Nitrous oxide (N₂O) is the third most important long-lived trace greenhouse gas after CO₂ and CH₄. The average atmospheric N₂O concentration has reached to 333.2 ppb in 2020, which is an increase of 23% with respect to pre-industrial levels (1). In natural ecosystems like forests and grasslands, N₂O is primarily produced during re-mineralization of organic matter via the soil microbial processes of nitrification and denitrification (2, 3). Soils are the largest natural source of N₂O, accounting for about 58% of the total natural source (4).

Natural soil N₂O emissions are mainly controlled by climatic and ecological variables (5). Carbon to nitrogen ratios (C/N) have been shown to be a good predictor for N₂O emissions from organic soils (6–9), while N₂O emissions from grassland soils were positively correlated with soil clay content (Clay) and temperature, but negatively correlated with annual precipitation (10). In Brazil, soil pH, soil organic carbon (SOC), and Clay were good predictors for N₂O emissions from unfertilized soils (11). Araujo et al. (2021) showed that annual temperature and soils properties (such as phosphorous, C/N ratio, clay and sand content, soil nitrate contents) were the main drivers for the spatial distribution of N₂O emissions from grassland and forest soils in Argentina (12). However, the principle factors dominating N₂O emissions have been shown to vary strongly among different regions (13), with the key drivers of global forest and grassland N₂O emissions still not clear.

Precise estimates of the global N₂O emissions from forests and grasslands are difficult to obtain owing to the high spatiotemporal heterogeneity of N₂O flux caused by the complex production and emission processes. Nevertheless, process-based models, with detailed description of the N cycling process exist and are considered powerful tools for estimating regional or global N₂O emissions (14). Various models have been developed to simulate N₂O emissions from soils, such as DNDC (15), LPJ (16), and DLEM (17). However, few process-based models have been validated against in situ observations across a wider domain with various climatic and ecological conditions. Consequently, it’s difficult to verify the reliability of the regional or global estimates, which always have a high uncertainty (18, 19). Although the reliability of regional estimates is gradually reduced with a better understanding of the N cycling process (4, 5, 13, 20, 21), the principle factors and the feedback mechanisms of N₂O emissions at regional scale are still highly uncertain (4, 5, 20, 22). Thus, identifying robust drivers of N₂O emissions for different ecological zones is critical for reducing the uncertainty of global estimates.

Field data-oriented analysis can be an effective tool to better understand regulating factors of N₂O emission from different ecosystems, if sufficient data is available (13). There have been data-oriented studies on forest and grassland N₂O emissions at global scale (13, 23, 24). Kim et al. (2013) found that only annual temperature had a significant correlation with N₂O emissions from natural soils (24). Zhuang et al. (2012) used the Artificial Neural Network (ANN) model to show that precipitation was the most sensitive variable for natural N₂O emissions (13). However, both of them didn’t take the effect of soil texture into account, which also is known to be an important factor for natural N₂O emissions (25). Stehfest et al. (2006) used linear mixed-effects model (LMM) to evaluate the effects of all possible continuous or discrete variables on N₂O emissions. Finally, it verified that SOC, pH, bulk density, drainage, and vegetation type have a significant influence on N₂O emissions from natural soils (23). LMMs are now widely used for analyzing datasets from multiple experiments (26). However, the observation length of 57% of the data used was less than 50 days, and only 12% of the data came from observations over more than 300 days. Thus, these datasets fail to accurately represent conditions over the entire vegetation cycle, resulting in large uncertainties. Since 1980s, a large number of long-term field N₂O experiments has been conducted throughout the world, providing sufficient representative data for field data-oriented analysis (27, 28).

Consequently, the aim of this study was to assess whether the magnitude of annual N₂O emissions from forest and unfertilized grasslands sites at global scale can be related to specific environmental factors (i.e. vegetation, soil pH, soil carbon content, soil nitrogen content, soil C/N ratio, soil bulk density, soil clay content, soil sand content, air temperature, precipitation, and N deposition etc.); and to explore whether the robust predictors are different for different land use and regions. To address these objectives, in-situ field measurements of annual N₂O emissions with detailed reports of soil properties were collected from forest and unfertilized grassland sites covering a wide range of environmental conditions that varied from 44.83S to 64.27N in the latitude direction at a global scale.

The primary measurement systems for N₂O emission are eddy covariance systems and the closed chamber technique (38, 39). Among those, the static chamber method has been widely used by collecting air samples in chambers and then analyzing with a gas chromatograph equipped with an electronic capture detector (40). Given that there is no systematic research on the difference between the two observation systems (41, 42), our study only considered the N₂O emission data that observed by the static chamber method in this study. In addition, only studies that reported emissions were considered, while those reporting soil N₂O uptake were neglected because N₂O uptake in natural soils is very weak and reported uptake rates often range within the detection limit of the measuring system. Moreover, highest uptake rates are generally found in wetland and peatland ecosystems (43), while in forest and grassland soils these are only of minor importance. Furthermore, the limited observations of N₂O uptake in forest and grassland soils are typically very episodic, driven by changes of N O 3 - during the year (44), while our analysis focuses on variations in annual emissions.

In this study, spearman analysis showed that the predominant variables (i.e. Clay, Sand, MAT, and MAP) that affected N₂O emissions were the same for both forest and grassland (Table 2). These factors also have been proven to have significant effects on N₂O emissions by the other studies (13, 24, 45). Moreover, the N₂O emissions from both forest and grassland had a negative relationship with pH, which was in consistent with previous studies (46, 47). But only the correlation in forest was significant (Table 2), most likely be due to the fact that forest soils are more acidic than grassland soils (48) (Figure S1A, FigureS2A). Soil C/N has been reported as good predictor for N₂O emissions from soils, especially for organic soils, with generally negative effects on N₂O emissions (8). However, our study indicated a positive effect of soil C/N on N₂O emissions (Table 2), which can probably be explained by the fact that our analysis only investigated a linear relationship, while the relationship between N₂O emissions and soil C/N is probably better explained by a parabola, and we only found the same negative effect in temperate zone (Table S1). A recent meta-analysis also found a Gaussian curve relationship between annual N₂O fluxes and C/N ratio in organic soils, which could explain the negative relationship when C/N was above 18 (9). However, soil C/N for most sites in our dataset was lower than 18 (Figure S1E, Figure S2E). In addition, atmospheric N deposition had no significant effect on annual N₂O emissions from both forest and grassland (Table 2). Because atmospheric N depositions of most sites were less than 30kg Nha^-1yr^-1, which presumably was too low to significantly affect N₂O emissions.

Soil N₂O productions and emissions result from the synergistic effect of temperature, substrate availability, microbial community, and gas transport process etc. Due to the sensitivity of N₂O production to these factors, N₂O emissions are highly heterogeneous in time and space, even at small scales (49). Our analysis showed that soil sand and clay content were important drivers that affected global N₂O emissions from both forest and grassland soils (Table 2). Soils with heavy clay content had higher N₂O emissions, and this positive effect was more significant in moist and wet regions where usually highest emissions are observed (Table 2; Table S1). This generally agrees with previous reports (37, 50–53) and can be explained by the larger volume of small pores in finer soil, making it more conductive to denitrification and N₂O production (25, 54). In addition, the clay particles are favorable for N mineralization rates (55), but not temperature sensitivity of N₂O emissions (56–59). Furthermore, our results showed that LMM including Clay or Sand for forest and grassland, respectively, had the minimum AIC (Table 3). That is to say, Clay is the best predictor for forest N₂O emissions, while Sand is the best predictor for grassland. However, there is currently no study on the importance of these two factors for different ecosystems at a global scale. We infer that this might be due to general differences in soil type of the two ecosystems. Because in our dataset, about 30% of grassland sites were classified as coarse textured soil, while only 16% of forest sites belonged to coarse groups (Table S2). This results in Sand being a more effective predictor for N₂O emissions from grassland. A similar relationship was also found between mineral-associated organic carbon and soil sand content, which was also more effective in grassland than in forest (60).

The seasonal dynamics of soil temperature and soil moisture could reflect the seasonal course of N₂O (61), while the annual dynamic of temperature and precipitation influence the spatial distribution and intern-annual dynamics of N₂O (4). Spearman correlation analysis showed that the explanations of MAT and MAP on N₂O emissions were relatively high (Table 2). Compared with temperature, precipitation is thought to be more important for N₂O emissions from natural ecosystems (5, 13, 62). In this study, the LMMs’ result showed that precipitation was the most effective factor for global grassland N₂O emission, although the fixed linear relationship was only significantly in dry and moist grassland (Table S1). The average N₂O emissions from wet tropical grassland was significantly lower than that from moist grassland (Table S2). However, there was no significant difference of N₂O emissions between the two zones for forests (Figure 3C). That can probably be attributed to the higher infiltration capacity of grassland soils resulting in more pronounced wetting and drying cycles and thus a stronger negative response of N₂O emissions to rainfall compared to the wet or flooded soil conditions in the finer textured forest soils (51, 63). Furthermore, we found that temperature was more important than precipitation for N₂O emissions from forest (Table 3). This is because of the significant direct or indirect effect of temperature on the soil enzyme activities and the resulting supply of N substrate of nitrification and denitrification (64). Increasing temperature could significantly favor N₂O emissions by a strong biotic regulation via ammonia-oxidizing bacteria amoA gene abundance, while the effect of rainfall reduction was not significant (65). However, some study found that increased or decreased precipitation could promote or suppress N₂O emissions to varying degrees (66). Sometimes, a decreasing soil water content could offset the positive effect of warming on N₂O emissions (67). So, there are still large uncertainties regarding the feedback of N₂O emission to the additive and interactive effects of temperature and precipitation in regional scale.

Up to now, estimates of simulation studies for N₂O emissions from global forest and grassland soils vary from 2.61 Tg N₂O-Nyr^-1 to 11.48 Tg N₂O-Nyr^-1 (4, 5, 13, 20, 21). Using the results of our data analysis we calculated global N₂O emissions for forest and grasslands by extrapolating average emissions for different groups, like climate, texture or vegetation classes without considering the effects of human management. According to the independent classify variable, we estimated that the N₂O emissions form forests and grasslands were 5.22-5.55 Tg N₂O-Nyr^-1, 1.45-1.93 Tg N₂O-Nyr^-1 (Table S2), respectively: This estimate is at the higher end of the estimates of Zhang et al. (2019) (Forest:3.62 ± 0.16 Tg N₂O-Nyr^-1, grassland:1.40 ± 0.03 Tg N₂O-Nyr^-1) (5). If we use the interaction variables to classify, the N₂O emissions form forests and grasslands were 4.11 Tg N₂O-Nyr^-1, 1.06 Tg N₂O-Nyr^-1 (Table S3), respectively, which was similar to the result of Tian et al. (2020) (natural soils, 5.6 Tg N₂O-Nyr^-1) (4). Although, these estimates have large uncertainties, they are well within the range of current estimates of global scale N₂O emissions, showing that the difference of key factors of N₂O emissions in different ecosystems or regions (Table 3; S1) can be a promising approach to estimate N₂O from these ecosystems.

This study highlighted that natural forest soils are a strong natural source of N₂O, with an average annual emission flux almost double that of grassland soils. For both ecosystems, soils with a high clay content in moist tropical climates appear to be a hotspot of N₂O. In addition, grasslands in moist region have relatively higher N₂O emissions than the other regions. Soil texture, annual mean temperature and precipitation are the most important factors that influence forest and grassland N₂O emissions at a global scale. However, the best predictors varied according to land use and region. While clay content was the best predictor for N₂O emissions from forest soils, especially in moist or wet regions, sand content predicted N₂O emissions from dry or moist grasslands in temperate and tropical regions best. However, MAP, with a significant positive effect on N₂O emissions from grassland soils in dry and moist regions, had the highest R² of fixed effected of grassland N₂O emissions.

Although the principle factors of soil N₂O emission are varying in different regions, these simple statistical models can help to derive global estimates for specific ecosystems when no detailed data for process-based simulation models is available. Moreover, this study provides an overall view of them. According to the result of this study, the process-based model estimates could do calibration and validation with more representative sites that capture different primary factors in different regions, which is benefit to increase the reliability and spatial applicability of the model. In addition, in order to accurately evaluate the estimates’ uncertainty, the sensitivity differences of the principle factors in various regions also should be taken into account in future studies.

The original contributions presented in the study are included in the article/Supplementary material. Further inquiries can be directed to the corresponding author.

LJ conducted all literature searches, data analyses, and writing of the manuscript. WZ helped with the project design. QZ and YT participated in the soil and climate information collection. WJ, SC, and TL all provided major feedback, comments, and suggestions in terms of both the overall direction and specific changes to the manuscript and associated literature searches and analyses. All authors contributed to the article and approved the submitted version.