Monthly Patterns of Ammonia Over the Contiguous United States at 2-km Resolution.

Monthly, high-resolution (∼2 km) ammonia (NH3) column maps from the Infrared Atmospheric Sounding Interferometer (IASI) were developed across the contiguous United States and adjacent areas. Ammonia hotspots (95th percentile of the column distribution) were highly localized with a characteristic length scale of 12 km and median area of 152 km2. Five seasonality clusters were identified with k-means++ clustering. The Midwest and eastern United States had a broad, spring maximum of NH3 (67% of hotspots in this cluster). The western United States, in contrast, showed a narrower midsummer peak (32% of hotspots). IASI spatiotemporal clustering was consistent with those from the Ammonia Monitoring Network. CMAQ and GFDL-AM3 modeled NH3 columns have some success replicating the seasonal patterns but did not capture the regional differences. The high spatial-resolution monthly NH3 maps serve as a constraint for model simulations and as a guide for the placement of future, ground-based network sites.

Introduction

Atmospheric ammonia (NH3) affects air quality, climate, and biodiversity through aerosol formation and composition and nitrogen deposition into the biosphere (Hauglustaine et al., 2014; Hill et al., 2019; Li et al., 2016; Malm et al., 2004; Phoenix et al., 2006; Reis et al., 2009). Atmospheric NH3 emissions are principally from agricultural activities, including the volatilization of agricultural waste and fertilizer application in managed croplands (Bouwman et al., 1997; Paulot et al., 2014). Agricultural NH3 emissions significantly degrade air quality with impacts on human health through ammoniated aerosol formation (Hill et al., 2019; Paulot & Jacob, 2014). With respect to climate, ammonium nitrate (NH4NO3) aerosols have a direct radiative forcing of −0.5 W m−2 over the central United States (Hauglustaine et al., 2014) and are increasingly important at the global scale (Paulot et al., 2018).

Despite the recognized importance of NH3, observations of the spatiotemporal variabilities of NH3 are limited, largely due to the extreme difficulties of measuring gas-phase NH3 (Fehsenfeld et al., 2002; von Bobrutzki et al., 2010). The Ammonia Monitoring Network (AMoN) (NADP, 2020; Puchalski et al., 2015) consists of the only routine measurements of biweekly NH3 across the United States (19 sites in 2010; 107 sites in January 2020). Large differences of NH3 magnitudes and seasonalities exist at short distances between stations (Nair et al., 2019). Satellite NH3 measurements are now available on a global scale from instruments such as the Infrared Atmospheric Sounding Interferometer (IASI), Cross-track Infrared Sounder (CrIS), Tropospheric Emission Spectrometer (TES), and Atmospheric Infrared Sounder (AIRS) (Clarisse olution (∼1 km) NH3 maps have been only provided on an annual basis (Van Damme et al., 2018), or relied et al., 2009; Shephard & Cady-Pereira, 2015; Shephard et al., 2011; Warner et al., 2016). However, high-res-on extra meteorological information to perform wind rotation on a point source at a local scale (Clarisse et al., 2019; Dammers et al., 2019). For seasonality studies, the finest spatial resolution was only on the order of 0.1° × 0.1° (Shephard et al., 2020; Van Damme et al., 2015; Warner et al., 2016), hindering the possibility of identifying small-scale NH3 hotspots and subseasonal variations.

Large discrepancies exist between the chemical transport model predictions of NH3 and observations on national and regional scales (Battye et al., 2019; Heald et al., 2012; Kelly et al., 2014, 2016, 2018; Nair et al., 2019; Zhu et al., 2013). Bottom-up NH3 emission inventories require detailed knowledge of spatially and temporally resolved farming practices that are rarely available (Paulot et al., 2014; Zhu et al., 2013). From a top-down perspective, Gilliland (2003), Gilliland et al. (2006), Pinder et al. (2006), and Paulot et al. (2014) used NH wet deposition data, and Henze et al. (2009) used sulfate and nitrate aerosol compositions to constrain NH3 emissions magnitude and seasonality, but all studies were limited by the sparse in-situ measurements. Chen et al. (2020) and Zhu et al. (2013) inverted satellite NH3 observations into NH3 et al. (2020) performed a 12-month inversion but was still limited by the coarse spatial resolution (∼30 km) emissions, but these were conducted at coarse scales (36–200 km) and only for three selected months. Cao of the chemistry model. The lack of accurate emission inventories results in uncertainties in NH4NO3 simulation and hence PM2.5 simulation (Holt et al., 2015; Kelly et al., 2018; Walker et al., 2012).

To this end, we have developed monthly, high-resolution (0.02° × 0.02°) maps of satellite NH3 columns over the contiguous United States (CONUS), southern Canada and northern Mexico—together one of the most productive agricultural regions in the world—to better understand NH3 spatiotemporal variabilities and the characteristics of hotspot regions themselves such as their locations, areas, magnitudes.

Methods and Data

The IASI v2.2R retrieval product data (2008–2017) were obtained from the MetOp/A (2008-current) and MetOp/B (2013-current) satellites (limited to cloud fraction ≤10%). Only the morning orbits (equatorial passing time ~ 09:30 local) were analyzed because of higher thermal contrast (sensitivity) vs. the evening overpasses (Clarisse et al., 2010). The v2.2R retrieval is based on an artificial neural network for IASI (Whitburn et al., 2016) with the European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA)-Interim reanalysis as its meteorological input (Van Damme et al., 2017). IASI NH3 observations have been validated on a daily and pixel basis and showed good agreement with in-situ data (Guo et al., 2021). A physical-based oversampling approach was used to average the satellite NH3 observations (level 2) to 0.02° × 0.02° (∼2 km) grid (level 3) across the CONUS through a generalized 2-D super Gaussian function (Sun et al., 2018). This algorithm weighs IASI measurements by their uncertainties (Sun et al., 2018), which includes varying sensitivities to thermal contrast. By using 10-years IASI data, we were able to achieve sufficient overlapped IASI pixels (Text S1 and Figure S1). We focused on the NH3 intraannual seasonality and magnitude. Any interannual variability is averaged out, aside from long-term trends of <10% (Warner et al., 2017; Yao & Zhang, 2016), which are generally much smaller than the seasonal variabilities.

For comparisons with models, the Community Multiscale Air Quality (CMAQ) model (version 5.2; www.epa.gov/cmaq) was used with NH3 emissions from 2014 National Emission Inventory (NEI 2014) for a 2014 simulation on a domain covering CONUS with 12-km horizontal grid resolution (Kelly et al., 2019; US EPA, 2018). The CMAQ simulation included bidirectional air-surface exchange for NH3 (Bash et al., 2013) based on the resistance parameterization (Massad et al., 2010). We also simulated 0.5° × 0.5° 2009–2013 NH3 columns by the Geophysical Fluid Dynamics Laboratory (GFDL)-AM3 chemistry-climate model (Donner et al., 2011; Naik et al., 2013; Paulot et al., 2016) using anthropogenic NH3 emissions from the Community Emissions Data System (CEDS) developed for the Coupled Model Intercomparison Project Phase 6 (CMIP6) (Hoesly et al., 2018). The emissions were monthly distributed according to the ECLIPSE v5 model based on European practices (Friedrich, 2004). Paulot et al. (2017) have a detailed description of the AM3 model configuration. CMAQ and AM3 NH3 column abundances were obtained at the nearest IASI overpass time. All data over water bodies were excluded. Annual CMAQ and AM3 NH3 columns are shown in Figure S2 along with their ratios relative to IASI. While key hotspots are broadly consistent with IASI, there remain differences in the areas and absolute and relative magnitudes between the hotspots.

Results and Discussion

High-Resolution Ammonia Annual Map

The IASI NH3 columns were oversampled for each month for all years to get the level 3 oversampled monthly map. Annual maps were made by averaging the monthly maps in equal weights. Figure 1 shows 0.02° × 0.02° 2008–2017 oversampled NH3 column abundances derived from IASI observations along with the location of AMoN sites in the CONUS. NH3 column abundances larger than the 95% (6.6 × 1015 mol cm−2) of the 10-year averaged level 3 map are defined as “hotspot” regions. This definition of hotspots differs from locating individual emission sources, though these are tightly correlated geographically due to the short lifetime of NH3. The 95% threshold concentration is not sensitive to the spatial resolution of the oversampling product as coarser resolutions (0.05° × 0.05°, 0.1° × 0.1°) have similar 95% values (within ±5%).

Figure 1.

2008–2017 averaged annual NH3 columns. Each active AMoN site (January 2020) is shown by a diamond “◊” (within 12 km of a hotspot) or a cross “+” symbol. Active AMoN sites are shown in white while inactive sites are shown in cyan. Labeled hotspots are: (a) Great Plains; (b) San Joaquin Valley; (c) Texas panhandle; (d) Snake River Valley; (e) west-central Ohio; (f) southeastern Pennsylvania, and (g) eastern North Carolina. For reference, the top 5% of annual NH3 columns (“hotspots”) correspond to 0.66 × 1016 mol cm−2.

By using the Hoshen–Kopelman algorithm to cluster adjacent grid points (Hoshen & Kopelman, 1976) above the 95% threshold, a total of 113 areal hotpots were identified (median area = 152 km2). Detailed hotspots locations and contours are shown in Figure S3. The square root of the median area yielded a characteristic hotspot length scale of 12 km (25th: 8 km, 75th: 24 km). The length scale was fairly insensitive to the oversampling product spatial resolution (0.05° × 0.05° = 13 km; 0.1° × 0.1° = 17 km), increasing slightly as expected due to the coarser grid resolution. The characteristic length scale was insensitive to the percentile as well (90th percentile = 5.3 × 1015 mol cm−2 yielded 11 km). In general, the characteristic length scale of IASI NH3 hotspots is on the order of 10 km, indicating that coarser resolution (either in model or observations) may miss or average these hotspots into levels more typical of the background.

The largest contiguous hotspot is in the central Great Plains (386,460 km2), and the San Joaquin Valley hotspot has the highest annual average column abundance (∼2.5 × 1016 mol cm−2). In terms of column-areal weighting, the most important hotspots are the Great Plains, San Joaquin Valley, Texas panhandle, and the Snake River Valley. Although the eastern United States has lower column abundances and fewer hotspot regions than the western United States, PM2.5 formation in the eastern United States is more sensitive to NH3 than the western United States (Holt et al., 2015). Important hotspots in the eastern United States include west-central Ohio, southeastern Pennsylvania, and eastern North Carolina. The locations of these high NH3 columns are consistent with those previously reported in-situ observations, satellite analyses, AMoN network, and intensive agricultural activities (Clarisse et al., 2009; Nowak et al., 2012; Schiferl et al., 2016, 2014; Shephard et al., 2011; Van Damme et al., 2018).

Among the 121 AMoN sites (Figure 1), only 12 AMoN sites are located within 12 km of a hotspot region. AMoN site placement is prioritized to study nitrogen deposition in sensitive ecosystems, and additional sites near hotspots would be valuable for constraining emissions. The high-resolution satellite maps can guide site placement choices in the future, depending upon the ultimate science goal of a site (e.g., emissions vs. downstream deposition).

Ammonia Seasonality Across the CONUS

While annual maps of NH3 constrain the locations of emission sources, the intraannual patterns of NH3 are critical in evaluating aerosol formation and nitrogen deposition. Monthly maps at a moderate (30–100 km) spatial resolution average out many hotspots (Van Damme et al., 2014; Warner et al., 2016) and also provide limited information about how different areas evolve during the year. Figure 2 shows the 2008–2017 monthly oversampled NH3 column concentrations over the CONUS. The general seasonality across the CONUS shows high NH3 columns in spring/summer, and low NH3 concentrations in winter, consistent with past studies (Henze et al., 2009; Paulot et al., 2014; Pinder et al., 2006). However, seasonal complexity exists due to regional differences in agricultural practices and climate. NH3 column maxima are observed in July or August in the western United States, while maxima occur in May/June in the eastern United States (Figure S4 in Supporting Information), in broad agreement with past work at coarser resolution (Van Damme et al., 2015).

Figure 2.

IASI 2008–2017 oversampled NH3 column concentration over CONUS from January to December.

The differences in monthly NH3 patterns are partially attributed to the dominant agricultural land use types for each region. The western United States is dominated by pasture lands (USDA, 2017), and livestock waste volatilizes to produce NH3 with increasing temperatures (Gyldenkærne et al., 2005). The eastern United States is featured by both pasture lands and croplands (USDA, 2017), and fertilizer and manure emissions lead to complex patterns across spring, summer, and fall. Note that the absolute differences between the monthly maxima and minima are large—on the order of 1016 mol cm−2—for most of the hotspot regions (Figure S5).

To better understand the NH3 seasonal patterns beyond the monthly maxima, the IASI NH3 seasonality was examined using k-means++ clustering (Arthur & Vassilvitskii, 2007; Forgy, 1965) of the monthly IASI NH3 columns. As an unsupervised learning algorithm, k-means++ groups observations so that the average squared distance between data in the same cluster is minimized (Arthur & Vassilvitskii, 2007; Forgy, 1965). The advantage of applying k-means clustering is the lack of any a priori assumption of regional seasonality. The data define the geographical regions that have similar seasonality, and k-means++ is independently performed on different datasets. Monthly NH3 data were standardized to have a mean of 0 and a variance of 1. Therefore, the clustering is not affected by differences in the mean or variance but is instead based on the correlation among all standardized monthly concentrations (Zhang et al., 2016). Because the optimal number of clusters is influenced by the underlying data and its patterns (Text S2), different optimal numbers of clusters were identified for each dataset (e.g., IASI, AMoN, model products).

Figure 3 shows the five geographic clusters (3a) identified for IASI NH3 and their seasonal dependencies (3b). The area from Central Plains to the Great Lakes resides in a cluster with a broad, May peak, and a secondary shoulder in September (cluster 1). To the southeast of that area across the South and along the Gulf Coast (cluster 2), there is a peak in May, a local minimum in July, and a secondary smaller peak in September/October. The interior western United States and adjacent high plains are subdivided into a narrower and more pronounced peak in July for the southwest (cluster 3) or August for the northwest (cluster 4). New England, boreal Ontario, and Québec, and the maritime provinces (cluster 5) have relatively small variations all year round.

Figure 3.

IASI, CMAQ, AM3, and AMoN NH3 seasonality clusters map (a), (c), (e), and (g) and standardized NH3 concentrations for each cluster (b), (d), (f), and (h).

Springtime maxima are consistent with the timing of fertilizer application, particularly the later start of the growing season as one moves northward (USDA, 2010). Fall maxima or shoulders are also consistent with fall-based fertilizer application (Goebes et al., 2003; USDA, 2010). In contrast, the narrower, summer peaks of the western United States and high plains are more consistent with the volatilization of livestock waste that strongly correlates with maximum air temperatures (Gyldenkærne et al., 2005). Feedlots and croplands are often adjacent to one another, making definitive agricultural use classifications difficult. Most hotspot regions (67%) fall into the group with a broad peak from spring to early fall (cluster 1), indicating contributions from both livestock waste and fertilizer application. The remainder (32%) fall into one of the two clusters (3 and 4) with the sharp, summer peaks associated more closely with feedlot emissions. Besides agricultural emissions, biomass burning may also contribute to some of the seasonality (Bouwman et al., 1997), especially in summer in the western United States and agricultural burning in the fall in the southeastern United States (Bray et al., 2018; Giglio et al., 2006; Luo et al., 2015).

Figure 3 also shows CMAQ (c, d) and AM3 (e, f) modeled NH3 seasonality clusters using k-means++. The clusters of CMAQ range from a strong summer peak (cluster i) to bimodal peaks in spring and fall (cluster ii). The differences between the CMAQ clusters fall within a narrower range than the differences among the observed IASI clusters. All AM3 clusters show bimodal peaks in spring and fall with varying relative magnitudes. The geographic locations of the model NH3 clusters are far more random and less consistent geographically with their neighbors than the IASI clusters (cf. Figure S6 shows CMAQ and AM3 seasonalities averaged to the IASI spatial clusters). Overall, the temporal evolution of the key clusters in AM3 and CMAQ shows agreement with IASI seasonality clusters, but the spatial patterns of the CMAQ and AM3 clusters strongly differ from those of IASI.

K-means++ clustering was also applied to the much smaller AMoN dataset. Three seasonality clusters were identified for 104 AMoN sites with sufficient (≥1-year record) seasonality measurements as shown in Figures 3g and 3h. Cluster γ, covering most of the eastern and Midwest U.S. sites, has a broad, single peak in June. Cluster α is featured by a peak in July and covers most regions in the western United States, in good agreement with the seasonal pattern classification with IASI. Cluster β shows a relatively insignificant seasonal variability except for a peak in November that is associated with seven sites and may be related to influences from unidentified local sources. IASI and AMoN patterns are not perfectly matched, but both show the similar spatial clusters of seasonalities between the western and Midwestern/eastern United States and also are broadly consistent on the temporal patterns of NH3 seasonalities for each geographic area.

The patterns of IASI and modeled NH3 seasonality are affected by many factors such as emissions, partitioning, and deposition. With respect to aerosol partitioning, gas-phase NH3 compromises much of the total NH in the warm season for both models (Figure S7 and S8). For deposition, only a small fraction (5–15%) dry deposits within 15 km (Dennis et al., 2010; Miller et al., 2015). Indeed, Nair et al. (2019) showed that the modeled NH3 concentrations have a strong spatial dependence on NH3 emissions. However, away from hotspots and during the cold season, deposition and partitioning likely become more important contributors to the seasonal patterns than the underlying emission inventory. Because CMAQ and AM3 modeling results are based on state or county level emission inventory statistics, satellite observations constraints can be used to ameliorate the effects of these geopolitical boundaries on model output.

Seasonality of Hotspots: Comparison of Model and Observations

Case studies of IASI and AMoN NH3 seasonality over hotspots were examined and compared to CMAQ and AM3 columns. Three 0.5° × 0.5° hotspot subregions were selected for comparison: (1) central Tulare County, California, the location of the highest IASI NH3 column (3.2 × 1016 mol cm−2) composed of 58% NH3 surface concentrations (∼15 μg/m3) in the network (58% cropland, 37% pastureland, USDA, 2017); cropland, 36% pastureland (USDA, 2017); (2) Cache County, Utah, the location of the highest annual AMoN and (3) Jo Daviess County, Illinois, a more cropland dominated hotspot (70% cropland, 14% pastureland) (USDA, 2017). Figures 4a, 4c, and 4e show the IASI oversampled NH3 column concentrations in Tulare, Cache, and Jo Daviess counties, respectively. Figures 4b, 4d, and 4f show the corresponding seasonality comparison between IASI, CMAQ, AM3 NH3 columns, and AMoN NH3 surface concentration. While the years included for AMoN, IASI, and model results differed amongst themselves at each site, the interannual variabilities are expected to be averaged out (Figure S9).

Figure 4.

Comparison of NH3 seasonality in hotspots regions. Panels (a), (c), and (e) are the annual averaged IASI oversampled NH3 column concentrations. The bold black dashed box indicates the selected hotspot regions. The black cross shows the nearby AMoN site. Black dashed lines represent AM3 grid boxes, and white dots represent the center of CMAQ grids. (b), (d), and (f) are the NH3 seasonality derived from the IASI oversampling NH3 column, CMAQ, and AM3 modeled NH3 columns (left axis), and AMoN NH3 concentration (right axis). Map data from Google Earth.

The three hotspots display distinctly different IASI NH3 seasonalities, showing that hotspot regions cannot be treated identically. For Tulare County, IASI, AMoN, and CMAQ all show a broad summer peak in Tulare County, while AM3 shows a bimodal peak. For Cache County, all four patterns differ between AMoN, IASI, CMAQ, and AM3. AMoN measures a relatively flat pattern across the year, IASI shows a broad summer peak, CMAQ has a stronger summer peak, and AM3 has bimodal peaks. While IASI identifies Cache site is only ∼100 m away from feedlots, a distance in which concentrations are strongly enhanced relative County as a hotspot, the AMoN site has the highest annual average in the CONUS. However, this AMoN to background levels (Golston et al., 2020; Miller et al., 2015). There are also intrinsic differences between a surface concentration and column amount that complicate comparisons between IASI and AMoN (e.g., boundary layer height). Meanwhile, Jo Daviess County exhibits a broad spring peak in IASI, but a bimodal structure in AMoN and AM3. IASI NH3 columns over hotspot regions allow one to test the relevant model parameters that impact seasonality (e.g., partitioning, emissions, deposition, transport).

Implications

Ammonia columns near source regions are very localized (∼12 km scale) and with strong spatial gradients. Because NH3 hotspots have a strong influence on the air quality and nitrogen deposition in nearby regions (Benedict et al., 2013), there is an urgent need to understand processes such as the underlying spatial pattern of emission inventories, deposition, transport, and partitioning at these same scales. Satellite data may help improve future site placement depending upon the desired objective (e.g., investigating hotspots emissions or characterizing downwind deposition). Ultimately, the differences in spatial and temporal scales between satellite observations (an instantaneous volume) and AMoN (two-week point measurement) require careful attention to many factors for robust comparisons (Kharol et al., 2018).

At monthly scales, the high-resolution NH3 maps provide improved, observational-based means to help constrain NH3 seasonality for improved regional scale modeling of PM2.5 and deposition. The differences between satellite and modeled NH3 seasonal patterns, especially for the global emission inventories developed for CMIP6, further demonstrate the importance of evaluating modeled NH3 with satellite measurements. Simply using the annual averages with a priori seasonalities applied is not accurate. To this end, recent work by Chen et al. (2020) is promising where IASI NH3 is inverted at 36-km resolution for April, July, and October with much different emissions spatially and temporally. Accurate modeling of boundary layer height, vertical profiles of temperature, humidity, and NH3 within the boundary layer, aerosol partitioning, and chemical lifetime are all needed to fully transform these column maps into accurate spatiotemporal emission inventories. Finally, additional validations (Guo et al., 2021) of satellite-derived NH3 columns are also needed to reduce biases at these scales (especially for conditions of temperature inversions during winter and in valleys).