from the sources within the state.The definition of a region encompassing Little Rock (e.g. 100 × 100 km centered at the site) would only add a significant amount of ambiguity to the model because every single trajectory ends at the site, thus the residence time would be the same.
The time that an air mass spent over these source regions was computed by summing the residence times of cells that fell within each region.The relationship between particle mass concentrations (y i , in µg m −3 ) and the time that air spends over each region (t j , in hours) was determined using the Tracer Mass Balance (TrMB) model (Eq.7) (Pitchford and Pitchford, 1985;Green et al., 2003;Xu et al., 2006): where C j (in µg m −3 ) is the contribution of j -source region on i-sample, β j (in (µg(m −3 h)) are the regression coefficients of the regions describing the combined outcome of emissions from the area (Q j ); aging and pollutants removed due to gravitational settling, turbulent mix-out, and wet deposition during transport to the receptor site (T j ); and the entrainment of particles from the j -sources to describe the dissociation between the trajectory and the transport (E j ).The units of the Q, T and E factors were in µg(m −3 h) within the region.The intercept, α, accounts for contributions from source regions outside the study domain.We estimated the source contributions by running the TrMB model with and without the intercept.In our study, we included all cells in which the air masses spent at least 72 hours over the study period; thus contributions from sources outside the regions described above may be negligible.Xu et al. (2006) showed that the computed contributions using the TrMB approach with and without the intercept are statistically insignificant and they represent the upper and lower estimates of the contributions, respectively.

Types of fine aerosol
Table 1 shows the values of concentration diagnostic ratios and the types of PM 2.5 aerosol in Little Rock for the 2002-2010 period.Sulfur (S) was present as sulfate (SO 2− 4 ) with sulfate-to-sulfur ratio of 3.59 ±0.35.The NH + 4 /SO 2− 4 molar ratio (2.06 ± 0.04) suggested that sulfate aerosols were in ammonium sulfate ((NH 4 ) 2 SO 4 ) form (Malm et al., 2002).The OC/EC ratios (6.12 ± 2.11) indicated a mixture of primary and secondary organic aerosol from various sources.OC/EC values lower than 1.1 were indicative of primary traffic emissions, while OC/EC values higher than 2.0 have been observed for coal and biomass combustion as well as secondary organic aerosol (Cachier et al., 1989;Chow et al., 1996;Watson et al., 2001;Turpin and Lim, 2001).Soluble potassium (K + ) accounted for 72 % of total K suggesting the significant impact of biomass burning emissions.Salts in soil also contributed about 20 % of K+.Ratios of Al/Si (0.48 ± 0.03) K/Fe (1.06 ± 0.06) and Al/Ca (1.76 ± 0.07) were comparable to those determined for samples collected at the Interagency Monitoring of Protected Visibility Environments (IMPROVE) sites in the Midwest and eastern United States (Hand et al., 2012).
The IMPROVE mass reconstruction scheme was applied to reconstruct aerosol mass into four major species, namely secondary inorganic, organic, light-absorbing carbon, and soil (Eqs.8-11) (Sisler 2000): [Fe] and [Ti] are the elemental carbon, organic carbon, nitrate, ammonium, sulfate, aluminum, silica, calcium, iron and titanium concentrations (in µg m −3 ), respectively.Nitrate may also be associated with coarse particles from neutralization of gas phase nitric acid with sea salt or calcium carbonate.This was typically observed in the western US (Malm et al., 2007).Organic carbon to organic mass (OC-to-OM) conversion factors varied from 1.7 ± 0.2 in IMPROVE background sites to 2.1 ± 0.2 for rural PM 2.5 aerosol to reflect the presence of oxygenated functional organic compounds formed during transport (Turpin and Lim, 2001;Malm and Hand, 2007).We assumed an OC-to-OM conversion factor of 1.6 which is typically used for urban PM 2.5 atmospheric aerosol (Turpin and Lim, 2001).Soil mass concentration [SOIL] was estimated as the sum of the elements present in the soil as oxides.Carbonaceous aerosol (OM and EC) accounted for 56 % of PM 2.5 mass with OM being the abundant component.Sulfate represented 29 % of PM 2.5 mass, while nitrate and mineral dust contributed approximately 8 % and 7 %, respectively.

Sources of fine aerosol
The good agreement between measured PM 2.5 and predicted (using the seven-factor PMF model) PM 2.5 mass concentrations (slope of 0.84 ± 0.02, an intercept of 0.8 ± 0.3 µg m −3 and R = 0.90; Fig. 2) was indicative of a physically meaningful solution explaining most of the variability of fine particles mass (%RMSE of 3.3 % for PM 2.5 mass and less than 15 % for individual chemical species).The difference was attributable to secondary organic aerosol formed from the condensation of biogenic (e.g.isoprene and terpenes) hydrocarbons which cannot be resolved using elemental tracers, ionic composition and total EC/OC concentrations (Hu et al., 2010).The profiles of the seven retained factors and their contributions to 24 h PM 2.5 mass concentrations are shown in Figs. 3 and 4, respectively.The mean contribution of each source on PM 2.5 mass concentration is presented in Table 2.The seven factors were attributed to specific sources of fine particles based on the loadings of individual chemical species.The profiles were comparable to those computed in other midwestern areas (Kim et al., 2005;Lee et al., 2006).
The first factor was assigned to primary particulate matter from gasoline and diesel vehicles with high contributions of OC (total and OC 1 , OC 2 , OC 3 and OC 4 ), EC, S, SO 2− 4 , K, K + and heavy metals (Zn, Cr, Co) typically found in tailpipe emissions (Lough et al., 2005).Soil elements (Al, Si, Fe, Ca) were also associated, indicating the possible influence of contaminated road dust released into the air by the friction between the tires and pavement as a vehicle travels.Primary traffic emissions were responsible for 0.3 ± 0.2 µg m −3 of PM 2.5 mass (Table 2) with no seasonal variation (Fig. 4a).The contributions to PM 2.5 mass were higher than 4 µg m −3 for a limited number of episodes in early summer of 2005 to 2009.
The second factor was attributed to secondary NO − 3 with high contributions to NO − 3 , NH + 4 , OC and SO 2− 4 .The mi-   influences.The contribution of particulate NO − 3 to PM 2.5 mass was 1.1 ± 0.3 µg m −3 , which was comparable to the reconstructed concentration of nitrate particles using the IM-PROVE scheme (Table 1).A clear seasonal profile with the highest being measured in the winter was observed due to lower ambient temperatures promoting the gasto-particles conversion of HNO 3 (Fig. 4b).
The third factor showed strong contributions to OC, SO 2− 4 , Ni, V, Fe, Mn and other heavy metals and was assigned to diesel emissions other than vehicles.This factor also demonstrated high contributions to OC, EC, S, SO 2− 4 , NO − 3 , Ni and V indicating the possible influence of transport.The impact of harbor and shipping emissions on fine particle levels in inland locations is previously observed.In our case, the Gulf Coast in Louisiana and eastern Texas (i.e.Houston) is characterized by increased marine traffic and many industrial operations (i.e. oil refineries).This may include diesel particles from rail engines and coal-fired power plants.It accounted for 1.2 ± 0.2 µg m −3 of PM 2.5 mass (Table 2) with no seasonal variability (Fig. 4c).Episodes of high contributions to PM 2.5 were mostly observed before 2007.
The high concentrations of Na, Na + and Cl on the fourth factor suggested the contribution of aerosol with marine origin (i.e.sea salt) possible from the Gulf Coast.While sea salt particles are typically found in the coarse mode, a fraction of them is associated with fine particles (Teinila et al., 2000).This factor also correlated with OC, EC, S, SO 2 4 and NO − 3 , indicating contamination during transport.This factor contributed, on average, 1.4 ± 0.4 µg m −3 on PM 2.5 mass concentration.Slightly higher contributions were computed in spring than those measured in winter and summer (Fig. 4d).
The fifth factor showed strong contribution to elemental S, SO 2− 4 , NH + 4 and to a lesser extent to EC and OC.It was attributed to secondary sulfate, the primary type of fine particles in Midwest.This source accounted for 37.5 % (4.8 ± 0.4 µg m −3 ) of PM 2.5 .A weak seasonal variability was identified  in [2002][2003][2004][2005] with contributions of up to 30 µg m −3 on PM 2.5 mass.
Mineral dust was identified as a source of PM 2.5 in Little Rock with high contributions to Al, Si, Ca, Fe, and Ti.In addition, fractions of OC, EC and other elements, such as Mg and Mn were also associated with this factor indicating a mixture of road and mineral dust.The contribution of road dust to PM 2.5 mass was 1.0 ± 0.1 µg m −3 with the highest contributions being observed in the summer.Prospero (1999) showed that the transatlantic transport of Saharan dust was manifested by synoptic scale systems affecting large areas in the southeastern US.The maximum dust concentrations (8.4-16.3µg m −3 ) in southern Florida were observed in summer months (June-August).The examination of individual 8-day trajectories showed transport from the Atlantic Ocean, while GOES imagery illustrated dust episodes in 2005 and 2008 over the Canary Islands.As a result, the seasonal trend of mineral dust in Little Rock suggested the possible influence of long-range transport (Fig. 4a and f).
Lastly, biomass burning was identified because of the high contributions to OC, EC and moderate amounts of K, K + , NO − 3 , S and SO 2− 4 .This source contributed 3.0 ± 0.5 µg m −3 to PM 2.5 mass with very little variability (from 1 to 9 µg m −3 ) throughout the year.The absence of a seasonal profile was corroborated by residential wood burning in the winter and the impacts of recreational, prescribed and wildland fires in spring, summer and fall.

Annual trends of fine particles and its sources
Ordinary least squares regression analysis of deseasonalized monthly average PM 2.5 mass and source contributions was used to determine the annual trends without the seasonal component (Jaffe and Ray, 2007).Table 3 presents the annual trends (absolute and relative compared to 2002) of PM 2.5 mass concentrations and the seven source (µg(m −3 yr)).The observed trends were statistically significant for PM 2.5 mass (−0.4µg(m −3 yr); −2.9 %), secondary nitrate (0.09 µg(m −3 yr); −7 %), secondary sulfate (0.2 µg(m −3 yr); −3.4 %) and diesel particles (0.11 µg(m −3 yr); −7.1 %).The mean annual concentrations of NO − 3 and SO 2− 4 dropped from 1.3 µg m −3 in 2002 to 1.0 µg m −3 in 2010 for NO − 3 and, from 6.0 µg m −3 in 2002 to 3.3 µg m −3 in 2010.The observed decrease for SO 2− 4 was comparable to the reductions of SO 2 emissions nationally (from 14,774 tons yr −1 in 2002 to 7478 tons yr −1 in 2010; 49 %) (EPA, 2012b).NO x emissions were reduced by 30 % (from 21 135 tons yr −1 in 2002 to 14 717 tons yr −1 in 2010).Pitchford et al. (2012) attributed the discrepancy between reductions in NO x emissions and particulate NO − 3 levels to the availability of gaseous NH 3 to react with HNO 3 and form NH 4 NO 3 particles and the thermodynamic coupling of the SO 2− 4 and NO − 3 formation mechanisms.They concluded that reductions of particulate SO 2− 4 (due to reduced SO 2 emissions) would cause an increase to particulate NO − 3 , but the overall PM 2.5 would be reduced.The significance of this non-linear relationship may be crucial for southern Midwest because of the high NH 3 emissions in southern states.
For mineral dust (−0.8 %), biomass burning (−1.2 %) and aged/contaminated sea salt (+3.9 %), statistically insignificant trends were computed (from −0.02 to +0.03 www.atmos-chem-phys.net/13/3721/2013/µg(m −3 yr)), because these sources may be influenced by weather patterns and climatology.Figure 5 illustrates the monthly contributions of biomass burning to PM 2.5 (and fitting) and the area burnt (in acres) by wildland fires in the US (data obtained from the National Interagency Fire Center) during the 2002-2010 period.While episodic high contributions from biomass burning were also computed for years with reduced burned areas by wildfires (2003 and 2009 in Fig. 5), on average, the trend of biomass contributions was comparable to that of the burned areas by fires in the US.During the 2004-2007 period, wildfires burned more than 8 million acres per year in the US as compared to less than 5 million acres in 2003 and 2008.This increase was partially attributed to sequencing of El Niño events (wet) and La Niña events (dry) by promoting the growth of fuels and the subsequently dry them out over a period of years (Crimmins and Comrie, 2004;Littell et al., 2009).During the study period, four El Niño and two La Niña events were identified.Weak El Niño events were observed in 2004 and 2006 and moderate El Niño events were identified in 2002 and 2009.In these climate events, the equatorial Pacific Ocean is warmer than usual, altering the weather conditions around the world.These conditions favored larger and destructive wildfires around the country.On the contrary, the colder Equatorial Pacific in La Niña events facilitates colder than normal conditions in northwest US and warmer than normal in southeast US, reducing the fire risks years (Crimmins and Comrie, 2004;Littell et al., 2009).

Regional contributions
Figure 6 shows the spatial variation of residence times in winter, spring, summer and fall.These maps show clear seasonal differences between air masses near the ground.More specifically, winter trajectories originated often from north/northwest over the upper Midwest, North Plains, and traveled through the Mississippi Valley prior to their arrival in central Arkansas.In spring and fall, trajectories demonstrated more local influences by spending most of their time in surrounding states.On the other hand, air masses in the summer extended from northeast to south/southeast covering a larger geographical area from the central midwest to the Gulf of Mexico and Cuba and were not as influenced by topographic restrictions.Thus, these distinctions indicated that air masses in winter may have substantially different compositional characteristics as compared to air masses for the other seasons, and particularly in the summer.
Figure 7 shows the mean (±error) of the contributions of each region on PM 2.5 mass and source contributions for the models with (lower line), and without (upper line) intercept.The differences of the contributions calculated for the two models were negligible (and statistically insignificant within 1 standard error) for NO − 3 , diesel particles, SO 2− 4 , mineral dust and biomass burning.Some differences were observed for primary traffic particle and aged/contaminated sea salt; however, in these cases, the overall estimated contributions for specific sectors were negligible (0.2 ± 0.4 µg m −3 ).These similarities suggested that PM 2.5 mass and its components originated from sources within the selected geographical do- main.The four regions around the site (NW, NE, SW and SE in Fig. 1), the Gulf Coast and southeast US contributed 42 % (5.6 ± 0.9 µg m −3 ), 16 % (2.1 ± 0.3 µg m −3 ) and 10 % (1.4 ± 0.2 µg m −3 ) of PM 2.5 mass, with sources within the southwest sector being responsible for 2.1 ± 0.3 µg m −3 of PM 2.5 .These sectors include the urban areas of Dallas in Texas, Oklahoma City in Oklahoma, Memphis in Tennessee and Baton Rouge in Louisiana and, point sources emitted cumulatively 731,262 tons of PM 2.5 in 2008 (48 % of national annual PM 2.5 emissions from point sources) (EPA, 2012b).Moderate contributions from the upper Midwest (0.8 ± 0.2 µg m −3 ) and central Texas (0.4 ± 0.2 µg m −3 ) were also computed.The same areas were responsible for 60 % of primary traffic particles followed by Pacific Northwest (16 % each).
A slightly modified spatial pattern was observed for secondary nitrate particles with 35 % of that being from the southern sectors (SW and SE), 19 % from Pacific Northwest and 11 % from the upper Midwest.Minor contributions (3-6 %) were also computed for central Texas and the NW sector.For the two sectors south of Little Rock, the increased nitrate contribution was due to the interaction of NH 3 -rich conditions in southern Arkansas and Louisiana with air mass loaded in NO x /gaseous HNO 3 from metropolitan areas in Texas to form NH 4 NO 3 (Pitchford et al., 2012).Furthermore, emissions of other agents such as soil mineral (Ca, K) and sea salt (Na, Mg) off the Gulf Coast and semi-arid areas in northeastern Texas may also neutralize HNO 3 and form stable salts in the aerosol phase.The significant contributions of the Pacific Northwest and North Plains domains on NO − 3 levels in central Arkansas were also attributed to neutralization of HNO 3 as the air masses travel over areas (North Plains, Nebraska, Kansas, Missouri, Minnesota, Iowa) with the highest NH 3 emissions in the US and the induced formation of nitrate particles due to typically lower ambient temperatures measured in northern US, as compared to those measured in the southern Mississippi valley.
About 80 % of secondary sulfates (3.7 ± 0.3 µg m −3 ) (SO 2− 4 ) and 62 % (0.7 ± 0.2 µg m −3 ) of diesel particles (other than diesel vehicles) were allocated to the four sectors around the receptor site, the Gulf Coast and the southeast US.These areas include a large number of SO 2 point sources including coal-fired power plants, oil refineries, offshore oil and natural gas platforms and ports.In 2008, these sources accounted for 55 % of annual SO 2 emissions in the US (33 % southeast US, 8 % NW, 5 % southeast, 4 % NW and 4 % SW).Two electrical power plants are located within Arkansas, emitting approximately 68,000 tons of SO 2 /year (less than 1 % of all SO 2 in the six regions) (one less than 100 km southwest of Little Rock) (EPA, 2012b).The Gulf Coast also appeared to be the primary regional contributor for sea salt (25 %; 0.3 ± 0.2 µg m −3 ) and mineral dust (55 %, 0.4 ± 0.1 µg m −3 ).The area of the Gulf Coast includes thirteen of the twenty busiest ports in the US, with Houston and New Orleans being among the top five (US Army of Engineers, 2010).Marine traffic within 400 km from the land in the western Gulf Coast ranks among the top five busiest areas in the world (more than 1/3 of vessel calls in the US mostly (∼60 %) from tanker and container ships) (Ward and Gallagher, 2010).These ships used oil with high sulfur content from 5 to 4500 ppm.The Gulf Coast region also includes approximately 3000 oil and natural gas platforms producing more than half of crude oil and natural gas production in the US.These platforms emit minor quantities of SO 2 (less than 0.1 % of national SO 2 emissions) but up to 1.5 % and 5.3 % of national NO x and VOCs emissions.
Approximately 45 % (1.3 ± 0.2 µg m −3 ) of biomass burning originated from the eastern (NE and SE) and SW sectors followed by Gulf Coast (9 %, 0.3 ± 0.1 µg m −3 ) and southeast US (8 %, 0.2 ±0.1 µg m −3 ).In these regions, most of the emissions are from residential wood-burning, recreational, agricultural and prescribed burns.Large fires are infrequently observed mostly in years with prolonged drought conditions.The combined contribution of the Pacific Northwest, North Plains and upper Midwest accounted for 20 % of biomass burning emissions (0.6 ± 0.2 µg m −3 ), mostly from large wildfires in Colorado Rockies forests and agricultural fires.Figure 8 shows the cumulative data of thermal anomalies caused by fire incidents as they were detected by Terra and Aqua MODIS satellites during the 2002-2010 period.The size of the aggregating cells is 0.25 degrees per side.These results indicated many temperature anomalies east and south of the receptor site and in southern Alabama and Georgia with weak/moderate thermal signatures which are typical of small-scale prescribed, agricultural and recreational burns.In addition, a smaller number of temperature anomalies associated with higher temperatures were observed in Kansas, Idaho, Oregon and Washington, indicating larger, long-lasting and more intense fire events.
The spatial distribution of regional contributions to PM 2.5 and its sources was comparable to those modeled for the Caney Creek IMPROVE site located in southwest Arkansas.It is found that sulfate from east Texas, southeast US, and upper Midwest; nitrate from central Midwest, and east Texas as well as organic aerosol from Houston and Arkansas were responsible for the observed reduction of visibility (ENVI-RON, 2007).

Conclusions
The sources of fine particles in central Arkansas for the 2002-2010 period were secondary sulfate (36 %), biomass burning (20 %), aged sea salt (10 %), diesel emissions (9 %), secondary nitrate (8 %), mineral dust (6 %), and primary particulate emissions from vehicles (2 %).The remaining unexplained fine particle mass was attributed to secondary organic aerosol.Strong seasonal variabilities were observed for nitrate (high in the winter) and mineral dust (high in early summer) and a weak seasonal profile (high in the summer) for sulfate for the 2002-2005 period.These trends were consistent with those observed in other urban environments and the dependence on meteorology and precursors emissions.The absence of seasonality of biomass burning emissions were due to contributions from residential wood burning in winter and wildland fires for the other seasons.
Fine particle mass concentrations and the contributions of secondary sulfate declined by 0.4 µg m −3 and 0.2 µg m −3 per year, which is consistent with the reductions on SO 2 emission from point and mobile sources.A slower declining trend was also observed for secondary nitrate despite the significant reductions of NO x emissions.This was explained by the abundance of NH 3 in the Midwest that favors the formation of stable forms of ammonium nitrate particles.The annual variability of biomass burning contributions to fine particle mass correlated very well with the burnt area by fires in the US.The impact of transport of fine particles in central Arkansas was assessed by regression against the residence time of air mass in pre-defined regions.The four regions around the receptor site accounted for large fractions of fine particle mass, primary traffic emission, secondary sulfate and biomass burning.The contributions of sources in the southeast US and western Gulf Coast were also significant, accounting for 30 % of secondary sulfate and contaminated sea salt particles.
In this study, we demonstrated that residence time regression analysis may be successfully used to identify the impacts of emissions from specific regions on particle mass and source contributions in urban receptor sites.Through this analysis, the effect of wildfires on PM 2.5 was identified and the role of meteorology and NH 3 emissions was observed.In addition, the impact of shipping activities in the western Gulf of Mexico and coastal cities on fine particles was assessed; a region of potential interest as emissions from point (i.e.power plants) and mobile sources are decreasing while marine traffic and associated emissions of gaseous precursors and particles will grow substantially.

25Figure 1 .
Figure 1.The predefined source regions for Little Rock, Arkansas.

Figure 5 .Fig. 5 .
Figure 5. Annual variation of PM2.5 from biomass burning in Little Rock, Arkansas and burnt 2 area by wildfires in the USA for the 2002-2010 period.3 4

Fig. 6 .
Fig. 6.Percentage of time the air mass parcel is in a horizontal grid cell (0.5 • × 0.5 • ) using backward trajectories at 500 m in winter (a), spring (b), summer (c) and fall (d).

Figure 8 .Fig. 8 .
Figure 8. Cumulative Terra and Aqua MODIS fire and thermal anomalies generated from 2 MODIS data for the monitoring period.The size of the grid cells is 0.5 o per side.3 Fig. 8. Cumulative Terra and Aqua MODIS fire and thermal anomalies generated from MODIS data for the monitoring period.The size of the grid cells is 0.5 • per side.

Table 1 .
Diagnostic ratio of PM 2.5 chemical species and major aerosol types in Little Rock, Arkansas.

Table 2 .
Source contributions to PM 2.5 in Little Rock, Arkansas.

Table 3 .
Absolute and relative (to 2002)annual trends for PM 2.5 and source contributions in Little Rock, Arkansas.