Improved accuracy and reduced uncertainty in greenhouse gas inventories by refining the IPCC emission factor for direct N2O emissions from nitrogen inputs to managed soils

Abstract Most national GHG inventories estimating direct N2O emissions from managed soils rely on a default Tier 1 emission factor (EF1) amounting to 1% of nitrogen inputs. Recent research has, however, demonstrated the potential for refining the EF1 considering variables that are readily available at national scales. Building on existing reviews, we produced a large dataset (n = 848) enriched in dry and low latitude tropical climate observations as compared to former global efforts and disaggregated the EF1 according to most meaningful controlling factors. Using spatially explicit N fertilizer and manure inputs, we also investigated the implications of using the EF1 developed as part of this research and adopted by the 2019 IPCC refinement report. Our results demonstrated that climate is a major driver of emission, with an EF1 three times higher in wet climates (0.014, 95% CI 0.011–0.017) than in dry climates (0.005, 95% CI 0.000–0.011). Likewise, the form of the fertilizer markedly modulated the EF1 in wet climates, where the EF1 for synthetic and mixed forms (0.016, 95% CI 0.013–0.019) was also almost three times larger than the EF1 for organic forms (0.006; 95% CI 0.001–0.011). Other factors such as land cover and soil texture, C content, and pH were also important regulators of the EF1. The uncertainty associated with the disaggregated EF1 was considerably reduced as compared to the range in the 2006 IPCC guidelines. Compared to estimates from the 2006 IPCC EF1, emissions based on the 2019 IPCC EF1 range from 15% to 46% lower in countries dominated by dry climates to 7%–37% higher in countries with wet climates and high synthetic N fertilizer consumption. The adoption of the 2019 IPCC EF1 will allow parties to improve the accuracy of emissions’ inventories and to better target areas for implementing mitigation strategies.


| INTRODUC TI ON
Nitrous oxide (N 2 O) is a potent greenhouse gas (GHG) whose atmospheric concentration's rate of increase has more than quintupled from 0.15 ppbv year −1 a century ago to 0.85 ppbv year −1 in 2001-2015 (Wells et al., 2018). The primary source of this increase is the land and not the oceans, as suggested by changes in nitrogen (N) isotopic composition of atmospheric N 2 O (Jia et al., 2019). According to modeling estimates and global databases, agriculture is accountable for about two-thirds of terrestrial emissions releasing over 6 Tg N 2 O year −1 in 2010-2016 (Jia et al., 2019). N 2 O emissions from the agricultural sector reported in national GHG communications include three main categories: manure management, managed soils, and biomass burning. Managed soils were estimated to contribute as much as 35%-86% to agricultural N 2 O emissions depending on the region  Klein et al. (2006) at 1% of the N either added and returned to soils or mineralized by soils with a confidence interval of [0.3%; 3%] according to findings by Bouwman and Boumans (2002), Bouwman et al. (2002b), Novoa and Tejeda (2006), and Stehfest and Bouwman (2006). The EF 1 emission factor has been criticized for having been derived from a dataset biased toward mid-latitude and temperate regions, being too uncertain, not accounting for differences in environmental conditions, management practices and land use systems, and assigning a linear response of N 2 O emissions to N inputs (Charles et al., 2017).
Emissions of N 2 O from soils result from complex interactions of production, consumption, and gas transport processes, which are controlled by biotic and abiotic factors .
Nitrous oxide is predominantly formed and consumed by oxidation of ammonium (NH 4 + ) through nitrification and reduction of N oxides (nitrate NO − 3 , nitrite NO − 2 ) via denitrification (Hergoualc'h et al., 2007). Rates of nitrification and denitrification at the cellular level are governed primarily by the availability of N, oxygen, and organic carbon (C; Firestone & Davidson, 1989). These controls are affected by numerous properties of the ecosystem and their dynamics (e.g., edaphic properties, climate, plant-microbe interactions) which can exert synergistic or antagonistic influences on the emissions Skiba & Smith, 2000). This complexity results in extreme spatiotemporal variability of N 2 O fluxes at the soil-atmosphere interface often leading to the presence of hot spots and occurrence of hot moments (Groffman et al., 2009;Hénault et al., 2012). Therefore, upscaling N 2 O emissions to national scales and developing emission factors for estimating national emissions with top-down commodity data, such as national fertilizer consumption statistics, remain a challenge Ogle et al., 2013).
The Tier 1 EF 1 allows countries to compute direct N 2 O emissions from managed soils using national data on synthetic and organic N applied to soils, N in crop residues returned to soils, and N mineralized in inorganic soils. This emission factor has been historically derived from experiments looking at the response of N 2 O emissions to N fertilizer application as they outnumber studies examining N 2 O emissions from SOM mineralization or from crop residues returned to soils. While the N application rate is recognized as the best single predictor of N 2 O emissions induced by N fertilization (Albanito et al., 2017;Shcherbak et al., 2014), factors such as climate, edaphic properties, or management practices under various land use systems may interact to a great extent. For instance, Charles et al. (2017) found that the EF 1 specific to organic N fertilizers increased by a factor of five as annual precipitation increased from below 250 mm to above 500 mm. The EF 1 was also found to be influenced by soil properties including C content, texture, and pH, both globally and in nationalscale analyses (Charles et al., 2017;Rochette et al., 2018;Shcherbak et al., 2014). Crop type and fertilizer type modulated the EF 1 computed from global data (Shcherbak et al., 2014), and data from the tropics (Albanito et al., 2017) and Mediterranean climates (Cayuela et al., 2017). Management practices including irrigation or the frequency of fertilizer application (Cayuela et al., 2017;Shcherbak et al., 2014) or parameters linked to the experimental design for measuring the fluxes such as the length of the experiment or the chamber size (Albanito et al., 2017;Shcherbak et al., 2014) were also found to influence the EF 1 . The literature, however, is divided on the type of response of the EF 1 to the N application rate. A response faster than linear has been highlighted at a global scale on yearly fluxes following the application of synthetic fertilizers to various crop types (Gerber et al., 2016;Philibert et al., 2012;Shcherbak et al., 2014) and at local scales for specific crops in the period following N application (Hoben et al., 2011;Oktarita et al., 2017). In contrast, findings by other studies conducted at regional scales (Tropics, Mediterranean climate) or national scales do not support the hypothesis of a nonlinear increase in the annual EF 1 as a function of the N applied (Albanito et al., 2017;Cayuela et al., 2017;Rochette et al., 2018).
The main objective of this research was to refine the IPCC Tier 1 EF 1 emission factor for N 2 O emissions making use of the most recent scientific literature, and considering the influence of climate, management practices, land cover, and edaphic properties. Our approach consisted in compiling and combining existing datasets of EF 1 and controlling variables, retaining only cases for which the EF 1 was based on an unfertilized control site. We classified climate as wet or dry according to the definition adopted by the IPCC (Reddy et al., 2019). Management practices included N fertilizer type (organic, synthetic, mixtures of synthetic and organic forms), N application rate, and irrigation in dry climate. Land cover entailed annual croplands, bare soils, and perennial systems. Edaphic properties included variables related to texture (fine vs. medium and coarse), C content, and alkalinity. We also tested the potential of the experimental length of individual observations to modulate the EF 1 . A second objective of this research was to assess the implications of using the EF 1 disaggregated by climate and fertilizer form from this research and adopted by the 2019 Refinement to the 2006 IPCC guidelines in place of the generic 1% value on direct soil N 2 O emissions from N inputs to global croplands.

| Selection of studies and extraction of data
We extracted all studies from the databases by Stehfest and Bouwman (2006;global  • Were from non-peer-reviewed publications, • Were conducted in the laboratory or greenhouses, and modeling studies (only field studies were selected), • Were conducted in flooded rice fields (emissions from N inputs in flooded rice are estimated using the IPCC EF 1FR ), • Related to grazed soils where urine and/or dung was deposited (emissions from urine/dung inputs in grazed soils are estimated using the IPCC EF 3PRP ), • Related to enhanced efficiency synthetic or organic fertilizer either treated with inhibitors or coated, and • Were conducted on drained and/or managed organic soils (the EF 1 serves for quantifying N 2 O emissions from SOM decomposition in mineral soils).
We further selected the cases from the source databases for which an emission factor was measured or could be computed from a control plot as: where N 2 O Ti is the N 2 O flux during the experimental period due to the application of inputs N i and other unquantified sources of N, and N 2 O Ci is the N 2 O flux during the experimental period at a control plot due to other sources of N than N i .

| Classification of variables influencing the emission factor
Among the variables that were present in the final database and deemed important controlling factors of the EF 1 , we selected those considered the most readily available to countries for conducting national inventories. These factors were related to climate, management practices, land cover, and edaphic properties in the topsoil, and were grouped into classes based on the following criteria.
• Climatic region: Wet or dry. Climate classification initially comprised four classes: temperate/boreal wet, temperate/boreal dry, tropical wet, and tropical dry. It was simplified by distinguishing dry climates from wet climates regardless of latitude since the EF 1 in temperate/boreal and tropical areas either wet or dry were not significantly different from each other (Table   S1). Temperate, boreal, and tropical zones correspond to those • N fertilizer type: Synthetic fertilizer and mixtures of synthetic and organic forms (further referred to as synthetic and mixed fertilizer) or organic fertilizer. The influence of the fertilizer type was first tested using three classes: synthetic, organic, and mixtures of synthetic and organic forms. As the classes synthetic fertilizer and mixtures of synthetic and organic forms yielded similar EF 1 values (Table S1), they were merged into a single class.
• Water management: Irrigation or the absence of irrigation in dry climate.
• Land cover: Annual croplands and bare soils or perennial systems. Bare soils included 70% of bare soils and 30% of crops classified as undefined in the original databases. Perennial systems encompassed perennial croplands, grasslands, agroforestry systems, tree plantations, and managed forests. A preliminary analysis demonstrated a similar response of the EF 1 for the classes of annual croplands, bare soils, and perennial systems (Table S1).
Because vegetation cover over time for annual croplands and bare soils are closer to each other than long-term vegetation cover in perennial systems, the first two classes were grouped into a single class.
• Soil texture class: Fine or medium coarse. Following the USDA classification system (USDA, 2017), fine-textured soils included sandy clay, silty clay, and clay; medium-textured soils were sandy loam, loam, silt loam, silt, clay loam, sandy clay loam, and silty clay loam; coarse-textured soils comprised sand and loamy sand. The EF 1 for medium-and coarse-textured soils were similar (Table S1); therefore, these classes were grouped together.
Several key controlling factors available at (sub)national level which were part of the original databases are not presented because either they had no significant influence on the EF 1 (e.g., soil C:N ratio) or they were seldom reported (e.g., cation exchange capacity).
Some studies noted an influence of sampling-related factors on the EF 1 . In particular, Albanito et al. (2017) found that the EF 1 decreased below 1% in studies longer than 6 months. Therefore, we tested the potential effect of the experimental length of individual experiments on the EF 1 . We considered the length intervals ≤120, (120; 180], (180; 240], (240; 300], and >300 days, according to data distribution ( Figure 1h) and following the classification by Albanito et al. (2017). Other sampling-related factors like chamber size or time elapsed since last N application could not be tested given the scarcity in reporting these variables in original databases.

| EF 1 data analysis
We used linear mixed-effect modeling (Gałecki & Burzykowski, 2013) for testing the response of the EF 1 emission factor to climate, management practices, land cover, edaphic properties, and F I G U R E 1 Frequency of the EF 1 in the dataset among geographical regions according to climate (a), N fertilizer form (b), N application rate (c), land cover (d), soil texture (e), soil C content (f), soil pH (g), and length of the experiment (h) experimental length. This approach was selected to account for lack of independence among data from individual sites compared to data from different sites. A location identification was assigned to all individual observations from experimental sites. Observations either with an identical coordinate or being from the same bibliographic reference with a same soil type and a same land cover were considered a unique location for the analysis.
The models included location identification as a random effect, and climate, management practice, land cover, edaphic property, or experimental length as fixed effects. Means for the fixed effects were compared using the LSD Fisher test. The 95% confidence interval of fitted values by the models was considered for uncertainty quantification of the EF 1 . For each model, we report the level of significance, the root mean square error (R 2 ), which indicates the coincidence between observed and simulated EF 1 values and the Akaike information criterion (AIC) for performance evaluation, where a smaller AIC is better. The statistical analysis was performed using the software Infostat (Di Rienzo et al., 2017).
The influence of controlling factors on the EF 1 was first evaluated independently for each variable. Thereafter, considering that climate is the most readily available information to countries, the influence of each individual factor was tested by climate. To maximize the statistical power and minimize the bias in the estimates and errors of the fixed effects, we limited the analysis to sample sizes >20 (Bell et al., 2010;Hox, 1998 Table   S2 in the paper by Mueller et al., 2012).  (Table S2).

F I G U R E 2
Relative frequency of the EF 1i emission factor (a), soil C content (b), and pH (c)

| Description and representativeness of the EF 1 dataset
The EF 1i (n = 848) were in the range [−0.016; 0.147] and were 70% below 0.01 (Figure 2a). The dataset was unbalanced in geographical coverage and representation of controlling variables. It was dominated by cases from Europe (34%) and North America (28%), followed by Asia (18%) while Africa, Central-South America, and Oceania formed an equal share of the dataset (6%−7%; Figure 1a).
Most studies (76%) were conducted in wet climates except for Africa where the trend was opposite.
Organic and synthetic fertilizers varied by form and rate. The share of research in the dataset evaluating the response of the EF 1 to organic fertilizer application was limited, except for Oceania ( Figure 1b). Organic fertilizers were 33% animal slurry, 31% solid manure, 15% wastewater, and the remaining included liquid manure, compost, crop residues, and other forms. Among the treatments in our dataset, 56% of them applied organic fertilizer in a liquid form and qualified as high risk by Charles et al. (2017), 40% applied organic fertilizer in a solid form (medium-low risk), and 4% were F I G U R E 3 Absolute difference (a) and percentage difference (b) between direct soil N 2 O emissions from global agricultural croplands using the Tier 1 method from the 2019 IPCC Methods Refinement to the 2006 IPCC National GHG Inventories Guidelines (MR; Figure S1a) and the 2006 IPCC National GHG Inventories Guidelines (GL; Figure S1b). The top figures display emissions difference from both synthetic and manure application (total), the middle and bottom figures refer to synthetic and manure application separately [Colour figure can be viewed at wileyonlinelibrary.com] unspecified. Synthetic fertilizers were 25% urea, 23% ammonium nitrate, 20% mixes, and the remaining encompassed anhydrous ammonia and other common mixes such as urea-ammonium-nitrate or calcium-ammonium-nitrate. In addition, 74% of N application rates in the dataset were below 200 kg N ha 1 ; however, Asia (especially China) displayed a greater proportion of studies with high N application rates (46% >200 kg N ha 1 ) in comparison with other regions (Figure 1c).
Perennial systems were not well represented ( Figure 1d) and mostly comprised grasslands for harvesting (88%) and tree plantations (12%, e.g. pine plantations). Annual crops were dominated by wheat (24%) and maize (23%), followed by barley and maize (10% each). The EF 1i were essentially from medium-and coarse-textured soils, though in Central and South America, texture was evenly distributed among classes (Figure 1e). Soil C contents varied from 0.03% to 13.3% with 63% below 2%, and all soils with C content >8% were Andosols ( Figure 2b). Observations from low C content soils were more common apart for North America (Figure 1f). The dataset included more measurements on acid soils than on basic soils except for Europe and Oceania (Figure 1g). Soil pH values ranged from 3.2 to 11.3, with 67% in the range [6; 8] ( Figure 2c). In terms of experimental design, 61% of studies were conducted over a period shorter than 180 days; longer studies were more frequent in Europe than elsewhere (Figure 1h).

| Controlling factors of the EF 1
Climate was a key control of the EF 1 with a mean three times higher in wet climates than in dry climates (Table 1). In terms of management practices, the EF 1 for synthetic and mixed fertilizers was double that of the EF 1 for organic fertilizers while the rate of N application had no effect on the emission factor (p = .0639). The land cover also influenced the EF 1 with a larger mean for annual croplands and bare soils than for perennial systems, but the level of significance of the model (p = .0235) was not as high as for the climate and fertilizer form models (<.0091). Edaphic properties modulated the EF 1 with values two times higher in fine-textured soils than in medium-and coarse-textured soils, in C-rich soils than in soils with low to medium C content, and in acid soils than in basic soils. The models for texture and soil C were highly significant (<0.0001) with an AIC below 3000.
Finally, the analysis indicated a significant but unspecific response of the EF 1 to the experimental length, with shortest (≤120 days) and longest (>300 days) experiments displaying a similar EF 1 (0.012-0.013) and no tendency toward lower EF 1 with increasing experimental length or vice versa. Each of the previously described models explained reasonably well the variation of the EF 1 (.4 ≤ R 2 ≤ .51).
Considering climate is the most accessible information to countries for conducting national GHG inventories, the influence of management practices, land cover, edaphic properties, and experimental design on the EF 1 was tested by climate (Table 2). For several factors (N application rate, land cover, soil C content, experimental length), the sample size was too small for the analysis of dry climates therefore for these variables, the analysis was restricted to wet climates. The form of the fertilizer substantially influenced the EF 1 in wet climates; with a similar response as when climates were aggregated, that is, a higher EF 1 for synthetic and mixed fertilizers than for organic fertilizers. The N application rate did not affect the EF 1 in wet climates, as indicated by the similarity in EF 1 means. In dry climates, irrigation induced a higher EF 1 than for rain-fed lands. This dry climate EF 1 in irrigated fields is very close to the dry climate EF 1 regardless of irrigation (Table 1) as most dry climate observations were from irrigated lands (63%). The larger EF 1 in fine-textured soils than in medium-and coarse-textured soils observed for all climates was persistent in wet climates whereas in dry climates, texture class did not significantly influence the EF 1 (p = .1876). Similarly, the higher EF 1 in C-rich soils than in lower C soils was also significant when the data were limited to wet climates. Soil alkalinity modulated the EF 1 in wet climates with higher values for acid soils, similarly as when climates were grouped together (Table 1). Interestingly, the pattern was opposite in dry climates, with a lower EF 1 in acid soils than in basic soils. Lastly, the experimental length displayed no clear pattern on the EF 1 in wet climates, similarly as for all climates (Table 1).
In wet climates, the most significant models with highest R 2 were the ones using texture class (p < .0001, R 2 = .49) or fertilizer form (p = .0002, R 2 = .48) as a fixed effect; the one with the lowest AIC (1909) was the soil C content model, but it explained less variation in the EF 1 (40%) than the aforementioned models. In dry climates, the model including irrigation was the most performant (AIC = 240) but displayed a relatively low R 2 (.30).
Considering data availability at national level and the performance of the models, the dry climate EF 1 (0.005, 95% CI 0.000-0.011, Table 1) and wet climate EF 1 for synthetic and mixed fertilizer   While manure represents one-third of total N application worldwide (Table S2), the EF 1i in the final dataset were essentially from experiments testing the response of N 2 O emission to synthetic and mixed N fertilizer (80%). Our dataset included most studies from the TA B L E 1 Sample size, mean, and uncertainty range of the EF 1 as influenced by climate, management practices (fertilizer form, N application rate), land cover, topsoil properties (texture class, C content, alkalinity), and experimental design (length of the experiment) Major global annual croplands like wheat, maize, and barley which all together account for 44% of global N inputs (West et al., 2014) were well represented in the dataset while there were relatively few soybean studies, which may not be surprising given the low amounts of fertilizer added to the N-fixing soybeans. Perennials were underrepresented, especially key global crops like sugarcane and oil palm which expand rapidly over the tropics (Phalan et al., 2013;Skiba et al., 2020).

| Implications of using the 2019 IPCC EF 1 disaggregated by climate and fertilizer form in place of the 2006 IPCC 1% EF 1 on direct soil N 2 O emissions from global agricultural croplands
Dominant edaphic properties in the dataset (medium and coarse texture- Figure 1e, low-medium C content- Figure 2b, and [6; 8] pH soils- Figure 2c) mirror characteristics of lands suitable and available for agriculture.
Some final conclusions drawn from the data compilation for this research point toward the necessity to maintain quality, credibility, and transparency standards in science. Several of the databases combined in the dataset were reduced, some of them to a great extent to meet the IPCC quality criteria of selecting peer-reviewed works published in scientific journals. Also, some syntheses that did not fully disclose data sources were discarded.

| The EF 1 , its controls, and its uncertainty
Variations in the EF 1 as affected by the environmental controls are consistent with our process understanding of soil-atmosphere N 2 O exchange. Soil N 2 O fluxes are largely controlled by soil moisture which regulates soil aeration and oxygen supply to microorganisms . N-oxides are emitted predominantly in the form of nitric oxide (NO) below a soil water-filled pore space (WFPS) of around 50%, above which N 2 O dominates over NO and reduces into N 2 at high WFPS (Davidson et al., 2000). The higher EF 1 in wet climates than in dry climates (Table 1) and in irrigated lands than in rain-fed lands of dry climates (Table 2) are consistent with corresponding average WFPS and mechanisms governing nitrification and denitrification. Even though studies seldom reported soil moisture, the average WFPS was significantly higher in wet climates (58%, n = 123) than in dry climates (50%, n = 42; p = .0128) and in irrigated lands (58%, n = 22) than in rain-fed lands (40%, n = 14) of dry climates (p = .0129). Furthermore, a higher EF 1 with increased precipitation is aligned with findings by Charles et al. (2017)  for manure (0.0187) based on a dataset also essentially from wet climates. The discrepancy in the result by Zhou et al. (2017) may lie in differences with the other datasets in the chemical composition and state (raw or composted) of the manure, its application mode (surface or subsurface), and edaphic properties which all have been observed to influence the EF 1 for organic fertilizer in different ways. A lower EF 1 for organic fertilizer than for synthetic and mixed fertilizers has been attributed to the supply of organic C enhancing both N immobilization (hereby reducing substrate supply for nitrification and denitrification) and denitrification reduction of N 2 O to N 2 (Zhou et al., 2017). This explanation supports the similarity in the EF 1 among fertilizer forms in dry climate where denitrification is limited, a result also found by Cayuela et al. (2017).
Contrary to some studies (e.g., Gerber et al., 2016;Philibert et al., 2012;Shcherbak et al., 2014), the EF 1 was not influenced by the rate of N application. Testing this response requires an EF 1i dataset with at least three different levels of N input per site (Shcherbak et al., 2014), which was not part of the objectives of our research.
Instead, we aimed at covering a large range of geographies, management practices, land covers, and edaphic properties to refine the EF 1 for use with national fertilizer consumption statistics. Nonetheless, we recommend countries with detailed fertilizer input rates test for an exponential response of the EF 1 to N inputs and develop their own emission factor response curve. Furthermore, countries with detailed data on N in plants may also test for a response to N surplus (i.e., N applied minus N uptaken by plants) which was found by several studies (Eagle et al., 2020;van Groenigen et al., 2010) to be a better predictor of soil N 2 O emissions than the rate of N application.
Our results suggest a higher EF 1 for annual croplands and bare soils than for perennial systems overall (EF 1 Annual = 0.014 vs. EF 1 Perennial = 0.009, Table 1) and in wet climates (EF 1 Annual = 0.017 vs. EF 1 Perennial = 0.010, Table 2). This result is aligned with findings by Abalos et al. (2016) in Ontario, Canada, who found EF 1 3.7, 3.1, and 1.3 times higher for annual crops than for perennial crops in three consecutive years. The difference in structure and functioning between annual and perennial systems induces distinct soil moisture and nutrient availability patterns and also affects soil microbial community composition Thompson et al., 2016). The permanence of perennial crops roots and their extended architecture maintains stable soil moisture levels over time (Vico & Brunsell, 2018) and favors soil organic matter buildup, which improves soil structure and reduces anaerobic microsites . The synergistic influence of these factors leads to overall lower soil N 2 O emissions in perennial than in annual croplands. This difference is reinforced by the continuous activity of perennial systems throughout the year which, compared to annual crops, reduces soil N availability for microbial conversion to N 2 O Gelfand et al., 2016). Finally, owing to some of the aforementioned mechanisms, distinct N-cycling microbial communities evolve in annual and perennial systems. A detailed description of differences in ammonia oxidizers and denitrifiers composition between annual and perennial croplands is provided by Thompson et al. (2016). As noted by Abalos et al. (2016), the potential for perennial systems to lower N 2 O emissions deserves further research attention; a conclusion greatly reinforced by the disproportion of annual versus perennial cropland studies in our dataset (Figure 1d).
Edaphic properties influence microbial nitrification and denitrification activity in several ways. Soil texture, in combination with soil bulk density and moisture, influences oxygen diffusion through the soil matrix . Generally, poorly drained fine-textured soils favor N 2 O emissions while well-drained coarse-textured soils favor NO emissions (Bouwman et al., 2002a).
This observation supports the larger EF 1 in fine-textured soils than in medium-and coarse-textured soils overall (Table 1) and in wet climates (Table 2), which is also consistent with findings by Charles et al. (2017) and Rochette et al. (2018) when organic fertilizer is applied. The texture effect on the EF 1 was insignificant in dry climates and potentially overridden by the climate effect leading to a dominance of NO emissions over N 2 O emissions regardless of texture.
Notwithstanding, this result is based on a limited number of studies and needs further research of fine-textured soil in dry climates in order to be conclusive.
Soil C plays a major role in N 2 O emissions as it serves as an electron donor for denitrification (Knowles, 1982), affects the water holding capacity and therefore the availability of oxygen in soils (Zhu et al., 2020), and stimulates heterotrophic respiration providing suboxic conditions for dissimilatory nitrate reduction pathways (Morley & Baggs, 2010). While some of these effects counter each other, the EF 1 has generally been found to increase as soil C content reaches higher levels (Charles et al., 2017;Rochette et al., 2018;Shcherbak et al., 2014), which is consistent with our findings.
The control that the pH exerts on soil N 2 O emissions is complex and dependent on nutrient status (Granli & Bøckman, 1996). Globally, the EF 1 increases with decreasing pH (Shcherbak et al., 2014;Wang et al., 2018), possibly as a result of the inhibition of N 2 O reduction into N 2 during denitrification (Hénault et al., 2019). Conversely, where nitrification is the main N 2 O production pathway, emissions tend to increase as the pH increases, at least in the pH range 6-8 (Granli & Bøckman, 1996). While denitrification is believed to be the main N 2 O-forming process, in dry climates, nitrification is likely to be more dominant. Therefore, the opposite response of the EF 1 to the pH in wet (EF 1 acid > EF 1 basic ) and dry (EF 1 acid < EF 1 basic ) climates (Table 2) is coherent with current mechanistic understanding of nitrification and denitrification.
In their review of studies in the tropics, Albanito et al. (2017) observed a decrease in the EF 1 below 1% in studies longer than 6 months and recommended to further evaluate the effect of study length on the response of N 2 O. Like Shcherbak et al. (2014) Table   S2). The difference lies in the manure dataset used by Gerber et al.
(2016) from Herrero et al. (2013) in which application rates are four times lower (7.8 Tg N) than the estimate by West et al. (2014), which we used in our analysis (33.9 Tg N, Table S2). According to a study on global N budget by Zhang et al. (2021), the West et al.'s (2014) data are on the higher end for manure N applied to cropland in the United States but are similar to the 2000 FAO data by Tubiello et al. (2013).
Furthermore, our estimates of direct soil N 2 O emissions from global agriculture (Table 3) are two times lower than the 2.0 Tg computed by Tian et al. (2020) for the year 2000. The later includes emissions from N applied to flooded rice, from crop residue inputs, and from the decomposition of drained organic soils which together may account for the difference in estimates (Gerber et al., 2016;Tubiello et al., 2013 which takes into account moisture regimes and topographic conditions (Canada, 2020), an approach similar to the 2019 IPCC MR. In general, the application of the 2019 IPCC MR will increase emission estimates for those countries with a predominantly wet climate and a large share of synthetic to manure fertilizer consumption such as France, compared to countries with a climate predominantly dry and a small share of synthetic to manure consumption, such as Mexico, and countries with dry climates such as Pakistan (Table 3; Table S2). The application of these factors in countries with a dry climate should be straightforward, while countries with wet or mixed climates, such as Indonesia and Brazil, will need their N consumption data to be disaggregated by fertilizer form and location. Rather than a constraint, however, this is an opportunity for countries to produce more accurate emission data, and better target mitigation strategies.

This research was conducted under CIFOR's Global Comparative
Study on REDD+ (www.cifor.org/gcs) and the Sustainable Wetlands guidelines. The constructive comments of the two anonymous reviewers greatly improved this manuscript and are much appreciated.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available at https://doi.org/10.17528/ CIFOR/ DATA.00273.