Changing patterns of global nitrogen deposition driven by socio-economic development

Abstract

Advances in manufacturing and trade have reshaped global nitrogen deposition patterns, yet their dynamics and drivers remain unclear. Here, we compile a comprehensive global nitrogen deposition database spanning 1977–2021, aggregating 52,671 site-years of data from observation networks and published articles. This database show that global nitrogen deposition to land is 92.7 Tg N in 2020. Total nitrogen deposition increases initially, stabilizing after peaking in 2015. Developing countries at low and middle latitudes emerge as new hotspots. The gross domestic product per capita is found to be highly and non-linearly correlated with global nitrogen deposition dynamic evolution, and reduced nitrogen deposition peaks higher and earlier than oxidized nitrogen deposition. Our findings underscore the need for policies that align agricultural and industrial progress to facilitate the peak shift or reduction of nitrogen deposition in developing countries and to strengthen measures to address NH₃ emission hotspots in developed countries.

Zhu et al. compile a global N deposition database and find a shift in N deposition, with developing countries emerging as new hotspots. A strong link between economic development and N dynamics is identified, with important policy implications.

Introduction

Intense human activity has significantly altered Earth’s nitrogen (N) cycle. Global emissions of reactive nitrogen (Nr) were estimated to be approximately 164 Tg in 1997 and 210 Tg in 2017¹. Sources of atmospheric Nr are dominated by ammonia (NH₃) mainly from agricultural production, and N oxides from fossil fuel combustion². After chemical transformation and physical transport in the atmosphere, NH₃ and NO_x are removed by wet and dry deposition (F_Wet, F_Dry)^2,3. N deposition can promote plant growth, crop yields and ecosystem carbon sinks^4,5. However, excessive N input causes soil and water acidification⁶, reduces soil buffering capacity⁷, decreases plant diversity⁸ and threatens human health⁹.

With the rapid development of industry, agriculture and urbanization, North America, Europe and East Asia became hotspots of global N deposition^3,10,11. However, N deposition flux (F_N) in developed countries [defined by the World Bank as those with a gross national income per capita above 14,005 United States Dollar ($)] or developed regions (which include groups of developed countries), including the United States and Europe, has decreased substantially^12,13, and F_N in China has stabilized or declined recently because of better environmental governance and economic structural transformation^14,15. The predicted global population increase, high demand for food and energy, industrial relocation and rapid trade development^1,16 are likely to change patterns of F_N. However, deposition observation networks are predominantly located in the United States, Europe and East Asia^12–15. Few observation sites exist in developing countries (those with a gross national income per capita below $14,005) and developing regions (groups of developing countries), and substantial gaps in records exist in Latin America and Africa¹⁷. Systematic reviews of global N deposition monitoring data are limited, and models of global N deposition have modest spatial resolution and are based primarily on highly uncertain emission inventories¹⁸.

Nr emission (E_Nr) distribution, the intensity of agricultural and industrial activity and climate are key factors affecting the spatiotemporal patterns of F_N¹⁵. E_Nr and agricultural N pollution are closely related to socioeconomic development, such as gross domestic product per capita (GDPpc)^19,20. Attempts to reconcile economic development and environmental governance have led to the introduction of better N management and technologies, and the relocation of industries¹. However, compared with developed countries, developing countries lag behind in terms of N use efficiency, N management and the widespread use of emission reduction technologies²⁰. It is important to clarify how socioeconomic development drives the dynamic evolution and changing global patterns of F_N in developing countries, which has significant implications for realizing the goal of halve nitrogen waste, as outlined in the Colombo Declaration²¹.

In this work, we compile data from international N deposition networks and 1390 published papers and construct a Monitoring-based Global Nitrogen Deposition (MGND) database for 1977–2021, encompassing more than 50,000 site-years of data (Supplementary Table 1 and Supplementary fig. 1). Based on the cascading network of GDPpc → E_Nr → satellite N column concentration (C_N, e.g., NH₃ or NO₂) → meteorological factors (e.g., mean annual precipitation, MAP) → F_N, which represents the driving mechanism and qualitative relationship of global N deposition, we develop a framework to generate a global N deposition grid dataset with a resolution of 0.125° × 0.125° for 2008–2020. We reveal the pattern and dynamic evolution of global N deposition, explore the mechanisms driving N deposition, including socioeconomic development, and propose measures for better N management in developing countries.

Results and discussion

Status of global N deposition in 2020

The global average total N deposition flux (F_Tot) to land areas in 2020 was 7.0 kg N ha⁻¹ yr⁻¹, of which ammonium (F_NHx) and nitrate (F_NOy) contributed 4.3 and 2.7 kg N ha⁻¹ yr⁻¹, respectively (Table 1). The global annual input of N through deposition to land in 2020 was approximately 92.7 Tg N, equivalent to 84% of the global agricultural N fertilizer use in that year (110.5 Tg N)²², and lower than global estimate of E_Nr (125.7–179.5 Tg N, Supplementary Table 2). Our estimate is comparable to the simulation results of multiple atmospheric chemical transport models (90.4 Tg N yr⁻¹ to global land in 2010²³) and measurement–model fusion work (87.2 Tg N yr⁻¹ to global land in 2010²⁴), and higher than that from results evaluated from the history of anthropogenic N inputs (63.9 Tg N yr⁻¹ to global land in 2010s²⁵).

Table 1.

Regional atmospheric N deposition flux and total input in 2020

Regions	FDry (kg N ha−1 yr−1)			FWet (kg N ha−1 yr−1)			FTot (kg N ha−1 yr−1)			N inputs (Tg N yr−1)
	NHx	NOy	NHx+NOy	NHx	NOy	NHx+NOy	NHx	NOy	NHx+NOy	NHx+NOy
Africa	2.79 ± 0.54	1.42 ± 0.18	4.21 ± 0.58	1.65 ± 0.05	1.19 ± 0.12	2.84 ± 0.17	4.45 ± 0.52	2.61 ± 0.30	7.05 ± 0.59	21.02 ± 1.76
Central America	3.30 ± 0.66	2.00 ± 0.33	5.3 ± 0.93	2.36 ± 0.11	1.72 ± 0.15	4.08 ± 0.24	5.65 ± 0.71	3.73 ± 0.41	9.38 ± 1.01	2.47 ± 0.27
Central Asia	1.49 ± 0.33	1.07 ± 0.09	2.56 ± 0.42	1.10 ± 0.31	0.78 ± 0.17	1.88 ± 0.48	2.59 ± 0.64	1.85 ± 0.25	4.44 ± 0.89	1.77 ± 0.36
East Asia	3.56 ± 0.09	2.22 ± 0.01	5.78 ± 0.09	3.19 ± 0.05	2.60 ± 0.11	5.79 ± 0.15	6.75 ± 0.08	4.82 ± 0.10	11.57 ± 0.17	13.33 ± 0.19
East Europe	0.63 ± 0.08	0.44 ± 0.04	1.07 ± 0.04	0.70 ± 0.07	0.54 ± 0.07	1.24 ± 0.15	1.33 ± 0.06	0.98 ± 0.11	2.31 ± 0.12	4.02 ± 0.21
Greenland	0.14 ± 0.08	0.05 ± 0.01	0.20 ± 0.09	0.06 ± 0.01	0.05 ± 0.004	0.11 ± 0.01	0.20 ± 0.06	0.10 ± 0.01	0.30 ± 0.08	0.06 ± 0.02
North America	1.44 ± 0.06	0.70 ± 0.03	2.13 ± 0.09	1.06 ± 0.04	0.75 ± 0.03	1.81 ± 0.06	2.5 ± 0.02	1.44 ± 0.05	3.94 ± 0.07	7.45 ± 0.13
Oceania	1.72 ± 0.70	0.73 ± 0.19	2.45 ± 0.75	0.95 ± 0.24	0.69 ± 0.16	1.64 ± 0.41	2.67 ± 0.85	1.42 ± 0.35	4.09 ± 1.06	3.41 ± 0.89
South America	3.52 ± 0.66	1.78 ± 0.34	5.30 ± 0.90	2.29 ± 0.19	1.66 ± 0.24	3.95 ± 0.38	5.82 ± 0.83	3.44 ± 0.57	9.25 ± 1.28	16.29 ± 2.25
South Asia	8.41 ± 0.48	4.14 ± 0.58	12.55 ± 0.83	5.17 ± 0.25	4.18 ± 0.28	9.34 ± 0.47	13.57 ± 0.23	8.32 ± 0.80	21.89 ± 0.84	9.40 ± 0.36
Southeast Asia	3.60 ± 0.32	2.38 ± 0.24	5.98 ± 0.52	3.61 ± 0.02	2.54 ± 0.20	6.15 ± 0.18	7.21 ± 0.34	4.92 ± 0.21	12.12 ± 0.37	5.18 ± 0.16
West Asia	1.46 ± 0.20	1.88 ± 0.24	3.35 ± 0.43	1.47 ± 0.29	1.08 ± 0.15	2.55 ± 0.45	2.93 ± 0.48	2.96 ± 0.37	5.89 ± 0.84	3.97 ± 0.56
West Europe	2.31 ± 0.24	1.39 ± 0.08	3.70 ± 0.31	2.68 ± 0.08	2.03 ± 0.02	4.71 ± 0.10	4.99 ± 0.32	3.42 ± 0.10	8.41 ± 0.41	4.31 ± 0.21
Global	2.46 ± 0.28	1.39 ± 0.13	3.85 ± 0.37	1.80 ± 0.05	1.35 ± 0.08	3.15 ± 0.12	4.27 ± 0.33	2.73 ± 0.21	7.00 ± 0.48	92.68 ± 6.39

The values are Mean ± SD among the three random forest models’ results. F_Wet, F_Dry, and F_Tot are wet, dry, and total N deposition, respectively.

The current spatial pattern of global N deposition is high in middle and low latitudes (30 °S–30 °N) and low in high latitudes (>50°N and 50°S) (Fig. 1a). F_N in South Asia, East Asia, Southeast Asia, and South America is higher than that in Western Europe and North America (Table 1). Especially high levels of F_N are concentrated in northern India, north China, and eastern China, with values of 40–60 kg N ha⁻¹ yr⁻¹. The spatial distribution patterns of F_NHx, F_NOy, F_Dry and F_Wet are the same as those of F_Tot (Supplementary Fig. 2). A surprisingly high level of F_NHx with value of 4.5 kg N ha⁻¹ yr⁻¹ is also found in Africa, which is far higher than previous results^23,26. Further analysis shows that C_NH3 in Africa is almost comparable with that in North America, East Asia, and Western Europe in recent years (Supplementary Fig. 3), consistent with previous research²⁷. The total amount of NH₃ emissions (E_NH3) from the emission inventory of Community Emission Data System (CEDS)²⁸ in Africa is also higher than that in North America and Western Europe. This suggests that previous studies may have greatly underestimated N deposition in Africa.

Fig. 1. Spatio–temporal patterns of global terrestrial N deposition.

a Spatial distribution of total N deposition in 2020. b Temporal dynamics of total, NHx, and NOy deposition from 1980 to 2020; the circles are direct observations and their error bars indicate SE (the variation among the monitoring sites); the triangles are estimated from random forest models and their error bars indicate SE (variations across three random forest models). Different colors represent different N deposition components. c Temporal dynamics of ratio of NHx to NOy deposition from 1980 to 2020. d Cumulative N deposition input from 1980 to 2020. Note: The Antarctic is not included. Source data are provided as a Source Data file.

Changes in global N deposition from 1980 to 2020

Global terrestrial F_Tot increased and then stabilized and slightly decreased during 1980–2020, reaching a peak in 2015 at 7.3 kg N ha⁻¹ yr⁻¹ (Fig. 1b). From 1980 to 1982, F_NHx was slightly lower than F_NOy, but the subsequent increase in F_NHx was much faster than that of F_NOy. F_NHx and F_NOy decreased slightly after 2015 (Fig. 1b). Over the last 40 years the ratio of F_NHx to F_NOy (R_NHx/NOy) has gradually increased from 0.81 in 1980 to a maximum of 1.73 in 2007. It then began to decline and stabilized at 1.5 after 2010 (Fig. 1c). The cumulative input of F_N to global land from 1980 to 2020 was approximately 3117 Tg N (Fig. 1d) whereas global N fertilizer use was approximately 3549 Tg N²². Thus, N inputs to terrestrial ecosystems through atmospheric deposition were almost equivalent to those of fertilizer N over that period.

Regional N deposition dynamics from 1980 to 2020

The regional dynamics of N deposition from 1980 to 2020 can be divided into three types (Fig. 2). Type 1 shows a decline, which predominantly occurs in developed countries such as North America, Western Europe, Japan and South Korea (Fig. 2a–c). Except in North America, F_Tot, F_NHx and F_NOy all decreased; in North America R_NHx/NOy increased or was approximately constant. Type 2 shows transition and mainly occurs in China, Russia, West Asia and other middle-income countries (grouped by income class according to the World Bank classification, gross national income per capita between $1146 and $14,005) (Fig. 2d–f). Here F_Tot, F_NHx and F_NOy all first increased and then stabilized or decreased. F_Tot reached a maximum and R_NHx/NOy decreased, as in China, or was approximately constant. It should be noted that, in contrast to our expectations, we found that N deposition in Africa also showed an increase and then stabilized (Supplementary Fig. 4). Further analysis revealed that the C_NO2 and PM_2.5 in Africa have both been decreasing (Supplementary Fig. 4), and GDPpc has experienced stagnant growth in the past decade (Supplementary Fig. 5). Although C_NH3 has slightly increased, it has not changed the overall decline of F_N in Africa. Type 3 shows an increase and predominantly occurs in low-income countries (gross national income per capita less than $1145) such as South Asia, Southeast Asia and South America. Here F_Tot, F_NHx, and F_NOy showed significant increases, and R_NHx/NOy also increased (Fig. 2g–i).

Fig. 2. Regional dynamics of atmospheric N deposition from 1980 to 2020.

a Dynamics in North America. b Dynamics in Western Europe. c Dynamics in Korea and Japan. d Dynamics in China. e Dynamics in Russia. f Dynamics in Western Asia. g Dynamics in South Asia. h Dynamics in Southeast Asia. i Dynamics in South America. The circles are direct observations and their error bars indicate SE (the variation among the monitoring sites). The triangles are results from the random forest models and their error bars indicate SE (variations across three random forest models). Different colors represent different N deposition components. The trend of N deposition is fitted using linear or binomial functions. R² is the coefficient of determination, ** represents the significance level at P = 0.01, and * represents the significance level at P = 0.05. The bar chart at the bottom of each plot shows the dynamics of the ratio of NHx to NOy deposition in the region. Source data are provided as a Source Data file.

Transfer of global N deposition hotspots from 2008 to 2020

The global terrestrial F_Tot has been relatively stable over the last decade (Fig. 1b) but, as noted above, F_N in developed countries decreased, whereas it increased in developing countries in South Asia, Southeast Asia and South America (Fig. 2). Our calculations suggest that global hotspots of F_N are moving from developed to developing regions. To verify the robustness of this conclusion, we analyzed trends in F_Tot from 2008 to 2020. F_Tot significantly increased in developing countries at middle and low latitudes in South Asia, Southeast Asia, and Brazil. However, it significantly decreased in developed countries or regions such as Europe, the eastern United States, and Japan. Meanwhile, N deposition in some regions of Africa, West Asia, and Argentina has also decreased (Fig. 3a). Decreases of C_NO2 and C_NH3 in those regions of Africa, West Asia and Argentina could directly contribute to the decline of F_Tot in these areas (Supplementary fig. 6).

Fig. 3. Trends in N deposition in developed and developing countries from 2008 to 2020.

a Trend analysis of total N deposition from 2008 to 2020. b Temporal dynamics and ratios of total N deposition in developed and developing countries (Mean ± SE). c Temporal dynamics of NHx and NOy deposition in developed and developing countries (Mean ± SE). The trend of N deposition is fitted using linear or binomial functions. R² is the coefficient of determination, ** and * represents the significance level at P = 0.01 and P = 0.05, respectively, and SE (error bar in figure) indicated the variation across the three random forest models. Source data are provided as a Source Data file.

We also separately analyzed trends in F_N in developed and developing countries from 2008 to 2020. The dynamics of changes in F_N in developing and developed countries were different, with the ratio of F_Tot between developing and developed countries showing a significant linear increase (R² = 0.43, P < 0.01, Fig. 3b). In developed countries, the slight overall downtrend in F_N was primarily due to a significant decrease in F_NOy (R² = 0.44, P < 0.05), while F_NHx showed only a slight and non-significant decrease (Fig. 3c). This emphasizes the need for developed countries to strengthen measures to reduce NH₃ emission hotspots. In contrast, the significant increase in developing countries is due to the significant increase of F_NHx (R² = 0.59, P < 0.01, Fig. 3c).

We analyzed trends at sites with continuous observations of wet deposition data from 2000 (Supplementary Fig. 7). This confirms that global hotspots of F_N have moved from developed to developing countries. It should be noted that these poorly monitored regions (i.e., Africa, Central Asia, Latin America, and Australia) still have the greatest uncertainty in F_N, and these highly uncertain regions include many of those with increasing trends, which requires further observation. Meanwhile, some long-term trends of indicators, such as foliar N content and surface water nitrate concentration, declined in United States^29–31 and Europe^32,33 and increased in tropical forest³⁴ and India³⁵, which potentially reflect the changing hotspots of F_N.

Drivers of global N deposition

E_Nr is an important mechanism driving the dynamic changes in F_N¹⁵. We analyzed the correlations between F_N and GDPpc, E_Nr, C_N and MAP in five major global regions at national and regional scales. E_NH3, NOx emissions (E_NOx), C_NH3 and C_NO2, and MAP were correlated with F_Tot, F_Wet and F_Dry. The strength and positivity or negativity of the correlations varied according to region (Supplementary Fig. 8) showing that F_N is strongly influenced by anthropogenic E_Nr and meteorological factors, which is consistent with previous research^11,15.

Past research has developed a framework of active factors → E_Nr → meteorological factors → F_N to analyze the factors and mechanisms influencing F_N^3,26. However, this does not clearly show the role of social and economic development in changes in F_N. We found that GDPpc had a significant negative correlation with F_N in North America, Western Europe and East Asia. We also observed a less pronounced, less significant positive correlation with F_N in Southeast Asia and Africa (Supplementary Fig. 8). We therefore built a cascade network of GDPpc → E_Nr → C_N → MAP → F_N using structural equation models. The factors in these models and the processes that influence them can explain 45–88% of the spatiotemporal variation (Supplementary Figs. 9 and 10), confirming the importance of socioeconomic development in determining changes in F_N and its components.

We then analyzed the relationships between GDPpc and F_N at the global scale using data from several countries or regions. The relationships between F_Tot, F_NHx, F_NOy, F_Wet, F_Dry and GDPpc of countries or regions at different stages in their economic development fitted perfectly on a normal distribution curve (Fig. 4a–c; Supplementary Fig. 11a, b). This is consistent with the classical environmental Kuznets curve (EKC) model^19,36. When we tested the logarithmic cubic equation model of EKC, the normal distribution curves were all significant (P < 0.01) (Fig. 4d–f; Supplementary Fig. 11c, d; Supplementary Table 3). The peak of F_Tot was at approximately $8800, and the peaks of F_Wet and F_Dry were at $8480 and $8762 GDPpc (Supplementary Table 3), respectively. The peak for F_NHx was at approximately $6600 and that of F_NOy at approximately $11,000 GDPpc. F_NHx was predominantly derived from agricultural activities and reached its peak before (in terms of time and/or economic development) F_NOy, which was mostly driven by industrial activities. This indicates that socioeconomic development, as expressed by GDPpc, is an important factor in determining the spatiotemporal pattern of F_N, and that the dynamic evolution of the relative contributions of agricultural and industrial activities determines the relationship between the peaks of F_NHx and F_NOy.

Fig. 4. Relationships between N deposition and gross domestic product per capita on a global scale.

Relationships between total N (a, d), NHx (b, e) and NOy (c, f) deposition and gross domestic product per capita (GDPpc). The data used in a–c are observations of N deposition and GDPpc; those in d–f are the logarithmic values of N deposition and of GDPpc. African countries are Niger, Mali, Cameroon and Cote d’Ivoire. Southeast Asian countries are Vietnam, Malaysia, Indonesia and Thailand. East Asian countries are China, South Korea and Japan. Western European countries are EU countries (EU27). North American countries are the United States and Canada. R² is the coefficient of determination, ** represents the significance level at P = 0.01. Source data are provided as a Source Data file.

Implications for global N management

The cumulative input of F_N to terrestrial ecosystems from 1980 to 2020 (3117 Tg N) was almost equivalent to the global N fertilizer application (3549 Tg N) over the same period. N deposition can significantly increase the ecosystem productivity for forests, grasslands and water bodies^37,38, so the impact of such a substantial amount of natural N fertilization on the carbon cycle and carbon sink in ecosystems requires research³⁹, and the impacts of N deposition on species diversity, greenhouse gas emissions, soil acidification, etc., need to be re-assessed. Hotspots, areas of intensive N deposition, have moved from developed to developing countries at middle and low latitudes. Plant growth in tropical ecosystems in low latitudes is often limited by phosphorus⁴⁰, and an increase in N input tends to aggravate this limitation and reduce tree productivity⁴¹, further threatening the structure and function of tropical ecosystems. Moreover, developed and developing countries have different dynamic trends of R_NHx/NOy (Fig. 2), which affect ecosystems differently: plants can show a strong preference for NH₄⁺ or NO₃^‒ 42,43 and the impacts of NH₄⁺ and NO₃^‒ deposition on soil acidification and greenhouse gas emissions also differ^44,45. Changes in R_NHx/NOy may therefore lead to changes in the species composition of natural and semi-natural ecosystems.

Reducing N emissions and deposition in developing countries requires global cooperation and a better understanding of industry and agriculture contributions. F_NHx remained the dominant role in global N deposition in 2020, but the relative contribution of changing F_NHx to changes in F_N was decreasing (Supplementary Fig. 12). Meanwhile, we found that F_NHx from agriculture peaks before F_NOy mainly from industrial activities, with higher peak value for F_NHx (Fig. 4). The marginal cost of agricultural E_NH3 reduction is substantially lower than industrial E_NOx reduction⁴⁶, making better N management in agriculture an economically efficient path for reducing N emissions while increasing crop yield and N use efficiency^47,48. Thus, developing countries should prioritize agricultural emission reduction technologies to peak F_NHx before F_NOy. However, most countries lack clear NH₃ emission reduction policies and technologies⁴⁹. Therefore, developing countries should fund agriculture to support rapid emission reduction technology applications alongside industrial development. By creating more efficient agriculture, developing and transitioning countries will avoid following the same path of N pollution as developed countries, and avoid the pollution events such as photochemical smog, acid rain, haze and soil acidification that Europe, North America and China have experienced^50,51.

Uncertainty analysis of global N deposition evaluation

Based on the MGND database, this study developed a framework to conduct a global grid dataset and systematically evaluate the current status, dynamic change, regional patterns, and hotspots transfers of global N deposition. This is a global evaluation independent of atmospheric chemical transport model simulations and based on extensive observation data. Although the estimated global total N deposition input to land (92.7 Tg N yr^-1) is higher than the results from atmospheric chemical transport model simulations (63.9–90.4 Tg N yr^-1)^23–25, by summarizing global natural and anthropogenic E_Nr, we found that total global E_Nr can reach ~125.7–179.5 Tg N yr^-1 (Supplementary Table 2), indicating that our N deposition estimate is within a reasonable range.

Anthropogenic emission inventories (i.e., NH₃ and NOx) are important driving data for atmospheric chemical transport models that simulate atmospheric N deposition. Currently, inventories from the CEDS²⁸ and Emissions Database for Global Atmospheric Research (EDGAR)⁵² are widely used in these models. These two datasets are estimated using a bottom-up approach based on emission activity information and emission factors. However, this approach is highly uncertain, especially in regions lacking socio-economic statistics (such as Africa) and for NH₃ components whose emission factors are highly variable and more complex⁵³. Recent studies using C_N to correct emission inventories have shown that bottom-up estimates of E_Nr are significantly underestimated, with E_NOx and E_NH3 underestimated by about 22% and 38%, respectively (Supplementary Table 2)^53,54, which may contribute to the lower F_N estimates from models compared to our estimates.

We systematically compared the F_N estimates in North America, Europe and China from other studies with our study, as well as the corresponding E_Nr, and found consistent results across all studies (Supplementary Table 4). The relative uncertainty of F_N was less than 10% over most of areas (Supplementary Fig. 13). The uneven distribution of observation sites in the global atmospheric N deposition network may introduce bias in the machine learning models, especially in Africa, Central Asia, Latin America, and Australia, where sites are sparse. Although the cascading network of GDPpc → E_Nr → C_N → meteorological factors→ F_N is globally applicable, it is crucial to strengthen N deposition observations in regions with limited data to reduce evaluation uncertainty.

Methods

Collection of atmospheric N deposition site observation data

We used three sources of N deposition data: (1) data from 43 sites monitoring forests, grasslands, croplands, wetlands, deserts and cities from 2013 to 2020 in the Chinese Wet Deposition observation network (ChinaWD)⁵⁵; (2) shared data from worldwide N deposition monitoring networks: European Monitoring and Evaluation Programme (EMEP) in Europe, the Clean Air Status and Trends Network (CASTNET), the Air Quality System (AQS), and the Ammonia Monitoring Network (AMoN) in the United States, the Canadian Air and Precipitation Monitoring Network (CAPMoN) and the National Air Pollution Surveillance Program (NAPS) in Canada, the Acid Deposition Monitoring Network (EANET) in East Asia, the International Network to Study Deposition and Atmospheric Composition in Africa (INDAAF), and the Nationwide Nitrogen Deposition Monitoring Network (NNDMN) from China Agricultural University¹⁴; (3) 1390 published papers reporting N deposition-related data from various locations (Supplementary Data 1). The criteria for selecting datasets from the literature were that the monitoring index had to include (i) ammonium, nitrate, total wet deposition (sum of ammonium and nitrate) or their concentrations in precipitation; (ii) the concentration or dry deposition of NH₃, NH₄⁺, NO₂, HNO₃ and NO₃⁻; (iii) daily, weekly or monthly observation frequencies; (iv) an observation period longer than one year.

The resultant MGND database includes site name, location, monitoring time, monitoring method, ecosystem type, annual rainfall, N concentration and flux of each component, and the data source. It spans 1977–2021 and includes 25,808 site years of wet deposition and 26,863 site years of dry deposition (Supplementary fig. 1 and Supplementary Table 1). Wet deposition comprised ammonium (F_Wet(NHx)) and nitrate (F_Wet(NOy)); dry deposition comprised gaseous NH₃ (F_Dry(NH3)), particulate NH₄⁺ (F_Dry(NH4+)), gaseous NO₂ (F_Dry(NO2)), gaseous HNO₃ (F_Dry(HNO3)) and particulate NO₃^‒ (F_Dry(NO3-)) (Supplementary Table 1).

Sources of auxiliary analysis data

The sources of auxiliary analysis data and detailed information used in this study are shown in Supplementary Table 5. Meteorological data mainly come from Climatic Research Unit (CRU, version cru_ts4.05)⁵⁶ and the reanalysis product of NECP-NCAR (NECEP-NCAR Reanalysis 1)⁵⁷, including MAP, mean annual temperature (MAT), Wet days (WET), Vapor pressure (VAP), net shortwave radiation flux (Nswrs), surface pressure (Pres), specific humidity (Shum) and wind speed (Wspd).

Anthropogenic pollutants emission inventories were obtained from CEDS (version CEDS v_2021_04_21)²⁸. We also used E_NH3 from EDGAR 6.1⁵² and Luo et al.⁵³. C_NH3 data were obtained from the Infrared Atmospheric Sounding Interferometer (IASI)⁵⁸. We used the standard monthly scale reanalysis of tertiary data products for 2008–2020. C_NO2 data were obtained from the Ozone Monitoring Instrument (OMI), Global Ozone Monitoring Experiment (GOME) and Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY)⁵⁹. The three satellite datasets were integrated into one dataset and covered 1996–2020¹⁵. SO₂ column concentration (C_SO2) were obtained from OMI/Aura SO₂ total column daily L3 data⁶⁰.

Data on the GDPpc and the population of each country were derived from the Maddison Project Database 2020⁶¹. The GDPpc were based on prices in 2011, which eliminated the impact of price changes and reflected the real values of product output over different periods. In addition, we also used PM_2.5⁶², night light⁶³, grid GDP⁶⁴, population⁶⁵, Normalized Difference Vegetation Index (NDVI)⁶⁶, global production-living-ecological space data⁶⁷, terrestrial human footprint dataset⁶⁸, crop-specific N fertilization dataset⁶⁹, and statistics data on N fertilizer application per unit area²² (Supplementary Table 5).

Analysis of temporal dynamic from site observation

We analyzed the variation of F_N in ten major countries or regions: Western Europe, North America, South America, Russia, Africa, Southeast Asia, South Asia, Western Asia, China, Japan and South Korea. The annual F_N in each area was calculated as the annual arithmetic mean and standard error of all the observed data for that area, including F_NHx, F_NOy, F_Wet, F_Dry, and F_Tot.

We used linear or nonlinear equations to analyze the trends in deposition from 1980 to 2020 in each area. We interpolated missing data based on the optimal fitted model for F_N with time for a specific area or period. Mean values and trends in global F_N were calculated using the weighted average method. We selected 352 observation sites with five years of continuous monitoring since 2000 and used the Mann–Kendall method to analyze F_Wet trends at the site scale.

All data were analyzed with SPSS version 13.0 statistical software. The correlations between F_N and GDPpc, E_Nr, C_N and MAP in five major global regions at national and regional scales were analyzed. The structural equation model was used to explore the predicators and influencing paths of F_N. All figures were drawn using SigmaPlot version 12.0 software. The spatial pattern figures for F_N were plotted with ArcGIS 10.0 software.

Construction of global N deposition grid dataset

We developed a framework to generate global grid F_N from site observation data between 2008 and 2020 (Supplementary Fig. 14). We did not extend the data to before 2008 due to insufficient data availability of some important variable (i.e., C_NH3). To minimize the influence of unevenly distribution observation sites on predicting global N deposition, we classified the global land into two categories: wilderness and human-modified area, based on global human footprint data, C_NH3 and C_NO2. The global human footprint reflects various aspects of human pressures using eight variables, including built environments, population density, nighttime light, croplands, pasture lands, roadways, railways, and navigable waterways⁶⁸. We defined wilderness areas as the intersection of regions where the global human footprint data is ≤1, C_NH3 is in the lowest 10%, and C_NO2 is in the lowest 2.5% for each year.

In general, wilderness areas are primarily located in high-latitude northern regions such as Alaska, Greenland, and Siberia, as well as the Sahara Desert in Africa, the Tibetan Plateau in China, and the desert regions of Australia. Our hypothesis is that these wilderness areas are less disturbed by anthropogenic activities, resulting in low levels of F_N. Areas with higher C_N have higher F_N. Therefore, F_N in these areas is estimated as Eq. (1):

$$F_{N,i,j}=\left({Nor}_{CNO2,j}+{Nor}_{CNH3,j}\right)\times 0.01$$ 1

where F_N represents the N deposition flux (kg N ha⁻¹ yr⁻¹); Nor_CNO2 represents normalization of C_NO2; Nor_CNH3 represents normalization of C_NH3; i represents N components; j represents years from 2008 to 2020; 0.01 is a unit conversion factor that considers pre-industrial N deposition levels (kg N ha⁻¹ yr^-1).

For human modified area, we used machine learning methods to upscale F_N from the site scale to the global grid scale. Only data from the main worldwide deposition observation networks was selected as the independent variable to achieve consistency and continuity of observations and methods. Notably, the random forest models demonstrated superior predictive accuracy, indicated by higher R² values both in the training and test sets, compared to support vector machine and BP neural network (Supplementary Table 6). We exclusively used random forest models to predict global N deposition in human-modified areas, employing three key pathways: the n6 model, the n22 best model, and the cascade model (Supplementary Table 7).

The randomForest package⁷⁰ in R software was used to build all the prediction model mentioned above. During model building, 70% of the data were randomly selected as the training set to evaluate the accuracy of the prediction model and 30% selected as the test set to evaluate the prediction performance of the model. The random forest models for F_Dry deposition were constructed using the ground monitoring concentrations of the different components. Deposition was estimated by multiplying the ground monitoring concentrations by the corresponding deposition velocity. We used recursive feature elimination (RFE) method to obtain the optimal variable combination in the n22 best model and the cascade model. Additionally, for all models, we used grid search method to select the best hyperparameters, such as the number of trees (between 100 and 1000), mtry (between 1 and the number of predictor variables or 1/3 of predictor variables for the n22 best models), and nodesize (the number of variables used at each node split, between 1 and the number of predictor variables), to maximize out-of-bag R² value. Shapley values (SHAP) were calculated to determine feature importance and analyze the sensitivity of the output to the input variable.

The n6 model was built with least variable combination based on the cascade network of GDPpc → E_Nr → C_N → MAP → F_N that was proven for China in our previous research¹⁵. The variation explained by the n6 models ranged from 72% to 83% (Supplementary Table 7). E_Nr was the most important predictor for most N deposition components while C_NO2 was the most important predictors for ground HNO₃ and NO₃^- concentration (Supplementary Fig. 15).

The n22 best model builds on the n6 model, further enhancing its explanatory and predictive abilities. Using the RFE method, we optimized the best variable combination from 22 variables. As expected, the n22 best models’ explanation rates are higher than those of the n6 models, except for ground HNO₃ and NO₃^- concentration (Supplementary Table 7). The most important predictor in the n22 best model was nearly the same as in the n6 models (Supplementary Fig. 16).

Given the large uncertainties in the emission inventory data, especially for E_NH3, we assumed this data would significantly impact the prediction results. Therefore, we designed the cascade model to first use C_N, economic activity, and land use to predict E_NH3. Then, we used the predicted NH₃ mission (pNH₃), combined with C_N, meteorological factors, and atmospheric pollutant emission data, to predict F_N (Supplementary Table 7). We extracted four sets of raster data (E_NH3 from CEDS, EDGAR, and two products in Luo et al.⁵³) with less than 10% variation in E_NH3, identified as the more accurate raster data for E_NH3 assessment, and used them as the dependent variable when predicting E_NH3 in cascade model. C_NH3 was the most important predictors for predicting E_NH3, and the variable pNH₃ was the most important predictors for F_N (Supplementary Fig. 17).

Finally, we combined the N deposition dataset for wilderness and human-modified area to generate a spatial dataset for global F_N from 2008–2020 with a spatial resolution of 0.125° × 0.125°. The global annual input of N through deposition to land in 2020 was highest according to the cascade model (98.0 Tg N yr⁻¹), followed by the n22 best model (94.4 Tg N yr⁻¹), and then the n6 model (85.6 Tg N yr⁻¹) (Supplementary Table 8). The relative uncertainty at each pixel were calculated across three models. The relative uncertainty was defined as the ratio of standard error to the mean value of three models. We also used the Theil–Sen Median (Sen slope estimate) to analyze the trend of global N deposition during 2008–2020. The Mann–Kendall nonparametric test was used to determine the significance of the trends.

Relating deposition to economic growth

We integrated the global data of five regions – East Asia, Southeast Asia, Africa, Western Europe and North America – that had relatively long-term N deposition observations at different stages of social development into one reginal dataset. Two methods were used to analyze the relationship between F_N and GDPpc, and to determine whether it conforms to the EKC. Firstly, based on scatter plots, a high-order equation was used to explore the relationship between GDPpc and F_N in each area. Secondly, a logarithmic cubic equation (Eq. (2))³⁶ was used to analyze the relationship between F_N and GDPpc. The logarithmic cubic equation fitting parameters are listed in Supplementary Table 5.

$$\ln F_{Ni}=\alpha +{\beta }_1*\left(ln{GDP}_{pc}\right)+{\beta }_2*{\left(ln{GDP}_{pc}\right)}^2+{\beta }_3*{\left(ln{GDP}_{pc}\right)}^3$$ 2

where GDP_pc is the real per capita GDP of each area in a certain year, F_N is the corresponding average N deposition flux, and i represents ammonium, nitrate, wet, dry or total deposition. β₀ is a constant and β₁, β₂, and β₃ are the estimated coefficients.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

G.Y. designed the research. J.Z., Y.J., N.H., Q.W., Z.C., H.H., X.Z., and P. L. conducted the research (collected the datasets and analyzed the data). J.Z., Y.J., and G.Y. wrote the manuscript. X.L., K.G., D.F., P.V. and F.Z. commented and revised on the manuscript.

Peer review

Peer review information

Nature Communications thanks Hanqin Tian, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The global N deposition grid dataset (2008-2020) and source data are available at 10.6084/m9.figshare.26778574. Monitoring data from EANET are obtained from https://www.eanet.asia/. Monitoring data from EMEP are obtained from https://emep.int/. Monitoring data from CASTNET are obtained from https://www.epa.gov/castnet/download-data. Monitoring data from AQS are obtained from https://www.epa.gov/outdoor-air-quality-data. Monitoring data from AMoN are obtained from nadp.slh.wisc.edu/networks/ammonia-monitoring-network/. Monitoring data from CAPMoN are obtained from https://www.canada.ca/en/environment-climate-change/services/air-pollution/monitoring-networks-data.html. Monitoring data from APQMP are obtained from https://open.canada.ca/data/dataset/ed1d9a68-fce1-4dbc-8158-67d38019aef8. Source data are provided with this paper.

Code availability

The primary code used in this study is available at 10.6084/m9.figshare.26778574.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-024-55606-y.