Soil Moisture, Soil NOx and Regional Air Quality in the Agricultural Central United States

Agricultural soils containing nitrogen‐rich fertilizers are a substantial source of reactive nitrogen to the atmosphere with potential to impact air quality. One form of reactive nitrogen, nitrogen oxides (NOx = NO + NO2), are a harmful air pollutant and form secondary pollutants, including particulate matter (PM) and ozone (O3). Soil nitrogen oxide emissions (SNOx) are heavily influenced by environmental conditions, however the understanding of the influence of environmental drivers on the behavior of SNOx is limited. Here, we implement a modified soil moisture‐dependent SNOx parameterization into the Weather Research and Forecasting model coupled with Chemistry (WRF‐Chem) and investigate the impact on regional air quality in the central U.S. Evaluating against TROPOspheric Monitoring Instrument (TROPOMI) column NO2 observations, WRF‐Chem columns better capture the TROPOMI column magnitudes earlier in the growing season when using the updated SNOx parametrization, with modeled column bias improved to −1.1% over the most heavily fertilized regions. Evaluating against Environmental Protection Agency (EPA) surface NO2 observations, the relationship between surface NO2 and soil moisture is better represented in agriculturally‐dominant regions when using the updated parameterization, with greatest surface NO2 concentrations at moderate soil moisture and lower concentrations at wetter or drier soil conditions. In simulations, these SNOx lead to increased O3 in select urban regions, with more than double the occurrences of O3 exceeding the EPA 8‐hr O3 standard of 70 ppb.


Introduction
Cropland soils are a known source of reactive nitrogen to the atmosphere, with the potential to substantially impact regional air quality (Chen et al., 2022;Geddes et al., 2022;Lu et al., 2021;Oikawa et al., 2015;Sha et al., 2021;Tan et al., 2023).Both natural and agricultural landscapes emit reactive nitrogen such as nitrogen oxides (NOx = NO + NO 2 ), however the application of nitrogen-rich fertilizers for food production leads to substantially larger soil NOx fluxes (Almaraz et al., 2018).Soil NOx emissions (S NOx ) are understood to be influenced by a number of environmental drivers, including temperature, soil moisture, soil texture, soil pH and nitrogen (N) content, but the relationships between S NOx and these environmental variables remain poorly constrained (Huber et al., 2023).As a result, our understanding of the influence of S NOx on ambient surface nitrogen dioxide (NO 2 ) concentrations and secondary pollutant formation, particularly in agricultural regions, remains limited.Independently, NO 2 is a harmful air pollutant with the potential to negatively impact human health and disrupt crop productivity even at low concentrations (Lobell et al., 2022;Qian et al., 2021).NO 2 additionally acts as a precursor to secondary atmospheric pollutants, including ozone (O 3 ) and particulate matter (PM).Like NO 2 , O 3 , and PM are harmful to human health when inhaled, and elevated O 3 poses a risk to ecosystem health as it can damage plants and reduce crop yield (Kampa & Castanas, 2007;McGrath et al., 2015).NOx concentrations are non-linearly related to O 3 concentrations, with the relative concentrations of both NOx and volatile organic compounds (VOCs) ultimately determining the concentration of O 3 at any given location.NOx can lead to gaseous nitrate (NO 3 ) formation, which can partition into particle NO 3 and ultimately elevate PM 2.5 , or aerosol particles with a diameter less than or equal to 2.5 μm.PM 2.5 is of particular interest in the context of public health, as particles of this size can easily become embedded within the lungs, leading to respiratory disease following repeated or prolonged exposure (Kampa & Castanas, 2007).In agricultural areas, S NOx plays an increasingly important role in determining the concentrations of secondary pollutants like O 3 and PM 2.5 , particularly as fossil NOx emissions decline (Geddes et al., 2022).
Atmospheric chemistry models are a valuable tool in assessing the impact of emissions on air quality.Within the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem; Grell et al., 2005), S NOx can be calculated online using version 2.04 of the Model of Emissions of Gases and Aerosols from Nature (MEGAN version 2.04; Guenther et al., 2006), which also estimates emissions of various biogenic volatile organic compounds.In MEGAN 2.04, S NOx are estimated as a function of temperature and a monthly-averaged emissions factor (EF) for different land cover types.Modeling S NOx was improved with the Berkeley Dalhousie Soil NOx Parameterization (BDSNP; Hudman et al., 2012), which estimates S NOx as a function of soil moisture instead of precipitation, improves the representation of pulsed emissions (Yan et al., 2005) and applies base biome-specific EFs from Steinkamp and Lawrence (2011).The BDSNP has since been implemented into the Goddard Earth Observing System Chemistry model (GEOS-Chem) as a default S NOx scheme.While not provided as a default emissions option in WRF-Chem, previous studies have implemented the BDSNP into WRF-Chem (Sha et al., 2021) and chemical transport models, including the Community Multiscale Air Quality Model (CMAQ; Rasool et al., 2016).
The impacts of S NOx on air quality have been extensively evaluated in parts of the western U.S. (Almaraz et al., 2018;Guo et al., 2020;Oikawa et al., 2015;Sha et al., 2021) and China (Lu et al., 2021;Shen et al., 2022Shen et al., , 2023)), but few studies have evaluated the impact of cropland S NOx on air quality in the central and eastern U.S. In this work, we run WRF-Chem for April-September 2019 using a central U.S. domain to evaluate the influence of S NOx on surface concentrations of NO 2 and secondary pollutants in cropland regions.We conduct sensitivity studies using three S NOx schemes: (a) MEGAN 2.04, (b) BDSNP, from here on referred to as the "standard BDSNP," and (c) an updated version of the BDSNP from Huber et al. (2023) with a revised soil moisture parameterization.We hypothesize the updated parameterization, which produces peak S NOx at wetter soil conditions, will improve agreement with observed spatial and seasonal NO 2 trends.We evaluate the modeled surface NO 2 concentrations and tropospheric NO 2 columns from each simulation against ground-and space-based NO 2 observations in both agricultural and urban domains, and discuss the impacts of these emissions changes on the formation of secondary pollutants.

Ambient Surface NO 2
Ambient surface concentrations of NO 2 are available from Environmental Protection Agency (EPA) groundbased observations (U.S. EPA, 2021).We exclusively choose sites located in or near agricultural land for analysis, as these regions are typically fertilized, leading to larger S NOx than from natural landscapes.Additionally, we select sites located more than 30 km from any metropolitan area with a population of more than 250,000 to reduce the influence from urban NOx emissions.We choose three EPA sites that meet these criteria: one in North Dakota (AQS site 38-101-0003) with data from 2017 to 2021, one in South Dakota (AQS site 46-127-0001) with data from 2015 to 2021, and one in Illinois (AQS site 17-117-0002) with data from 2018 to 2021 (Figure 1).

TROPOMI NO 2 Column Concentrations
The TROPOspheric Monitoring Instrument (TROPOMI) is a spectrometer developed by the European Space Agency (ESA) onboard the polar-orbiting Sentinel-5 Precursor (S5P) satellite, which provides NO 2 vertical column densities (VCDs) derived from observations in the near UV spectrum (Veefkind et al., 2012).TROPOMI was launched in late 2017, with data first becoming available starting 30 April 2018.TROPOMI provides near global daily coverage of a variety of trace gases, with nadir NO 2 VCDs provided at a spatial resolution of 3.5 × 7 km 2 before, and 3.5 × 5.5 km 2 after 6 August 2019.We use the Level 2 version 2.4 NO 2 product and apply a data quality filter, using only pixels with "qa_value" ≥0.75 to omit pixels with unreliable retrievals due to clouds or other albedo artifacts.The finer resolution Level 2 data are oversampled to the coarser 12 km WRF grid for comparison with WRF-Chem model output used in this study.

Soil N, Soil Moisture and Temperature
We use the Trajectories Nutrient Data set for Nitrogen (TREND-Nitrogen; Byrnes et al., 2020a) to determine soil N availability as input for the BDSNP.TREND-Nitrogen provides annual county level N additions for the contiguous U.S. for 1930-2017 from a variety of sources and sinks, which can be combined using a mass balance approach to determine total available soil N.For the mass balance source terms, we use fertilizer N, atmospheric N deposition, biological N fixation and livestock manure N.For the mass balance sink term, we use crop uptake of N. We disaggregate the county level data to a standard grid for use in WRF-Chem.2), occurring in mid-May.Gray contours delineate the Corn Belt boundaries (see Section 2.2.4).Black triangles represent selected Environmental Protection Agency (EPA) surface site locations with ambient surface NO 2 observations.Urban areas discussed in Section 6.2 are denoted with crosses.
We use Soil Moisture Active Passive (SMAP) Level 4 soil moisture data (9 km horizontal resolution) for evaluation of WRF-Chem output at the grid cell closest to the EPA surface NO 2 sites (Section 5.1).SMAP L4 soil moisture is produced using Goddard Earth Observing System version 5 (GEOS-5) meteorological and land surface data, as well as assimilated observed SMAP satellite L-band brightness temperatures as inputs for an ensemble Kalman filter (Reichle et al., 2017).For surface temperature data, we use ERA5 2 m temperature (Hersbach et al., 2020).

WRF-Chem
WRF-Chem is a meteorological model fully coupled with online chemistry (Grell et al., 2005).Unlike chemical transport models that require meteorological data as input, WRF-Chem models the meteorological and chemical variables at each time step, accounting for physical and radiative feedbacks between meteorology and atmospheric chemistry.We use WRF-Chem version 3.9.1 to simulate a model domain located over the central United States (U.S.) with a horizontal spatial resolution of 12 × 12 km 2 on a Lambert Conic Conformal grid (Figure 1).The model contains 35 vertical pressure levels from the surface to 50 hPa, with the lowest model layer height approximately 60 m.Model simulations were conducted for a 6-month period from 25 March 2019 to 30 September 2019, with the first week used for model spin up.We select 2019 as a year that was wetter than average in the Central and Eastern U.S. (Yin et al., 2020), providing a test case for the updated S NOx scheme (Huber et al., 2023) designed to minimize dry conditions that produce maximum S NOx in the original parameterization.The 6-month time period simulates the cropland growing season, including fertilizer application that heavily influences S NOx .
We use the Mozart gas phase chemical mechanism (Emmons et al., 2010) and the MOSAIC 4-bin option with aqueous chemistry (Zaveri et al., 2008) for aerosol phase chemistry.Land surface physics are represented using the CLM4 option (Lawrence et al., 2011).We use the Morrison 2-moment microphysics scheme (Hong et al., 2004), the BMJ parameterization (Betts & Miller, 1986;Janjić, 1994) for cumulus convection, and the YSU option (Hong et al., 2006) for the boundary layer scheme.The 2014 National Emissions Inventory provides anthropogenic emissions and MEGAN 2.04 (Guenther et al., 2006) provides biogenic VOC emissions.Meteorological initial and boundary conditions are provided from the North American Model (NCEP, 2015), and chemical initial and boundary conditions are provided from CAM-Chem (Buchholz et al., 2019).

MEGAN Version 2.04
The MEGAN v2.04 biogenic emissions model (Guenther et al., 2006) scales S NOx based on a temperature function used for non-isoprene compounds (Morichetti et al., 2022): where A is a monthly-average land use-specific EF, β = 0.11 is a temperature dependent parameter for nitric oxide emissions, and T is air temperature in Celsius.Many previous WRF-Chem studies that account for biogenic NO emissions implement MEGAN v2.04, which can be used with any land surface scheme.

BDSNP: Standard and Updated
The second and third S NOx schemes are the online BDSNP, including: (a) the standard BDSNP (Hudman et al., 2012) and (b) an updated BDSNP that accounts for variable soil moisture impacts (Huber et al., 2023).The standard BDSNP estimates S NOx using four terms: where A′ biome is the base S NOx EF as a function of soil N availability (N Avail ), f(T ) is an exponential function that accounts for soil temperature effects and g(WFPS) is a Poisson function that scales S NOx based on water-filled pore space (WFPS), which is volumetric soil moisture (VSM) divided by soil porosity and ranges from 0 to 1.
Soil NOx pulse emissions are represented in Equation 2with P(L dry , t), which depends on the length of the preceding dry period L dry and time since pulse was initiated t, with L dry defined as the amount of time since VSM has been below 17.5%.Equation 3 is used for non-arid grid cells, where g(WFPS) peaks at WFPS = 0.3 cm 3 cm 3 , and Equation 4 is used for arid grid cells, where g(WFPS) peaks at WFPS = 0.2 cm 3 cm 3 .
We updated the BDSNP soil moisture function (Huber et al., 2023) to produce a spatially dynamic relationship between soil moisture and S NOx by changing: (1) The Poisson function g(WFPS) to reach a maximum (i.e., g(WFPS) = 1) at the median April-September soil moisture value for each grid cell; and (2) Using normalized volumetric soil moisture (NVSM), a normalized soil moisture metric ranging from 0 (driest) to 1 (wettest), as the independent soil moisture variable with a revised g(NVSM) function in Equation 2: ⎧ ⎨ ⎩ Each model grid cell has a unique combination of coefficients a, b, and c for Equation 5or Equation 6 optimized to produce g(NVSM) = 1 at the median NVSM and g(NVSM) = 0.02 at the relative wettest and driest conditions for a grid cell.While NVSM is a similar metric as WFPS, the range of modeled WFPS values at any given location throughout a year can be limited to a relatively narrow range, which can influence the magnitude of g(WFPS) and therefore modeled S NOx (Huber et al., 2023).

WRF-Chem Tropospheric NO 2 Column Concentration Quantification
WRF-Chem tropospheric NO 2 column concentrations, or VCDs, are quantified using the averaging kernels (AKs) and the ratio of total to tropospheric air mass factors (AMFs) associated with the TROPOMI NO 2 column retrieval for each overpass (Douros et al., 2023).This accounts for instrument sensitivity in modeled columns, creating a more direct comparison with the TROPOMI product (e.g., Poraicu et al., 2023).The AKs and AMFs are derived from the a priori NO 2 profile from the TM5-MP chemical transport model (Williams et al., 2017): where AMF and AMF trop are the TROPOMI-provided total and tropospheric air mass factors (AMFs), respectively, A v is Avogadro's number, g is the force due to gravity, M is the molecular weight of air, n is the WRF model layer, and n tot is the total number of model layers within the troposphere, with the tropopause determined by the last model layer with a temperature lapse rate >2 C km 1 .[NO 2 ] n represents the WRF-Chem layer average NO 2 concentration in ppb, [AK] n is the weighted TROPOMI-provided AK for a WRF-Chem layer after aligning the differing TM5 and WRF vertical levels, and ΔP n is the difference in air pressure between the bottom and top of each WRF level.
We use all available TROPOMI data coinciding with the model domain and simulation period, typically with either two or three TROPOMI overpasses each day, and calculate simulated VCDs using WRF-Chem model output that is closest to each TROPOMI overpass time.

U.S. Corn Belt Definition
The "Corn Belt" refers to an approximate region of the Midwestern U.S. extending roughly from Nebraska to Ohio, characterized by high agricultural productivity (Green et al., 2018).Within the context of this study, the Corn Belt extent is determined by the magnitude of the BDSNP EF, A′ BIOME .We identify all grid cells with annual maximum A′ BIOME > 0.5 mol N km 2 hr 1 , and refer to this region with high soil N content as the Corn Belt (Figure 1, gray contours), with averaged results shown for this region (Sections 3-5).Using A′ BIOME > 0.5 as the threshold ensures the majority of the region with largest fertilizer inputs lies within the boundary, while excluding nearby urban centers.

Simulated Soil NOx Emissions
Simulated daily average S NOx (Figure 2d) in the Corn Belt is influenced by the main drivers of the emissions parameterization: available soil N (Figure 2a), temperature (Figure 2b) and WFPS (Figure 2c).Starting in mid-April, 75% of the fertilizer for the region is applied over the course of a month, elevating A′ BIOME , after which the remaining 25% of fertilizer is applied over the following 4 months (Figure 2a).A′ BIOME decays with time, with an assumed fertilizer N lifetime of 4 months (Hudman et al., 2012).Daily average temperature increases gradually from <10°C in early April to a maximum of 32°C in mid-July, then generally remains at or above 20°C while gradually decreasing until the end of the simulation (Figure 2b).Soil moisture generally remains at or above 60% WFPS from April to early July, after which values remain between 40% and 60% WFPS, with the median for the 6-month period at 0.61 cm 3 cm 3 (Figure 2c).
MEGAN S NOx are a function of temperature only, so fluctuations in daily average S NOx scale with temperature (Figure 2d, blue line).MEGAN S NOx are the lowest of the three schemes, with daily average values in the Corn Belt generally <5 ng N m 2 s 1 and only 2 days reaching 10 ng N m 2 s 1 in mid-July.Standard BDSNP S NOx (Figure 2d, red line) gradually increase with temperature until early August, when average WFPS decreases to its lowest value of 0.43 cm 3 cm 3 .S NOx from the standard BDSNP are greatest at this point, as values are closest to the BDSNP WFPS optimum of 0.3 cm 3 cm 3 .From early August onward, WFPS increases again, reducing the standard BDSNP emissions.Updated BDSNP S NOx (Figure 2d, gray line) increase from early April through June, when the maximum daily average S NOx of 30 ng N m 2 s 1 is reached coinciding with soil moisture near the median WFPS of 0.61 cm 3 cm 3 .Once soil moisture decreases below the median through July, S NOx also decreases.Generally, MEGAN S NOx will be greatest when temperatures are highest, standard BDSNP emissions will be largest when mean WFPS is closest to 0.3 cm 3 cm 3 (which is never reached in the Corn Belt for this time period), and updated BDSNP S NOx will be largest near the median WFPS and lower during wetter and drier soil moisture conditions.

Influence of Soil NOx Scheme on Tropospheric NO 2 Column
Space-based observations of NO 2 provide valuable insight into NO 2 variability where surface NO 2 observations may be limited, including fertilized cropland.We oversample TROPOMI NO 2 VCDs to the WRF domain and show the 15 April to 14 July average, when the majority of N-rich fertilizers elevates soil N levels (Cao et al., 2018) and S NOx may be detectable from satellite observations (Figure 3a).TROPOMI NO 2 VCDs are largest near urban areas and point sources, with additional large spatial enhancements in agricultural regions, including the Corn Belt and south along the lower Mississippi River Valley (Figure 3a).Broad spatial TROPOMI NO 2 VCD enhancements are present in the eastern Corn Belt region in Illinois, Indiana and Ohio, however these states also have numerous urban areas and NO 2 point emission sources from power generation surrounded by extensive cropland, making it difficult to isolate the relative contribution from fossil NOx versus S NOx from TROPOMI observations alone.
When using the MEGAN S NOx scheme, average simulated NO 2 VCDs for mid-April to mid-July exhibit a negative spatial bias relative to TROPOMI throughout much of the model domain (Figure 3b).This negative bias is generally reduced when using the standard (Figure 3c) and updated (Figure 3d) BDSNP schemes.In the Corn Belt, the simulations using MEGAN, standard BDSNP and updated BDSNP exhibit a spatial R 2 of 0.66, 0.59 and 0.72, respectively, and spatial normalized mean bias (NMB) of 15.9%, 6.5%, and +4.7%, respectively.The improved spatial R 2 and NMB in the simulation using the updated BDSNP, which has the largest S NOx of all three schemes, is evidence that much of the broad enhancements in TROPOMI NO 2 observed over agricultural regions could be due to S NOx .Additionally, TROPOMI NO 2 over non-Corn Belt cropland located between NO 2 point sources in Illinois, Indiana and Ohio generally exhibit better agreement with WRF-Chem VCDs when using the updated BDSNP than when using the MEGAN or standard BDSNP schemes (Figure S1 in Supporting Information S1), providing further evidence that the elevated TROPOMI VCDs could be due to cropland S NOx .
Outside of the Corn Belt, all three WRF-Chem simulations have negative column NO 2 biases over southeastern portions of the domain, extending from Mississippi into South Carolina (approximately 90 to 83°W and 28 to 35°N) coincident with mostly forested regions.Conversely, there are positive biases in all three WRF-Chem simulations, but particularly when using either BDSNP scheme, over southwestern portions of the model domain (approximately 102 to 94°W and 28-40°N).These biases are likely due to under-or overestimates of the EFs used in the S NOx schemes.
Monthly average TROPOMI tropospheric NO 2 VCDs in the Corn Belt (Figure 4a) increase from April to a maximum in June (1.68± 0.82 × 10 15 molecules cm 2 ), followed by a pronounced decrease into July and August and slight increase in September.The WRF-Chem simulations using MEGAN (Figure 4b) and updated BDSNP (Figure 4d) exhibit an increasing trend in NO 2 VCDs from April to a maximum in July (1.65 ± 0.50 × 10 15 and 1.93 ± 0.47 × 10 15 molecules cm 2 , respectively), with slight VCD decreases into August and September.The WRF-Chem simulation using the standard BDSNP (Figure 4c) produces gradually increasing NO 2 VCDs until August, when a maximum monthly average of 1.94 ± 0.82 molecules cm 2 is reached.Relative to TROPOMI, the daily average WRF-Chem VCDs exhibit a mid-April to mid-July NMB (R 2 ) of 19.4% (0.04), 11.4% (0.05) and 1.1% (0.09) and an April-September NMB (R 2 ) of 4.2% (0.00), +8.5% (0.00) and +11.7% (0.02) when using MEGAN, standard BDSNP and updated BDSNP respectively.
The simulations generally do not capture the seasonality in the observed April-September NO 2 column (Figure 4a).One possible explanation is the assumed seasonality in soil N content, as previous studies have shown that soil N concentrations can decay from peak values more rapidly following fertilization and planting than occurs in the S NOx parameterizations (Dong et al., 2022;Kabala et al., 2017).In MEGAN v2.04, the maximum EF in the Corn Belt occurs during July when temperatures exceed 30°C, resulting in peak annual S NOx and column NO 2 .In both BDSNP schemes, the EF (A′ BIOME ) peak in May, but assume a fertilizer N lifetime of 4 months that elevates A′ BIOME late into the growing season (Figure 2a).Additionally, the standard BDSNP soil moisture scaling term increases S NOx into July and August as soils continue to dry closer to the 0.3 cm 3 cm 3 WFPS optimum, causing NO 2 VCDs to be the largest of the three simulations (Figure 3c).Finally, the known overestimation of modeled NO 2 in the free troposphere, combined with TM5 AKs that scale free tropospheric concentrations even higher, has been shown to contribute to simulated column biases (Douros et al., 2023;Jung et al., 2022;Shah et al., 2023).

EPA Surface NO 2 Concentrations Versus Soil Moisture
To assess the impact of soil moisture on surface NO 2 , we investigate the relationship between surface NO 2 concentrations, soil moisture and temperature at three EPA sites in cropland within the model domain (Figure 5).Daily mean surface NO 2 concentrations at all three sites exhibit the largest April-September surface NO 2 at moderate soil moisture and lower concentrations at higher and lower soil moisture.The median WFPS at the North Dakota, South Dakota and Illinois sites is 0.27, 0.50, and 0.65 cm 3 cm 3 , respectively (Figures 5a-5c), which vary due to differing climates and general soil properties.The greatest daily mean NO 2 concentrations are during May and June when fertilization and nearby crop planting typically occur.For all sites, September exhibits the lowest NO 2 concentrations, which is a typical crop harvest time and when soil N would be expected to be lowest, as most cropland will not have been fertilized in the recent past and crops will have consumed the majority of available soil N (Pang & Letey, 2000).We do not show data for October-March, due to anthropogenic sources dominating in late fall and winter months (Jaeglé et al., 2005) and reduced S NOx influence by lower temperatures, depleted soil N, and relatively higher soil moisture.
At the South Dakota EPA site (Figure 5b), surface NO 2 exhibits a strong response to soil moisture, with higher daily mean NO 2 concentrations near the median soil moisture and lower concentrations at wetter and drier soil moisture.We further separate the South Dakota site data by month to assess the seasonal cycle of surface NO 2 (Figure 6).Monthly average daily mean NO 2 concentrations increase from 1.67 ± 1.42 ppb in April (Figure 6a) to a maximum of 3.52 ± 2.53 ppb in June (Figure 6c), after which concentrations decrease again through September (Figure 6f; 0.99 ± 0.80 ppb).The soil moisture-NO 2 behavior differs by month, depending on the range of soil   6a and 6b), lower NO 2 concentrations coincide with higher WFPS leading to a negative NO 2 -WFPS relationship.In June and July, the NO 2 -WFPS relationship is negative at abovemedian WFPS, but positive at below-median WFPS (Figures 6c and 6d).As the typical growing season winds down into August and September (Figures 6e and 6f), NO 2 concentrations remain relatively low while WFPS remains below the median.The North Dakota and Illinois EPA sites exhibit a similar NO 2 -WFPS relationship and monthly NO 2 trends as the South Dakota site, with concentrations increasing from April until June and then decreasing again (Figure S2 in Supporting Information S1).
In a recent study, Wang et al. (2023) identified a similar relationship between TROPOMI NO 2 VCDs and soil moisture in California cropland, with lower VCDs at wetter and drier soil moisture and higher VCDs near the median WFPS of approximately 0.23-0.35cm 3 cm 3 , which is near the optimum WFPS used in the standard BDSNP.Huber et al. ( 2020) identified a satellite-based relationship in Mississippi, U.S., with maximum VCDs occurring at a WFPS closer to 0.6 cm 3 cm 3 .This suggests that both the standard and updated BDSNP produce peak emissions for similar soil moisture conditions in California cropland (with median near 30% WFPS), while in the Mississippi region, and likely other regions within our central U.S. model domain, the updated BDSNP would produce peak emissions and therefore larger NO 2 concentrations, earlier in the season when soil conditions are wetter than the standard BDSNP.
Previous studies have found S NOx exhibits an exponential relationship with temperature, with emissions increasing with temperature until approximately 30-40°C, at which point emissions are understood to plateau (Wang et al., 2021(Wang et al., , 2023)).At the South Dakota EPA site in April and May, lower NO 2 concentrations coincide with lower temperatures, with a positive NO 2 -temperature relationship (Figures 6a and 6b).Starting in June, the NO 2 -temperature relationship becomes more ambiguous, as higher temperature days do not always coincide with higher NO 2 concentrations (Figure 6c).However, evaluating all months and years shows a positive NO 2 -temperature relationship, with the highest NO 2 concentrations at any given temperature typically occurring at moderate soil moisture values (Figure S3 in Supporting Information S1).
To our knowledge, a similar relationship between soil moisture and ambient NO 2 concentrations has not been shown in the literature.The observed EPA NO 2 -WFPS relationships presented in this study appear to be unique to agricultural EPA sites and are not found to the same extent at other natural, non-agricultural sites (Figure S4 in Supporting Information S1), serving as further evidence that the relationships shown in Figures 5 and 6 are predominantly driven by cropland S NOx .

WRF-Chem Surface NO 2 Versus Soil Moisture
When analyzing the modeled NO 2 -soil moisture relationships at the grid cells closest to the EPA sites, the modeled and observed NO 2 -soil moisture relationships generally do not compare well for an individual year (Figure S5 in Supporting Information S1).This is likely due to discrepancies between the true available soil nitrogen and site-specific A′ BIOME factors used in the BDSNP, as well as differences in simulated and observed precipitation at the EPA site locations which influence modeled S NOx .Therefore we examine the simulated daily NO 2 -soil moisture relationship averaged over the Corn Belt to allow the inclusion of more data points, as the simulation time period is too short to produce the observed relationships at just one grid cell (Figure 5).The modeled data are subset for the Corn Belt and daily averages of WFPS, 2 m temperature and surface NO 2 concentration are calculated.
When using MEGAN S NOx , there is no discernible NO 2 -WFPS relationship in the Corn Belt, with concentrations increasing slightly as soils dry (Figure 7a).When using the standard BDSNP S NOx , concentrations increase almost linearly as soils dry, with the greatest daily mean NO 2 concentrations at the driest soil conditions (Figure 7b).This behavior is due Equation 3 of the standard BDSNP, which produces the peak S NOx scaling factor at 0.3 cm 3 cm 3 WFPS in non-arid regions.The lowest simulated daily average WFPS averaged for the Corn Belt is approximately 0.41 cm 3 cm 3 (Figure 7), and soils would need to further dry before the standard BDSNP would reach its maximum emissions potential.Additionally when using the standard BDSNP, days with the highest simulated surface temperature also have the highest simulated surface NO 2 concentrations, behavior not observed at the South Dakota EPA site (Figures 6c-6e).
When using the updated BDSNP S NOx (Figure 7c), surface NO 2 increases with decreasing soil moisture until the median soil moisture of 0.61 cm 3 cm 3 is reached, at which point concentrations begin to decrease with decreasing soil moisture.Surface NO 2 concentrations are highest at higher temperatures at any given soil moisture value, however days with higher surface temperature do not necessarily coincide with greater daily mean surface NO 2 concentrations, for example, multiple days with lower daily mean NO 2 but higher temperature and WFPS less than the median value (Figure S6 in Supporting Information S1).This reflects the observed behavior in Figures 6c-6e, with days with lower soil moisture but high temperature exhibiting lower NO 2 concentrations.
We note that when using the standard BDSNP in WRF-Chem, the choice of land surface model likely plays an outsize role in determining the magnitude of S NOx .Soil porosity for each soil type remains the same regardless of the land surface option being used in WRF-Chem, however different land surface options can result in drastically different soil moisture magnitudes and variability (Xu et al., 2021).Therefore, the choice of a different land surface option than that used in this study (CLM4) would result in different WFPS values and therefore different S NOx magnitudes.When using the updated BDSNP in WRF-Chem, the use of NVSM as the soil moisture metric largely removes uncertainty associated with the choice of land surface model, as the WFPS no longer needs to be calculated and therefore the absolute VSM value does not dictate the magnitude of S NOx (Huber et al., 2023).Conversely, MEGAN v2.04 S NOx are a function of temperature only (Oikawa et al., 2015), and therefore the choice of land surface model should not have a substantial influence on MEGAN v2.04 S NOx magnitudes.

Regional Impacts
To examine the impacts of S NOx changes on air quality, we analyze a 3-month average (mid-April through mid-July) when fertilizer is applied and soil N concentrations and subsequent air quality impacts would likely be largest.We show simulated surface concentrations of NO 2 , O 3 , particulate NO 3 and PM 2.5 for the updated BDSNP simulation and differences with MEGAN and standard BDSNP runs (Figure 8).The updated BDSNP simulates largest surface NO 2 concentrations near urban areas and point sources, with moderate spatial enhancements across much of the Corn Belt (Figure 8a).Surface NO 2 concentrations are larger over regions of intensive agriculture when using the updated BDSNP versus the standard BDSNP, but concentrations are lower in regions with less intensive agriculture (Figure 8e).Similarly, the largest differences in surface NO 2 between the updated BDSNP and MEGAN simulations are over cropland, although average NO 2 concentrations are larger throughout the entire domain (Figure 8i).Surface NO 2 concentrations averaged for the Corn Belt are 43% (99%) higher during this period when using the updated BDSNP versus the standard BDSNP (MEGAN) scheme.8b).Ozone differences between simulations (Figures 8f  and 8j) vary spatially from the NO 2 differences (Figures 8e and 8i), with the largest O 3 impacts confined to the Ohio River Valley and lower Mississippi River Valley and more modest O 3 increases across much of the upper Midwest (Figure 8f).As with NO 2 , the standard BDSNP simulates more O 3 over regions with less cropland.Differences between the updated BDSNP versus MEGAN simulate greater O 3 differences in the Ohio River Valley, but O 3 increases are simulated throughout the domain (Figure 8j).Surface O 3 concentrations averaged for the Corn Belt are only 2% (7%) higher when using the updated BDSNP versus the standard BDSNP (MEGAN) scheme, with greater O 3 differences further east, that is, in the Ohio River Valley.
Changes in gas-phase NOx can also influence the formation of particulate nitrogen (NO 3 ).Surface NO 3 concentrations are largest along a strip extending northward from Texas and into the Corn Belt (Figure 8c).Surface PM 2.5 spatial patterns follow some of the NO 3 concentrations, however greater PM 2.5 are simulated throughout much of the southeastern domain, as well as near apparent point sources (Figure 8d).Both NO 3 and PM 2.5 exhibit similar spatial differences when using the updated BDSNP versus standard BDSNP, with increases throughout most of the northern portions of the model domain and concentration decreases across the south (Figures 8g and 8h).NO 3 and PM 2.5 are larger throughout the entire model domain when using the updated BDSNP versus MEGAN, with the largest differences typically in the Corn Belt region (Figures 8k and 8l).Surface NO 3 concentrations averaged for the Corn Belt are 25% (58%) and surface PM 2.5 is 5% (13%) higher when using the updated BDSNP versus the standard BDSNP (MEGAN) scheme.

Urban Impacts
The spatial differences in modeled atmospheric concentrations of criteria pollutants when using different S NOx schemes indicate that S NOx have the potential to not only impact agricultural areas but also impact downwind population centers.We evaluate the influence of S NOx on mid-April to mid-July urban air quality (using metrics of NO 2 , O 3 , and PM 2.5 ) for three central U.S. urban cities in the vicinity of cropland: Des Moines, IA, Indianapolis, IN and Louisville, KY (Figure 9).Histograms show simulated instantaneous concentrations (output every 4 hr) for points in the 9 × 9 grid cells surrounding the urban area.In Des Moines, IA, a small city and the Iowa state capital, modeled surface NO 2 is lowest when using the MEGAN S NOx scheme and highest when using the updated BDSNP scheme (Figure 9a).Relative to simulations using the MEGAN scheme, the updated (standard) BDSNP produces approximately 80% (20%) more surface NO 2 occurrences greater than 5 ppb during mid-April to mid-July (Figure 9a).In Indianapolis and Louisville, both larger cities in which urban NO 2 emissions are considerably larger than in Des Moines and with urban populations >1,000,000, the impact of the S NOx parameterization on surface NO 2 is reduced.Relative to MEGAN simulations, the updated (standard) BDSNP simulates approximately 20% (10%) more NO 2 occurrences greater than 5 ppb in both Indianapolis (Figure 9d) and Louisville (Figure 9g) in mid-April to mid-July.For all three cities, surface O 3 concentrations are highest when using the updated BDSNP, with larger increases seen in surface O 3 than in surface NO 2 (Figures 9b,9e,and 9h).Relative to the simulation with MEGAN, the updated (standard) BDSNP simulates approximately 270% (210%) more and 220% (150%) more occurrences of O 3 greater than the EPA 8-hr standard of 70 ppb in Indianapolis (Figure 9e) and Louisville (Figure 9h), respectively.In Des Moines, the simulation using MEGAN produced O 3 above 70 ppb only once, while this occurred 64 (18) times with the updated (standard) BDSNP runs.In Indianapolis and Louisville when using the updated BDSNP, the occurrence of numerous greater O 3 concentrations (Figures 9e-9h) but lesser coincident change in NO 2 (Figures 9d-9g) is evidence that the elevated O 3 could be drive by S NOx -induced O 3 formation in upwind cropland subsequently transported into the cities.While we did not save continuous hourly output to accurately calculate an 8-hr average, the increased number of O 3 concentrations at or above 70 ppb in all three cities (Figures 9b, 9e, and 9h) when using the updated BDSNP suggests that more days would likely exceed a 70 ppb standard of a maximum 8-hr average when using the updated BDSNP at these locations (Figure S7 in control on S NOx , carries significance for understanding the air quality impacts of this emissions source in heavily fertilized agricultural regions.The soil moisture-S NOx relationship largely determines the timing and magnitude of S NOx , and has downstream impacts on NO 2 , O 3 and PM 2.5 concentrations.Including a new soil moisture dependent scheme from Huber et al. (2023) increases simulated S NOx in the Midwest and shifts S NOx earlier in the growing season than when using the standard BDSNP and MEGAN S NOx schemes.When using the standard BDSNP, which produces peak S NOx at 30% WFPS for most of the model domain, S NOx peak at lower values and remain higher later in the growing season.S NOx are consistently lower when using the MEGAN version 2.04 scheme.When using the updated BDSNP scheme, mid-April to mid-July bias in WRF-Chem tropospheric NO 2 VCDs is reduced relative to TROPOMI NO 2 for regions in or adjacent to active cropland.However, in July-September, modeled NO 2 columns are greater than TROPOMI, potentially due to the fertilizer lifetime assumed in the BDSNP, which elevates NO 2 late into the growing season.Further, at EPA surface sites in or directly adjacent to agricultural fields, observed surface NO 2 concentrations exhibit a distinct relationship with soil moisture, with highest surface NO 2 concentrations at moderate soil moisture values and lower concentrations at wetter and drier soil conditions, a feature that is well captured in WRF-Chem when using the updated BDSNP.These findings underscore the necessity to address the impact of soil moisture on S NOx , and that temperature alone cannot explain observed variability.Further, these findings highlight the value in long-term, multi-year ambient air quality observations, particularly in understudied agricultural regions, to better constrain uncertain area emission sources that contribute to background pollutant concentrations.
The impacts of this parameterization change are largest for surface NO 2 concentrations, with simulated 2019 mid-April to mid-July surface NO 2 concentrations in the Corn Belt 43% higher when using updated BDSNP versus the standard BDSNP and 99% higher than when using the MEGAN scheme.The impact on secondary pollutant concentrations is less pronounced.When using the updated BDSNP, concentrations of O 3 (PM 2.5 ) in the Corn Belt are 2% (5%) larger than when using the standard BDSNP, and 9% (13%) larger than when using the MEGAN S NOx scheme, with secondary pollutant concentrations are higher further downwind in the Ohio River Valley.This work highlights the importance of improving the representation of S NOx to understand the impact on air quality in agricultural areas and in population centers downwind.In select cities, at least twice as many hourly simulated O 3 concentrations are above the EPA 8-hr standard of 70 ppb when using the updated BDSNP S NOx scheme.This suggests that cropland soils may be important for U.S. regional and urban air quality policy, as there are numerous urban areas within the study region that are in nonattainment for O 3 , for example, the Ohio River Valley (Geddes et al., 2022).Our results suggest that when using MEGAN or the standard BDSNP to estimate S NOx , modeled surface O 3 is likely underestimated in and directly upwind from regions that are currently in nonattainment for O 3 standards, and therefore atmospheric chemistry models using these S NOx schemes may not be properly informing pollution mitigation efforts.

Figure 1 .
Figure 1.Weather Research and Forecasting model coupled with Chemistry domain (solid black line) centered over the Central U.S. Colors represent the annual maximum Berkeley Dalhousie Soil NOx Parameterization soil NOx emissions factor (A′ BIOME , mol N km 2 hr 1 ; Equation2), occurring in mid-May.Gray contours delineate the Corn Belt boundaries (see Section 2.2.4).Black triangles represent selected Environmental Protection Agency (EPA) surface site locations with ambient surface NO 2 observations.Urban areas discussed in Section 6.2 are denoted with crosses.

Figure 2 .
Figure 2. Simulated daily (a) Berkeley Dalhousie Soil NOx Parameterization (BDSNP) emissions factor A′ BIOME , (b) 2 m air temperature, (c) water-filled pore space (WFPS) and (d) S NOx with Model of Emissions of Gases and Aerosols from Nature (MEGAN) (blue), standard BDSNP (red) and updated BDSNP (gray) S NOx schemes averaged in the Corn Belt (Figure 1, gray contour).The dashed line in panel (c) represents the median Corn Belt WFPS.

Figure 3 .
Figure 3. Average mid-April to mid-July 2019 tropospheric NO 2 vertical column densitie (VCD) for the central U.S. model domain.VCDs are shown for (a) TROPOspheric Monitoring Instrument (TROPOMI), (b) Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) with Model of Emissions of Gases and Aerosols from Nature (MEGAN) S NOx , (c) WRF-Chem with standard Berkeley Dalhousie Soil NOx Parameterization (BDSNP) S NOx and (d) WRF-Chem with updated BDSNP S NOx .Gray contours delineate the Corn Belt (see Section 2.2.4).

Figure 4 .
Figure 4. Tropospheric NO 2 vertical column densities (VCDs) for April-September 2019 averaged for the Corn Belt (Section 2.2.4) determined from (a) TROPOspheric Monitoring Instrument (TROPOMI), (b) Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) with Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.04, (c) WRF-Chem with standard Berkeley Dalhousie Soil NOx Parameterization (BDSNP) and (d) WRF-Chem with Huber et al. (2023) updated BDSNP S NOx schemes.Lines show daily average (thin lines), monthly average simulated (solid thick lines) and monthly average TROPOMI VCDs (dashed thick lines).

Figure 5 .
Figure 5. Observed April-September daily mean surface NO 2 concentrations versus Soil Moisture Active Passive L4 soil moisture at three Environmental Protection Agency sites in (a) North Dakota (2017-2021), (b) South Dakota (2015-2021) and (c) Illinois (2018-2021) (see Figure 1).Colors show daily average ERA5 2 m air temperature.Vertical dashed lines show the median water-filled pore space for each location.

Figure 6 .
Figure 6.Same as Figure 5, but for the South Dakota Environmental Protection Agency (EPA) site and separated by month of observation.

Figure 7 .
Figure 7. Simulated daily mean surface NO 2 concentration versus daily mean water-filled pore space (WFPS) in the Corn Belt (see Section 2.2.4) for April-September 2019.Results shown when using (a) Model of Emissions of Gases and Aerosols from Nature (MEGAN), (b) standard Berkeley Dalhousie Soil NOx Parameterization (BDSNP) and (c) updated BDSNP S NOx parameterizations.Vertical dashed line shows the simulated median WFPS for this region.

Figure 8 .
Figure 8. (a-d) 15 April to 15 June 2019 average simulated surface concentrations of NO 2 , O 3 , NO 3 , and PM 2.5 , respectively, when using the updated Berkeley Dalhousie Soil NOx Parameterization (BDSNP) scheme.(e-h) Differences in surface concentrations when using the updated BDSNP versus standard BDSNP S NOx scheme and (i-l) differences when using the updated BDSNP versus Model of Emissions of Gases and Aerosols from Nature (MEGAN) S NOx scheme.The colorbar associated with each panel is located below each panel.

Figure 9 .
Figure 9. Histograms of mid-April to mid-July 2019 simulated surface concentrations of NO 2 (ppb), O 3 (ppb) and PM 2.5 (μg m 3 ) for the nine closest grid cells to (a-c) Des Moines, IA, (d-f) Indianapolis, IN and (g-i) Louisville, KY, colored by S NOx scheme.The vertical dashed line shows the Environmental Protection Agency 8-hr O 3 standard of 70 ppb.