Observed concentrations of aerosols are obtained from the EPA-AQS (http://www.epa.gov/airquality/airdata/). The EPA-AQS daily PM2.5 mass concentrations are available over 1999–2013 at about 1200 sites, and the speciated aerosol concentrations, including those of SO42-, NO3-, NH4+, BC, and OC, are available for 2000–2013 at about 300 sites.

The measurements of aerosol concentrations from the EPA-AQS were carried out at various time intervals (for example, with measurements every 1, 3, or 6 days) at different sites, and there were plenty of missing values at many sites. The observed concentrations are pre-processed following the three steps: (1) For a specific site, the observations are used in our analyses if the site had at least 5 months of observations and there were at least five observation records within each month. (2) The mean seasonal cycle in aerosol concentrations in the months of November–March is removed (the aerosol concentrations without seasonal cycles are defined as Ci,j′=Ci,j-1n∑i=1nCi,j, where Ci,j is the aerosol concentrations in month j of year i, n is the number of years examined). Such a deseasonality approach was used in previous studies that examined the monthly variations in mineral dust aerosol (Cakmur et al., 2001; Mahowald et al., 2003), the decreasing trends in observed PM2.5 concentrations and satellite AOD in the southeastern US over 2000–2009 (Alston et al., 2012), and the monthly variations in global AOD (Li et al., 2013). (3) Since the observed aerosol concentrations exhibited a significant decreasing trend from 1999 to present in the US due to the reductions in emissions of aerosols and aerosol precursors (Alston et al. (2012); http://www3.epa.gov/airtrends/aqtrends.html#comparison), the long-term linear trend in concentrations is identified by the least-square fit and then removed from the wintertime observed concentrations for each site. The EPA-AQS sites with measurements that meet the criteria described in (1) are shown in Fig. 1.

We also examine the impacts of PNA on simulated aerosol concentrations in the US by using the GEOS-Chem model (version 8-2-1; http://acmg.seas.harvard.edu/geos). The GEOS-Chem model is a global chemical transport model driven by the assimilated meteorological fields from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). The version of the model we use has a horizontal resolution of 2∘ × 2.5∘ and 30 hybrid σ-pressure layers from the surface to 0.01 hPa altitude. The model has a fully coupled simulation of tropospheric O3–NOx-VOC (volatile organic compound) chemistry and aerosols including SO42-, NO3-, NH4+, BC, OC (Park et al., 2003, 2004), mineral dust (Fairlie et al., 2007), and sea salt (Alexander et al., 2005). Considering the large uncertainties in chemistry schemes of secondary organic aerosol (SOA), SOA in our simulation is assumed to be the 10 % carbon yield of OC from biogenic terpenes (Park et al., 2003) and 2 % carbon yield of OC from biogenic isoprene (van Donkelaar et al., 2007; Mu and Liao, 2014). We mainly examine simulated anthropogenic aerosols from the GEOS-Chem simulation, since mineral dust concentrations in winter are very small (Malm et al., 2004; Zhang et al., 2013) and sea salt is not a major aerosol species in the US (Malm et al., 2004).

The model uses the advection scheme of Lin and Rood (1996), the deep convective scheme of Zhang and McFarlane (1995), the shallow convection scheme of Hack (1994), the wet deposition scheme of Liu et al. (2001), and the dry deposition scheme of Wesely (1989) and Wang et al. (1998). The instantaneous vertical mixing in the planetary boundary layer (PBL) is accounted for by the TURBDAY mixing scheme (Bey et al., 2001). The average PBL height in wintertime in the US is about 480 m and occupies the lowest 3–6 vertical model layers.

We simulate aerosols for the years 1986–2006 driven by the GEOS-4 reanalysis data. The years of 1986–2006 are chosen for chemistry–aerosol simulation because these are the years that the GEOS-4 data sets are available. Global anthropogenic emissions are from the Global Emissions Inventory Activity (GEIA) (Park et al., 2004, 2006; Zhu et al., 2012; Yang et al., 2015). Anthropogenic emissions over the US are overwritten by the US EPA National Emission Inventory for 1999 (NEI99), which have monthly variations in emissions of precursors including SO2, NOx, and NH3. Monthly biomass burning emissions are taken from the Global Fire Emissions Database version 2 (GFED-2) (Giglio et al., 2006; van der Werf et al., 2006). During the simulation of aerosols for the years 1986–2006, the global anthropogenic and biomass burning emissions of aerosols and aerosol precursors are fixed at year 2005 levels, so that the variations in aerosol concentrations are caused by variations in meteorological parameters alone. non-methane volatile organic compounds (NMVOCs) and NOx from lighting and soil, are allowed to vary over 1986–2006 following the variations in the GEOS-4 meteorological parameters. Biogenic NMVOC emissions are calculated using the module of Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther et al., 2006). Lightning NOx emissions are described by Sauvage et al. (2007) and Murray et al. (2012). Soil NOx emissions are calculated using the algorithm proposed by Yienger and Levy (1995).

The GEOS-Chem simulation of aerosols in the US have been evaluated extensively by previous studies (Park et al., 2003, 2004, 2005; Heald et al., 2006, 2008; van Donkelaar et al., 2006, 2008; Liao et al., 2007; Fu et al., 2009; Drury et al., 2010; Leibensperger et al., 2011; Zhang et al., 2012). These studies have shown that the GEOS-Chem model can capture the magnitudes and distributions of aerosols in the US. The OH concentrations in GEOS-Chem were examined by previous studies (Holmes et al., 2013; Wu et al., 2007) by calculating the methane lifetime, which agreed closely with the lifetime of 11.2 ± 1.3 years constrained by methyl chloroform observations by Prather et al. (2012). Previous studies also showed that aerosol concentrations were not so sensitive to OH concentrations in the GEOS-Chem simulations (Heald et al., 2012).

The PNA index (PNAI) is commonly used to quantify the changes in PNA phase (Wallace and Gutzler, 1981; Leathers et al., 1991). This study follows the definition of PNAI by Leathers et al. (1991). In order to examine the monthly variations in PNA, the mean seasonal cycle of geopotential height at 700 hPa is removed for the months of November, December, January, February, and March (NDJFM) in the studied years. Such a deseasonality approach has been used in the analyses of the growth and decay of the PNA phase in NDJFM (Feldstein, 2002), the development of NAO (Feldstein, 2003), the influence of NAO on precipitation in Europe (Qian et al., 2000), and the variations in Madden–Julian oscillation (Wheeler and Hendon, 2004). If we are concerned with the PNAI during n years, the monthly PNAI in month j (j is one of the 5 months of NDJFM) of year i is calculated by PNAI=13-Zi,j*′47.9∘N,170∘W+Zi,j*′47.9∘N,110∘W-Zi,j*′(29.7∘N,86.3∘W) where Zi,j*′=Zi,j′1n×5∑i=1n∑j=15Zi,j′2 and Zi,j′=Zi,j-1n∑i=1nZi,j. Therefore, Zi,j′ denotes the removal of seasonal cycle, and Zi,j*′ denotes the standardized anomaly of geopotential height at 700 hPa in month j of year i with seasonal cycle removed.

The PNAI is calculated by using both the National Center of Environmental Prediction – Department of Energy Reanalysis 2 data (NCEP-2; horizontal resolution 2.5∘ × 2.5∘ globally; http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis2.html) for the years 1986–2013 (referred to as NCEP2-PNAI) and the GEOS-4 assimilated meteorological data (referred to as GEOS4-PNAI) for 1986–2006 (Fig. 2). Both series of PNA index show strong monthly variations (Fig. 2), and the GEOS4-PNAI agrees with NCEP2-PNAI over 1986–2006 with a high correlation coefficient of 0.99, indicating that the NCEP-2 and GEOS-4 data sets are consistent in representing the monthly variations of PNAI.

There are n×5 PNAI values for n years, since we calculate PNAI for the months of NDJFM of each year. These n×5 PNAI values are classified into three categories for our composite analyses of aerosol concentrations and meteorological parameters: the positive PNA months (PNA+) that are 25 % of the n×5 PNAI months with the highest positive PNAI values, the negative PNA months (PNA-) that are 25 % of the n×5 PNAI months with the highest negative PNAI values, and the rest months that are referred to as the transitional months (Fig. 2).

Figure 4a shows the simulated surface-layer concentrations of PM2.5 (the sum of SO42-, NO3-, NH4+, BC, and OC) and each aerosol species averaged over NDJFM of 1999–2006. These years are selected because they are the common years of model results and EPA-AQS observation data sets. The simulated PM2.5 concentrations were higher in the eastern US than in the western US. The maximum surface PM2.5 concentrations reached 14–16 µg m-3 in Ohio and Pennsylvania. PM2.5 concentrations in the western US were generally less than 4 µg m-3, except for California where PM2.5 concentrations were 2–6 µg m-3. The distribution of SO42- was similar to that of PM2.5, with higher concentrations in the eastern US (1–8 µg m-3) than in the western US (0–3 µg m-3) due to the coal-fired power plants in the Midwest (Park et al., 2006). The concentrations of NO3- and NH4+ were the highest over and near the eastern Midwest, with values of 3–4 and 2–3 µg m-3, respectively. The maximum OC concentrations were simulated to be 2–3 µg m-3 in two regions, from Ohio to Massachusetts and from Alabama to South Carolina. The simulated BC concentrations in the US were 0–1 µg m-3, except for the contiguous New York where BC concentrations reached 1–2 µg m-3. The magnitudes and geographic distributions of SO42-, NO3-, NH4+ concentrations simulated in our work are similar to those simulated by Park et al. (2006) and Pye et al. (2009), and our simulated OC and BC were similar to those reported by Park et al. (2003).

Figure 4b presents the scatter plots of the simulated concentrations versus the EPA-AQS observations. The simulated PM2.5 concentrations had normalized mean bias (NMB =∑m=1M(Sm-Om)/∑m=1MOm×100%, where Sm and Om are the simulated and observed aerosol concentrations in month m, respectively. M is the total number of winter months examined) of -30 % over the US, and the correlation coefficient between simulated and observed PM2.5 concentrations was 0.57. The simulated wintertime SO42-, NO3-, and NH4+ had NMBs of 37, 4, and 26 %, respectively. Similar bias in simulated SO42- in December–January–February (DJF) was reported Park et al. (2006), as the GEOS-Chem model results were compared with observations from the Clean Air Status and Trends Network (CASTNET). The high bias in our simulated NH4+ was associated with the overestimation of SO42-. Our model underestimates PM2.5, SO42-, NO3-, NH4+, OC, and BC in the western US (Fig. 4b), which can be explained in part by the relatively high aerosol concentrations observed for this region from the EPA-AQS. Hand et al. (2014) compared the observed concentrations of aerosols from the EPA-AQS with those from the Interagency Monitoring of Protected Visual Environments (IMPROVE) for 2008–2011, and showed that the ratios of wintertime aerosol concentrations of ammonium sulfate, ammonium nitrate, OC, and BC from the EPA-AQS to those from the IMPROVE were, respectively, 2.3, 7.7, 8.3, and 13.1, as the concentrations were averaged over the western US. Liu et al. (2004) also attributed the high EPA-AQS concentrations in the western US to the relative sparse urban sites that were heavily influenced by strong local sources such as automobiles and wood fires. The low model biases in the western US may also be caused by the biases in emissions in the model.

Since this study is dedicated to examine the influence of PNA phase on the month-to-month variations of aerosol concentrations during wintertime, Fig. 4c compares, for each aerosol species, the deviation from the mean (DM) of observed concentration with that of simulated concentration for each winter month. The DM is defined as DMm=Cm-1M∑m=1MCm/1M∑m=1MCm, where Cm is the simulated average aerosol concentration over the US in month m, and M is the number of winter months examined (we consider the months of NDJFM over 1999–2006 for PM2.5, and the months of NDJFM over 2000–2006 for SO42-, NO3-, NH4+, BC, and OC). The model captures fairly well the peaks and troughs of DMs for PM2.5, SO42-, NO3-, and NH4+, with correlation coefficients of 0.61, 0.45, 0.33, and 0.65, respectively. The model does not capture well the monthly variations of DMs of concentrations of OC and BC, because both anthropogenic and biomass burning emissions are fixed at year 2005 levels during our simulation over 1986–2006 to isolate the impacts of variations in meteorological parameters (PNA phases) on aerosols (see Sect. 2.2). Since the biomass burning emissions, which contribute largely to carbonaceous aerosols, have large interannual variations (Duncan et al., 2003; Generoso et al., 2003; van der Werf et al., 2006), we also show in Fig. 4c the time series of DMs of biomass burning emissions of OC and BC by using biomass burning emissions in NDJFM over 2000–2006 from GFED v2. The correlation coefficients between biomass burning emissions and observed concentrations of OC and BC were 0.36 and 0.34, respectively, indicating that the observed variations in OC and BC were influenced by monthly and interannual variations in biomass burning. We have also calculated the temporal correlation coefficient between EPA-AQS observations and GEOS-Chem model results at each site for each aerosol species (Fig. S4). The temporal correlations were statistically significant for PM2.5, SO42-, NO3-, NH4+ at most sites in the US, especially over and near the eastern Midwest where largest increases in aerosol concentrations were identified in the PNA+ months relative to the PNA- months.

We have performed the GEOS-Chem simulation for the years 1986–2006, in which there were 35 PNA+ and 35 PNA- months (Fig. 2). Figure 5a shows the concentrations of PM2.5 and each aerosols species (SO42-, NO3-, NH4+, BC, and OC) averaged over the PNA- months of 1986–2006. The magnitudes and geographic distributions of aerosol concentrations in PNA- months were similar to those averaged over NDJFM of years 1999–2006 in Fig. 4.

The simulated absolute and relative differences in aerosol concentrations between PNA+ and PNA- months are shown in Fig. 5b and c. The PM2.5 concentrations over the US are simulated to increase in PNA+ relative to PNA- months. The maximum enhancement in PM2.5 concentrations in PNA+ months was 1.8–2.4 µg m-3 (20–40 %), located in the juncture of Tennessee and Arkansas. Note that the pattern of simulated differences in PM2.5 between PNA+ and PNA-months was similar to that of observations (Fig. 3), except that the simulated differences were not large in California, mainly due to the underestimation of OC in California as compared to EPA-AQS data (Fig. 4b). The simulated PM2.5 concentrations were higher by 0.6 µg m-3 (12.2 %), 0.3 µg m-3 (14.0 %), and 0.9 µg m-3 (10.8 %) over the whole of western and eastern US, respectively, in the PNA+ months than in PNA- months (Table 2). The simulated relative difference in PM2.5 was close to that from observations (Table 1) in the western US, but the simulated relative differences in PM2.5 were larger than those from observations in the eastern and whole of US.

Figure 5 also shows the differences in simulated surface-layer concentrations of SO42-, NO3-, and NH4+ between the PNA+ and the PNA- months. The differences in concentrations of SO42- were larger in the western than in the eastern US, with maximum enhancements of 0.4–0.8 µg m-3 (30–50 %) over the western north-central states (South Dakota, Nebraska, Minnesota, Iowa, and Missouri). The differences in concentrations of NO3- and NH4+ between PNA+ and PNA- months had similar geographical patterns, with increases in concentrations in a large fraction of the eastern US and over a belt region along the Rocky Mountains in the western US. The increases in concentrations of NO3- and NH4+ over the eastern US in PNA+ months relative to PNA- months agreed very well with those seen in observations in most regions of the US (Fig. 3). The model underestimates the differences in NO3- in California and the southeastern US, because the model does not capture well the temporal variations of NO3- in these two regions (Fig. S4). The differences in concentrations of SO42-, NO3-, and NH4+ in the eastern US between the PNA+ and PNA- months were 0.2 µg m-3 (4.0 %), 0.4 µg m-3 (33.5 %), and 0.2 µg m-3 (13.2 %), respectively (Table 2).

The differences in concentrations of OC and BC between PNA+ and PNA- months had similar geographical patterns, with large increases in concentrations over and near the eastern Midwest and the region from northwestern US to Texas. The maximum differences reached 0.2–0.4 µg m-3 (10–20 %) and 0.1–0.2 µg m-3 (20–30 %) in Illinois, Indiana, and Ohio for OC and BC, respectively. The magnitudes of the differences in OC and BC were statistically significant but were smaller than the observations (Tables 1 and 2). The absolute differences in OC were less than 0.1 µg m-3 in the western, eastern, and whole of US due to the underestimation of OC in the simulation.

In summary, model results agreed with observations in that the concentrations of all aerosol species of SO42-, NO3-, NH4+, BC, and OC averaged over the US were higher in PNA+ months than in PNA- months. Relative to the PNA- months, the average concentration of PM2.5 over the US was higher by about 8.7 % (12.2 %) based on observed (simulated) concentrations. Furthermore, simulated geographical patterns of the differences in PM2.5 and each aerosol species between the PNA+ and PNA- months were similar to those seen in observations in most areas of the US (except for California), with the largest increases in aerosol concentrations in PNA+ months over and near the eastern Midwest.

The transboundary transport of pollutants to and from the US depends largely on winds in the free troposphere (Liang et al., 2004). Figure 6 shows the horizontal winds at 700 hPa averaged over the winter months of NDJFM of 1986–2006 and the corresponding differences between PNA+ and PNA- months on the basis of the GEOS-4 meteorological fields. Strong westerlies prevailed over the US in wintertime (Fig. 6a). Relative to the PNA- months, anomalous northeasterlies occurred over a large fraction of US in the PNA+ months. Anomalous anti-cyclonic circulation occurred near the northwestern US and anomalous cyclonic circulation occurred near the southeastern US (Fig. 6b), corresponding to the large positive and negative differences in geopotential height in these two regions (see Fig. S1), respectively.

We also calculate mass fluxes of PM2.5 at the four lateral boundaries (from the surface to 250 hPa) of the US for different PNA phases (Table 3). The domain of the box to represent the US is (75–120∘ W, 28–49∘ N), as shown in Fig. 6b. For PM2.5 in wintertime, the inflow from the west boundary and the outflow from the east boundary had the largest absolute values (Table 3). Relative to the PNA- months, the inflow from the west boundary and the outflow from the east boundary in PNA+ months exhibited reductions of 16.1 and 13.5 kg s-1, respectively (Table 3). The inflow flux from south boundary decreased by 15.8 kg s-1, and the inflow flux from north boundary increased by 31.7 kg s-1, leading to a net increase of inflow flux of 13.3 kg s-1 in PNA+ months. Therefore, the transboundary transport has an overall effect of increasing PM2.5 aerosols in the US in PNA+ months relative to PNA- months. The relative change in net flux was 9.9 % ((PNA+ minus PNA-) × 100 % / PNA-) (Table 3), coinciding well with the enhancement of 12.2 % in surface-layer PM2.5 concentration averaged over the US (Table 2). Note that because the GEOS-Chem model underestimates PM2.5 concentrations in the western US (Fig. 4b), the net outflow flux from the selected box might have been underestimated, but this should not compromise our conclusions about the relative differences in net flux between PNA+ and PNA- phases.

The PNA pattern is also associated with variations in meteorological variables such as temperature (Leathers et al., 1991; Konrad II, 1998; Notaro et al., 2006; Knight et al., 2008; Liu et al., 2015; Ning and Bradley, 2014, 2015), precipitation (Leathers et al., 1991; Henderson and Robinson, 1994; Coleman and Rogers, 2003; Notaro et al., 2006; Archambault et al., 2008; Myoung and Deng, 2009; Ning and Bradley, 2014, 2015; Wise et al., 2015), and humidity (Sheridan, 2003; Coleman and Rogers, 2003; Knight et al., 2008) in US, which are expected to influence aerosol concentrations within the US through chemical reactions, transport, and deposition.

Figure 7 shows the composite differences, between the PNA+ and PNA- months, in surface air temperature (T), precipitation rate (PR), relative humidity (RH), surface wind speed (WS), and planetary boundary layer height (PBLH), based on the reanalyzed GEOS-4 data sets. Relative to PNA- months, temperatures were higher by 1–3 K over the northwestern US and lower by 1–4 K in the southeastern region. Such geographic distributions of temperature anomalies were attributed to the maritime warm air in the northwestern US accompanied by the enhanced tropospheric geopotential height in North America (Leathers et al., 1991; Sheridan, 2003; Liu et al., 2015; Ning and Bradley, 2015) (see also Fig. S1d) and the more frequent outbreak of cold air in southeastern US accompanied by the depressed geopotential height (Konrad II, 1998; Liu et al., 2015) (see also Fig. S1d). The differences in precipitation between PNA+ and PNA- months reached -1.6 to -2.4 mm day-1 (-32 to -48 %) over and near the eastern Midwest, -2.4 to -3.2 mm day-1 (-48 to -64 %) in the northwestern US, and 1.6–2.4 mm day-1 (16–32 %) in the southeastern US. These effects of PNA on precipitation were similar to those obtained from wintertime station data by Leathers et al. (1991), Coleman and Rogers (2003), and Wise et al. (2015). With respect to RH, the values in the eastern US were generally lower in the PNA+ months than in PNA- months, as a result of the reduced moisture flux from the Gulf of Mexico to the eastern US (Coleman and Rogers, 2003), where RH showed maximum reduction of -3 to -9 %. The enhancement of RH of up to 6-9 % in Texas, Oklahoma, and New Mexico was due to the anomalous easterlies over the south central US (see Fig. 7), which diminished the influence of the dry air from the deserts of the southwestern US and northwestern Mexico in PNA+ months (Sheridan, 2003). The surface WS showed reductions in PNA+ months relative to PNA- months in two regions, one was the belt region along the Rocky Mountains from the northwestern US to Texas (with the maximum reductions of 1.5–2.0 m s-1 (48–64 %)) and the other region, with a northeast–southwest orientation, was from Ohio to Louisiana (with maximum reductions of 0.5–1.0 m s-1 (16–32 %)). The differences in PBLH between PNA+ and PNA- months were statistically significant in the western US and in the belt region from the northeastern US to the eastern Midwest, with maximum reductions in PBLH of 75–100 m (15–20 %) and 75–100 m (10–15 %), respectively. The changes of PBLH were possibly associated with the frequent outbreak of cold air in the US in PNA+ phases (Leathers et al., 1991; Archambault et al., 2010), which induced cool air at the surface and clouds over the intermountain and the eastern Midwest regions (Sheridan, 2003).

The comparisons of Fig. 5b with Fig. 7a indicate that the increases in PM2.5 concentrations over and near the eastern Midwest in PNA+ months relative to the PNA- months can be attributed to the decreases in PR, WS, and PBLH in these locations, since the changes in these three variables depressed wet deposition, local horizontal diffusion, and vertical diffusion of the surface aerosols, respectively. The increases in SO42- in the western US (Fig. 5b) corresponded to the decreases in PR and PBLH and the increased temperature. The increases in temperature enhance chemical reaction rates (Aw and Kleeman, 2003; Dawson et al., 2007; Kleeman, 2008). In the eastern US, although PR, WS, and PBLH decreased over and near the eastern Midwest, the cooling in the eastern US might have offset the effects by PR, WS, and PBLH, inducing practically no changes in SO42- in the eastern US. The large increases in NO3- and NH4+ in the southeastern US (Fig. 5b) can be attributed to the reduced surface temperature, which was favorable to the wintertime formation of NO3- and NH4+ (Dawson et al., 2007). The differences in concentrations of OC and BC between PNA+ and PNA- corresponded well with the reduced PR, WS, and PBLH.

In order to quantify the impacts of anomalies in meteorological parameters driven by PNA on concentrations of different aerosol species, the pattern correlation coefficients (PCC; http://glossary.ametsoc.org/wiki/Pattern_correlation) are calculated and shown in Table 4. These pattern correlation coefficients denote the relationship between the geographical distribution of anomalies of each of T, PR, RH, WS, and PBLH (Fig. 7b) and that of the differences in concentration of each aerosol species between PNA+ and PNA- months (Fig. 5c). As shown in Table 4, over the whole US, the PNA influenced SO42- concentrations mainly through changes in PBLH, PR, and T, with the highest PCC values of -0.43, -0.38, and +0.26, respectively. For NO3-, the PNA-induced variations in temperature had a strong negative correlation (PCC = -0.59) with the PNA-induced differences in concentrations, indicating that surface temperature was the dominant meteorological factor to influence NO3- concentrations. For NH4+, OC, and BC, PR and PBLH were the two variables that had the largest negative PCC values (Table 4).