Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Carousel with three slides shown at a time. Use the Previous and Next buttons to navigate three slides at a time, or the slide dot buttons at the end to jump three slides at a time.

Martin Hartmann & Johan Six

Ziheng Peng, Chunling Liang, … Shuo Jiao

Shilai Zhang, Guangfu Huang, … Fengyi Hu

Antonio P. Camargo, Rafael S. C. de Souza, … Paulo Arruda

Joséphine Demay, Bruno Ringeval, … Thomas Nesme

Geoffrey K. Kinuthia, Veronica Ngure, … Luna Kamau

Stefano Menegat, Alicia Ledo & Reyes Tirado

Genevieve L. Noyce, Alexander J. Smith, … J. Patrick Megonigal

Johannes Lehmann, Deborah A. Bossio, … Matthias C. Rillig

Nature Food (2022 )Cite this article

The internal soil nitrogen (N) cycle supplies N to plants and microorganisms but may induce N pollution in the environment. Understanding the variability of gross N cycling rates resulting from the global spatial heterogeneity of climatic and edaphic variables is essential for estimating the potential risk of N loss. Here we compiled 4,032 observations from 398 published 15N pool dilution and tracing studies to analyse the interactions between soil internal potential N cycling and environmental effects. We observed that the global potential N cycle changes from a conservative cycle in forests to a less conservative one in grasslands and a leaky one in croplands. Structural equation modelling revealed that soil properties (soil pH, total N and carbon-to-N ratio) were more important than the climate factors in shaping the internal potential N cycle, but different patterns in the potential N cycle of terrestrial ecosystems across climatic zones were also determined. The high spatial variations in the global soil potential N cycle suggest that shifting cropland systems towards agroforestry systems can be a solution to improve N conservation.

Reactive nitrogen (N) supplies N to soil microorganisms and plants but has a negative impact on the environment by affecting the quality of air and water, which in turn affects human health1. We thus need to maximize the benefits of reactive N while minimizing its negative impact on the environment1. The fate of soil N is affected by the rate of N fluxes and by the chemical form of N2, among a large number of other factors. Soil gross N cycling rates provide an understanding of the internal N cycle. A process-based understanding of global gross N transformations remains paramount to explaining how the internal soil N cycle contributes to sustained N losses from terrestrial ecosystems. Given the importance of soil gross N cycling rates for estimating the potential risk of N loss, it is critical to understand the variability of soil gross N cycling rates resulting from the global spatial heterogeneity of climatic and edaphic variables. However, our understanding of the global spatial variations of soil gross N transformation rates is still insufficient. Conceptual frameworks and empirical studies have been suggested during the past few decades to characterize the soil N cycle. For example, the conceptual model of Davidson et al.3 suggests that soil where nitrate (NO3−) dominates over ammonium (NH4+) has excess N and a ‘leaky’ N cycle (that is, high NO3− losses through denitrification or leaching), whereas soil where NH4+ dominates over NO3− is characterized by a ‘conservative’ N cycle. Experimentally, Corre et al.4 found a conservative N cycle in boreal forests, where soil immobilization rates of NO3− (\(I_{{\mathrm{NO}}_3}\) , the conversion of NO3− into organic N) and NH4+ (\(I_{{\mathrm{NH}}_4}\) , the conversion of NH4+ into organic N) were comparable to rates of gross nitrification (GN, the microbial oxidation of organic N or NH4+ to NO3−) and gross N mineralization (GNM, the conversion of organic N into inorganic N), respectively. However, in tropical forest soils, a leaky N cycle has been observed where GNM and GN are greater than \(I_{{\mathrm{NH}}_4}\) and \(I_{{\mathrm{NO}}_3}\) , respectively5. In temperate grasslands, a leaky N cycle was observed in China, whereas a conservative N cycle was observed in other regions6. Croplands in the different regions are usually also characterized by a leaky N cycle2,7. It is unlikely that a general pattern will emerge from these conceptual frameworks and individual experiments that can be applied to a broad range of ecosystems. However, the findings of the individual experiments can be pooled to show a general tendency of ecosystem N cycling patterns. So far, global gross N transformation rates have not been assessed to explain the pattern of soil internal N cycling and its contribution to potential N losses in different ecosystem types. Furthermore, previous global-scale studies reported that soil gross N cycling rates are mainly driven by a combination of soil attributes and climate8,9, but these studies neglected the connection between gross N cycling rates. The last data synthesis on these processes dates back almost 20 years8 and did not draw firm conclusions about the global pattern of the soil internal N cycle due to the lack of data. There is an urgent need for a global synthesis to clarify how ecosystem-wide, land use, edaphic and climatic factors influence the internal soil N cycle, taking into account the relationship between gross N transformation rates.

To fill these knowledge gaps, we compiled 4,032 observations from 398 published 15N pool dilution and tracing studies (Supplementary References and Supplementary Data 1) incorporating gross N cycling rate data across various ecosystems (Supplementary Fig. 1a,b) to characterize the spatial patterns of global soil N cycling. We also analysed the impacts of soil and climate attributes and their interactions on controlling global soil gross N cycling rates, as well as the relationship between gross N cycling rates. Our synthesis aimed to answer three questions. First, what are the global patterns and spatial variations of soil gross N cycling rates, and do they differ across terrestrial ecosystems and climatic zones? Second, how do soil and climate variables interact with gross N cycling rates globally, and what is the connection between gross N cycling rates? Third, what are the implications of the above relationships for the spatial variations of the global soil N cycle? To answer these questions, we first calculated the average gross N transformation rates across ecosystem types and analysed global-scale patterns in the data (Supplementary Tables 1–10). We then predicted the distribution of soil gross N transformation rates globally by five machine learning models using a global database of soil and climatic variables (Supplementary Figs. 2a–c and 3a–h). Next, we conducted structural equation modelling (SEM) to estimate the factors that directly and indirectly control soil N cycling. Finally, we calculated the ratios of gross autotrophic nitrification (GAN, the microbial oxidation of NH4+ to NO3−) to \(I_{{\mathrm{NH}}_4}\) and of soil NO3− to NH4+, and we used mixed-effects meta-regression models to investigate the main factors affecting these ratios. These ratios are utilized as indicators of the potential risk of N losses.

In our dataset, most incubation periods for gross N transformation rates ranged from 24 to 48 h, because gross N rate estimates based on 15N isotopic pool dilution after a 48 h incubation can lead to inconsistent estimates8,10. Although many studies suggested an incubation period of 24 to 48 h to minimize the effect of remineralization on computed GNM11, other studies suggested that GNM is overestimated during short incubation periods12. Estimates of soil gross N cycling rates in our analysis should therefore be interpreted with caution. Moreover, most of our data were based on laboratory studies, which do not necessarily reflect the in situ conditions of soil N cycling8,13,14,15. Hence, we recognize that a number of the soil N cycling rates used in our study are possibly more in line with potential rates, a circumstance that also applies to all other studies of this kind. However, to avoid further inconsistencies, we do not use the term ‘potential’ here either, but we point out that the data should be interpreted with the appropriate caution.

The global averages (±standard errors) of GNM, GAN, gross heterotrophic nitrification (GHN, the microbial oxidation of organic N to NO3−), \(I_{{\mathrm{NO}}_3}\) , \(I_{{\mathrm{NH}}_4}\) and dissimilatory nitrate reduction to ammonium (DNRA) were 8.63 ± 0.55, 3.04 ± 0.33, 1.77 ± 0.44, 1.93 ± 0.31, 8.17 ± 0.94 and 0.44 ± 0.09 mg N kg–1 d–1, respectively (Fig. 1a). Soil GN was dominated by GAN (63%; Fig. 1a). GHN was also an important N transformation process, representing 37% and 17% of the total production of NO3− and of mineral N, respectively (Fig. 1a). However, recent studies have shown that GHN is stimulated in the presence of plants16. Hence, since most of the studies included in our analysis were laboratory studies, it can be expected that the fraction of NO3− produced via GHN would be higher than what was demonstrated by our study. Soil \(I_{{\mathrm{NH}}_4}\) dominated (81%) gross N immobilization rate (GI) and consumed 90% of the total NH4+ production, manifesting high NH4+ retention globally (Fig. 1a). This is consistent with previous studies, indicating a preferential microbial uptake of NH4+ (ref. 17). Soil microorganisms prefer NH4+ because of the additional energy requirement for \(I_{{\mathrm{NO}}_3}\) and NO3− reduction and also because NH4+ can suppress \(I_{{\mathrm{NO}}_3}\) (ref. 18). However, most of our results are based on laboratory studies, so this preference may not be absolute but influenced by other factors. Under high plant NH4+ demand, for example, plants outcompeted microbial NH4+ acquisition, resulting in a switch towards \(I_{{\mathrm{NO}}_3}\) (ref. 16). We also cannot ignore that NO3− moves more easily than NH4+ in soil solution by diffusion and mass flow to the root surface. A recent study found that plants take up less labelled NH4+ than NO3−, while soils retain more NH4+ than NO3− (ref. 19). Furthermore, previous studies suggested that sieving stimulates soil \(I_{{\mathrm{NH}}_4}\) but inhibits \(I_{{\mathrm{NO}}_3}\) as a result of evenly distributing NH4+, resulting in an underestimation or overestimation of the gross N transformation rates13. As laboratory studies are probably limited in capturing the full soil gross N cycling rate dynamics, our global estimates of soil gross N cycling rates should be interpreted with appropriate caution.

a, Pattern of global soil gross N cycling and N2O emission rates. b, Pattern of soil gross N cycling and N2O emission rates in the organic soil in forests. c–e, Patterns of gross N cycling rates in the mineral soil in forests (c), croplands (d) and grasslands (e). f, Conceptual diagram of global soil N cycle under different land uses. Differences in GNM (P < 0.0001), \(I_{{\mathrm{NH}}_4}\) (P = 0.045), \(I_{{\mathrm{NO}}_3}\) (P = 0.408), GAN (P < 0.0001), GHN (P = 0.393), DNRA (P = 0.004) and N2O emission (P = 0.01) rates among mineral soil horizons of forests, croplands and grasslands were tested using one-way analysis of variance with least significant differences. The different letters next to the numbers indicate significant differences in gross N transformation and N2O emission rates across terrestrial ecosystems at P < 0.05, while the values in parentheses are the number of observations. The P values were obtained by two-tailed tests. The comparisons among terrestrial ecosystems were here confined to mineral soil horizons. SOM, soil organic matter.

The fact that soil NO3− is more likely to be lost to the environment indicates the need to maximize the global NO3− consumption processes (\(I_{{\mathrm{NO}}_3}\) and DNRA). Although previous studies have demonstrated that the contribution of \(I_{{\mathrm{NO}}_3}\) to GI was negligible20, we found that \(I_{{\mathrm{NO}}_3}\) represents 19% of global GI and 40% of total NO3– production (Fig. 1a). \(I_{{\mathrm{NO}}_3}\) in the soil temporarily converts NO3−-N into microbial biomass, where it can later be converted into stable organic N or remineralized, decreasing the risk of N loss from the soil21. We also found that DNRA accounts for 18.5% of the global NO3− consumption (Fig. 1a). Although we noticed that the processes of \(I_{{\mathrm{NO}}_3}\) and DNRA occur, they are still low and consume less than 50% of the global NO3− production, demonstrating a lower global NO3− retention. However, we cannot disregard recent studies indicating the critical role of plant root exudates in stimulating DNRA in soil22, suggesting that gross N cycling rates based on laboratory studies in our analysis may be different in the presence of plants. As a result of low NO3− retention, high ratios of soil NO3− to NH4+ (5.30) and GAN to \(I_{{\mathrm{NH}}_4}\) (1.73) were observed at the global scale, indicating a leaky N cycle (Fig. 1a), and thus there is a high potential risk of N loss2. A relatively high average nitrous oxide (N2O) emission rate (40 ± 8.0 µg N kg−1 d−1, n = 136) was observed globally (Fig. 1a). However, we observed high spatial variations in the global N cycle (Fig. 2) as its pattern changes from a conservative cycle in forests to a less conservative one in grasslands and a leaky one in croplands (Fig. 1), as discussed below.

a–d, The global spatial variations of GNM (a), DNRA (b), gross ammonium immobilization (c) and gross nitrate immobilization (d).

A decoupled N cycle was observed in croplands: \(I_{{\mathrm{NH}}_4}\) rates were somewhat lower than GNM rates, GN rates were six times those of \(I_{{\mathrm{NO}}_3}\) , the GAN-to-\(I_{{\mathrm{NH}}_4}\) ratio was 2.84 ± 0.73 and the NO3−-to-NH4+ ratio was 12.9 ± 1.71, indicating a leaky soil N cycle (Fig. 1d), which is in line with previous findings23. Soils with a low GAN-to-\(I_{{\mathrm{NH}}_4}\) or a low soil NO3−-to-NH4+ ratio have a lower potential for N losses than those with high ratios2. High ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and NO3− to NH4+ in croplands resulted in high N2O emissions (Fig. 1d and Supplementary Fig. 4). Our study revealed that GAN, GN, and the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and NO3− to NH4+ in grasslands and forests were significantly lower than those in croplands (Fig. 1c–e), which is in line with previous studies8,9,23,24. Agricultural practices result in different soil pH conditions, leading to a different function and structure of the community of soil microorganisms. For example, the high rate of GN in croplands may be associated with high nitrifier activity24. Generally, ammonia-oxidizing bacteria (a type of nitrifying bacteria that oxidizes ammonia to NO3−) cannot grow in soil with pH less than 5.0–5.5 (ref. 25). In our dataset, croplands have an average pH of 6.26, conditions that favour ammonia-oxidizing bacteria25. Long-term N supply would promote GN through enhancing the abundance and activity of ammonia-oxidizing bacteria26. However, agricultural practices increase soil aeration by damaging soil structure, which accelerates carbon (C) decomposition27. Additionally, high rates of mineral N additions block the production of humus-degrading enzymes by soil microorganisms and thus inhibit GNM4 and ultimately GI. Among the climatic zones, the highest rates of GN and the highest ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and NO3− to NH4+ were found in humid subtropical croplands (Supplementary Figs. 5 and 6). In support of this, our global predictions revealed higher rates of GAN and GN as well as higher ratios of GN to \(I_{{\mathrm{NH}}_4}\) in croplands in tropical and subtropical regions (Fig. 3a,b,d).

a–d, The global spatial variations of GN (a), GAN (b), GHN (c) and the ratio of GN to gross ammonium immobilization (d).

We found a coupled N cycle between the organic and mineral layers of forest soils: \(I_{{\mathrm{NH}}_4}\) and \(I_{{\mathrm{NO}}_3}\) rates were comparable to GNM and GN rates, the ratios of GN to \(I_{{\mathrm{NH}}_4}\) and of GAN to \(I_{{\mathrm{NH}}_4}\) were 0.31 ± 0.08 and 0.45 ± 0.10 in the organic and mineral layers, respectively, and the ratios of NO3− to NH4+ were 0.51 ± 0.13 and 1.67 ± 0.20 in the organic and mineral layers, respectively, manifesting a conservative soil N cycle (Fig. 1b,c), which is consistent with earlier findings23. GNM and \(I_{{\mathrm{NH}}_4}\) in croplands were significantly lower than those in grasslands and forests (Fig. 1c–e), which is again consistent with previous studies8,9,23,24. Our global predictions are in line with our observed patterns of gross N transformation rates across ecosystem types (Fig. 2a,c,d), as forest and grassland soils mostly had high rates of GNM and \(I_{{\mathrm{NH}}_4}\) across various climatic zones, and most had high \(I_{{\mathrm{NO}}_3}\) rates in tropical and subtropical zones. Former regional-scale to global-scale studies reported that GNM, \(I_{{\mathrm{NH}}_4}\) and \(I_{{\mathrm{NO}}_3}\) were best explained by soil microbial biomass8,9, which is consistent with our findings (Supplementary Tables 1, 6 and 7). Soil total C and N, which are key sources of energy for soil microorganisms, were higher in grasslands and forests than in croplands, thus promoting soil microbial biomass24. In support of this, the higher availability of soil substrates to microorganisms in forest organic soil horizons enhances microbial activity and ultimately GNM, \(I_{{\mathrm{NH}}_4}\) and \(I_{{\mathrm{NO}}_3}\) (P < 0.01; Fig. 1b and Supplementary Fig. 7). However, due to the limited availability of substrates in mineral layers of forest soils, microbial activities were restricted9, and thus gross N transformation rates in mineral soil layers also decreased (Fig. 1c). In contrast, significantly higher soil C/N ratios in forests increase the microbial N demand and thus reduce the substrate (NH4+) availability for nitrification, which explains the observed lower rates of GAN and GN (Fig. 4a and Supplementary Tables 2 and 4). Moreover, the rapid recycling of NH4+ in forests may leave little chance for nitrifiers to compete for available NH4+. In our dataset, forests had an average pH of 4.86, so nitrification in forest soils was probably limited by low pH25. This also may explain why GAN in grasslands was higher than in forests (Fig. 1e), as the average pH of grasslands was 6.17 in our dataset. We thus noted a decoupled N cycle in grasslands; total NO3− consumption represents 57% of total NO3− production, and the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and NO3− to NH4+ were 2.08 ± 0.61 and 1.77 ± 0.37, respectively, manifesting a leaky N cycle (Fig. 1e). However, the soil N cycle in grasslands was less leaky than that in croplands; the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and NO3− to NH4+ in grasslands were 1.36 and 7.29 times less than those in croplands.

a, SEM revealing the influences of MAP, MAT, soil pH, soil total N and soil C/N ratio on gross N transformation rates (GNM, \(I_{{\mathrm{NH}}_4}\) , \(I_{{\mathrm{NO}}_3}\) , GAN, GHN and DNRA) and net NH4+ and NO3− production. The black and red arrows indicate significant positive and negative relationships, respectively, where the significance level was set at α = 0.05. *P< 0.05; **P < 0.01; ***P < 0.001, based on two-tailed tests. The values beside the arrows are standardized coefficients. R2 refers to the proportion of the variance explained by endogenous variables. b,c, Model-averaged importance of the predictors of the effect of the variable on the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) (b) and soil NO3− to NH4+ (c). The importance is based on the sum of Akaike weights derived from the model selection process using Akaike’s information criterion corrected for small samples. The cut-off is set at 0.8 (dashed line) to differentiate between important and unimportant predictors.

Furthermore, we analysed a subset of data for sites that measured the full N cycle or most variables of soil N processes (Supplementary Data 2) to test whether the number of observations affected the global pattern of the soil N cycle. The results of this subset confirmed our findings that the soil N cycle pattern changes from conservative in forests to leaky in croplands, as indicated by the increasing ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and of soil NO3– to NH4+ from 0.48 ± 0.10 and 1.72 ± 0.34 in forests to 2.06 ± 0.35 and 14.2 ± 2.94 in croplands, respectively (Supplementary Fig. 8).

Arctic ecosystems are generally expected to be limited by the availability of nutrients, including N. When soil freezes, microbial activity (which is the main stimulator of GNM8) is inhibited as the temperature decreases because the liquid water film, which is a prerequisite for biological activity, is reduced28. This reduction in liquid water films prevents soil substrate diffusion and soil microorganism and enzyme activities29, ultimately reducing GNM9. Moreover, the space of air-filled pores in the soil may decrease as a result of the expansion of water during freezing, causing less oxygen diffusion and the microbial depletion of oxygen remaining in those pores, thereby suppressing aerobic respiration28. Hence, the cold climate in the Arctic slows down the activities of decomposers, reducing GNM. The high C/N ratio is also a major reason for the low N mineralization rates in Arctic soils30. In contrast, our global predictions showed that GNM rates in the Arctic are higher than in some of the most productive black soils on Earth, which is hard to imagine. However, given that the gross N transformation rates included in our global analysis are often measured under laboratory conditions, it is not surprising that C-rich soils would have higher gross N rates than soils with lower C (for example, boreal forests versus croplands)8,9. In contrast to field studies, soil moisture and temperature conditions are precisely controlled in the laboratory, which may affect the activities of the decomposers and ultimately the GNM. For instance, Rustad et al.31 found that a temperature increase of 2.4 °C improves soil N mineralization by 46%. In addition, microbial access to substrates is driven by the availability of water in the frozen soil32. In dry tundra, the effect of snow depth on the increase of soil N availability was less pronounced than that in moist tundra33. Therefore, the difference between field and laboratory flux measurements may be the main reason for the high rate of GNM in the Arctic in our global predictions. Gross N cycling rates in our global analysis should thus be interpreted with caution and need to be validated under field conditions. However, we should not ignore the studies that reported that soil GNM increases with increasing snow depth, which is due to enhanced soil organic C availability and abundance of N mineralization genes34. This increase in organic C substrate availability may have resulted from the increased breakdown of soil organic macromolecules or C and N input through microbial cell turnover or killed roots33. During winter, deepened snow increases the underlying soil thermal insulation, causing higher soil temperatures34. For example, increased snow depth from 30 to 150 cm increased the soil surface temperature by 6 °C35, which may enhance soil organic matter decomposition and gross N transformation rates33,34. There is therefore an ongoing debate about soil N cycling rates in the Arctic, and there is still an urgent need for more field studies to resolve this controversy.

Total soil N content was the most important factor influencing GNM (Fig. 4a). Soils with a higher total N content typically contain more microbial biomass9 and exhibit greater GNM rates8,9 (Supplementary Table 1). This relationship between soil total N and GNM is maintained across terrestrial ecosystems and climatic zones (Fig. 5a and Supplementary Fig. 9a). Precipitation can also influence global GNM (Fig. 4a) by altering plant community composition and related litter fall input, which increases soil substrate availability, thus promoting soil microbial biomass9. In support of this, the highest rates of GNM were observed in tropical forests (Supplementary Fig. 5a) with high rainfall and abundant soil substrates. We also found that GAN is mainly controlled by soil C/N ratio, soil pH and GNM, with standardized coefficients of −2.11, 0.80 and 0.38, respectively (Fig. 4a and Supplementary Table 4). The requirement of microorganisms for inorganic N increases during organic C decomposition in soils with a high C/N ratio, thus decreasing the substrate NH4+ for nitrifiers and resulting in a low abundance of ammonia-oxidizing bacteria, which use NH4+ as a substrate36. We found that GAN increased with increasing ammonia-oxidizing bacteria (R2 = 0.31) and overall bacteria (R2 = 0.52) abundances (P = 0.001; Supplementary Table 4). However, our study showed that soil C/N ratio controls GAN only in natural ecosystems (forests and grasslands) (Fig. 6b) and in all climatic zones except the continental zone (Supplementary Fig. 9c). Free ammonia rather than NH4+ is the substrate of ammonia-oxidizing bacteria. A higher soil pH shifts the equilibrium between NH4+ and ammonia towards ammonia, thus increasing ammonia availability and ultimately GAN (Fig. 4a and Supplementary Table 4). This significant and positive influence of soil pH on GAN is maintained across different terrestrial ecosystems (Fig. 6c), but it has been shown only in the continental, humid subtropical and Mediterranean regions (Supplementary Fig. 9d). Although the stimulated effect of GNM on GAN is plausible because the mineralization process is the master producer of NH4+, which is the main substrate for soil nitrifiers9, our study showed that GNM was a controlling factor of GAN in forest and croplands but not in grasslands (Fig. 6a). In addition, GNM and GAN were correlated only in the continental and humid subtropical zones (Supplementary Fig. 9b).

a, The regression relationship between GNM and soil total N across terrestrial ecosystems. b, The regression relationship between \(I_{{\mathrm{NH}}_4}\) and GNM across terrestrial ecosystems. c, The regression relationship between GHN and GNM across terrestrial ecosystems. d, The regression relationship between GHN and soil pH across terrestrial ecosystems. e, The regression relationship between GHN and soil total N across terrestrial ecosystems. f, The regression relationship between GHN and MAT across terrestrial ecosystems. The solid lines are the slopes, the grey areas indicate the 95% confidence intervals around the regression lines and n is the number of observations. Statistical significance was obtained with a two-tailed Student’s t-test.

a, The regression relationship between GAN and GNM across terrestrial ecosystems. b, The regression relationship between GAN and soil C/N ratio across terrestrial ecosystems. c, Regression between GAN and soil pH across terrestrial ecosystems. d, The regression relationship between \(I_{{\mathrm{NO}}_3}\) and GHN across terrestrial ecosystems. e, The regression relationship between \(I_{{\mathrm{NO}}_3}\) and soil total N across terrestrial ecosystems. f, The regression relationship between DNRA and MAP across terrestrial ecosystems. The solid lines are the slopes, the grey areas indicate the 95% confidence intervals around the regression lines and n is the number of observations. Statistical significance was obtained with a two-tailed Student’s t-test.

Previous studies suggested that GHN in acidic soils contributes to GN37. This is consistent with our SEM, which found that soil pH is a negative factor controlling global GHN (Fig. 4a). Lowering soil pH could enhance soil fungal abundance, which in turn stimulates GHN37. There were positive relationships between GHN and the abundance of fungi (R2 = 0.55, P < 0.001) and the fungi-to-bacteria ratio (R2 = 0.54, P = 0.002) at the global scale (Supplementary Table 3). Soil GHN is more closely related to fungal activity due to their lower N demand per unit C and their higher acid tolerance than bacteria6,37,38. Soils with high organic C and low pH therefore exhibit relatively higher fungal activity. Our study showed that total soil C is positively associated with GHN (P < 0.001) and negatively correlated with GAN (P = 0.11; Supplementary Tables 3 and 4). Unexpectedly, this inverse relationship between soil pH and GHN existed only in croplands (Fig. 4d) and in humid subtropical zones (Supplementary Fig. 10f), while GHN increased significantly with increasing soil pH in forests (Fig. 5d). Furthermore, we found positive relationships among GHN, total N and GNM rate (P < 0.01; Supplementary Table 3), but our SEM (Fig. 4a) revealed that GHN was more closely related to GNM than to total N, indicating the importance of NH4+ as a substrate for heterotrophic nitrifiers. Our global predictions also confirmed the importance of NH4+ as a substrate for heterotrophic nitrifiers, as higher GHN rates were observed in the tropics with higher rates of GNM (Fig. 3c). By examining these relationships in different terrestrial ecosystems, we found that GNM plays a central role in controlling GHN only in forests (Fig. 5c), but soil total N controls GHN in croplands and grasslands (Fig. 5e). In contrast, GHN decreased (R2 = 0.54, P = 0.007, n = 12; Fig. 5c) with increasing GNM in grasslands. Our analysis also revealed that mean annual temperature (MAT) was a driving factor of GHN globally (Fig. 4a and Supplementary Fig. 5f), which is consistent with the finding of Liu et al.39, who reported that high temperatures decrease GHN. Soil fungi, which control GHN (Supplementary Table 3), are more active than bacteria at lower temperatures40. Additionally, our SEM (Fig. 4a) revealed that high temperatures reduce soil total N content, which is a substrate for GHN and GNM. The highest rate of GHN was recorded in the continental climate zone, confirming the negative effect of temperature on GHN (Supplementary Table 3 and Fig. 4a). Our global predictions also revealed high rates of GHN in the continental climate zone (Fig. 3c). However, the effect of MAT on GHN was inconsistent across climatic zones. For example, GHN increased significantly with decreasing MAT in the humid subtropical and Mediterranean regions but with increasing MAT in the continental regions (Supplementary Fig. 9h), suggesting that the effect of temperature on GHN is not universal but has a threshold. The highest average GHN (4.00 ± 2.07 mg N kg−1 d−1, n = 24) in our dataset was recorded when MAT was in the range of 11–15 °C. However, our study is inconsistent with other studies that reported that heterotrophic nitrifiers in hot, semiarid grasslands can nitrify best at 40 °C, a value far above the optimal temperature for heterotrophic nitrifying activities (25 °C) in forested environments with a rainy and warm climate41. We must not ignore that the GHN rates in our study are often estimated under laboratory incubation conditions.

Global GNM is the key factor driving \(I_{{\mathrm{NH}}_4}\) (Fig. 4a and Supplementary Fig. 4b), a relationship that is well established6,8, and this relationship is maintained across terrestrial ecosystems and climatic zones (Fig. 5b and Supplementary Fig. 9i). This is also confirmed by the highest rates of \(I_{{\mathrm{NH}}_4}\) in tropical forests with higher GNM rates (Supplementary Fig. 5d). Soil GHN and total N were the main stimulators of global \(I_{{\mathrm{NO}}_3}\) (Fig. 4a). The stimulating effect of GHN on \(I_{{\mathrm{NO}}_3}\) is plausible, as both require high C availability, which is an unfavourable condition for GAN (Fig. 4a and Supplementary Table 4). GHN is thus a major source of NO3− under these conditions, stimulating global \(I_{{\mathrm{NO}}_3}\) . Moreover, soils with a higher total N content often contain more microbial biomass9 and exhibit greater \(I_{{\mathrm{NO}}_3}\) (ref. 8). Soil microbial biomass stimulates both GNM and GN globally8,9, which are positively correlated with \(I_{{\mathrm{NO}}_3}\) (P < 0.001; Supplementary Fig. 10c,g), as they are responsible for providing a NO3− substrate to soil microorganisms. Positive associations of GHN and total soil N with \(I_{{\mathrm{NO}}_3}\) were observed in croplands and forests but not in grasslands (Fig. 6d,e). Moreover, GHN controlled \(I_{{\mathrm{NO}}_3}\) only in the humid subtropical areas, but soil total N controlled the \(I_{{\mathrm{NO}}_3}\) rate in all climatic zones except for the continental regions (Supplementary Fig. 9j). We also found that the DNRA rate is primarily driven by mean annual precipitation (MAP) (Figs. 4a and 6f), which is in line with previous studies42 and is shown by the higher DNRA rates in the tropical and subtropical regions in our global predictions (Fig. 2b). Soil oxygen depletion as a result of increased moisture content leads to low redox potential, and then NO3− is used as an electron acceptor, facilitating the reduction of NO3− to NH4+ (ref. 42). By testing this relationship across climatic zones, we observed this connection in the marine west coast and tropical wet regions only (Supplementary Fig. 9l), and this was consistent with our global predictions (Fig. 2b). However, we did not observe significant differences in DNRA rates in terrestrial ecosystems across different climatic zones (Supplementary Fig. 5c). In addition, the highest rates of \(I_{{\mathrm{NO}}_3}\) and DNRA were reported from humid subtropical zones, which may be due to the high ratio of soil NO3− to NH4+ in this region compared with other regions (Supplementary Fig. 11), as NO3− is the substrate for both processes2. The high precipitation rate in humid subtropical regions can increase the availability of soil substrate (for example, total N and C) and thus stimulate microbial activity2,9. Furthermore, we found that higher net NH4+ production rates are observed with enhanced GNM and DNRA and are suppressed by increasing \(I_{{\mathrm{NH}}_4}\) . In contrast, net NO3− production rates are stimulated by enhanced GHN, GAN, net NH4+ production and soil pH and are suppressed by increasing \(I_{{\mathrm{NO}}_3}\) (Fig. 4a).

A more detailed understanding of the global N cycle in response to various controls is of great interest to a wide readership, as this ultimately determines important processes such as the ecosystem response to climate change (for example, progressive N limitation theory). It is critical to understand the variability of soil gross N cycling rates resulting from the global spatial heterogeneity of climatic and edaphic variables, which is important for estimating the potential risk of N loss. Our study aimed to predict the global spatial variations of soil gross N cycling rates and highlights promising areas for future 15N gross transformation studies. The type of data used in this study has also been used in previous meta-studies, the last one almost 20 years ago8. Although most gross N transformation rates in our synthesis are possibly not representative of in situ rates (owing to laboratory investigation and the lack of plants), this study provides an overview of the current state of knowledge on gross N rates that goes far beyond previous studies with a limited number of observations8. Unlike previous studies, we are able to draw firm conclusions. Our study shows that soil NO3− retention is lower overall, with wide ratios of soil NO3− to NH4+ and of soil GAN to \(I_{{\mathrm{NH}}_4}\) , indicating a leaky N cycle. The global patterns of the soil N cycle change from conservative in forests to leaky in croplands. We also found a difference in the global N cycle across climatic zones (Supplementary Fig. 11). This underlines the importance of forests in the global N cycle and the need for further insights on NO3− retention in croplands, as well as the potential effect of climate change on the global soil N cycle, as discussed below.

Our study revealed that land use was the most important factor affecting the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and of soil NO3− to NH4+ (Fig. 4b,c). We did not observe significant differences in GAN rates across climatic zones, confirming that land use was more important in controlling GAN rates than climate23. Land use is thus likely to be the controlling factor of the potential risk of global soil N losses. The low ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and of soil NO3− to NH4+ in forests imply that GNM and GI are tightly coupled (Fig. 1b,c), indicating that forests can effectively conserve reactive available N4,23. In contrast, nitrification was the main fate of NH4+ from GNM in croplands (Fig. 1d). Converting croplands to forests would improve soil N retention and minimize N losses to the environment, but this may be difficult to achieve given the need to maintain food security for a rapidly growing population. Instead, we suggest that expanding agroforestry can be a solution that can improve N conservation, among many other benefits. In agroforestry systems, the deep tree rooting can catch and recycle subsoil inorganic N leached below the rooting zone of linked croplands, causing a more efficient interception of the leaked N43. Moreover, NH4+ consumption in tree-based systems is higher, leaving less NH4+ N for nitrification and thus lowering soil N losses, compared with cropland systems44. A recent meta-analysis reported that soil organic C and N storage and available N increased by 21%, 13% and 46%, respectively, under agroforestry compared with crop monocultures43. Increased soil organic C content in agroforests compared with cropland makes the soil N cycle more conservative43,44 and is therefore an important factor for climate-smart agricultural systems.

Although we noticed that \(I_{{\mathrm{NO}}_3}\) and DNRA occur, they are low and consume only 20% of the total NO3− production in croplands, demonstrating a lower NO3− retention in croplands than in other land use systems. DNRA and \(I_{{\mathrm{NO}}_3}\) are favoured by increasing levels of soil C (Supplementary Tables 7 and 11), a condition that restricts GAN (Supplementary Table 4). A rapid depletion of NH4+ was observed when C (for example, in the form of crop residues) was added to soil, causing microorganisms to immobilize NO3− to maintain their growth, which in turn promotes \(I_{{\mathrm{NO}}_3}\) ideally with negligible denitrification N loss45. Concurrently, C supplies electrons through respiration or fermentation, which facilities the reduction of NO3− to NH4+, providing energy to DNRA bacteria46. Exogenous organic C additions can thus promote \(I_{{\mathrm{NO}}_3}\) and DNRA while restricting GAN (Fig. 4a), reducing soil NO3− accumulation in croplands.

Our global analysis revealed that higher temperatures directly reduce GHN and indirectly reduce GNM, GAN, \(I_{{\mathrm{NO}}_3}\) and \(I_{{\mathrm{NH}}_4}\) via reducing soil total N (Fig. 4a). Soil microbial maintenance costs increase with higher temperatures, causing higher energy requirements and lower microbial C use efficiency, which results in lower microbial biomass and gross N transformation rates9,47. Global warming may thus reduce gross N transformation rates in the long run while stimulating it in the short term. Furthermore, the global hydrological cycle is intensifying and will continue to do so in the future, with a global redistribution of precipitation (wet sites become wetter, dry sites drier) and more intense rainfall accompanied by longer and more intense droughts. Soil microbial biomass and N2O emissions decrease with increasing drought intensity, while decreasing precipitation significantly increases extractable NH4+ (ref. 48). Sustained N processing during drought could thus lead to greater N losses during subsequent wetting events. This can also be seen in the increasing soil δ15N values, which are due to the effects of drier conditions49.

Finally, our SEM showed that 53%, 79%, 87%, 73%, 36% and 90% of the variation in GNM, \(I_{{\mathrm{NH}}_4}\) , \(I_{{\mathrm{NO}}_3}\) , GHN, GAN and DNRA, respectively, is still unexplained, which may be because the influence of microbial community structures was not included in the analysis. Although the regression analysis (Supplementary Tables 1, 3 and 4) revealed a vital role of soil microorganisms in controlling soil gross N cycling rates globally, our SEM did not include the effect of soil microorganisms due to the paucity of data. Climate change, N deposition and/or anthropogenic disturbances affect soil microbial community composition50, which might affect soil gross N cycling rates and, eventually, soil N availability and loss. Consequently, future studies should focus more on the effects of microbial community composition on soil gross N cycling rates, which will improve the prediction of soil N cycling under future global changes.

Although our dataset is much larger than those used in previous syntheses, large uncertainties in our estimate of the global N cycle pattern in terrestrial ecosystems still exists. There are three sources of uncertainties.

Most studies included in our dataset were based on laboratory experiments conducted under controlled conditions using disturbed soils and often without plants; these are not necessarily representative of in situ conditions8,13. For example, GNM rates were up to five times lower in the laboratory than in the field14. Soil disturbance such as sieving can alter soil bulk density, aggregate structure, soil aeration, nutrient availability, and the abundance and activity of microorganisms, with consequences for soil N transformations. This is seen, for instance, in an immediate increase in GNM and \(I_{{\mathrm{NH}}_4}\) and suppression of \(I_{{\mathrm{NO}}_3}\) , but with no effect on GN13. Even if GNM is unaffected, a redistribution of substrate may enhance the contact with immobilizers10. In addition, the form in which N fertilizers are applied (for example, as granules in the field or as a liquid in the laboratory) affects the availability of N. Laboratory studies are often carried out with only soil. However, the interactions of plants via C rhizodeposits can affect the soil microbial community and consequently the associated N transformations16. For example, GNM and GHN were stimulated by the presence of wheat15, whereas soil microorganisms switched to assimilate NO3− in the presence of NH4+-preferring plants16. Soil moisture and temperature conditions in the laboratory are carefully controlled, but these conditions are variable in the field. In particular, the fluctuating effect of soil moisture due to precipitation creates conditions referred to as hot moments, where N transformations are different compared with controlled conditions51, which is also reflected by the high temporal variability of N2O emissions52. Soil N2O emissions also vary significantly with fertilizer application mode53, crop type54, irrigation pattern55 and tillage practice53. Since these agricultural factors can influence N2O emissions in the field but not in the laboratory, laboratory studies should be interpreted with caution and, if possible, validated under field conditions.

Our dataset contained results from 15N pool dilution and tracing techniques. These techniques are the most commonly used methods for measuring gross N cycling rates; however, the 15N label addition can increase the size of soil N pools (that is, NH4+ and NO3−), which can stimulate the gross N consumption rates56. This seems to be less of a problem if only low amounts of 15N are applied56,57, which is also dependent on the ecosystem (that is, if it is used to only low N amounts)14. In temperate grassland soils, for example, larger amounts of highly enriched 15N have often been applied58, whereas in low-fertility arable soils, smaller N amounts have been used59. In forest soils, typically 5% of the initial pool size is applied60. Most N additions used in the 15N isotopic pool dilution technique in our dataset were small (0.001–5.0 mg N kg−1), but our synthesis also included studies that used large additions of N, which probably led to an overestimation of \(I_{{\mathrm{NH}}_4}\) and \(I_{{\mathrm{NO}}_3}\) . The combination of the 15N isotopic pool dilution technique with an estimate of the net N turnover in separate samples that do not receive 15N additions (that is, the reformed difference approach) can be an improvement8,57. Furthermore, 15N tracing studies running for long enough that mineral N pools return to background values can evaluate the stimulation effect of N additions61. Advanced natural 15N abundance techniques can also be useful for studying N dynamics without any N application.

Machine learning typically relies on the variance of predictions made by ensembles of models62, such as the random forests (RF) algorithm. Each tree in RF is a model of an ensemble, and the variation in predictions between individual trees is utilized to estimate uncertainties. One issue with these approaches is that information for unknown environments is unavailable, because they do not take into account dissimilarities in the predictor space between new and training data63. The recent suggestion by Meyer and Pebesma63 to add the area of applicability to the modeller’s standard toolkit and to report a map of dissimilarity-index-dependent performance estimates alongside prediction maps may provide improved uncertainty estimates. The area of applicability was not estimated in our study, but our standard deviation maps clearly reveal areas where the models perform poorly or extrapolate with higher standard deviations than the root mean square error (RMSE) (for example, deserts, polar regions and other regions where no observations were available) (Figs. 2 and 3). Our analysis thus highlights promising areas for future 15N gross transformation studies—that is, areas characterized by high uncertainties. Generally, the close correlation between modelled and observed gross N transformation (close to the 1:1 line) confirmed the usefulness of the ensemble machine learning (Supplementary Fig. 12).

All peer-reviewed publications published before December 2020 that examined soil gross N transformation rates were systematically collected by searching Google Scholar and the Web of Science database. We also searched within these publications for references. Studies that have been included in previous meta-analyses of gross N transformation rates were also included in our synthesis8. We used the following search terms: ‘gross nitrogen rates’, ‘soil gross nitrogen transformation’, ‘gross nitrogen mineralization’, ‘gross nitrification’, ‘gross nitrogen immobilization’ and ‘gross dissimilatory nitrate reduction to ammonium’. We followed PRISMA guidelines to conduct the literature search (Supplementary Fig. 13). We employed the following criteria for compiling gross N transformation rate data: (1) soil gross N transformation rates were quantified using the topsoil samples (0–20 cm), (2) most of the incubation periods for gross N transformation rates ranged from 24 to 48 h and (3) the 15N isotopic pool dilution technique and/or tracing models were used to measure gross N transformation rates. In total, 398 studies met these criteria (Supplementary Fig. 13 and Supplementary References). The dataset of gross N transformation rates was created by compiling 4,032 observations representing data from isotope tracing assays in different ecosystems. We evaluated a total of 1,065, 434, 413, 437, 240, 171, 903, 233 and 136 observations for GNM, \(I_{{\mathrm{NH}}_4}\) , \(I_{{\mathrm{NO}}_3}\) , GI, GAN, GHN, GN, DNRA and N2O emission, respectively. The global distribution of study sites for gross N transformation rates included in our study is shown in Supplementary Fig. 1a,b. In large-scale pattern analysis, measurements from organic, mineral and mixed (organic + mineral) soil horizons or from disturbed and intact soils were included; however, we only used data from disturbed mineral soil horizons to compare ecosystem types. Most of the collected studies (312 studies) were conducted under controlled laboratory conditions.

Two authors performed data extraction independently, aiming to extract from the eligible studies the detailed site information such as climatic zone, latitude, longitude, ecosystem type, MAT, MAP and soil chemical (pH, total C and N, C/N ratio, and extractable NH4+-N and NO3−-N) and biological (microbial biomass C and N, fungi-to-bacteria ratio and the abundances of bacteria, ammonia-oxidizing bacteria, ammonia-oxidizing archaea and fungi) attributes, along with soil gross N transformation rates (GNM, GAN, GHN, GN, \(I_{{\mathrm{NO}}_3}\) , \(I_{{\mathrm{NH}}_4}\) , GI and DNRA). The data on the emission of N2O were also collected from the original articles. The ratios of GAN to \(I_{{\mathrm{NH}}_4}\) , GAN to GNM and NO3− to NH4+ were computed and included in the analysis. We also calculated the net NH4+ and NO3− production rates. GetData (v.2.22) (http://getdata-graph-digitizer.com) was used to extract the data contained in graphs. All geographical regions except Antarctica are represented in our dataset, with a wide range of MAP (266–7,000 mm yr−1) and MAT (−4.80 to 28.5 °C). Terrestrial ecosystems in our dataset included forests (58%), grasslands (15%) and croplands (25%). We coded climatic zones as marine west coast, the Mediterranean, tropical wet, continental and humid subtropical according to the Köppen classification system.

We checked the normality of the data using the Kolmogorov–Smirnov test. If the data did not show a normal distribution, a transformation to the natural logarithm was performed to approximate normality and stabilize the distribution.

We calculated the average (±standard error) gross N cycling rates globally and across soil layers and different types of ecosystems. Since there were insufficient data for GAN and GHN in forest organic layers, we used GN to compare mineral and organic soil layers in forests. Differences in gross N cycling rates among soil layers and ecosystem types were tested using analysis of variance with least significant differences for multiple comparisons. Moreover, regression analysis was used to analyse the relationships between soil and climatic variables and gross N transformation rates and between gross N transformation rates and each other (Supplementary Tables 1–10).

The variance inflation factor was used to estimate the collinearities among variables, and variables with a variance inflation factor value of >5 were excluded. We then conducted an SEM using the lavaan package64 in R to test how gross N transformation rates (GNM, GHN, GAN, \(I_{{\mathrm{NH}}_4}\) , \(I_{{\mathrm{NO}}_3}\) and DNRA) and net NH4+ and NO3− production rates are impacted by soil variables (for example, pH, total N and C/N ratio) and climatic variables (MAT and MAP) and by each other. The conceptual SEM included the direct impacts of soil properties and climatic variables on gross and net N transformation rates as well as the effects of gross N transformation rates on each other and on net N production rates. It also included the indirect effects of climatic variables on gross N transformation rates via changing soil attributes. To evaluate the conceptual models, we used goodness-of-fit statistics (comparative fit index, 0.94; Tucker–Lewis index, 0.90). Furthermore, we tested the effect of soil variables (for example, pH, total N and C/N ratio) and climatic variables (MAT and MAP) and/or land use on the ratios of GAN to \(I_{{\mathrm{NH}}_4}\) and of soil NO3− to NH4+ in a mixed-effects meta-regression model using the glmulti package65 in R. We estimated the importance of each variable as the sum of Akaike weights for models that included this variable, which is considered as the overall support for each variable across all models. To explore the most important variables, we set the cut-off to 0.8.

The network of gross N transformation rates was predicted using five machine learning models: RF, support vector machine (svmRadial), generalized boosted regression models (gbm), stepwise regression (leapSeq) and generalized linear models (glmnet). Three of these machine learning methods (RF, gbm and svmRadial) are based on decision trees and boosting approaches, while two are linear regression models (leapSeq and glmnet). The caret66 and caretEnsemble67 packages were used to combine the five approaches, and the single best prediction model was built from these five base models. We created gross N transformation rate models using environmental variables, including climatic factors, soil attributes (pH, N, C, bulk density and clay content) and land use cover (Supplementary Fig. 3a–h). Recursive feature elimination was utilized to estimate the number of covariates that should be included in the model fit. Once a predetermined number of covariates had been reached, the least significant explanatory variable was gradually eliminated to reduce computational load and ensure that the average resolution of all covariates was equal with a spatial resolution of 1 km. These variables were collected from a worldwide collection of soil and climatic property information. The soil property database, which has a geographical resolution of 1.0 km, was derived from the International Soil Reference and Information Centre’s World Inventory of Soil Emissions database (www.isric.org)68. Climatic data with a resolution of 0.5° was obtained using the getdata function from the raster package69. The land use system was sourced from the Food and Agriculture Organization of the United Nations70. The freely available world base map data were downloaded from the Global Administrative Areas Database (https://gadm.org/index.html). The worldwide distribution maps were created using the ESRI ArcGIS program71. All gross N transformation rate predictions in this investigation were made using R v.4.1.1, and the R scripts that were used are available on Figshare72.

We assessed the prediction of gross N transformation rates using tenfold cross-validation with five repeats. The whole database was subsampled into ten subsamples, nine of which served as training data and one as test data. We averaged the test results from each subsample to estimate the model’s performance. The RMSE, the regression coefficients of determination (R2) and the mean of the absolute value of errors (MAE) are three extensively used validation indicators that were calculated according to the following formulas:

where n is the number of samples, and Pi, Oi and \(\bar O\) are the predicted, observed and mean of observed values, respectively.

The model with the lowest MAE, lowest RMSE and greatest R2 for each gross N transformation rate was selected as the best (Supplementary Fig. 2a,b). We used the selected model to map the global gross N transformation rates. The R2 values of the best models are shown in Supplementary Fig. 12. To evaluate the uncertainty of the produced maps, we characterized the distributions of mean, median and quantile values (upper and lower). These four values were computed for each N transformation predicted, and the standard deviation of each map was then derived from these values (mean, median, and upper and lower boundaries). The standard deviation map was used for evaluating the uncertainties of the produced maps. These quantiles were used to express the uncertainties of the global soil map73.

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

The data supporting the findings of this study are available in Supplementary Data 1 and 2. The data underlying Figs. 2 and 3 are available on Figshare (https://doi.org/10.6084/m9.figshare.21406731.v4). Source data are provided with this paper.

The R (v.4.1.2) code used to generate the results and figures reported in this study is available on Figshare (https://doi.org/10.6084/m9.figshare.21406731.v4).

Galloway, J. N. & Cowling, E. B. Reflections on 200 years of nitrogen, 20 years later. Ambio 50, 745–749 (2021).

Zhang, J., Cai, Z. & Müller, C. Terrestrial N cycling associated with climate and plant‐specific N preferences: a review. Eur. J. Soil Sci. 69, 488–501 (2018).

Davidson, E. A., Keller, M., Erickson, H. E., Verchot, L. V. & Veldkamp, E. Testing a conceptual model of soil emissions of nitrous and nitric oxides: using two functions based on soil nitrogen availability and soil water content, the hole-in-the-pipe model characterizes a large fraction of the observed variation of nitric oxide and nitrous oxide emissions from soils. Bioscience 50, 667–680 (2000).

Corre, M. D., Beese, F. O. & Brumme, R. Soil nitrogen cycle in high nitrogen deposition forest: changes under nitrogen saturation and liming. Ecol. Appl. 13, 287–298 (2003).

Rütting, T. et al. Leaky nitrogen cycle in pristine African montane rainforest soil. Glob. Biogeochem. Cycles 29, 1754–1762 (2015).

Wang, J. et al. Soil N transformations and its controlling factors in temperate grasslands in China: a study from 15N tracing experiment to literature synthesis. J. Geophys. Res. Biogeosci. 121, 2949–2959 (2016).

Liu, S. et al. Importance of matching soil N transformations, crop N form preference, and climate to enhance crop yield and reducing N loss. Sci. Total Environ. 657, 1265–1273 (2019).

Article  ADS  CAS  Google Scholar 

Booth, M. S., Stark, J. M. & Rastetter, E. Controls on nitrogen cycling in terrestrial ecosystems: a synthetic analysis of literature data. Ecol. Monogr. 75, 139–157 (2005).

Elrys, A. S. et al. Patterns and drivers of global gross nitrogen mineralization in soils. Glob. Change Biol. 27, 5950–5962 (2021).

Murphy, D. et al. Gross nitrogen fluxes in soil: theory, measurement and application of 15N pool dilution. Adv. Agron. 69, 69–118 (2003).

Bengtson, P. & Bengtsson, G. Bacterial immobilization and remineralization of N at different growth rates and N concentrations. FEMS Microbiol. Ecol. 54, 13–19 (2005).

Braun, J. et al. Full 15N tracer accounting to revisit major assumptions of 15N isotope pool dilution approaches for gross nitrogen mineralization. Soil Biol. Biochem. 117, 16–26 (2018).

Booth, M. S., Stark, J. M. & Hart, S. C. Soil-mixing effects on inorganic nitrogen production and consumption in forest and shrubland soils. Plant Soil 289, 5–15 (2006).

Harty, M. A. et al. Gross nitrogen transformations in grassland soil react differently to urea stabilisers under laboratory and field conditions. Soil Biol. Biochem. 109, 23–34 (2017).

Article  ADS  CAS  Google Scholar 

He, X. et al. Plants with nitrate preference can regulate nitrification to meet their nitrate demand. Soil Biol. Biochem. 165, 108516 (2022).

He, X. et al. Plants with an ammonium preference affect soil N transformations to optimize their N acquisition. Soil Biol. Biochem. 155, 108158 (2021).

Wang, J. et al. The influence of long-term animal manure and crop residue application on abiotic and biotic N immobilization in an acidified agricultural soil. Geoderma 337, 710–717 (2019).

Article  ADS  CAS  Google Scholar 

Rice, C. W. & Tiedje, J. M. Regulation of nitrate assimilation by ammonium in soils and in isolated soil microorganisms. Soil Biol. Biochem. 21, 597–602 (1989).

Gurmesa, G. A. et al. Retention of deposited ammonium and nitrate and its impact on the global forest carbon sink. Nat. Commun. 13, 880 (2022).

Article  ADS  CAS  Google Scholar 

Tahovská, K. et al. Microbial N immobilization is of great importance in acidified mountain spruce forest soils. Soil Biol. Biochem. 59, 58–71 (2013).

Davidson, E. A., Hart, S. C. & Firestone, M. K. Internal cycling of nitrate in soils of a mature coniferous forest. Ecology 73, 1148–1156 (1992).

Biswal, B. K. & Chang, J. in Impact of COVID-19 on Emerging Contaminants (eds Kumar, M. & Mohapatra, S.) 211–229 (Springer, 2022).

Lang, M. et al. Soil gross nitrogen transformations are related to land-uses in two agroforestry systems. Ecol. Eng. 127, 431–439 (2019).

Cookson, W. et al. Controls on soil nitrogen cycling and microbial community composition across land use and incubation temperature. Soil Biol. Biochem. 39, 744–756 (2007).

De Boer, W. & Kowalchuk, G. A. Nitrification in acid soils: micro-organisms and mechanisms. Soil Biol. Biochem. 33, 853–866 (2001).

Venterea, R. T. et al. Nitrogen oxide gas emissions from temperate forest soils receiving long‐term nitrogen inputs. Glob. Change Biol. 9, 346–357 (2003).

Anderson, T.-H. & Domsch, K. H. Soil microbial biomass: the eco-physiological approach. Soil Biol. Biochem. 42, 2039–2043 (2010).

Tucker, C. Reduction of air- and liquid water-filled soil pore space with freezing explains high temperature sensitivity of soil respiration below 0 °C. Soil Biol. Biochem. 78, 90–96 (2014).

Schimel, J. P. & Mikan, C. Changing microbial substrate use in Arctic tundra soils through a freeze–thaw cycle. Soil Biol. Biochem. 37, 1411–1418 (2005).

Paré, M. C. & Bedard-Haughn, A. Landscape-scale N mineralization and greenhouse gas emissions in Canadian cryosols. Geoderma 189–190, 469–479 (2012).

Rustad, L. E. et al. A meta-analysis of the response of soil respiration, net nitrogen mineralization, and aboveground plant growth to experimental ecosystem warming. Oecologia 126, 543–562 (2001).

Article  ADS  CAS  Google Scholar 

Jusselme, M. D. et al. Variations in snow depth modify N-related soil microbial abundances and functioning during winter in subalpine grassland. Soil Biol. Biochem. 92, 27–37 (2016).

Schimel, J. P., Bilbrough, C. & Welker, J. M. Increased snow depth affects microbial activity and nitrogen mineralization in two Arctic tundra communities. Soil Biol. Biochem. 36, 217–227 (2004).

Xu, W. et al. Deepened snow enhances gross nitrogen cycling among Pan-Arctic tundra soils during both winter and summer. Soil Biol. Biochem. 160, 108356 (2021).

Morgner, E., Elberling, B., Strebel, D. & Cooper, E. J. The importance of winter in annual ecosystem respiration in the High Arctic: effects of snow depth in two vegetation types. Polar Res. 29, 58–74 (2010).

Xiao, R., Ran, W., Hu, S. & Guo, H. The response of ammonia oxidizing archaea and bacteria in relation to heterotrophs under different carbon and nitrogen amendments in two agricultural soils. Appl. Soil Ecol. 158, 103812 (2021).

Martikainen, P. J. Heterotrophic nitrification—an eternal mystery in the nitrogen cycle. Soil Biol. Biochem. 168, 108611 (2022).

Pedersen, H., Dunkin, K. A. & Firestone, M. K. The relative importance of autotrophic and heterotrophic nitrification in a conifer forest soil as measured by 15N tracer and pool dilution techniques. Biogeochemistry 44, 135–150 (1999).

Liu, R., Suter, H., He, J., Hayden, H. & Chen, D. Influence of temperature and moisture on the relative contributions of heterotrophic and autotrophic nitrification to gross nitrification in an acid cropping soil. J. Soils Sediments 15, 2304–2309 (2015).

Pietikäinen, J., Pettersson, M. & Bååth, E. Comparison of temperature effects on soil respiration and bacterial and fungal growth rates. FEMS Microbiol. Ecol. 52, 49–58 (2005).

Cookson, W. R., Müller, C., O’Brien, P. A., Murphy, D. V. & Grierson, P. Nitrogen dynamics in an Australian semiarid grassland soil. Ecology 87, 2047–2057 (2006).

Pandey, C. et al. DNRA: a short-circuit in biological N-cycling to conserve nitrogen in terrestrial ecosystems. Sci. Total Environ. 738, 139710 (2020).

Article  ADS  CAS  Google Scholar 

Muchane, M. N. et al. Agroforestry boosts soil health in the humid and sub-humid tropics: a meta-analysis. Agric. Ecosyst. Environ. 295, 106899 (2020).

Kaur, B., Gupta, S. & Singh, G. Soil carbon, microbial activity and nitrogen availability in agroforestry systems on moderately alkaline soils in northern India. Appl. Soil Ecol. 15, 283–294 (2000).

Wang, J. et al. Organic amendment enhanced microbial nitrate immobilization with negligible denitrification nitrogen loss in an upland soil. Environ. Pollut. 288, 117721 (2021).

Van Den Berg, E. M., Boleij, M., Kuenen, J. G., Kleerebezem, R. & van Loosdrecht, M. DNRA and denitrification coexist over a broad range of acetate/N-NO3− ratios, in a chemostat enrichment culture. Front. Microbiol. 7, 1842 (2016).

Dai, Z. et al. Elevated temperature shifts soil N cycling from microbial immobilization to enhanced mineralization, nitrification and denitrification across global terrestrial ecosystems. Glob. Change Biol. 26, 5267–5276 (2020).

Homyak, P. M., Allison, S. D., Huxman, T. E., Goulden, M. L. & Treseder, K. K. Effects of drought manipulation on soil nitrogen cycling: a meta‐analysis. J. Geophys. Res. Biogeosci. 122, 3260–3272 (2017).

Craine, J. M. et al. Global patterns of foliar nitrogen isotopes and their relationships with climate, mycorrhizal fungi, foliar nutrient concentrations, and nitrogen availability. New Phytol. 183, 980–992 (2009).

Castro, H. F., Classen, A. T., Austin, E. E., Norby, R. J. & Schadt, C. W. Soil microbial community responses to multiple experimental climate change drivers. Appl. Environ. Microbiol. 76, 999–1007 (2010).

Article  ADS  CAS  Google Scholar 

Fidel, R., Laird, D. & Parkin, T. Effect of biochar on soil greenhouse gas emissions at the laboratory and field scales. Soil Syst. 3, 8 (2019).

Lan, T., Han, Y., Roelcke, M., Nieder, R. & Cai, Z. Processes leading to N2O and NO emissions from two different Chinese soils under different soil moisture contents. Plant Soil 371, 611–627 (2013).

Liu, X., Mosier, A. R., Halvorson, A. D. & Zhang, F. S. Tillage and nitrogen application effects on nitrous and nitric oxide emissions from irrigated corn fields. Plant Soil 276, 235–249 (2005).

Gerber, J. S. et al. Spatially explicit estimates of N2O emissions from croplands suggest climate mitigation opportunities from improved fertilizer management. Glob. Change Biol. 22, 3383–3394 (2016).

Han, B. et al. The effects of different irrigation regimes on nitrous oxide emissions and influencing factors in greenhouse tomato fields. J. Soils Sediments 17, 2457–2468 (2017).

Davidson, E., Hart, S., Shanks, C. & Firestone, M. Measuring gross nitrogen mineralization, and nitrification by 15N isotopic pool dilution in intact soil cores. J. Soil Sci. 42, 335–349 (1991).

Hart, S. C., Nason, G. E., Myrold, D. D. & Perry, D. A. Dynamics of gross nitrogen transformations in an old-growth forest: the carbon connection. Ecology 75, 880–891 (1994).

Hatch, D. J., Jarvis, S. C., Parkinson, R. J. & Lovell, R. D. Combining field incubation with 15N labelling to examine N transformations in low to high intensity grassland management systems. Biol. Fertil. Soils 30, 492–499 (2000).

Murphy, D. V., Fillery, I. R. P. & Sparling, G. P. Method to label soil cores with 15NH3 gas as a prerequisite for 15N isotopic dilution and measurement of gross N mineralisation. Soil Biol. Biochem. 29, 1731–1741 (1997).

Wessel, W. W. & Tietema, A. Calculating gross N transformation rates of 15N pool dilution experiments with acid forest litter: analytical and numerical approaches. Soil Biol. Biochem. 24, 931–942 (1992).

Müller, C. & Clough, T. J. Advances in understanding nitrogen flows and transformations: gaps and research pathways. J. Agric. Sci. 152, 34–44 (2014).

Bastin, J.-F. et al. The global tree restoration potential. Science 365, 76–79 (2019).

Article  ADS  CAS  Google Scholar 

Meyer, H. & Pebesma, E. Predicting into unknown space? Estimating the area of applicability of spatial prediction models. Methods Ecol. Evol. 12, 1620–1633 (2021).

Rosseel, Y. lavaan: an R package for structural equation modeling. J. Stat. Softw. https://doi.org/10.18637/jss.v048.i02 (2012).

Calcagno, V. & de Mazancourt, C. glmulti: an R package for easy automated model selection with (generalized) linear models. J. Stat. Softw. https://doi.org/10.18637/jss.v034.i12 (2010).

Kuhn, M. Building predictive models in R using the caret package. J. Stat. Softw. https://doi.org/10.18637/jss.v028.i05 (2008).

Deane-Mayer, Z. A. & Knowles, J. caretEnsemble: Ensembles of caret models. R package version 2.0.1 https://github.com/zachmayer/caretEnsemble (2019).

Hengl, T. et al. SoilGrids250m: global gridded soil information based on machine learning. PLoS ONE 12, e016974 (2017).

Fick, S. E. & Hijmans, R. J. WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. Int. J. Climatol. 37, 4302–4315 (2017).

Latham, J., Cumani, R., Rosati, I. & Bloise, M. FAO Global Land Cover (GLC-SHARE) Beta-Release 1.0 Database, Land and Water Division (FAO, 2014).

Elrys, A. et al. A global dataset of soil gross nitrogen transformation rates. figshare https://doi.org/10.6084/m9.figshare.21406731.v4 (2022).

Arrouays, D. et al. Global soil map: toward a fine-resolution global grid of soil properties. Adv. Agron. 125, 93–134 (2014).

We thank all the researchers whose data were used in this global synthesis. We also acknowledge the University of Berkeley, Museum of Vertebrate Zoology, the International Rice Research Institute, R. Hijmans, N. Garcia, J. Kapoor, A. Rala, A. Maunahan and J. Wieczorek for the world base map data. We also acknowledge FAO Global Land Cover (GLC-SHARE) Beta-Release 1.0 Database, Land and Water Division, J. Latham, R. Cumani, I. Rosati and M. Bloise for the global land use and land cover dataset. Financial support for this work was provided by the National Natural Science Foundation of China (grant nos 42122055 (Y.C.), 42150410380 (A.S.E.) and 41977081 (Y.C.)).

School of Geography, Nanjing Normal University, Nanjing, China

Ahmed S. Elrys, Yanhui Zhang, Zhao-xiong Chen, Hui-min Zhang, Yi Cheng, Jin-bo Zhang & Zu-cong Cai

College of Tropical Crops, Hainan University, Haikou, China

Ahmed S. Elrys & Lei Meng

Soil Science Department, Faculty of Agriculture, Zagazig University, Zagazig, Egypt

Ahmed S. Elrys & Mohamed K. Abdel-Fattah

College of Natural Resources and Environment, Northwest A&F University, Yangling, China

Department of Agriculture, Faculty of Agriculture, Environmental Management and Renewable Energy, University of Technology and Arts of Byumba, Byumba, Rwanda

Co-Innovation Center for Sustainable Forestry in Southern China, Nanjing Forestry University, Nanjing, China

Key Laboratory of Karst Dynamics, MLR & Guangxi, Institute of Karst Geology, Chinese Academy of Geological Sciences, Guilin, China

Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing, China

Key Laboratory of Virtual Geographic Environment (Nanjing Normal University), Ministry of Education, Nanjing, China

Department of Renewable Resources, University of Alberta, Edmonton, Alberta, Canada

Institute of Plant Ecology, Justus Liebig University Giessen, Giessen, Germany

School of Biology and Environmental Science and Earth Institute, University College Dublin, Dublin, Ireland

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

You can also search for this author in PubMed  Google Scholar

A.S.E. and Y.C. designed the study. A.S.E., Y.U., M.K.A.-F. and J.W. gathered the data and performed the analysis. A.S.E. and Y.C. took the lead in writing the manuscript. All authors contributed to discussing the results and writing and editing the paper.

The authors declare no competing interests.

Nature Food thanks Wolfgang Wanek and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Tables 1–10, Figs. 1–13 and References.

All the data included in our study.

A subset of data for sites that measured the full N cycling rates or most variables of soil N processes.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Elrys, A.S., Uwiragiye, Y., Zhang, Y. et al. Expanding agroforestry can increase nitrate retention and mitigate the global impact of a leaky nitrogen cycle in croplands. Nat Food (2022). https://doi.org/10.1038/s43016-022-00657-x

DOI: https://doi.org/10.1038/s43016-022-00657-x

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Nature Food (Nat Food) ISSN 2662-1355 (online)

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.