X‐ray computed tomography to predict soil N2O production via bacterial denitrification and N2O emission in contrasting bioenergy cropping systems

While renewable biofuels can reduce negative effects of fossil fuel energy consumption, the magnitude of their benefits depends on the magnitude of N2O emissions. High variability of N2O emissions overpowers efforts to curb uncertainties in estimating N2O fluxes from biofuel systems. In this study, we explored (a) N2O production via bacterial denitrification and (b) N2O emissions from soils under several contrasting bioenergy cropping systems, with specific focus on explaining N2O variations by accounting for soil pore characteristics. Intact soil samples were collected after 9 years of implementing five biofuel systems: continuous corn with and without winter cover crop, monoculture switchgrass, poplars, and early‐successional vegetation. After incubation, N2O emissions were measured and bacterial denitrification was determined based on the site‐preference method. Soil pore characteristics were quantified using X‐ray computed microtomography. Three bioenergy systems with low plant diversity, that is, corn and switchgrass systems, had low porosities, low organic carbon contents, and large volumes of poorly aerated soil. In these systems, greater volumes of poorly aerated soil were associated with greater bacterial denitrification, which in turn was associated with greater N2O emissions (R2 = 0.52, p < 0.05). However, the two systems with high plant diversity, that is, poplars and early‐successional vegetation, over the 9 years of implementation had developed higher porosities and organic carbon contents. In these systems, volumes of poorly aerated soil were positively associated with N2O emissions without a concomitant increase in bacterial denitrification. Our results suggest that changes in soil pore architecture generated by long‐term implementation of contrasting bioenergy systems may affect the pathways of N2O production, thus, change associations between N2O emissions and other soil properties. Plant diversity appears as one of the factors determining which microscale soil characteristics will influence the amounts of N2O emitted into the atmosphere and, thus, which can be used as effective empirical predictors.


K E Y W O R D S
bacterial denitrification, computed microtomography, particulate organic matter, plant diversity, sitepreference analysis, soil pore size distributions
The majority of N 2 O production in soils is driven by biologically mediated nitrogen transformations, among which, nitrification and denitrification are generally regarded as the most influential processes in majority of terrestrial ecosystems (Barnard, Leadley, & Hungate, 2005;Fowler et al., 2009). Out of the two, denitrification is frequently found to be highly spatially variable and more difficult to predict, while often the dominant source of soil N 2 O production (Li, Sorensen, Olesen, & Petersen, 2016;Ostrom et al., 2010;Robertson & Tiedje, 1987). Oxygen deficiency is a necessary condition for denitrification to occur, and soil water-filled pore space (WFPS) levels in a 70%-80% range are regarded as optimal for denitrification (Butterbach-Bahl et al., 2013). Due to soil heterogeneity, anoxic conditions suitable for denitrification can occur in isolated microsites even within a well-aerated soil (Keiluweit, Gee, Denney, & Fendorf, 2018;Keiluweit, Nico, Kleber, & Fendorf, 2016;Keiluweit, Wanzek, Kleber, Nico, & Fendorf, 2017;Kravchenko et al., 2017;Loecke & Robertson, 2009;Sexstone, Revsbech, Parkin, & Tiedje, 1985). The size of such local anoxic spots can increase or decrease in response to not only physical drivers, that is, O 2 inflow from the atmosphere, but also biological activities of soil microorganisms consuming O 2 . Emission of N 2 O out of the soil also depends on the severity of O 2 depletion, which can facilitate N 2 O reduction to N 2 , and on the presence of N 2 O escape routes, both governed by gas diffusion (Balaine et al., 2013;Ball, 2013;Mutegi, Munkholm, Petersen, Hansen, & Petersen, 2010). Further complicating matters are patterns of soil wetting/drying, resulting in differences in water distributions in the pore space for the same water contents, thus affecting gas diffusion rates in different parts of pore space and causing trapping of N 2 O in air-disconnected pores (Guo, Drury, Yang, Reynolds, & Fan, 2014;Rabot, Henault, & Cousin, 2014, 2016. Yet another complication is added by influences on N 2 O production from highly spatially variable N and C sources, such as decomposing plant residues (Kravchenko et al., 2017;Parkin, 1987). N 2 O production in soils is a consequence of the activity of nitrifying and denitrifying microorganisms and potentially abiotic processes. Definitive resolution of production pathways remains challenging; however, isotopic site-preference (S P ) analysis can provide considerable insight. S P is defined as the difference in δ 15 N between the central and outer N atoms in N 2 O and has been shown to be constant during microbial production even though δ 15 N and δ 18 O vary or isotopically fractionate (Frame & Casciotti, 2010;Haslun, Ostrom, Hegg, & Ostrom, 2018;Sutka et al., 2006). Specifically, pure culture studies demonstrate that S P values of 33‰-37‰ and −10 to 0‰, respectively, can be used to indicate N 2 O production from hydroxylamine oxidation/fungal denitrification and bacterial denitrification (Sutka et al., 2006;Sutka, Ostrom, Ostrom, Gandhi, & Breznak, 2004). Based on these values, the proportion of N 2 O derived from bacterial denitrification can be determined from the S P value of soil-derived N 2 O (Ostrom & Ostrom, 2011). Reduction of N 2 O during denitrification has the potential to alter S P ; however, the magnitude of this effect is small (Jinuntuya-Nortman, Sutka, Ostrom, Gandhi, & Ostrom, 2008;Ostrom et al., 2007). Opdyke, Ostrom, and Ostrom, (2009 estimated that if 10% of produced N 2 O was reduced, the shift in S P would only be 0.7‰ and, further, would result in an underestimate of the importance of production from bacterial denitrification. The potential for abiotic production of N 2 O in soils is increasingly being recognized and may be particularly important in Fe-rich soils and in the presence of nitrite (Zhu-Barker, Cavazos, Ostrom, Horwath, & Glass, 2015). Although potentially variable (Buchwald, Grabb, Hansel, & Wankel, 2016), the S P of N 2 O produced by abiotic processes is generally in the range of 26‰-35 ‰ (Grabb, Buchwald, Hansel, & Wankel, 2017;Heil et al., 2014;Heil, Liu, Vereecken, & Bruggemann, 2015) and thus similar to the range expected for production via hydroxylamine oxidation and fungal denitrification. Bacterial denitrification largely stands alone with S P values primarily in the range of 0‰-10‰ which provides a basis to constrain this process relative to other production pathways (Ostrom & Ostrom, 2011).
The goal of this study was to assess the contribution of soil pore characteristics and POM to N 2 O production via bacterial denitrification and to N 2 O emissions from soils under several contrasting bioenergy cropping systems. We measured pore characteristics by µCT scanning and production of N 2 O from bacterial denitrification using S P analysis. The studied systems are continuous corn with and without winter cover crop, monoculture switchgrass, poplars, and early-successional vegetation. We hypothesize that longterm growth (9 years) of different plant species affects soil C and N levels as well as the presence, connectivity, and size distributions of soil pores. These characteristics, in turn, will influence C and N availability to microorganisms and will govern the ability of soil to retain water and to enable gas exchange, and thus will both directly and indirectly affect the N 2 O production from denitrification and the emissions of N 2 O overall.
Our first objective was to study presence, connectivity, and size distribution of soil pores with radii >30 µm, under the assumption that their absence leads to anoxic conditions, while presence contributes to O 2 and N 2 O in-and outflows. The second objective was to explore the role of POM on N 2 O production and emission. We hypothesized that greater presence of POM surrounded by soil with prevalence of large pores will lead to greater N 2 O emissions. POM is a source of C and N, and its active decomposition can lead to local anaerobic conditions-both factors favoring N 2 O production-while the presence of large pores in its vicinity would enhance air diffusion and therefore cause larger emissions of produced N 2 O.
In addition, we explored the role of water distribution patterns in N 2 O production and emission. Soil moisture and matric potential are the characteristics that define which soil pores and to which extent are filled with water. Due to the hysteresis of soil water retention, to which extent soil pores are filled with water at any moisture content and matric potential level depends on the history of soil wetting/drying. Generally, at the same matric potential, soil water content is higher, when an initially saturated soil is drained, as compared with a dry soil after rewetting. Because of hysteresis in soil water retention, the spatial distributions of water within the pore space vary. On the one hand, drying of the wet soil forms pockets of trapped water, which reduces gas diffusion and enhances local anoxic conditions. On the other hand, rewetting of dry soil results in trapping air in dead-end pores, which also affects the gas diffusion and local anoxic conditions in soils. To which extent the hysteresis of soil water retention affects N 2 O production and N 2 O emissions from soil is still unknown. We hypothesized that even though soil subjected to the wetting-up mode will have the same soil moisture or the same matric potential level as the soil subjected to the drying-down mode, it will still have lower N 2 O production via denitrification and lower N 2 O emissions.

| Experimental site and studied bioenergy systems
The experimental site is Great Lakes Bioenergy Center's Biofuel Cropping System Experiment located at Kellogg Biological Station, Michigan, USA, on well-drained Alfisol of Oshtemo and Kalamazoo series (mesic Typic Hapludalf; Robertson & Hamilton, 2015). The field experiment has been set up in 2008 as a randomized complete block design with five replications and with bioenergy systems assigned at random to 0.12 ha experimental plots within each replicated block. The five studied systems are two agronomic treatments: continuous corn (Zea mays L.) and continuous corn with winter cover crop of cereal rye (Secale cereale L.), a monoculture switchgrass (Panicum virgatum L.), a hybrid poplar (Populus nigra × P. maximowiczii "NM6") with herbaceous understory (Sprunger & Robertson, 2018), and an early-successional community abandoned from agriculture in 2008. The experimental site was plowed prior to establishment, and no further plowing took place in any of the systems. The two continuous corn systems were managed as no-till. In the two corn systems, 168 kg N/ha of nitrogen fertilizer is applied annually, with 32 kg N/ha at planting and 136 kg N/ha as side-dress. Annually, 56 kg N/ ha is applied to switchgrass and early-successional systems. The poplar system received a single application of 155 kg N/ha in 2010 and no annual applications of N. It should be noted that while, in general, these systems can be grown for several practical purposes, in this study, during the entire field experiment, all the systems were managed exclusively for bioenergy production. Thus, we refer to them as bioenergy systems. Detailed descriptions of the agronomic protocols, biomass harvest, and aboveground residue management are provided by Sanford et al. (2016) and Sprunger, Oates, Jackson, and Robertson (2017).

| Soil sampling
In this study, we collected soil samples only from four of the five replicated blocks of the field experiment. Intact soil cores (5 cm in diameter, 5 cm in height) were collected from four experimental plots of each cropping system, with 2-3 locations sampled within each plot, resulting in a total of 10 soil cores per cropping system. For each system, two cores were used for hysteresis measurements, and eight cores, that is, two cores from each experimental plot, were used in the incubation experiment. The cores were taken from depth of 5-10 cm in February of 2017, weighted, wrapped in an aluminum foil, and stored at 4°C.

| Water retention hysteresis
In order to determine hysteresis in soil of the studied treatments, water retention was measured using a modified evaporation method in two soil cores per cropping system (Wind, 1968). Round bottom ceramic cups (0.95 cm OD, 2.858 cm length, type 0652X07-B01M3, Soilmoisture Equipment Corp., Santa Barbara, CA) connected via pressure transducers (PX26-030DV, Omega Engineering, Inc., Stamford, CT) to a panel meter (DP25B, Omega Engineering, Inc., Stamford, CT) were installed into each core to record pressure head in the soil. Soil cores were, first, capillary saturated by keeping them for 48 hr on water-saturated coarse sand and then gradually fully saturated by increasing the water level around the cores. Saturated cores were weighed and subject to air-drying from the upper surface of the cores. The changes in the pressure head were recorded hourly. Once the change in the pressure head value approached 4-5 kPa, the cores were covered with a plastic cap to stop evaporation and kept for 24 hr to let water pressure equilibrate with the water content in the soil core. Then, soil cores were weighed, and the evaporation procedure was repeated. As a result, we obtained 15-20 approximately evenly spaced data points for the soil water retention curve. The measurements of the drainage curve of the water retention were stopped at pressure head h = −70 kPa.
The saturation limb of the soil water retention was measured in the same samples by incremental addition of water to the soil, followed by recording the pressure heads and weights of cores upon equilibrating pressure heads. The experiment was conducted until full soil saturation, that is, zero pressure head registered by the tensiometers. Then, soil water content was measured gravimetrically. The changes in the core weights were used for the soil moisture calculations at each pressure head.

| Incubation experiments
The remaining soil cores were split into three groups to study the effect of hysteresis in water retention on N 2 O emission rates from the soil. The first group consisted of the cores that after full saturation were brought to −10 kPa pressure using a 5 bar pressure plate extractor (Soilmoisture Equipment Corp., Santa Barbara, CA) and are referred to as drying treatment (Dry). The second and third groups included soil cores that were, first, subject to drainage at −70 kPa and then rewetting, and are referred to as treatments rewetted to same pressure (WetPr) and to the same water content (WetWC). The WetPr treatment was rewetted to −10 kPa pressure head, while the WetWC treatment received water in the same amounts that were lost during the core drainage from −10 to −70 kPa. Therefore, presumably Dry and WetPr had the same pressure heads, but different water contents, while the Dry and WetWC treatments had similar water contents, but different pressure heads (Supporting information Figure S1). The number of replicated cores was 3 in the WetWC and Dry treatments and 2 in the WetPr treatment for each cropping system. Soil preparation for the incubation experiment was conducted in a cold room at 4°C to reduce the microbial activity and took from 7 to 10 days, depending on the time needed for saturation.
After soil cores were brought to their respective hysteresis conditions and weighed, a 1 cm layer was nondestructively cut from each core and used for water content determination. The cores were then placed into 16 oz Mason jars with two sealed rubber stoppers in the jar caps. A vial with 10 ml of water was placed into each jar to maintain high humidity in the jar and to prevent water evaporation from the soil. The seal in the lids was tested using compressed air to prevent N 2 O losses from the jars during incubation. N 2 O content and S P were measured after 72 hr incubation in the dark at temperature of 20°C, as described below. The soil samples were weighed to assure the absence of water content losses and then wrapped and stored in the cold room till X-ray µCT analysis.

| N 2 O and S P analysis
The relative importance of bacterial denitrification (including nitrifier denitrification) to total N 2 O production was determined as described by Ostrom et al. (Ostrom et al., 2010). Specifically, pure culture studies demonstrate that S P values of 33 to 37 and −10 to 0 ‰, respectively, indicate N 2 O production from hydroxylamine oxidation + fungal denitrification and bacterial denitrification (Sutka et al., 2006). Based on these values, the proportion of N 2 O derived from bacterial denitrification can be determined from the S P value of soil-derived N 2 O (Ostrom & Ostrom, 2011). We provide two estimates of the proportion of N 2 O derived from bacterial denitrification: conservative and nonconservative based on endmember S P values of −10‰ and 37‰ and 0‰ and 33‰, respectively (Kravchenko et al., 2017). The N 2 O obtained from the soil cores was analyzed using a Trace Gas System (Elementar) interfaced to an Elementar Isoprime 100 mass spectrometer for determination of bulk δ 15 N, δ 18 O, and S P (Sutka et al., 2006). Within the Trace Gas System, water and CO 2 are removed using chemical scrubbers (magnesium perchlorate and Carbosorb, respectively), and N 2 O is chromatographically separated from the residual CO 2 on a Porapak Q column that is interfaced to the mass spectrometer (Sutka et al., 2006). We applied corrections for the contribution of 17 O to masses 31 and 45 and for a small degree of rearrangement of 15 N between the α and β positions within the ion source (Toyoda & Yoshida, 1999). The concentration of N 2 O is determined from the peak area of the m/z 44 trace during isotopic analysis with a reproducibility (1 SD) of 3% or better (Ostrom et al., 2010

| X-ray scanning
After incubation experiments, the cores were subjected to X-ray scanning using a GE Phoenix v|tome|x at the Institute of Soil and Environment at the Swedish University of Agricultural Sciences in Uppsala. The X-ray scanner was equipped with a 240 kV tube, a tungsten target, and a 16″ flat panel detector with 2,048 × 2,048 detector crystals (GE 1600). Each 3D X-ray µCT image was reconstructed from 2,000 projections acquired at a tube voltage of 130 kV and an electron flux of 200 µA. Each projection was obtained from the average of three consecutive radiographs recorded at the individual projection angles. The exposure time per radiograph was set to 200 ms. No optical filters were used during the image acquisition. 3D µCT X-ray images were reconstructed from the projections using the GE software datos|x. Each image had a resolution 29 µm in all directions.

| Image analysis
The X-ray µCT images were processed in ImageJ/Fiji software (Schindelin et al., 2012). A 2 voxel radius 3D median filter was applied to all images to remove random noise. A region of interest with a diameter and height of 4.5 cm was selected from the central portion of each µCT for the following analyses to avoid sampling artifacts close to the column walls.
For POM determination, we, first, visually identified a subset of 5-12 POM fragments from each core based on the grayscale value, size, and shape characteristics. The range of gray scale values was obtained from the central portion of each fragment. The average of the minimum and maximum gray scale values were then used as a threshold for initial POM identification. Then, a set of erosion/dilation steps was applied to eliminate boundary artefacts, followed by 3D Gaussian filter, and subsequent segmentation. Particle Analyser plugin (BoneJ) was then used to select only POM fragments with volume exceeding 0.016 mm 3 .
Visible gravel/stone fragments >2 mm in size were delineated on the images using single threshold segmentation. Due to their higher attenuation coefficients, most stones are much brighter, that is, have higher grayscale values, than the other soil components, and thus are easy to segment. After thresholding, the >2 mm fraction was identified using Particle Analyser plugin with particle size >8 mm 3 .
The X-ray resolvable soil pore network was obtained by first removing the areas delineated as POM and gravel from the X-ray images. The resulting images exhibited histograms in which gray scale values associated with air-filled pores were easily distinguished from ones pertaining to the soil matrix by using the minimum method. Pore size distributions with spatial locations of different pore radii were obtained using the maximal inscribable sphere method of Pore size distribution tool of Xlib plugin for ImageJ (Münch, 2008).
Given the scanning resolution of our µCT analysis, we could identify pores with radii >30 µm. In the subsequent description and discussion of the results, we interchangeably refer to such pores as either >30 µm pores or as visible pores.
To assess the size of the soil matrix that was not in close contact with >30 µm pores, we delineated a 180 µm wide zone around each µCT visible pore. It was assumed that due to proximity to large water-free pores, the soil within these zones has better aeration, than the remaining soil matrix. While choosing 180 µm for this delineation, we explored the patterns of changes in soil volume for a 30-810 µm range of distances from the pores (Supporting information Figure S1). The studied systems differed from each other the most in a 180-630 µm range, and 180 µm was consistent with spatial correlations reported for biological characteristics and soil C at microscales (Nunan et al., 2003;Quigley, Rivers, & Kravchenko, 2018); however, its choice is somewhat arbitrary in terms of gas diffusion. Delineation of 180 µm zone was conducted as a series of 3D dilations using 3D Dilate tool of ImageJ. The difference between the number of soil matrix voxels and the number of voxels belonging to the within the 180 µm zone was used in further data analyses as an indicator of the size of the volume of poorly aerated soil, and is referred to as Vol-180.
In order to assess POM's aeration status, we overlaid the 3D POM images with the images of µCT visible soil pores and identified the POM fragments that were or were not connected to the soil surface by the µCT visible pores. These two groups of POM fragments are referred to, respectively, as connected and unconnected POM.

| Statistical analysis
Comparisons among the bioenergy systems and hysteresis treatments in terms of the studied soil and N 2 O measurements were conducted using a statistical model with the fixed effects of bioenergy system, hysteresis treatment, and their interaction and with the random effects of the two blocking factors involved in the experiment, that is, the experimental field blocks and the laboratory-processed blocks. Normality of the residuals and homogeneity of variances were checked for each studied variable. In case of marked deviations from normality, the data were logtransformed, for example, cumulative amount of emitted N 2 O, while in case of variance heterogeneity, unequal variance analysis was performed (Milliken & Johnson, 2001). When inherent characteristics of the soil cores were correlated with N 2 O measurements, for example, with WC during saturation, we conducted analysis of covariance (ANCOVA) and compared the effects of systems and hysteresis treatments after accounting for variations in these additional soil core characteristics (Milliken & Johnson, 2009). The analyses were conducted using PROC MIXED procedure of SAS (SAS Inc,9.4).
Relationships among the studied continuous variables were first explored using correlation analyses conducted using the PROC CORR procedure of SAS and then were followed by fitting the relationships that were nonlinear with polynomial regression models using either PROC REG or PROC MIXED procedures. Comparisons among the studied systems in terms of parameters of the regression equations relating N 2 O and soil variables, for example, regression slopes, were conducted using ANCOVA approach in PROC MIXED (Milliken & Johnson, 2009). Path analysis for examining influences on N 2 O emissions was conducted as described in Wuensch (2016).

| Soil characteristics
The soils of the studied bioenergy systems differed in a number of characteristics (Table 1). The two systems with high diversity of plant communities, that is, poplars and early-successional system, had significantly larger soil organic C and N contents than the two no-till corn-based systems. Poplar and early-successional systems had markedly higher total porosity and higher presence of visible pores. However, in terms of the presence of pores <30 µm the studied systems did not differ from each other.
The three systems with perennial vegetation, that is, switchgrass, poplars, and early-successional community, tended to have higher levels of POM than the annual corn systems, and continuous corn had significantly lower POM than the rest of the systems (p < 0.1). The volume of POM connected to the atmosphere by >30 µm pores in poplar and early-successional systems tended to be higher than that in corn with cover crops and switchgrass and exceeded that in the continuous corn.
The Vol-180, that is, the volume of poorly aerated soil, was the lowest in soils from poplar, followed by early-successional system, with the two not significantly different from each other, and was higher in the two corn systems and the switchgrass. At the time of field sampling, the systems did not differ in their water content levels, but the change in water content from the time of field sampling to the level at full saturation tended to be higher in poplars and early-successional systems. All systems had similar WFPS levels.
Soils from the two biodiverse systems, poplars and early succession, substantially differed from the other three systems in terms of their pore size distributions (Figure 1). These soils had higher abundance of 60-270 µm pores than the other three systems. The differences were particularly pronounced for pore sizes between 60 and 180 µm. The differences among corn and switchgrass systems in pore size distributions were not statistically significant, and even numerically the differences were very minor across the entire range of the studied pore sizes.   Table S1). These low SP values are consistent with a strong domination of N 2 O production from bacterial denitrification (Ostrom & Ostrom, 2011). Based on these SP values, the proportion of N 2 O derived from bacterial denitrification ranged from 55% to 91.9% or 64.3 to over 100% depending on whether the conservative or nonconservative model is used (Supporting information  Table S1).

| Effect of bioenergy systems on N 2 O
The bioenergy system effect on the proportion of N 2 O generated via bacterial denitrification (N 2 O-BD) was not statistically significant (p = 0.11). Numerically, corn with cover crop and early succession had lower N 2 O-BD levels than T A B L E 1 Soil characteristics of the studied bioenergy systems. Shown are the means and the standard errors obtained in an ANOVA for each studied soil property along with p values from an F test for the bioenergy system effect (n = 8) the other three systems (Figure 2a). Likewise, the bioenergy systems did not differ in the cumulative amount of N 2 O emitted during 3 day incubations (CumN 2 O). Numerically, the two biodiverse systems had somewhat lower CumN 2 O values than the low-diversity systems (Figure 2b), which is consistent with past field observations from this experimental site .

| Soil variables associated with proportion of N 2 O generated via denitrification and with cumulative emitted N 2 O
The five bioenergy systems separate into two distinct groups in terms of the relationships between soil characteristics and two N 2 O variables, that is, N 2 O-BD and CumN 2 O. The first group consists of the two corn systems and switchgrass, which had lower soil C, low porosity, and low abundance of pores in 60-270 µm range. The second group consists of two biodiverse systems, poplars and early succession, which have high soil C levels, high porosity, and high abundances of pores in 60-270 µm range. Since the main management difference between these two groups is the diversity of their plant communities, we will refer to them as low-diversity and high-diversity systems, respectively.
In soils of low-diversity systems N 2 O-BD was positively correlated with Vol-180 and negatively correlated with presence of pores in 30-120 µm size range ( Table 2). The relationship between N 2 O-BD and Vol-180 was nonlinear with quadratic regression explaining 22% of N 2 O-BD variability (Figure 3a).
In soils of high-diversity systems, N 2 O-BD was related to WFPS and, especially, to saturation of small pores (Table 2), however not to Vol-180 (Figure 3b). N 2 O-BD in the high-diversity systems' soil was negatively related to the change in water content from field level to saturation (Table 2). Lower N 2 O-BD levels were observed in the soil cores where the change in water content was large, while higher N 2 O-BD levels tended to be present in the cores where such change was small. To some extent, this trend was present in low-diversity systems as well, excluding continuous corn, the system where a very narrow and low range of water content change values was observed (Figure 4).
In low-diversity systems, the relationships of CumN 2 O with soil variables were very similar to those of N 2 O-BD, and included positive correlation with Vol-180 and negative correlations with presence of pores in 30-60 µm size range (Table 2). In high-diversity systems, CumN 2 O was strongly associated with several soil variables, including some of the studied water content variables, Vol-180, and pores in the 30-150 µm size group. While in both lowand high-diversity systems, CumN 2 O was positively related to Vol-180, WFPS, and saturation of small pores, the relationships were much stronger in high-diversity systems (Table 2, Figure 3). Contrary to the expectations, neither the total POM nor connected or unconnected POM was related to N 2 O-BD in either low-or high-diversity systems (Table 2). In low-diversity systems, CumN 2 O was not associated with either total POM or connected/unconnected POM. However, in high-diversity systems, the unconnected POM was strongly positively correlated with CumN 2 O ( Figure 5).
The relationships between the two studied N 2 O variables themselves, that is, N 2 O-BD and CumN 2 O, markedly differed between the low-and high-diversity systems ( Figure 6). In low-diversity systems, the emitted CumN 2 O linearly increased as N 2 O-BD increased from 60% to 80%, and slightly decreased as N 2 O-BD increased to 90%. In high-diversity systems, CumN 2 O was not related to N 2 O-BD at the range 60%-80%. Note that N 2 O-BD values >80% were not observed in high-diversity systems.

| Effect of moisture regime
Numerically, the mean values of the N 2 O-BD were the highest in Dry, followed by WetWC and WetPr water treatments, but the differences were not statistically significant (Figure 7a). The highest amount of N 2 O was emitted from the Dry treatment (drying of wet soil), followed by WetWC (wetting dry soil to the same water content as Dry), and the lowest in WetPr (wetting dry soil to pressure similar to that of Dry, but higher than WetWC; Figure 7b).

| DISCUSSION
After 9 years of implementation, the bioenergy systems diverged into two distinct groups not only in terms of their organic matter and pore characteristics, but also in the factors contributing to N 2 O production and emission. The latter is reflected by the differences between these two groups in terms of strengths and directions of the relationships between soil characteristics and N 2 O variables. Specifically, in the low-diversity systems, corn and switchgrass, which had low C and low pore abundances, the emission of N 2 O was strongly associated with the proportion of N 2 O generated via bacterial denitrification. There both the proportion of N 2 O generated via bacterial denitrification and the N 2 O emission were greater at greater volumes of poorly aerated soil (Vol-180). However, water and pore characteristics explained only a small portion of variability in N 2 O emissions in these systems. On the contrary, in the high-diversity systems, poplar and early succession, which had high C and high pore abundances, N 2 O emission was not related to the proportion of N 2 O generated via bacterial denitrification. Yet, a number of water and pore characteristics, notably Vol-180 and the abundance of unconnected POM, explained a substantial amount of variability in N 2 O emissions in high-diversity systems.
Our data supported the hypothesis that long-term implementation of biofuel crops affects soil pore and water characteristics ( Figure 1 and Table 1). Based on their pore characteristics, low-and high-diversity systems can be regarded, respectively, as those prone to high and low oxygen deficiency. Due to higher presence of >30 µm pores and lower Vol-180, the high-diversity systems had more favorable conditions for soil aeration and gas diffusion. On average, 88%-96% of POM in high-diversity systems was connected to the atmosphere by visible pores. On the contrary, the lower presence of >30 µm pores and higher Vol-  Notes. Correlation analysis was conducted separately in corn systems and switchgrass (low diversity) (n =~24) and poplar and early-successional (high diversity) (n =~16) systems. Correlation coefficients in bold and italic are significantly different from zero at p < 0.05 and p < 0.1, respectively 180 of the low-diversity systems suggest greater proneness to anoxic conditions. On average, only 70%-80% of POM in low-diversity systems was connected to the atmosphere by visible pores. Moreover, soil moisture conditions at the time of sampling demonstrated that indeed, low-diversity systems were less aerated, requiring smaller changes in water content at the time of sampling to reach full saturation (Table 1). This suggests that for a long time before the sampling, that is, during fall and winter months when evapotranspiration was low, the conditions within low-diversity systems could have been potentially more conducive to denitrification than in the high-diversity systems. The results only partially supported the hypothesis regarding the importance of water retention hysteresis for N 2 O emissions. It appeared that in this study, the actual water content, and not the mode in which it was reached, that is, wetting versus drying, mattered the most. The N 2 O emissions from the soil that was dried and then brought to −10 kPa pressure were significantly lower than those from the soil that was dried to −10 kPa pressure after saturation as well as from the soil that, after drying, was wetted to the same water content.
We formulated a path model to relate aeration conditions, history of saturation, bacterial denitrification, and N 2 O emissions and fitted it separately to data from low-diversity and from high-diversity systems (Figure 8). The model hypothesizes that the aeration conditions (represented by Vol-180) and a history of saturation (represented by the change in water content from field to saturation) influence N 2 O-BD directly and influence CumN 2 O both directly and indirectly through their effect on N 2 O-BD.

| Soil characteristics related to proportion of N 2 O produced via bacterial denitrification
The effect of anoxic conditions on N 2 O-BD differed in low-diversity and high-diversity systems. In low-diversity  Figure 8). This result is consistent with dominance of bacterial denitrification in soil with higher water content and lower aeration reported by Well, Kurganova, Gerenyu, and Flessa (2006). Kravchenko et al. (2017) observed that in the absence of incorporated plant leaves, the proportion of N 2 O from bacterial denitrification was higher in the soil with a prevalence of small (<10 µm) pores, the outcome driven by less aerated conditions in soils dominated by small pores and consistent with the results from low-diversity systems of this study. However, none of the indicators of low aeration associated with N 2 O-BD in low-diversity systems were related to N 2 O-BD in the high-diversity systems: There was no negative relationship of N 2 O-BD with abundance of 30-120 µm pores (Table 2) and no positive relationship of N 2 O-BD with Vol-180 ( Figure 3b, Table 2). Moreover, Vol-180 had a significant negative direct effect on N 2 O-BD in high-diversity systems (Figure 8). Contrary to the expectations, the presence of POM was not related to N 2 O-BD in either low-or high-diversity systems. Kravchenko et al. (2017) reported that when plant leaves were added to the soil, the proportion of N 2 O generated via denitrification was higher in the soil with prevalence of large (>35 µm) pores than in the soil with prevalence of small (<10 µm) pores. They explained this result by formation of local anoxic conditions not within the soil, but within the decomposing leaves themselves. Hence, we hypothesized that in the soils of high-diversity systems, due to their higher amounts of POM and large pores, N 2 O-BD also would be positively associated with POM and pore abundance. Yet, this hypothesis was not supported by the data and greater anoxic conditions, which could be surmised to occur within POM fragments due to enhanced decomposition (Kravchenko et al., 2017;Negassa et al., 2015), did not translate into greater N 2 O production via bacterial denitrification.
The historic conditions in terms of water content levels preceding soil sampling appeared to be one of the few soil WetPr (red) is wetting soil to 10 kPa, and WetWC (green) is wetting soil to the same water content initially present in Dry); shown are means adjusted for the change in the water contents during saturation values. Letters mark significant differences among water application modes (p < 0.05) characteristics significantly associated with N 2 O-BD in high-diversity systems. In high-diversity systems, the change in the water content at the time of sampling needed to reach full saturation was negatively correlated with N 2 O-BD (Table 2), and in the studied path model, it had statistically significant negative effect on N 2 O-BD (Figure 8). Note that in the low-diversity systems, these associations also had a negative sign, but were not statistically significant. The closer the soil was to full saturation at the time of sampling, the greater the proportion of N 2 O generated via bacterial denitrification. This relationship suggests that historic presence of denitrification-favorable conditions plays a positive role in current denitrification processes. It can be inferred that if the soil was held for a long period of time at conditions close to saturation, then denitrifying communities may have proliferated and expanded from small pores and aggregate centers, where they typically reside (Ebrahimi & Or, 2015;Tiedje, Sexstone, Parkin, Revsbech, & Shelton, 1984). Soil samples for this study were collected in February, after~3 months of high soil moisture levels. Their denitrifying communities were not likely affected by sampling or sample preprocessing; thus, they continued to actively function during the incubation experiment of this study. The potential importance of adaptation history in defining the ability of microbial communities to produce N 2 O has been expressed before (Jungkunst, Freibauer, Neufeldt, & Bareth, 2006;Krause et al., 2017;Lagomarsino et al., 2016), while the history of soil moisture conditions was found to play an unexpectedly greater role than current soil water levels for other soil processes, for example, soil respiration (Smith, 2017).

| Soil characteristics related to N 2 O emission
Despite some similarities between low-and high-diversity systems in terms of associations between emitted N 2 O and the soil aeration-related variables, overall, the two systems differed substantially, and the difference was most remarkable for CumN 2 O versus N 2 O-BD relationship. In the lowdiversity systems, greater proportion of N 2 O from bacterial denitrification directly translated into greater N 2 O emissions ( Figure 6a, Figure 8). The magnitude of anoxic conditions, as indicated by (a) greater area of unaerated soil volume Vol-180 and (b) lower presence of 30-60 µm pores, appeared to be the main physical factor driving both the proportion of bacterial denitrification and N 2 O emissions in these soils. In the high-diversity systems, bacterial denitrification had a statistically nonsignificant, but numerically also positive, direct effect on emitted N 2 O ( Figure 8). However, the straightforward pathway of poor aeration leading to greater contribution from bacterial denitrification to greater N 2 O emissions, which was clearly observed in low-diversity systems, was not present in high-diversity soils.
In soils of high-diversity systems, poor aeration conditions led to substantially greater N 2 O emissions directly. The positive effect of poor aeration was manifested through (a) negative relationship of CumN 2 O with presence of 30-120 µm pores (Table 2) (Table 2). Yet, while poor aeration did favor increased N 2 O emissions, it did so without a concomitant increase in the proportion of N 2 O produced via bacterial denitrification. In fact, the direct effect of Vol-180 on CumN 2 O (0.99) and its indirect effect on CumN 2 O via N 2 O-BD mediation (−0.31) (calculated as a product of −0.85 and 0.46) had opposite signs. This inconsistency suggests that in these soils, other anaerobic processes besides bacterial denitrification contributed to N 2 O production, perhaps fungal denitrification. The S P of N 2 O produced by fungi has been shown to be~37‰ (Rohe et al., 2014;Sutka, Adams, Ostrom, & Ostrom, 2008), which is substantially greater than the values of −10 to 0‰ associated with production from bacterial denitrification (Frame & Casciotti, 2010;Sutka et al., 2006;Toyoda & Yoshida, 1999). Fungi have been demonstrated to denitrify at more oxic conditions than bacteria and to be a substantial contributor to N 2 O production, especially in woody soils, but also in grasslands and soils of agricultural systems (Chen  Chen, Mothapo, & Shi, 2014;Crenshaw, Lauber, Sinsabaugh, & Stavely, 2008;Laughlin & Stevens, 2002). Fungi were found to be in a greater abundance in poplar and early-successional, however also in switchgrass, systems of this study than in corn-based systems (Jesus et al., 2016). In soils of high-diversity systems, it is possible that fungal denitrification activity may be more responsive than bacterial denitrification to decreases in soil aeration, while in soils of low-diversity systems, an increase in fungal denitrification upon decreasing aeration was less noticeable, due, in part, to the lower abundance of fungi (Jesus et al., 2016). Note that denitrification by fungi only results in N 2 O, but not N 2 production, since fungi lack the capability of N 2 O to N 2 reduction (Prendergast-Miller, Baggs, & Johnson, 2011;Sutka et al., 2008). Since N 2 O is the terminal product of denitrification by fungi, then fungal denitrification can be expected to result in more straightforward associations between N 2 O emissions and the factors contributing to it, as the additional nonlinear effect of reduced N 2 O production via its reduction to N 2 is eliminated. This could be a potential explanation for better predictive powers of soil water and pore characteristics observed in high-diversity systems of this study. It should be also noted that N 2 O reduction to N 2 was more likely to take place in poorly aerated soils of low-diversity systems, and, due to tendency for fungi to live in larger pores with greater air diffusion, was likely more affecting N 2 O of bacterial than fungal origin. This would explain lack of differences in the average values of CumN 2 O and N 2 O-BD among the studied systems. Further assessment of fungal denitrification and fungal presence in these soils is needed to test this hypothesis.
It should be also noted that reduction of N 2 O can alter SP; however, reduction shifts the relationship between δ 18 O and δ 15 N and δ 18 O and δ 15 N α shift toward 2.6 and 1.9, respectively, when reduction occurs (Ostrom et al., 2006;Jinuntuya-Nortman et al, 2008). Based on the data in Supporting information Table S1, we observed relationships between δ 18 O and δ 15 N and δ 18 O and δ 15 N α of 0.45 and 0.46, respectively, that are not consistent with a strong influence of N 2 O reduction in our samples (Ostrom & Ostrom, 2011). Further, as discussed in the Introduction section, reduction of N 2 O equal to 10% of its production will only shift S P by 0.7 ‰ (Opdyke et al., 2009), which has a minor effect on our estimates of the importance of bacterial denitrification.
The findings indicated that several years of implementing bioenergy systems with contrasting plant diversities led to substantial changes in soil pore characteristics, along with soil C and N levels. These changes were likely one of the contributors to the observed differences in associations between N 2 O emissions and soil microscale variables in bioenergy systems with low and high plant diversities.
Contrary to the expectations, none of the studied pore or POM characteristics worked as an effective predictor of N 2 O emissions across the entire set of the systems. However, saturation of <30 µm pores, presence of air-filled 30-90 µm pores, size of poorly aerated soil volume, and volume of POM unconnected to the atmosphere appeared to be feasible microscale soil predictors of N 2 O emissions in soils from bioenergy systems with high plant diversity. Yet, in the systems with low plant diversity, N 2 O production via bacterial denitrification appeared to be the only feasible predictor of N 2 O emissions, while predictive capabilities of microscale pore and POM data were relatively weak. The possible cause is that long-term differences in implementing contrasting bioenergy systems may have affected the relative contributions of different processes, for example, bacterial and fungal denitrification and nitrification, to N 2 O production and further reduction. Differences in influences from physical microscale soil characteristics on these processes likely led to the observed differences in capabilities of different soil characteristics in predicting N 2 O emissions.