Global inorganic nitrogen dry deposition inferred from ground- and space-based measurements

Atmospheric nitrogen (N) dry deposition is an important component in total N deposition. However, uncertainty exists in the assessment of global dry deposition. Here, we develop empirical models for estimating ground N concentrations using NO2 satellite measurements from the Ozone Monitoring Instrument (OMI) and ground measurements from 555 monitoring sites. Global patterns and trends in the fluxes of NO2, HNO3, NH4+, and NO3− were assessed for 2005–2014. Moreover, we estimated global NH3 dry deposition directly using data from 267 monitoring sites. Our results showed that East Asia, the United States, and Europe were important regions of N deposition, and the total annual amount of global inorganic N deposition was 34.26 Tg N. The dry deposition fluxes were low in Africa and South America, but because of their large area, the total amounts in these regions were comparable to those in Europe and North America. In the past decade, the western United States and Eurasia, particularly eastern China, experienced the largest increases in dry deposition, whereas the eastern United States, Western Europe, and Japan experienced clear decreases through control of NOx and NH3 emissions. These findings provide a scientific background for policy-makers and future research into global changes.

The main methods for evaluating dry deposition fluxes at regional or global scales are geostatistical methods and model simulation. For example, the geostatistical method has been used to evaluate the spatial patterns of dry deposition in Europe, the United States and China 5,18 ; Dentener et al. 3 and Vet et al. 2 simulated global total N deposition including dry and wet forms based on multiple atmospheric chemistry transport models. Recently, a new method was developed to study the spatial pattern of dry deposition by applying satellite observations 20,21 . Cheng et al. used Global Ozone Monitoring Experiment (GOME) and Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) observations to determine the spatial and temporal characteristics of NO 2 dry deposition based on the empirical relationship between NO 2 columns and rural NO 2 in situ measurements in eastern China 20 . Nowlan et al. characterized global NO 2 dry deposition fluxes using satellite measurements from the Ozone Monitoring Instrument (OMI) in combination with simulations from the GEOS-Chem model 21 . This new method features certain major advantages: First, satellite observations can be used to evaluate spatially and temporally continuous NO 2 fluxes 20,21 . Second, these observations can provide results with a higher spatial resolution than model simulations at the global scale 21 . Third, fewer parameters are needed in this method than in the model simulations 20 . Consequently, it is worthwhile to develop a theory and methodology for evaluating the spatio-temporal patterns of global dry deposition using satellite observations. This study compiled a worldwide dataset of atmospheric inorganic N concentrations from 555 ground monitoring sites (Fig. 1), including 7,424 site-year data, downloaded OMI NO 2 column standard products between 2005 and 2014, and data on dry deposition velocities from 163 sites worldwide. Based on the chemical transformations between airborne reactive N, we developed methods that can, for the first time, determine the 2005-2014 global patterns and trends in dry deposition fluxes directly from ground-and space-based data.

Results
Spatial patterns of global dry N deposition fluxes. The magnitude and spatial patterns of global dry deposition fluxes differed significantly by region and N species (Fig. 2). In summary, eastern China, Western Europe, and the eastern United States were the three global hotspots for NO 2 , HNO 3 , NH 4 + , and NO 3 − fluxes. According to the site results for NH 3 (Fig. 3). These results indicated that the dry deposition fluxes increased or decreased in some regions annually and exhibited a weak positive trend worldwide. The significant increases were located in the western United States and Eurasia, particularly eastern China, and the significant decreases occurred in the eastern United States, Western Europe, and Japan. Globally, HNO 3 was the most abundant N species in the dry deposition flux increases because of its high deposition velocity, followed by NO 2 , NH 4 + , and NO 3 − . Global and regional total dry N deposition. Table 1 shows the global and regional total dry deposition.
Based on the results inferred from the OMI NO 2 columns, NO 2 was the most abundant N species in dry deposition globally, followed by HNO 3 , NH 4 + , and NO 3 − . Asia and Africa received the largest volume of deposition, based on the sum of these four N species ("Subtotal" in Table 1), followed by North America, South America, Europe, and Oceania. Asia and Africa were also the regions with the greatest deposition based on a single N species.
In the present study, the regional deposition of NH 3 was calculated as the product of the averaged fluxes based on site measurements and a regional area. Because crop sites represented a large portion of the collected NH 3 sites (approximately 1/3), the regional NH 3 deposition results in this study may be overestimated. According to  (Table 1), the global deposition of NH 3 was 22.28 Tg N a −1 , and Asia and Africa were the regions with the greatest deposition, followed by South America, North America, Europe, and Oceania. Summing all five N species, the global total deposition was approximately 34.26 Tg N a −1 .

Discussion
Dry deposition and dry/wet deposition ratios from different studies. Studies on dry deposition at a large scale are still limited, and they have primarily focused on regions with high N deposition, i.e., the United States, Europe, and China. The results of this study and previous studies on both global-and regional-scale dry deposition are listed in Table 2. Our results are comparable to previous studies, and the large-scale dry deposition results that differ are on the same order of magnitude (Table 2). Additionally, certain differences exist among the results from different methods because each method has its own uncertainty. The uncertainty in the atmospheric chemistry transport models is primarily derived from the assessment of NO x and NH 3 emissions and the dry deposition parameterizations 2,3 . The accuracy of geostatistical methods depends on the number, distribution and types of monitoring sites 5,18 . However, globally, N deposition monitoring sites are still rare except in Europe, North America, and Asia 2 . The method used in the present study is based on data from ground and satellite measurements, and the uncertainty in this method arises from these two data sources. Although all types of methods have their own uncertainties, the development of multiple methods will make the evaluation of dry deposition more accurate and comprehensive. Compared to other methods, the method used in this study has two advantages. First, this method has a simpler structure and requires fewer parameters, which reduces computation time and decreases uncertainty associated with the multiple data sources. Second, this method can conveniently provide continuous results for trend analysis of dry deposition.
Based on previous studies of wet deposition (Table 2), we calculated dry/wet deposition ratios in the United States, Europe, and China and found the average ratios to be 0.93, 0.55, and 0.56, respectively. In the United States, Europe, and China, dry deposition contributed 48%, 35%, and 36% to total deposition, respectively. Vet et al. estimated that the global total deposition was 79.5 Tg N a −1 based on multiple models 2 . However, they did not note the specific value of dry deposition. If we assume that 40% of the total deposition is deposited via dry deposition, then the global dry deposition would be 31.8 Tg N a −1 according to the total deposition result from Vet et al. This value is close to our result of 34.26 Tg N a −1 . The above analysis corroborated dry deposition's important role in global N deposition. However, the majority of ecological field experiments on N enrichment to date have focused on wet deposition fluxes; therefore, the investigation of how dry deposition affects ecosystem structures and functions is an important ecological issue.

Key hotspots of dry deposition changes. According to the results of our trend analysis between 2005
and 2014 (Fig. 3), the eastern United States, Western Europe, and Japan show a clear declining trend in dry deposition, corresponding to monitoring site reports from the United States and Europe 6,22 . These findings suggest that dry deposition is still high in these regions but has declined significantly in recent years. As a result of the Cross-State Air Pollution Rule, NO x emissions from electrical generation are expected to have decreased by over 50% from 2005 to 2014 in the eastern United States 23 . In the 28 EU countries, NO x emissions and NH 3 emissions   Table 1. Global and regional dry deposition (Tg N a −1 ). Note: Dry deposition per region was calculated by multiplying the average fluxes by the regional area. "Subtotal" represents the sum of NO 2 , HNO 3 , NH 4 + , and NO 3 − dry deposition per region. "Total" represents the summed total of the dry deposition of all five N species per region.
Scientific RepoRts | 6:19810 | DOI: 10.1038/srep19810 decreased on average 51% and 28%, respectively, from 1990 to 2012 through the control of air pollution emissions 22 . These policy examples suggest that N deposition can clearly be decreased by controlling NO x and NH 3 emissions, which is important for weakening the potential detrimental effects of N saturation on ecosystems [24][25][26] .
In sharp contrast to the above regions, eastern China not only experienced high dry deposition fluxes but also featured the greatest increase in dry deposition fluxes over the past decade ( Fig. 3) and the most expected hotspots of N deposition. These results agree with the continuous measurements of wet and dry deposition at ten sites in this area 7 . According to those results, the total N deposition at the ten sites ranged from 28.5 to 100.4 kg N ha −1 a −1 , with an average value of 60.6 kg N ha −1 a −1 . Of this total, 40% was deposited via precipitation, and the remaining 60% was deposited by dry deposition. Large NO x and NH 3 emissions are the reason for the ongoing high N deposition in this region. Between 1980 and 2010, NO x and NH 3 emissions in China grew approximately linearly and increased from 1.4 to 6.3 Tg N a −1 and from 5.7 to 14.5 Tg N a −1 (ref. 27 ), respectively, resulting in inevitably high quantities of deposited N. Although N deposition can increase ecosystem carbon sequestration to a certain extent 28 , excessive N results in negative impacts on soil, water, and biological diversity [24][25][26] . In recent years, worsening smog-related weather conditions in China have created a threat to public health, and the Chinese government enacted pollution control and management regulations and strengthened measures to control pollutant emissions. N deposition is expected to decrease with the promulgation and implementation of these regulations.
Scientific basis for establishing remote sensing empirical models. In this study, we established remote sensing empirical models to estimate ground NO 2 , TNO 3 (the sum of HNO 3 and NO 3 − ), and NH 4 + concentrations using OMI satellite measurements and ground measurements. Although they are empirical models, there is a scientific basis for establishing them. The logical framework for the method of determining dry N deposition is shown in Fig. 4.
Blond et al. noted that NO 2 ground measurements performed in urban areas cannot be used to validate remote sensors with relatively low spatial resolutions due to strong concentration gradients in urban areas 29 . However, NO 2 ground concentrations at rural sites, where measurements can represent large areas, are significantly positively correlated with NO 2 columns. Based on this positive correlation, Cheng et al. established a remote sensing empirical model to estimate NO 2 dry deposition in eastern China 20 . We improved Cheng's model by developing a global NO 2 model and modifying the parameterization and validation methods.
When NO 2 is released into the atmosphere, it is converted into gaseous HNO 3 or particulate NO 3 − (ref. 4 ). The conversion processes of NO 2 to HNO 3 or NO 3 − and the processes of mutual conversion between HNO 3 and NO 3 − are shown in Fig. 4. Because NO 2 is the source of HNO 3 and NO 3 − , we inferred that a positive relationship should exist between the number of sources and sinks, and our study demonstrated this assumption. We tested the relationships using data from monitoring sites observing the concentrations of NO 2 , HNO 3 , and NO 3 − and found that NO 2 concentrations have a strong positive correlation with the sum of HNO 3 and NO 3 − concentrations ( Supplementary Fig. S1). This is the scientific basis of the TNO 3 model.
In this study, the NH 4 + model was the result of an initial attempt, but validation subsequently indicated that this model is reliable. We surmise that this model can evaluate ground NH 4 + concentrations using NO 2 columns because NH 4 NO 3 is the main form of NO 3 − in aerosols and because NO 2 is the source of the NO 3 − in NH 4 NO 3 (see Fig. 4). Thus, a strong linear positive correlation exists between NH 4 + and NO 3 − concentrations 30 . Additionally, we also attempted to establish an NH 3 model using NO 2 columns, but the result was not satisfactory. The main reason for this is that NH 3 and NO 2 come primarily from agricultural and industrial activities, respectively; thus, no restrictive relationships exist in their chemical transformation due to their different sources.
Uncertainty analysis. Although our findings are reliable based on the site data validation (see Supplementary Fig. S2), they are still uncertain to some extent. There are four possible contributors to the uncertainty. First, some error is from the OMI NO 2 column products, derived mainly from the calculation of air mass factors (AMF), and the uncertainty in the AMF is approximately 10-40% (ref. 31 ). Second, error may come from ground monitoring data. The monitoring data collected in this study were derived from different monitoring networks or the literature, and some errors may arise from researchers using different methods of sample collection and different measuring instruments. Third, some uncertainties are attributed to the estimation of deposition velocity, and previous studies have suggested that there is great uncertainty in this estimation 11,18 . Here, we attempt to reduce this uncertainty by collecting deposition velocity values from the published literature instead of calculating them directly. Furthermore, the regional assessment of NH 3 deposition in our study contains uncertainties. The regional deposition of NH 3 was calculated from the site-based NH 3 fluxes averaged over the region, and crop sites represented a large proportion of the collection sites (approximately 1/3). Thus, the regional results of NH 3 deposition in this study may be overestimated. We note that NH 3 columns were retrieved from the IASI satellite 32,33 , and we expect that it can be used to calculate NH 3 ground concentrations in the future. This calculation will be helpful in evaluating spatial patterns of ground-level NH 3 concentrations more precisely, thus improving the spatial resolution of NH 3 dry deposition.

. Logical framework for the method used to determine dry N deposition and atmospheric N-related processes, including N emissions and chemical transformation processes.
Based on chemical transformations between inorganic N species, the OMI NO 2 columns were used to estimate the ground concentrations of NO 2 , TNO 3 (HNO 3 + NO 3 − ), and NH 4 + by establishing remote sensing empirical models, and then dry deposition fluxes were calculated using the inferred method. Note: The red arrows represent N emissions from natural and anthropogenic sources; the black arrows represent the chemical transformation processes between atmospheric inorganic N species, which are discussed in the literature 4,10 ; the solid and dotted black arrows are the primary and secondary processes, respectively; the blue arrows represent the logical framework for the evaluated method in this study; and the symbols "F", "C", and "V d " represent dry deposition flux, ground concentration, and deposition velocity of inorganic N species, respectively. Scientific RepoRts | 6:19810 | DOI: 10.1038/srep19810 measuring air N are the ion chromatography and spectrophotometric methods. Although many different methods for sampling and chemical measurement exist, the results from different methods are highly consistent [13][14][15] . This agreement is the basis for our analysis of the global data from different studies. To study global dry deposition at the annual scale, the criteria for collecting data were as follows. First, the land use type of the monitoring site must be clearly described, e.g., forest, crop, grassland, wetland, etc. Second, we imposed no restrictions on the sampling and measuring methods, but the sampling frequency must have been on the day, week, or month scale, and the sampling period must have been longer than one year. Third, the atmospheric concentrations of one or several species, i.e., NO 2 , NH 3 , HNO 3 , NH 4 + , and NO 3 − , must have been measured. After we collected the data, certain processes were performed to make the data available, including data collation, data unit transformation, and abnormal value elimination.
Our datasets included the following: the name of the monitoring site, location of the monitoring site, monitoring period, monitoring method, land use type, NO 2 -N concentration, NH 3 -N concentration, HNO 3 -N concentration, NH 4 + -N concentration, NO 3 − -N concentration, and the literature source. After rigorous data screening and quality control, we obtained a total of 555 sites and 7,424 site-year data for atmospheric inorganic N concentrations. There are 265 sites in North America, 124 in Europe, 98 in Asia, 32 in Africa, 23 in Oceania, and 13 in South  America, and 1,588, 1,015, 1,692, 1,437, and 1,692 site-year data for NO 2 , NH 3 , HNO 3 , NH 4 + , and NO 3 − concentrations, respectively. The monitoring sites were distributed worldwide (Fig. 1) and among the major terrestrial ecosystems, including forest, grassland, crops, shrub, desert, wetland, and tundra. The OMI has three spectral channels with a spectral range between 270 and 500 nm and is used to measure trace gases, including O 3 , NO 2 , SO 2 , HCHO, BrO, and OCIO. It has a spatial resolution of 13 km× 24 km and provides nearly global coverage every day. The details of the OMI can be obtained in Levelt et al. 34 .
NO 2 vertical tropospheric columns are derived from the DOMINO v2.0 OMI NO 2 product provided by the Tropospheric Emission Monitoring Internet Service (TEMIS, www.temis.nl). The details of this product can be found in Boersma et al. 31 . The unit of this product is 10 15 molec./cm 2 with a spatial resolution of 0.125° × 0.125°. In this study, we downloaded the global monthly product of NO 2 columns between January 2005 and December 2014 in the format of an ESRI grid. Then, the annual NO 2 column mean was calculated by averaging the monthly NO 2 columns. Dry deposition fluxes. In the inferential method 12 , the dry deposition flux (F dry ) is typically estimated by multiplying the atmospheric N ground concentration (C), including gaseous N and particulate N, by the deposition velocity (V d ). The F dry can be expressed by the following equation: Unlike other N species, NH 3 presents obvious bi-directional fluxes, i.e., NH 3 can be deposited from the atmosphere onto land, but it can also be emitted from the land into the atmosphere 35 . Thus, a gaseous NH 3 "canopy compensation point" likely exists, and deposition occurs only when the measured NH 3 concentration is higher than the compensation point 36,37 . Accordingly, unlike the other four N species, the equation of F dry for NH 3 is as follows: where C 0 is the canopy compensation point of NH 3 . The values of C 0 for various ecosystems are obtained from previous studies 38,39 .
According to equations (1) and (2), the calculation of F dry for atmospheric inorganic N requires information on C and V d .

Ground concentrations (C). Based on global ground monitoring concentrations of atmospheric inorganic
N and OMI NO 2 columns, we developed remote sensing empirical models at an annual scale to determine the global ground NO 2 , TNO 3 , and NH 4 + concentrations. Because we used the same modelling approach for these three N species, we describe the NO 2 model as an example here. The specific approach was as follows. First, NO 2 columns were extracted according to the locations of the monitoring sites using ArcGIS 10.0 software. The NO 2 ground concentration at each monitoring site and the corresponding NO 2 column were treated as a pair of data. Second, the linear model (y = a + bx) was selected as the regression model, where x was the NO 2 column and y was the corresponding in situ NO 2 ground concentration. Third, 2/3 of the pairs of data were selected to establish the model, and the other 1/3 of the data was used for model validation. Statistics of fit and validation were also calculated. Fourth, the previous step was repeated 500 times through a random and non-repeated sampling method to decrease the random error due to certain fitted data, and the averaged statistics were used to evaluate the fit and validation of the model.
In this study, the validation statistics included the coefficient of determination (R 2 ), root mean square error (RMSE), and modelling efficiency (EF). The calculation and meaning of the statistics can be seen in the Supplementary Information. The final equations for estimating ground NO 2 , TNO 3 , and NH 4 + concentrations are shown below (equations (3)-(5)), and the averaged statistics of model parameterization and validation are shown in Supplementary Table S1.  Fig. S3a-c). Because no significant correlation existed between NH 3 ground concentrations and the NO 2 columns, we could not establish an empirical model to estimate NH 3 ground concentrations globally. Instead, we collected 267 NH 3 monitoring sites from the literature and monitoring networks to assess the global   Table S2). The main land uses of these sites included forest, grassland, crop, shrub, wetland, desert, and water. Previous studies have suggested that land use was the dominant factor for dry deposition velocities 16,17 . The results of dry deposition velocities for different forms of N in various land uses are listed in Table 3. The results from different researchers indicated that dry deposition velocities obviously differ between different land uses. Accordingly, the average deposition velocities for the five N species in various land uses were calculated. Then, the deposition velocities of the five N species were mapped to the global land cover map according to land use types ( Supplementary Fig. S4). In this present study, we used a global land cover map published by the European Space Agency (Globcover 2009) 40 and resampled it to 0.125°× 0.125°.
Calculation and validation of dry deposition fluxes (F dry ). Based on the estimated global ground concentrations and the corresponding V d in the above sections, we estimated the global spatial patterns of NO 2 , HNO 3 , NH 4 + , and NO 3 − fluxes using equation (1). Because of the large difference between HNO 3 V d and NO 3 − V d ( Supplementary Fig. S4), it was necessary to separate TNO 3 concentrations into HNO 3 and NO 3 − concentrations to calculate their fluxes. Due to an insufficient number of monitoring sites, we separated TNO 3 concentrations at the continental scale using the following specific method. We calculated the average HNO 3 /NO 3 − ratio for each continent using monitoring sites with simultaneous observations of ground HNO 3 and NO 3 − concentrations. Using the ratios and the ground TNO 3 concentrations, the global ground HNO 3 and NO 3 − concentrations were calculated. The average HNO 3 /NO 3 − ratios were 0.60, 1.72, 1.67, 1.84, and 0.66 for Europe, Asia, North America, Africa, and South America, respectively. The ratio for Oceania was assumed to be 1.00 due to a lack of monitoring sites. Additionally, we calculated NH 3 fluxes using equation (2) based on concentration measurements from 267 sites and their V d values. Then, arithmetic averages were used to represent the magnitude of regional and global NH 3 fluxes.
To verify the dry N deposition fluxes estimated in this study, we collected the site reported fluxes in the references or observing networks and compared the reported fluxes and corresponding simulated fluxes in this study ( Supplementary Fig. S2). With the exception of NO 3 − fluxes, the model fluxes of NO 2 , HNO 3 , and NH 4 + showed good correlation with their reported fluxes (r ≈ 0.60). The averaged model fluxes of the four N species were close to their reported fluxes. Statistically, 71%, 70%, 78% and 62% of the model fluxes agreed within ± 50% of the reported fluxes for NO 2 , HNO 3 , NH 4 + , and NO 3 − , respectively. Alternatively, we also noted that certain sites plotted far from the 1:1 line in the scatter plots ( Supplementary Fig. S2), indicating that the model results were underestimated or overestimated to some extent at certain sites. Nonetheless, these findings demonstrated that the results of our model agree well with the majority of the reported results.