The Greater Yellowstone Area (GYA) (see Fig. 1) of the United States, with Yellowstone National Park (YNP) and Grand Teton National Park (GTNP) at its core, is one of the largest remaining intact ecosystems in the northern temperate zone and features diverse wildlife, alpine lakes, forests, and geologic wonders (Keiter and Boyce, 1994; NPS, 2017). Increasing concentrations of reactive nitrogen (Nr) compounds in air, rain, and snowpack samples over the GYA have been reported in the past 30 years and linked to Nr emissions from human activities (Clow et al., 2003; Blett et al., 2011; IMPROVE, 2011; Sullivan et al., 2011; USGS, 2014; NADP, 2016; Nanus et al., 2017; also, see Fig. S1). The inorganic wet Nr deposition rates measured at high-elevation National Trends Network (NTN) sites within the GYA in 2010 were 2.5–3.5 kg N ha-1 yr-1, compared with 1.5–2.5 kg N ha-1 yr-1 in 2000 (NADP, 2016). This is relevant to the long-term conservation of the area because as Nr deposition levels increase, they can cross critical load (CL) thresholds, at which negative effects to sensitive ecosystem components can occur (Porter et al., 2005; Pardo et al., 2011). Additional concerns posed by enhanced Nr deposition include lake acidification, loss of lichen biodiversity, and eutrophication (Baron, 2006; Blett et al., 2011; NADP, 2016). While ecosystem changes due to excess Nr deposition over Class I areas including the GYA have been documented (e.g., Baron et al., 2011; Saros et al., 2011; Sullivan et al., 2011; Spaulding et al., 2015; Nanus et al., 2017), the origins, chemical composition, and spatial and temporal changes in the deposition over this region are not as well understood.

Total Nr is a mix of oxidized and reduced inorganic nitrogen (N) and organic N compounds that are chemically and biologically active in the Earth's biosphere and atmosphere and deposited through wet and dry processes. These compounds arise from a variety of sources, with inorganic oxidized N primarily emitted as nitrogen oxides (NOx) from fossil fuel combustion, with approximately 25 % from power plants, 50 % from automobiles, and 10 % from other mobile sources, based on annual county-level estimations (EPA, 2015a). Atmospheric reactions of NOx result in nitric acid (HNO3), particulate nitrate (PNO3), and other compounds. Reduced N arises primarily from ammonia (NH3) gas emissions from agricultural activities, which can react with acidic aerosols to form ammonium (NH4+) compounds (Galloway et al., 2004). Mobile sources are also an important source of NH3 and can be the primary emitter in urban areas (Sun et al., 2014, 2017). Emissions from this sector have large uncertainties and a recent study suggests that on-road NH3 emissions in the 2011 National Emissions Inventory (NEI) were underestimated by a factor of 2.9 (Fenn et al., 2018). There are hundreds of organic N compounds, including reduced (e.g., amines) and oxidized forms (e.g., alkyl nitrates). Sources of organic N are less well known, but increasing evidence suggests that biomass burning and agriculture (AG) are significant contributors, as are atmospheric reactions of NOx with volatile organic compounds (Cape et al., 2011; Reay et al., 2012). With the steady decline of NOx emissions in the United States during past decades as a result of the implementation of the Clean Air Act, the importance of reduced N to the total N deposition budget has increased (Li et al., 2016). Specific to the GYA, local anthropogenic emissions are small, but upwind sources, including agricultural activities in the Snake River valley and northern Utah, wildfires throughout the western United States, energy development in the Upper Green River basin, and anthropogenic activities at urban centers such as Salt Lake City, are larger and likely to be significant contributors to regional N emissions (Prenni et al., 2014).

To better understand the levels and composition of the Nr compounds deposited in the GYA and to help guide strategies to reduce N deposition, the National Park Service (NPS) initiated the Grand Teton Reactive Nitrogen Deposition Study (GrandTReNDS), which included spatially and temporally detailed measurements of N compounds during April to September 2011 (Benedict et al., 2013a; Prenni et al., 2014). It was found that during summer months at the high-elevation sites (e.g., Grand Targhee; see Fig. 2), 62 % of the N deposition was from reduced N and about equally split between dry and wet deposition, and oxidized N only accounted for 27 % of the N deposition budget, with the remaining N in the form of wet-deposited, organic N. Study findings indicate a significant west-to-east gradient in atmospheric NH3 concentrations, with higher concentrations west of the Teton mountain range. Concurrently measured concentrations of HNO3, PM2.5 (particulate matter with an aerodynamic diameter of less than 2.5 µm) nitrate, and NH4+ showed relatively small west-to-east gradients inside GTNP (Benedict et al., 2013a; Prenni et al., 2014).

The origins of Nr transported to the GYA and other remote locations in the western United States have been examined in past modeling studies. Back trajectory analyses have shown that air mass transport to GTNP is predominantly from the west through the Snake River valley and from the southwest through northern Utah (Prenni et al., 2014). Zhang et al. (2012) applied the global chemical transport model (CTM) GEOS-Chem (Bey et al., 2001) using zero-out sensitivity simulations and found that in 2006 natural sources, including lightning and wildfires, contributed more than 10 % of the total Nr deposition over the Teton area. Lee et al. (2016) used the adjoint version of GEOS-Chem to quantify the sources of Nr deposition in eight selected federal Class I areas in 2010 and found a nonnegligible footprint (>20 %) of Nr deposition in the western United States, including GTNP and Rocky Mountain National Park (RMNP), attributed to long-range transport from sources in California, especially during summertime. Mobile NOx and livestock NH3 were also found to be major sources of Nr deposition in GTNP. Similar modeling studies focusing on RMNP also suggested the important contributions of distant sources including those from California and other counties and the fact that the contributions from sources of reduced Nr were larger than those from sources of oxidized Nr (Thompson et al., 2015; Malm et al., 2016).

In this work, we add to the growing body of Nr modeling source apportionment studies by conducting a detailed analysis using the Particulate Source Apportionment Technology (PSAT) module within the CAMx (Comprehensive Air Quality Model with extensions) (ENVIRON, 2014) CTM to quantify the seasonal contributions from different source regions and source sectors to Nr throughout the GYA. Compared with previous Nr deposition simulation studies in United States, this work uses tagged reactive tracers to attribute the contributions from four designated emission sectors and 27 designated emission regions to Nr deposition in the GYA with a much higher horizontal grid resolution (12 km) and an up-to-date emission inventory instead of using a zero-out approach (e.g., Zhang et al., 2012) or an adjoint model (e.g., Lee et al., 2016). The model simulation of Nr and its constituents were first evaluated against routine measured data as well as the unique data measured during the GrandTReNDS campaign period (Benedict et al., 2013a; Prenni et al., 2014). Nr deposition from CAMx simulations was also compared with total deposition maps (TDEPs), which were developed for deposition trend analysis and ecological impact assessment (Schwede and Lear, 2014). The detailed source apportionment results are presented here, focusing on seasonal variations and the relative importance to CL exceedance in sensitive ecosystems within the GYA. The discussion of identified model bias and uncertainties in the interpretation of source apportionment results, including the model lateral boundary conditions, the impact of model precipitation to wet deposition simulation, and the impact of ammonium dry deposition velocity on dry deposition are also presented.

Modeling simulations for 2011 were conducted using the CAMx version 6.10 (ENVIRON, 2014) with two nested grids. The outer domain (36 km) covered the contiguous United States (CONUS), as well as portions of Canada and Mexico, while the inner domain (12 km) encompassed the western United States and focused on states within the Western Regional Air Partnership (WRAP) (see Fig. 1).

The hourly meteorological inputs for 2011 were generated by the Weather Research and Forecasting (WRF) model (WRF-ARW, version 3.5.1) (Skamarock et al., 2008) and were obtained from the Intermountain West Data Warehouse (IWDW) (http://views.cira.colostate.edu/tsdw/, last access: 30 August 2018). This meteorological simulation performed in comparison to other recent prognostic model applications used in air quality planning (UNC-Chapel Hill and ENVIRON, 2014a).

The emission inventory used by CAMx was primarily derived from the 2011 NEI version 2 (NEI2011v2) (EPA, 2015b) with the Sparse Matrix Operator Kernel Emissions (SMOKE) processing system version 3.0 (Houyoux et al., 2002) for anthropogenic emissions, the Model of Emissions of Gases and Aerosols from Nature (MEGAN) version 2.10 (Guenther et al., 2012) for biogenic emissions, and the WRAP Windblown Dust Model (WRAP-WBD) to estimate wind-driven dust emissions (UNC-Chapel Hill and ENVIRON, 2014b). Emissions from the oil and gas (OG) sector were further updated by the IWDW to represent the best-available inventory for OG activity in the western United States at the time of modeling (UNC-Chapel Hill and ENVIRON, 2014b). The emissions for fire activities (Fire) include agricultural fires, prescribed fires, and wildfires and were generated by the Particulate Matter Deterministic and Empirical Tagging and Assessment of Impacts on Levels (PMDETAIL) study (Moore et al., 2012). PMDETAIL developed 2011 fire emissions using satellite data, ground detection, and burn scar and estimated the plume rise, depending on fire size and type. The hourly, nonsurface fire emissions were allocated to the proper CAMx vertical layers based on the model-predicted planetary boundary layer (PBL) height and the spanning of the plume top and bottom above the ground (Mavko and Morris, 2013).

The boundary conditions for the 36 km domain were estimated from a 2011 global model run using the Model for Ozone and Related chemical Tracers (MOZART) version 4.6 (Emmons et al., 2010). The simulation year of 2011 was preceded by 15 days of spin-up time to minimize the effects of initial conditions. A more-detailed description of the WRF–SMOKE–CAMx modeling platform applied in this study is summarized in Table S1 in the Supplement as well as the 2011 Three-State Air Quality Study (3SAQS) (UNC-Chapel Hill and ENVIRON, 2014b).

For the source apportionment estimates, 27 source regions (Fig. 1), as well as the lateral boundary conditions (BCs), were “tagged” in the CAMx PSAT simulation. In addition, the emissions for each region were further subdivided into four source sectors: (1) AG, (2) OG, (3) Fire, including wildfires and prescribed fires, and (4) the remaining sources labeled as Other. The Other source sector primarily comes from mobile and large point sources, with smaller contributions from natural sources such as lightning. Table S2 provides the annual NH3 and NOx emissions used in this modeling study with a breakdown by tagged source regions and source sectors. Figure 2 provides the annual emissions of NH3 in the inner 12 km domain as well as the monitoring sites or receptor areas used for the model evaluation and analysis. For NH3 emissions, the AG sector contributed 84.1 % of the total emissions within the 12 km domain, while the OG, Fire, and Other sectors contributed 0.1 %, 4.5 %, and 11.4 %, respectively (Table S2). In the Snake River valley, the AG sector emissions dominate the emission budget. For NOx emissions, the contribution rankings from the four tagged emission sources are Other (83.8 %), OG (12.8 %), Fire (3.2 %), and AG (0 %). The regions were selected to highlight important source sector contributions to Nr deposition in the GYA. For example, the state of Wyoming was partitioned into five regions (YNP, Jackson, Upper Green River, eastern Wyoming, and western Wyoming) to differentiate the possible source impacts from urban activity in Jackson from energy development in southwestern Wyoming (Blett et al., 2011; NPS, 2017). Significant agricultural operations in the Snake River valley in Idaho, northern Utah, and northeastern Colorado were tagged due to their high ammonia emissions (see Fig. 2) associated with fertilizer application and confined animal feeding operations (Fenn et al., 2003; Clarisse et al., 2009; Prenni et al., 2014). Lastly, wildfires are episodic events (http://wrapfets.org/map.cfm, last access: 30 August 2018) that can have large intermittent contributions to Nr deposition, but they can mask important contributions from other sources that are significant in nonfire years.

CAMx–PSAT treats nitrogen-containing compounds as one of seven species: gaseous NH3; particulate ammonium (PNH4); reactive gaseous nitrogen (RGN), which includes primary emissions of NOx, nitrous acid (HONO), nitrate radical (NO3), and dinitrogen pentoxide (N2O5); gaseous nitric acid (HNO3); gaseous peroxy nitrogen (TPN), including peroxyacetyl nitrate (PAN) and peroxynitric acid (PNA); gas-phase organic nitrate (NTR); and particulate nitrate (PNO3). PSAT maintains the source-group identity (i.e., source region and source sector) by apportioning the secondary species to the precursor emissions (ENVIRON, 2014). In the source apportionment comparison results, we report the reduced Nr deposition as the sum of NH3 and PNH4 and the oxidized Nr deposition as the sum of RGN, HNO3, PNO3, TPN, and NTR in units of kg N ha-1.

Acceptable model performance of the regional air quality modeling system is a prerequisite for a credible source apportionment interpretation (Boylan and Russell, 2006; EPA, 2014; Emery et al., 2017). In this work, the CAMx simulation was extensively evaluated against routine monitoring data as well as data collected in the GrandTReNDS special field study (Benedict et al., 2013a; Prenni et al., 2014) and against the nitrogen deposition estimates from the National Atmospheric Deposition Program (NADP, http://nadp.slh.wisc.edu/, last access: 30 August 2018) TDEP hybrid modeling results (Schwede and Lear, 2014). Performance metrics recommended by the EPA's modeling guidance for ozone, PM2.5, and regional haze attainment demonstrations (Yu et al., 2006; EPA, 2014) were used (see Table 1).

The variables and routine monitoring networks used in the model evaluation were NH3 concentrations from the Ammonia Monitoring Network (AMoN) (http://nadp.sws.uiuc.edu/AMoN/, last access: 30 August 2018); nitric acid (HNO3), PNO3, and PNH4 concentrations as well as estimated dry deposition fluxes from the Clean Air Status and Trends Network (CASTNet) (https://www.epa.gov/castnet, last access: 30 August 2018); PNO3 and PNH4 concentrations from the Chemical Speciation Network (CSN) (https://www3.epa.gov/ttnamti1/speciepg.html, last access: 30 August 2018); PNO3 concentrations from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network; and wet-deposited inorganic oxidized (NO3-) and reduced (NH4+) nitrogen and associated precipitation rates from the NADP NTN sites. Each network had a unique sampling frequency and duration (Table 1). The hourly CAMx outputs were aggregated to match the timescales of the measured data. All measurement data flagged as questionable, either due to mal-operation or to insufficient samples to calculate representative values, were excluded from the analysis. Table 1 reports the percentage of valid measurements used for statistical analysis during evaluation time. For most of the N species, the percentage of valid samples is more than 80 %. In general, the CASTNet, IMPROVE, AMoN, and NADP networks sample in rural areas, while the data from the CSN network primarily represent the air quality in urban and suburban settings. Although organic N species were also measured in the GrandTReNDS campaign, we focus on the inorganic N budget comparison, given the large uncertainties for organic N prediction (Jickells et al., 2013) and its incomplete treatment in the model's chemical mechanism. For example, the modeling system does not account for primary emissions of organic N compounds but does include the formation of organic N from the alkylperoxy radical and secondary alkoxy radical (ENVIRON, 2014).

The seasonal modeled Nr deposition budgets averaged over the GYA are presented in Fig. 6. As shown, the total Nr deposition rates peaked in the summer (1.12 kg N ha-1 season-1) with somewhat lower rates in the spring (0.91 kg N ha-1 season-1) and fall (0.81 kg N ha-1 season-1) and with winter rates (0.29 kg N ha-1 season-1) being about a factor of 3 smaller than in the other seasons. These patterns are similar to the measured and modeled data presented in Fig. 3. In total, the annual model Nr deposition was 3.13 kg N ha-1 yr-1, with wet deposition accounting for only ∼40 %. Reduced N compounds were the largest contributor, except in winter, which is consistent with past studies (Li et al., 2017). Contributions from organic N compounds are not measured in routine monitoring programs. Together they accounted for <10 % of the Nr deposition, suggesting a small but significant contribution. This is also less than has been measured in field studies conducted at GTNP (Benedict et al., 2013a; Prenni et al., 2014) and in RMNP (Benedict et al., 2013b), where the GrandTReNDS study showed on average a 8 %–18 % contribution from organic N to total Nr deposition budgets during the whole campaign period and up to 39 % in June at the NOAA Climate Station site (Fig. 7 in Benedict et al., 2013a).

The relative contributions from the four modeled source sectors (AG, OG, Fire, and Other) and the BCs averaged over the GYA are presented in Fig. 7, while Fig. 8 presents the seasonal and spatial patterns of their contributions over the GYA. As shown in Table S2, the AG source sector was composed of almost all reduced N compounds (>99 %), while the Other source sector was primarily composed (97 %) of oxidized N compounds, with about 88 % originating from anthropogenic combustion emissions, including point and mobile sources, and the remainder originating from the natural emissions from soil and lightning. Contributions from the Fire and the BCs sectors were more evenly split between reduced and oxidized N contributions.

Reduced N from the AG source sector was the largest contributor in the spring (40 %) and fall (41 %) seasons, while oxidized N from the Other source sector was the largest contributor in summer (29 %) and winter (44 %) (Fig. 7). In terms of geographic impact (Fig. 8), AG emissions contributed as much as 80 % of the total Nr deposition in the western portion of the GYA during the spring and fall, which was associated with the outflow from the Snake River valley. In the model, NH3 from regional agriculture activities was treated as being from surface area sources (i.e., emitted into the first model layer, which is approximately 24 m thick). These low-level emissions can be quickly deposited to the surface unless there is sufficient vertical mixing to inject the NH3 into the upper levels of the atmosphere (Ferm, 1998; Fenn et al., 2003) or if it reacted with acidic gases and aerosols. Consequently, it is likely that a higher fraction of the modeled NH3 emissions from the AG sector will be deposited in the lower-elevation periphery of the GYA near the agricultural lands and not impact the more distant mountainous interior (Fig. 2). The incorporation of the bidirectional NH3 flux could extend the NH3 emission footprint (Bash et al., 2013; Zhu et al., 2015).

The OG source sector contributed only about 1 % of the total Nr deposition over the GYA, with contributions of 10 % or more occurring during winter in the southeastern corner of the GYA where nearby OG activity in the Jonah Field and Pinedale Anticline was taking place. Wildfires are episodic and their locations and magnitudes vary significantly from year to year (Westerling and Swetnam, 2003; Parisien et al., 2012). In 2011, fire events contributed on average 18 % of the total Nr deposition in the GYA. Most of the wildfire happened in summer and fall, while agriculture and prescribed burning occurred in winter and spring. Near the fire activities, the contribution to Nr deposition could be more than 90 %, as seen in Fig. 8. The footprint of fire emission impacts depends on the simulated injection height of the fire plumes. The emissions from fires that occurred within the GYA during the summer and fall likely remained within the mixed layer and had less of a chance to be transported far downwind to impact more distant areas (Fig. S4). The Other source sector had relatively uniform contributions throughout the GYA, indicative of contributions from regional sources. The Other sector accounted for 26 % of the annual Nr deposition, with its largest absolute contributions in the summer, but had the highest relative contribution in the winter at 44 % when AG contributions were at their lowest. Finally, the BCs had high contributions, often over 20 %, with the highest contributions occurring in the northern part of the GYA and at higher-elevation sites.

The seasonal contributions from the modeled source regions and sectors to the average total Nr deposition over the GYA are summarized in Fig. 9. As shown, the Snake River valley in Idaho was the largest contributor (in all seasons), with annual mean contributions of 38 % and a maximum contribution of 43 % in fall. Most (74 %) of the Nr from this region was from the AG source sector and was composed of reduced N (Table S4). The next four largest contributors, on average, were the BCs (21 %), western Wyoming (8 %), California (7 %), and northern Utah (6 %). The impact of emissions from Wyoming on the GYA during summer and fall (14 % and 16 %, respectively) was more pronounced than winter and spring (5 % and 7 %, respectively). The contributions of long-range transport from California and the BCs were higher during spring and winter.

Seasonal source apportionment results of the average dry and wet Nr deposition over the GYA are shown in Figs. 7 and 9. Compared to the results for total Nr deposition, the dry Nr deposition had higher contributions from closer sources, such as the Snake River valley (46 % for dry vs. 38 % for total), with emissions primarily from AG sources. Similarly, contributions to dry Nr deposition from Wyoming were 15 % compared to 12 % for total Nr deposition and ranked as the second-largest contributor. The contributions from distant source regions decreased. For example, the BCs decreased from 21 % for total Nr deposition to 12 % for dry Nr deposition.

The opposite pattern is seen for wet Nr deposition, where the contributions from the distant source regions increased relative to the neighboring ones. The annual contributions from the BCs increased to 34 % and peaked in spring and summer at 37 %, associated with higher precipitation amounts than the other two seasons. Annual contributions from sources in California (10 %) and Utah (8 %) surpassed Wyoming (7 %). Furthermore, the seasonal variation for wet Nr deposition was different from dry and total Nr deposition, with the highest deposition rates occurring in spring as opposed to summer.

The GYA has been the focus of several ecological assessments of the response of ecosystems to changing Nr deposition levels (Spaulding et al., 2015; Nanus et al., 2017). Figure 10 presents the source attribution results for 10 sites within the GYA where either ecosystem response studies or deposition monitoring has been conducted for lichen diversity, alpine lake chemistry, and snowpack analysis. In Table 2, the CL values are provided as a range of lower-end and upper-end estimates of the annual total inorganic Nr deposition values (Lynch et al., 2015) with confidence levels (Pardo et al., 2011). The simulated Nr deposition exceeded the lower CL values at three of the 10 sites, specifically, Holly Lake, Pinedale, and Tower Falls. Comparatively, the 2011 TDEP Nr deposition results exceeded the CL at 6 out of 10 sites (Black Joe Lake, Biscuit Basin, Holly Lake, Jedediah Smith Wilderness, Pinedale, and Twin Island). As shown in Fig. 10, the sites that exceeded the CL tend to be in high-alpine locations, with four of these sites on the western slope of the mountains, which are downwind of the Snake River valley. These results are consistent with another modeling study to access CL exceedances in Class I areas using GEOS-Chem (Ellis et al., 2013; Lee et al., 2016). In addition, in one study (Nanus et al., 2017) over 30 % of the GYA was estimated to potentially exceed lower Nr deposition CL thresholds, with the greatest impacts in sensitive high-elevation basins, including areas within national parks and wildernesses.

In terms of emission sectors and source regions contributing to the total annual Nr deposition at CL exceedance sites, emission sources from the Snake River valley were the largest contributors (27 %–32 %), and AG emissions were the largest source of this subset. The next three largest contributors were transport from the BCs (23 %–25 %) and emissions from northern Utah (8 %–15 %) and California (7 %–8 %). Wyoming emissions associated with the OG and Fire emission sectors contributed around 3 %–5 % and 14 %–23 %, respectively, of the Nr budget for receptor sites at the southeastern corner of the GYA.

It is evident from the results in Sect. 4 that the attribution of total Nr deposition to source regions and sectors is sensitive to NH3 dry deposition rates, the relative contributions of dry and wet deposition, and the concentrations of N compounds from the BCs. However, the model evaluation revealed a significant underestimation of NH3 concentrations and overestimation of HNO3 concentrations and precipitation rates; thus, these modeling errors could bias the source attribution results. To better understand the potential effects of these biases, sensitivity analyses of the source attributions to changes in NH3 dry deposition rates and average precipitation rates as well as potential biases in the BCs were evaluated.

To test the sensitivity of the apportionment to NH3 dry deposition rates, the deposition velocities were reduced by increasing the NH3 resistance scaling factor by 10 %, following the methodology used in Thompson et al. (2015). The Zhang et al. (2003) dry deposition scheme was used in the CAMx simulations (Table S1), and this resistance scaling factor is designed to address the rapid removal of “sticky” compounds such as HNO3 and NH3 and can yield a nonlinear response in the estimated dry deposition velocity. July and August 2011 were simulated using the modified deposition velocity, and these results will be referred to as “DV_0.1”. The 10 % change in the resistance factor slowed the NH3 deposition velocity from 2.5–4 to 1–1.5 cm s-1 over the GYA, resulting in values more comparable to those used in the GrandTReNDS study (Benedict et al., 2013a; Prenni et al., 2014). The simulated NH3 concentrations for the DV_0.1 case increased throughout the GYA compared to the base case. This resulted in better agreement with NH3 measurements at the Grand Targhee and NOAA climate station sites but poorer agreement at the Driggs monitoring site (Fig. S5). The slower dry deposition velocities result in a longer NH3 lifetime, allowing it to travel farther from nearby source regions, e.g., the Snake River valley, into the GYA and cause a more homogeneous concentration pattern throughout the GYA (Fig. S6). As shown in Fig. 11, the slower deposition velocities also somewhat altered the source attribution results. The contribution from the AG emission sector increased with the DV_0.1 simulation to 23 % compared to 19 % in the base case, with a smaller decrease in the contributions from the Other and the Fire sectors. This change was due to small increases in the contributions from the Snake River valley and northern Utah and decreases from Wyoming. Overall, decreasing the NH3 dry deposition rate by about a factor of 2 had only a small impact on the Nr deposition budget and source apportionment results in the GYA. It is important to note that, although this was a significant reduction in the simulated dry deposition velocity for NH3, it still represents a relatively rapid removal rate compared to other species, and NH3 is quickly lost from the atmosphere in either case. It is known that NH3 deposition in many environments is a bidirectional as opposed to a unidirectional process, and modeling the NH3 flux as a bidirectional process may further decrease the bias for ambient NHx concentration simulations (Bash et al., 2013; Wen et al., 2014; Whaley et al., 2018). The key process in air quality models to represent the re-emission of NH3 from soil and plants to the atmosphere is the estimation of the available soil NHx pool and the parameterization of compensation points for the conditions to re-emit NH3 (Zhang et al., 2010; Whaley et al., 2018). In the CMAQ model, the bidirectional NH3 deposition was realized by coupling with the United States Department of Agriculture's (USDA) Environmental Policy Integrated Climate (EPIC) agroecosystem model to provide the fertilization timing, rate, and composition (Bash et al., 2013). There is no similar parameterization available in the current CAMx model. Furthermore, the CAMx source apportionment tools cannot properly account for the origin of NH3 concentrations at a receptor that has been deposited and then re-emitted.

The CAMx simulation overestimated the wet Nr deposition at measured sites, which was likely associated with an overestimation in the precipitation rates from WRF, especially at high-elevation sites. This precipitation rate bias was large, with the annual precipitation over the GYA more than 73 % higher than the PRISM estimates. We used the Noah land-surface model and Kain–Fritsch scheme cumulus parametrization in the WRF simulations (Table S1), and those physical module configurations were reported to have the tendency to overestimate precipitation (Warrach-Sagi et al., 2013). To evaluate the impact of the overestimation in precipitation on the source attribution results, the seasonal wet deposition rates were scaled to the measured precipitation rates at all NADP NTN and GrandTReNDS monitoring sites, following the procedures by Appel et al. (2011). This was equivalent to scaling the modeled wet deposition rates by the ratio of the measured to modeled precipitation rates. This approach assumes that the concentrations of Nr in the precipitation were the same in the model and measured data, which was not the case. After the precipitation adjustment, the correlation between the simulated and measured Nr wet deposition improved (Fig. S7). Within the GYA, however, the scaled Nr wet deposition underestimated the measured N deposition by about a factor of 2 and significantly underestimated the ratio of wet to dry deposition. Consequently, scaled wet deposition results were not used in this assessment. The overestimation of HNO3 concentrations in the GYA is another reason for the wet Nr deposition overestimation. However, its impact on source apportionment results was not conducted here due to unclear reasons for the model bias (emission, chemistry, meteorology, deposition scheme) and limited computational resources.

The BC used in this work was derived from a MOZART global model simulation. An alternative set of BCs from the GEOS-Chem global model was also evaluated. Both sets of BCs resulted in high contributions to the total Nr deposition in the GYA, with the GEOS-Chem results having a slightly higher average contribution of 23 % compared to 21 % for MOZART (Figure S8). However, the GEOS-Chem BCs resulted in higher relative contributions of oxidized N to the total Nr deposition rate compared to the MOZART BCs (51 % and 45 %, respectively). The poor correspondence in the oxidized to reduced Nr split is reflective of the large uncertainties in the BC contributions to the Nr deposition and suggests that more evaluation of the global model results is warranted.

To examine the potential bias in the BC contributions, the simulated PNO3 concentrations were compared to measurements from the IMPROVE monitoring program over the western United States for 2011. This comparison is shown in Fig. 12, in which the ratio of the simulated to measured PNO3, i.e., an estimate of the bias, is plotted against the relative fraction of the contribution of the BCs to the simulated PNO3. The data were first segregated by the fractional contribution of the BCs and then averaged together. As shown, for the MOZART BCs, the bias increased with larger relative contributions from the BCs, and when the BC fraction was 60 %, the bias was more than a factor of 2. This suggests that at least the particulate nitrate concentrations from the BCs are overestimated and possibly other Nr compounds from the BCs as well. In a CMAQ simulation using BCs derived from a GEOS-Chem simulation, Baker et al. (2015) also found that the contributions from the BCs to PNO3 were overestimated when compared to IMPROVE data.

The CAMx model and its PSAT source apportionment tool were used to examine and quantify the contributions of four different source sectors and 27 source regions and the boundary conditions (BCs) to the 2011 total inorganic Nr deposition within the GYA. The source sectors were agriculture (AG), oil and gas activities (OG), wild and prescribed fires (Fire), and remaining contributions labeled as Other. The Other sector was primarily composed of oxidized N originating from anthropogenic combustion sources, including mobile and point sources, and the AG sector was almost entirely composed of reduced N compounds. Fire and the BCs were a mix of reduced and oxidized N compounds. This assessment focused on only the inorganic N fraction. There is measured evidence that organic N (Benedict et al., 2013a; Prenni et al., 2014) is a significant contributor to Nr deposition, and the inability to assess its origin in the current CTM is an important uncertainty in this work. Nevertheless, this Nr source apportionment work is the first thorough analysis of the origin of inorganic Nr in the GYA using a regional air quality modeling platform. The detailed source sector and source region configurations in PSAT enabled quantitative, though uncertain, estimates of their relative importance. This information is needed by stakeholders and regulators to understand the causes of excess Nr deposition in the GYA, monitor changes in Nr deposition, and develop possible future mitigation strategies.

Overall, the model simulation had a reasonable capacity to reproduce the measured seasonal and annual total Nr deposition levels throughout the GYA. However, the model simulation underestimated the measured NH3 concentrations by 65 % on average and overestimated the measured HNO3 by 108 %. Therefore the model tended to overestimate the contributions from oxidized N compounds and underestimate those from reduced N compounds to total Nr deposition. In addition, both reduced and oxidized Nr wet depositions were overestimated by 20 %–30 %, which was due, at least partially, to the simulated precipitation frequency and magnitude being too high in the model. These biases suggest that the modeled contributions from the AG emission sector were underestimated, while those from the Other sector's activities were overestimated.

The simulated annual total Nr deposition over the GYA in 2011 was 3.13 kg N ha-1 yr-1 and exceeded the CL estimates for lichen and lake chemistry primarily at high-elevation sites on the western slope and southern portion of the GYA. This finding is consistent with other studies using global models. Ellis et al. (2013) used the GEOS-Chem model to estimate the Nr deposition to Class I areas for 2006 and showed that the simulated total Nr deposition at GTNP (2.9 kg N ha-1 yr-1) and YNP (2.6 kg N ha-1 yr-1) exceeded the low end of CL for lichens (2.5 kg N ha-1 yr-1).

Emissions from the AG sector within the modeling domain were the largest contributor to the GYA total Nr deposition budget at 34 % year-1. The contributions from the Other sector were also large at 26 %. The OG emission sector generally had a small contribution, except at the southern edge of the GYA, where it could contribute over 10 % of the total Nr deposition during winter months, with almost half of the OG contributions originating from emissions in the neighboring Jonah Field in western Wyoming. The Fire emission sector also had a significant contribution of 18 % over the year. This was due to regional contributions from fires throughout the west and large contributions (>90 %) in areas within the GYA where several wildfires occurred (Fig. 8). The large impact from fires within the GYA is notable since the episodic nature of fire will result in differing year-to-year contributions from this uncontrollable sector.

The largest impact from the AG emission sector originated from sources relatively close to the GYA, and the Snake River valley accounted for 74 % of the annual agricultural contribution. The agricultural contribution from Wyoming was 7 %, and more-distant source regions in northern Utah, California, and the northwestern United States each accounted for 4 %–5 % of the agricultural contribution. Nearly half (45 %) of the Nr deposition from the OG emission sector originated within Wyoming, especially the Upper Green River (27 %). The largest impact from the Fire emission sector originated from the Snake River valley (33 %) and within the GYA (25 %). The Other emission sector was more evenly distributed among near and distant regions, with the Snake River valley 23 %, Wyoming 17 %, and northern Utah, California, and the northwestern United States accounting for 14 %–16 % of the Nr deposition.

Long-range transport of N species from the BCs, which primarily originated from international sources, contributed 21 % of the total Nr deposition within the GYA during 2011 and had the largest absolute contribution during the summer. Several studies have shown the importance of international source contributions to particulates and N deposition within the continental United States (Park et al., 2004; Brewer and Moore, 2009; Zhang et al., 2012; Fann et al., 2013; Baker et al., 2015; Thompson et al., 2015). However, the BCs contribution in this work is on the high end of the reported values. For example, in a similar modeling study by Thompson et al. (2015), the estimated contribution of BCs to Nr deposition in Rocky Mountain National Park in 2009 was 13 %. Zhang et al. (2012) used the GEOS-Chem model to evaluate N deposition in the United States during 2006–2008 and showed that foreign anthropogenic contributions were generally <10 % but could rise up to 30 % near the Canadian and Mexican borders. In addition, our evaluations of the BCs suggest that the contribution of the BCs to ambient PNO3 and possibly other Nr compounds was overestimated (Fig. 12), clearly suggesting that more research is needed on the role of distant emission sources in impacting N deposition in remote areas, as well as further investigations into model biases.

The observed precipitation in 2011 was ∼30 %–50 % higher than the historical average (NOAA, 2012), with the largest bias occurring at the eastern sites in the GYA (Fig. S9). This suggests that dry deposition of NH3 may be a more important contributor to total Nr deposition during spring than that observed during GrandTReNDS. Also, considering that the wet deposition in the GYA tended to be overestimated and the precipitation amount in 2011 was anomalously high, the source regions identified as having a higher weighting on the annual wet Nr deposition budget (e.g., California) may not have such a significant impact as the current PSAT results suggested.

As discussed, source apportionment assessments of Nr and its deposition to remote, ecologically sensitive areas such as the GYA have large uncertainties. Many of these uncertainties are known to the air quality modeling community, including the challenges of simulating precipitation in complex terrain, adequately characterizing NH3 emissions from agricultural operations, the occurrence of wildfires, and the difficulty in simulating the NH3 bidirectional flux and the deposition flux of the other Nr compounds. Contributions from long-range transport of international emissions can also play a significant role in deposition in remote locations in the western United States. Further refinement in all of these areas is required to better understand and estimate the relative contributions of emission sources to excess N deposition within the GYA. Nevertheless, the modeling assessment showed that reduced N contributed more than 50 % of the total Nr deposition over the GYA, with >90 % of the NH3 emissions originating from agriculture sources. In addition, the Snake River valley in Idaho accounted for 74 % of the agricultural contribution to the total Nr deposition. Significant contributions from more distant sources, e.g., California and international sources, to both oxidized and reduced Nr deposition illustrate the regional nature of the Nr deposition problem. Emissions of oxidized N compounds are projected to continue to decrease, while emissions of ammonia are projected to remain relatively constant or increase (Li et al., 2016). This will further increase the importance of the AG sector. However, exceedances of CL are still relatively small, and it is possible that decreased oxidized N deposition could reduce the Nr deposition sufficiently to bring total Nr deposition below the CL in some GYA ecosystems.