Nitrogen oxides (NO x = NO + NO 2 ) play a crucial role in tropospheric chemistry, which influence the oxidizing capacity of the troposphere by directly reacting with hydroxyl radicals (OH) and catalyzing the formation of ozone (O 3 ). 1 Most studies and regulatory policies in many countries, including the United States (U.S.), have focused largely on limiting anthropogenic NO x emissions from motor vehicle and fossil fuel combustion. Previous studies have suggested soils as a significant source of NO x emissions, accounting for one-fourth of the total global NO x budget and even larger fractions over high-temperature fertilized agroecosystems and other dryland ecosystems following irrigation or precipitation events; 2-7 thus, SNO x may have a contributing role in recent changes in air quality trends.
The U.S. Environmental Protection Agency (EPA) National Emission Inventory (NEI) reported a steady decrease in NO x emissions from anthropogenic sources over the U.S. during the 2005-2018 period with a rate of 0.11 Tg N yr -1 or 54% overall. 8 However, the trend of tropospheric NO 2 column densities (columns) observed by satellites and nationwide NO 2 concentrations predicted by an ensemble of models are both inconsistent with the sustained decrease in NO x emissions reported by the NEI, which stopped decreasing after the year of 2009. 9,10 Silvern et al. 11 separated OMI observations into winter and summer as well as urban and rural and found that OMI NO 2 in rural summer during the 2005-2017 period had no significant reduction trend. Furthermore, an increase in daily nonpeak O 3 concentration was observed in many parts of the U.S. [12][13][14][15] Recent studies suggest that this enhancement of O 3 can be mainly attributed to the temperature-driven increase in NO x emission, mostly from soils. 2,16 Consequently, soils may be an important source of NO x that has been overlooked in previous studies and regulatory frameworks but has a potentially increased impact on tropospheric NO x budget and O 3 pollution.
Regional air quality models are often used to investigate the impact of emission sources on air quality and evaluate the effectiveness of emission control strategies. [17][18][19][20] SNO x varies nonlinearly with region-specific agricultural management, soil conditions, and meteorology and in drylands may predominantly be emitted as a pulsed flux in response to irrigation/ precipitation-drying cycles; 5,21,22 however, these relationships are not well constrained in models. Most models predict SNO x as a function of surface air temperature, soil moisture, and ecosystem type, such as the Yienger and Levy model (YL95), 23 or the Model of Emissions of Gases and Aerosols from Nature (MEGAN, the scheme widely used in WRF-Chem), 24,25 which generally underestimates SNO x fluxes and neglects the irrigation/precipitation-induced emission pulses from dry soils. Oikawa et al. 2 found that SNO x calculated by MEGAN in WRF-Chem was underestimated by a factor of 10 in comparison to NO x chamber measurements in rural California. Many studies have reached an agreement that numerical models generally underestimate SNO x and misrepresent some key spatial and temporal features, which could be attributed to several uncertainties in the model settings, such as inaccurate emissions coefficients, poor soil moisture data, derivation of soil temperatures from surface air temperatures, neglect of nitrogen deposition, and lack of inclusion of emission pulses. 4,6,21,[26][27][28] We address the uncertainty in the role of SNO x on regionalscale atmospheric chemistry through a combination of new satellite observations of tropospheric NO 2 distributions (TROPOMI) and revision of an SNO x scheme that is subsequently added in the WRF-Chem model. The default SNO x scheme in WRF-Chem, MEGAN v2.04, was replaced by adding the Berkeley Dalhousie Soil NO Parameterization (BDSNP) scheme with modifications to better represent land cover distributions, soil temperature representation, and emission pulses, as well as include fertilizer N emissions from agricultural soils (hereafter the Berkeley-Dalhousie-Iowa Soil NO parameterization or BDISNP). Within the U.S., the state of California has the highest agricultural output, as well as extensive agricultural and natural drylands. In croplands, where nitrogen-rich fertilizers are applied to soils and have regular irrigation, NO x emissions can be significantly enhanced in comparison to the urban regions. 29 Additionally, California has been experiencing warmer temperatures and increasing droughts. 30,31 Some rural regions, such as the Imperial Valley, San Joaquin Valley, and South Coast, also suffer from O 3 pollution that regularly exceeds government standards. 2,15 We thus choose California as a case study region and predict that SNO x could contribute to both NO x and O 3 distributions in the atmosphere. Our results provide insights needed for developing more effective emission reduction strategies to improve the air quality of California and other regions, especially in rural areas with a high prevalence of respiratory illnesses.
2.1. Model Configurations. The Weather Research and Forecasting (WRF) model coupled with online chemistry (WRF-Chem) version 3.8.1 was used in this study. 32 The simulation was performed on one domain over the western U.S. with a grid spacing of 12 km and 74 vertical levels. The physical schemes include the Morrison 2-moment microphysical scheme, Grell 3-D cumulus scheme, 33 RRTM for longwave radiation, 34 and Goddard scheme for shortwave radiation, 35 Yonsei University planetary boundary layer scheme, 36 and Noah land surface model. 37 The Regional Acid Deposition Model, Version 2 (RADM2) for gas-phase chemistry, 38 the Modal Aerosol Dynamics Model for Europe (MADE) 39 and the Secondary Organic Aerosol Model (SORGAM) aerosol modules with some aqueous reactions were chosen. 40 The 0.625°× 0.5°Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis data provide the meteorological initial and boundary conditions. 41 MERRA-2 is produced using the Goddard Earth Observing System (GEOS) atmospheric data assimilation system and uses observations to correct the model simulated precipitation over tropical and midlatitude land areas (60°S-60°N). 42 The 0.25°× 0.25°Global Land Data Assimilation System (GLDAS) data provides the initial and boundary conditions of soil properties (e.g., soil moisture and temperature). 43 Anthropogenic emissions were imported from the U.S. EPA NEI in 2011. Biomass burning emissions are from Fire Locating and Modeling of Burning Emissions Inventory (FLAMBE). [44][45][46] The simulation was conducted from 29 June to 31 July 2018 with the first 2 days as the spinup period. The model output from 1 July to 31 July was analyzed.
2.2. Implementation of BDISNP in WRF-Chem. The BDISNP scheme is based on the BDSNP scheme 4 with a few changes to improve its adaptation to WRF-Chem. Within the BDISNP, the base emission coefficient is composed of two parts: one is the biome emission factor depending on 20 MODIS land cover types, and the other is the available nitrogen in soils including fertilizer and deposition N, which is also used to adjust the base emission coefficients for each biome. It also considers the nonlinear change of SNO x flux with multiple environmental and meteorological factors including soil temperature, soil moisture, the precipitationinduced emission pulse from dry soils, and canopy effects. The function of SNO x flux (detailed in SIs) can be expressed as
where F SNO x (mol km -2 h -1 ) is the SNO x flux, A b ′ is the base emission coefficient, and N biome (kg N m -2 s -1 ) and N avail (kg N m -2 s -1 ) are the wet/dry biome emission factor and nitrogen source availability in soils, respectively. The adjusting factors include soil temperature and moisture factor (f(T), g(θ)), pulsing factor (P(l dry )), and canopy reduction factor (CRF). T (°C) and θ (unitless) are the soil temperature and water-filled pore space (WFPS, defined as the ratio of the volumetric soil moisture content to the porosity), respectively. l dry (h) is the length of the dry period, which is determined by the variation of soil moisture rather than the amount of precipitation.
As one of the important input data of the SNO x scheme, the N fertilizer emissions account for the timing and distribution of N fertilizer on the basis of the MODIS-derived seasonality of the canopy. Since the total N fertilizer use in 2017 in the U.S. (11649324 tons; the data in 2018 are not available, http:// www.fao.org/faostat/en/#data/RFN) is similar to that in 2006 (11625400 ton in U.S., the baseline year of N fertilizer data used by Hudman et al. 4 ) and California N fertilizer sales plateaued in the early 2000s, 47 we use the same fertilizer data from Hudman et al. 4 in the BDISNP.
In comparison to the BDSNP scheme, the BDISNP framework has three major modifications: (1) updating the default land cover data in the WRF model by using the Moderate Resolution Imaging Spectroradiometer Land Cover Type (MCD12Q1) Version 6 data (https://lpdaac.usgs.gov/ products/mcd12q1v006/) in 2018 with a spatial resolution of 500 m to reproduce more a realistic biome type in BDISNP (Figure S1a,b), (2) using the GLDAS data to predict the initial and boundary condition of soil moisture and temperature and directly adopting the soil temperature at the top layer to simulate SNO x rather than using 2 m air temperature (T2) as a proxy for soil temperature (e.g., soil temperature on dry soils with WFPS < 0.3 estimated as T2 + 5 °C at all times) in the BDSNP scheme, and (3) using the modeled green vegetation fraction (GVF) to determine the distribution of arid (GVF ≤ 30%) and nonarid (GVF > 30%) regions instead of using the static climate data as in the BDSNP scheme because the response of the soil moisture factor depends on climate zones and can vary by year.
2.3. Model Experiment Design. To show the improvement in model performance after updating the SNO x scheme in WRF-Chem and evaluate the sensitivity of air quality to soil NO x sources in rural California, we conducted four experiments: i.e., Default, BDSNP, BDISNP, and NoSNOx (Table 1). Default is the base simulation with the MEGAN scheme.
BDSNP is the simulation with the BDSNP scheme. BDISNP is the updated simulation with the BDISNP scheme, updated land types, and better soil temperature representation. NoSNOx is the same as BDISNP but without the soil NO x emission.
2.4. Satellite-Based Observations. The TROPOMI (TROPOspheric Monitoring Instrument) instrument, aboard the European Space Agency (ESA) Sentinel-5 Precursor (S-5P) satellite, was launched on 13 October 2017. It provides almost daily global coverage of tropospheric column densities (denoted as columns) of NO 2 with an unprecedented horizontal spatial resolution of 3.5 × 7 km 2 , has a better signal to noise ratio, and overpasses at about 13:30 local time (LT). 48,49 We use the level-2 daily gridded TROPOMI NO 2 data with quality controls: cloud-screened (cloud fraction below 30%) and quality-assured (qa_value above 0.50). 50 The averaging kernels (AK, defined as the altitude-dependent air mass factor) used in the retrieval algorithms are applied in the intercomparison between TROPOMI and WRF-Chem tropospheric NO 2 columns. Due to satellite data having irregular grid boxes, TROPOMI NO 2 was oversampled to the model grid (12 × 12 km 2 ).
The soil moisture observations were obtained from the Soil Moisture Active Passive Level 4 Soil Moisture (SMAP L4_SM) product, which merged lower-level SMAP data with the Goddard Earth Observing System-5 (GEOS-5) Catchment land surface model in the GEOS-5 ensemble-based land data assimilation system. 51 This product has a 9 × 9 km 2 horizontal resolution and is available twice daily (6:00 am and 6:00 pm LT).
Global Precipitation Measurement (GPM) provides observation data of precipitation every 3 h at a 0.25°× 0.25°spatial resolution. The Integrated Multi-satellitE Retrievals for GPM (IMERG) is the unified algorithm that provides rainfall estimates combining data from all passive-microwave instruments in the GPM Constellation. 52 2.5. In Situ Measurements of NO 2 and O 3 . Hourly surface NO 2 and O 3 measurements in California during July 2018 were obtained from the U.S. EPA Air Quality System (AQS) (https://www.epa.gov/aqs) to explore the implication of SNO x to air quality. Seven NO 2 sites and 17 O 3 sites were selected to compare with the model simulations; the distribution of measurement sites is shown in Figure S1.
3.1. Soil NO x emissions. Figure 1b,c compares the distribution of simulated monthly mean SNO x fluxes from the Default and BDISNP simulations. The implementation of the BDISNP scheme leads to SNO x in July being 9 times higher than that of Default in California. The cropland regions (shown as yellow land types in Figure S1b), which include both high rates of fertilizer application and regular irrigation, show the largest SNO x with monthly emissions of 3.6 Gg N mon -1 in BDISNP, while there is only 0.5 Gg N mon -1 in Default. Our results are consistent with those of Oikawa et al., 2 which suggests that multiplying default soil NO x emission rates in WRF-Chem by a factor of 10 can reach a level similar to the measurements of mean SNO x in the Imperial Valley of California. The much greater SNO x calculated by BDISNP reflects the improvements in the model that better reflect more diverse land covers, soil properties, agricultural management, and pulse emissions.
In comparison to BDSNP, BDISNP simulated monthly SNO x in California decrease by 0.95 Gg N mon -1 (Figure S2). As the types of land covers in California have not changed much in the past 25 years (land cover types in Default are in 1993), only the area of certain land types has expanded or decreased. The higher SNO x in BDSNP is mainly ascribed to its overestimation of soil temperature by assuming that the soil temperature is 5 °C higher than T2 for all land cover types on
Table 1. Description of Model Experiments experiment description Default simulation uses MEGAN v2.04 to calculate soil NO x emissions BDSNP simulation uses BDSNP to calculate soil NO x emissions BDISNP simulation uses BDISNP to calculate soil NO x emissions, including updates of land type to the year of 2018 and directly adoption of soil temperature at the top layer NoSNOx simulation is the same as BDISNP except that the NO x emissions from soils are turned off Environmental Science & Technology dry soils (WFPS < 0.3) at all times. However, the WRF-Chem simulated daytime soil temperature is only on average 1 °C higher than T2 in California, and the difference between the soil and air temperature is much more dynamic than the constant 5 °C difference used in BDSNP (Figure S3). Indeed, the soil temperature in northern California covered with forests and savannas in average is 25.7 °C, 1.8 °C lower than T2, which in turn causes the soil temperature factor to increase from 14 (BDISNP) to 22 (BDSNP). Using T2 as a proxy for soil temperature in BDSNP can lead to large uncertainties in daily or hourly SNO x estimation that are key to the hourly and daily O 3 prediction.
3.2. Tropospheric NO 2 Columns. Satellite-based observations of NO 2 have a wide spatial coverage in comparison to in situ measurements. TROPOMI with a finer spatial resolution is able to capture horizontal gradients and smallscale features, thus providing a good opportunity to evaluate the improvement of the BDISNP scheme in simulating NO 2 columns and detecting the NO x emissions from soils. Here, we compare model simulations (Default and BDISNP) with TROPOMI NO 2 columns during July 2018 in California (Figure 1d-f). Default and BDISNP can reproduce the hot spots of NO 2 columns in urban regions shown in the TROPOMI NO 2 columns (e.g., San Francisco, Los Angeles), but both underestimate the monthly mean NO 2 columns to some extent by 1.4 (1.9) × 10 15 , 0.75 (1.2) × 10 15 , and 0.94(1.7) × 10 15 molecules cm -2 for TROPOMI, Default, and BDISNP averaged over California (croplands), respectively. However, BDISNP shows improved performance in simulating tropospheric NO 2 columns in comparison to Default with a decreasing relative mean bias from 52.3% to 39.8% (Figure S4) and RMSE from 0.7 × 10 15 to 0.6 × 10 15 molecules cm -2 in California (Figure S5). The improvements over cropland are even more obvious; BDISNP reduces the mean bias and RMSE by nearly 23% and 38%, respectively, and increases the R value from 0.74 to 0.78, leading to a good agreement with the TROPOMI NO 2 columns (Figure S6). Soil temperature is a major factor in the SNO x scheme, and high-temperature fertilized soils can emit much higher NO x levels. 2 We find that BDISNP can reproduce the observed response of daily NO 2 columns to temperature in rural areas but the Default could not (Figure S7). Although the Environmental Science & Technology instantaneous uncertainty of TROPOMI tropospheric NO 2 columns at the pixel level is 25-50% or can be up to 0.7 × 10 15 molecules cm -2 , 53 averaging over a larger area or for a longer time (such as 1 month) can largely reduce the noise and improve the precision of TROPOMIN NO 2 columns. 49,54 Therefore, in reference to the monthly TROPOMIN NO 2 columns, the improvement of NO 2 columns in BDISNP is credible, and the BDISNP scheme has the fidelity needed to study the implication of SNO x to air quality in California.
We also investigate the impacts of SNO x on tropospheric NO 2 columns in California, calculated as the difference between BDISNP and NoSNOx simulations (Figure S8). SNO x causes the monthly mean NO 2 columns to increase by 0.2 × 10 15 molecules cm -2 (34.7%) in California by following a distribution similar to that for modeled SNO x . The largest impact is in croplands and drylands (shown as gray land types in Figure S1b, also called desert), where monthly mean NO 2 columns increase by 0.53 × 10 15 molecules cm -2 (53.3%) and 0.31 × 10 15 molecules cm -2 (57.2%), respectively.
3.3. Rain-Induced Emission Pulse. Pulsed SNO x occurs when very dry soils are wetted by precipitation/irrigation, resulting in a reactivation of water-stressed bacteria, but most models do not consider this enhancement in SNO x . The BDISNP scheme adopts the same approach of Hudman et al., 4 in which pulsing activates once soils dry to a WFPS of 0.3 or less for at least three consecutive days prior to soil wetting. In this section, we evaluate the ability of the WRF-Chem model with the BDISNP scheme to characterize the pulsed emission in drylands: the Sheephole Valley of California (Figure S1c), which is in the Mojave Desert, experiences infrequent precipitation during the summer and is isolated from the urban NO 2 plumes. Due to the short photochemical lifetime of NO x (<1 day) and high NO 2 /NO x ratio (>0.8) in the boundary layer, TROPOMI NO 2 with unprecedented resolution allows for SNO x processes to be evaluated using observed NO 2 columns enhancements at spatiotemporal scales unresolvable with previous satellite-based NO 2 products. 29,[55][56][57] Moreover, the contribution of lightning-generated and biomass-burning NO x is shown to be minimal in Southern California in July 2018; [58][59][60] thus, the enhancement of TROPOMI NO 2 columns in the Sheephole Valley can therefore be mostly attributed to SNO x .
We analyzed the multisatellite data with high temporal resolution, including daily TROPOMI NO 2 columns, 3 hourly GPM precipitation, and twice a day SMAP soil moisture observations and found that the observed precipitation was accompanied by the enhancement of soil moisture in the Sheephole Valley (the location of black circles in Figure 2a,b) on 10 July and there was no precipitation in this region before that date. Consequently, TROPOMI NO 2 columns increased on 10 July over the same region (Figure 2c). We hypothesize that this enhancement of NO 2 columns is due to the raininduced NO x emission pulse from dry soils.
As a test of the pulse emission hypothesis, we find that the BDISNP simulation can reproduce the enhancement of NO 2 columns and the pulsed emission from dry soils in the Sheephole Valley on 10 July (Figure 2d). The modeled peak SNO x after the first precipitation can reach 114 ng N m -2 s -1 (Figure 3a), showing a similar level of peak NO x flux postwetting in the Colorado Desert as measured by Eberwein et al. (the median value of ∼100 ng N m -2 s -1 ). 22 These results suggest that the BDISNP scheme can characterize the rain-induced pulse, an improvement from the Default scheme. Such considerable SNO x supported by both simulation results and field measurements in the Imperial Valley 2 and Colorado Desert 22 indicates that rural regions (including croplands and drylands) are major components of total NO x emissions in California.
While a clear improvement against the Default simulation is found, the BDISNP-simulated NO 2 columns in the Sheephole
Valley on 10 July are 65% higher than that of TROPOMI (Figure 3b). This may be because the simulated precipitation began on 9 July, which caused the first NO x pulse in the Sheephole Valley after a long dry period. However, even with this precipitation, WFPS on July 9 is below 0.3 (soil moisture threshold to determine the timing of NO x pulsing). Hence, when the simulated precipitation still appeared on 10 July, the model simulates the second NO x pulse, causing the BDISNP to estimate greater emissions for multiple days and overestimate NO 2 columns on 10 July (Figure 3c,d). Huber et al. 29 suggested that a threshold of 0.3 for WFPS in BDSNP may overestimate emissions at lower soil moisture and underestimate emissions at higher soil moisture for some cropland soils. Therefore, the threshold of WFPS can be optimized further by comparing with ground-based measurements of NO x fluxes in future studies. Furthermore, the BDISNPsimulated precipitation and soil moisture have a certain bias in comparison with the observations. Accurate meteorological fields are critical to simulate the timing and distribution of SNO x when emissions are dominated by pulsing processes and require further study.
3.4. Impact of Soil NO x Emissions on Air Quality. With the implementation of BDISNP in WRF-Chem showing an improved simulation of atmospheric NO x distribution, we quantify the effects of SNO x on air quality in California. Figure 4 shows the proportion of SNO x to total NO x emissions in July and the change in monthly mean surface NO 2 and O 3 concentrations due to the effects of SNO x , calculated as the amount of differences between BDISNP and NoSNOx simulations. We found that the substantial NO x emissions from soils in California, a previously overlooked source, can contribute to 40.1% of the state's total NO x budget (Figure 4a). Over croplands with high fertilizer application, such as the Central Valley and Imperial Valley, the NO x from soils rivals anthropogenic contributions, which account for 50.7%. A larger proportion of SNO x is found over drylands in Southern California in comparison to croplands, suggesting that wetting dry desert soils after precipitation to produce large emission pulses could cause SNO x to exceed anthropogenic sources, accounting for 76.1%. Our results are consistent with a prior study on SNO x estimates by using bottom-up models and spatially and temporally limited airborne measurements, 61 suggesting that agricultural soils could contribute to 20-51% of California's total NO x emissions. Such large amounts of NO x emissions from soils have significant impacts on air quality, which increase the monthly mean surface NO 2 concentrations by 1.2 ppbv (176.5%) in California, 3.0 ppbv (114.0%) in croplands, and 1.1 ppbv (183.8%) in drylands. The monthly mean surface O 3 concentrations also increase by up to 8.4 ppbv (23.0%) in California, 7.3 ppbv (23.2%) in croplands, and 9.5 ppbv (24.8%) in drylands (Figure 4b,c and Figure S9).
On consideration that SNO x has such a large influence on surface NO 2 and O 3 concentrations in rural California, we compared the diurnal variation of modeled NO 2 and O 3 with EPA observations over the downwind area of Los Angeles (the pink rectangle in Figure S1d), which has a high air temperature (>40 °C) during the summer. The simulated SNO x fluxes calculated by BDISNP and Default are also shown in Figure 4. It is seen that the implementation of the BDISNP scheme leads to an SNO x flux 12 times higher than that of the Default in this rural region, with the peak occurring in the daytime. BDISNP with the elevated SNO x flux significantly increases the surface NO 2 concentrations in the early morning and predicts a diurnal variation similar to the observation (R values for the diurnal variations of 0.86 and 0.93 in Default and BDISNP, respectively). Within the Default scheme, the model underestimates O 3 concentrations in the daytime and estimates average monthly O 3 at 41.8 ppbv in this region. However, the BDISNP scheme increases surface O 3 concentrations by 9.3% (3.9 ppbv) and shows a better agreement with the observed diurnal variation. These results suggest that the atmospheric chemistry in this rural region is NO x -limited and the air quality is sensitive to SNO x . Therefore, the intensive agricultural practices and dry desert soils associated with high SNO x in rural regions likely contribute to poor air quality in California by elevating O 3 concentrations.
Nevertheless, even after the SNO x scheme in WRF-Chem is updated, the simulated tropospheric NO 2 columns and surface NO 2 concentrations in the afternoon are still lower than those observed. There are a few factors that could lead to model underestimations, including the uncertainties in the SNO x scheme, underestimations of NO x emissions from other sources, deviations of simulated meteorological fields, and TROPOMI retrieval errors.
For the SNO x scheme, BDISNP assumes that NO x emissions increase exponentially with soil temperature until the temperature reaches 30 °C. However, previous research suggested that SNO x continues to increase with a nonlinear response to soil temperature when it is above 30 °C on the basis of NO x chamber measurements in the Imperial Valley, California, and found that SNO x can increase by 38% on average as the soil temperature increase from 30-35 °C to 35-40 °C. 2 It is thus necessary to improve the response of SNO x in different land types to the soil temperature factor under high-temperature conditions. BDISNP also accounts for the loss of NO x to the plant canopy on the basis of the work of Jacob and Bakwin. 62 However, its default canopy reduction scheme is not mechanistic in nature and may not accurately represent the temporal and spatial variability in canopy effects. We thus stress that future users of the model should implement a more appropriate canopy reduction scheme for their application, which can be achieved by using stomatal uptake to calculate the CRF through analyzing the laboratory measurements of stomatal NO 2 deposition to local vegetation. 63 In addition, the biome emission factors (e.g., grassland, savannas, and needleleaf) based on the work of Steinkamp and Lawrence 26 and the emission factor associated with fertilization (set to 2.5% in BDISNP) are uncertain and may be underestimated. Consequently, a more intensive evaluation of the BDISNP scheme is needed when ground-based measurements of NO x flux are available to improve the parameterization in future studies.
Underestimating NO x sources from anthropogenic emissions, lightning, and biomass burning can also account for the discrepancies. Furthermore, the EPA NEI used in this study is from 2011, which is believed to have an overestimation of NO x emission by up to a factor of 2 in summer months. [64][65][66][67] Although lightning is rare [58][59][60] and there were no large fire activities occurring in July 2018 in California (https:// worldview.earthdata.nasa.gov/), nevertheless the uncertainty of NO x emissions from lightning and the lack of biomass burning in the model may thus lead to underestimating tropospheric NO 2 columns. SNO x is also dependent on accurate meteorological fields in the model; a mischaracterized meteorology therefore could lead to these discrepancies. Additionally, because surface variables, such as soil moisture and temperature, are dependent on land cover types and are highly sensitive to the choice of land surface models, 68 updating land cover types and improving the performance of the land surface model in the future can better simulate SNO x fluxes. On the other side, the KNMI-DOMINO product determines the stratospheric portion of NO 2 columns by assimilating slant columns in the TM5-MP chemistry transport model, but the stratospheric NO 2 columns can be lower than ground-based measurements by up to 0.15 × 10 15 molecules cm -2 . 69 The tropospheric averaging kernels archived in TROPOMI, which use NO 2 profile information coming from the chemistry transport model and data assimilation system to convert slant columns to vertical columns, could also have uncertainties. While the KNMI product is known to compare well with aircraft-and ground-based measurements of NO 2 columns, [70][71][72] these retrieval errors can nevertheless also lead to the discrepancies between model simulations and TROPOMI observations. In summary, our results highlight that SNO x is an important source of atmospheric NO x in California, contributing ∼40% on a state average and more than 50% in rural regions (slightly larger than 50%) with high fertilizer application and in minimally managed native drylands. Therefore, soil NO x emission should be included in regulations to reduce adverse effects to air quality and human health.
The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.0c06834. Overview of the BDISNP scheme, response of daily NO 2 columns to soil temperature, emission factors for 20 soil biomes, distribution of land cover types, location of the Sheephole Valley, and measurement sites of surface NO 2 and O 3 , differences in simulated SNO x flux between BDISNP and Default (BDSNP), frequency and distribution of the differences between soil and 2 m air temperatures, bias of simulated NO 2 columns in comparison to the TROPOMI, scatter plots of observed and simulated NO 2 columns in California, scatter plots of observed and simulated NO 2 columns in croplands, temporal variation of soil temperature and daily NO 2 columns, change in NO 2 columns by the effects of SNO x , and distribution of simulated surface NO 2 and O 3 and relative changes in surface NO 2 and O 3 by the effects of SNO x (PDF)
https://dx.doi.org/10.1021/acs.est.0c06834 Environ. Sci. Technol. XXXX, XXX, XXX-XXX C
https://dx.doi.org/10.1021/acs.est.0c06834 Environ. Sci. Technol. XXXX, XXX, XXX-XXX D
https://dx.doi.org/10.1021/acs.est.0c06834 Environ. Sci. Technol. XXXX, XXX, XXX-XXX E