Denitrifying pathways dominate nitrous oxide emissions from managed grassland during drought and rewetting

Isotopic measurements showed that N₂O production during drought is unexpectedly dominated by denitrification of organic nitrogen.

Abstract

Nitrous oxide is a powerful greenhouse gas whose atmospheric growth rate has accelerated over the past decade. Most anthropogenic N₂O emissions result from soil N fertilization, which is converted to N₂O via oxic nitrification and anoxic denitrification pathways. Drought-affected soils are expected to be well oxygenated; however, using high-resolution isotopic measurements, we found that denitrifying pathways dominated N₂O emissions during a severe drought applied to managed grassland. This was due to a reversible, drought-induced enrichment in nitrogen-bearing organic matter on soil microaggregates and suggested a strong role for chemo- or codenitrification. Throughout rewetting, denitrification dominated emissions, despite high variability in fluxes. Total N₂O flux and denitrification contribution were significantly higher during rewetting than for control plots at the same soil moisture range. The observed feedbacks between precipitation changes induced by climate change and N₂O emission pathways are sufficient to account for the accelerating N₂O growth rate observed over the past decade.

INTRODUCTION

Nitrous oxide (N₂O) is a strong greenhouse gas and the most important stratospheric ozone-depleting substance emitted in the 21st century (1–3). The primary global source of N₂O is production during N cycling by microbiota in soils. Anthropogenic activities, in particular fertilizer use leading to an increase in reactive N (N_r) (4–7), have caused an increase in N₂O mole fraction from 260 to 270 nmol mol⁻¹ in the preindustrial era to 331.7 nmol mol⁻¹ today (5, 8). Particularly concerning is the steep acceleration in growth rate from around 0.75 to above 1.0 nmol mol⁻¹ a⁻¹ over the past 10 years (9), which corresponds to an increase in emissions of 2 ± 0.4 Tg N₂O-N a⁻¹ (section S1.1) (8, 10). The causes of this acceleration, which is not explained by any steep rise in fertilizer use (9), are poorly known and could relate to the impact of climate change on emission processes. Climate change feedbacks could account for the recent increase in N₂O growth rate, as well as lead to further acceleration in the coming decades, affecting our ability to balance optimal fertilizer use for emission reductions and continuing food security (7, 11, 12).

N₂O is produced by soil microbes primarily during nitrification—an oxic process—and from denitrifying pathways, primarily under low oxygen conditions. Denitrifying pathways include heterotrophic denitrification (13), nitrifier denitrification (where ammonia oxidizer bacteria reduce nitrite) (14), and fungal denitrification, as well as minor pathways such as chemodenitrification (abiotic) and codenitrification (biotic), where an N atom from ${\text{NO}}_2^{-}$ or NO combines with another N species—typically organic—to form N₂ or N₂O (15–17). These denitrifying pathways are hereafter referred to as “denitrification,” with the specific subpathway given when relevant. In addition, heterotrophic denitrification can proceed to completion, consuming N₂O and reducing it to N₂ under low oxygen conditions (18, 19). The balance between nitrification and denitrification in soils influences N₂O emission strength as well as affecting total reactive nitrogen loss, leaching, and fertilizer nitrogen (N) use efficiency, and thus controls environmental impacts of fertilizer N (9, 20). Isotopic measurements may show an increase in the contribution of denitrification to global anthropogenic N₂O emissions over the past half-century with no well-identified cause (21). However, as the nitrification and denitrification pathways respond differently to environmental drivers, climate change impacts are challenging to predict, and models are currently unable to provide a robust assessment of emission feedbacks in the coming decades (9).

Climate change is expected to have strong impacts on precipitation regimes and thus causes major soil moisture anomalies—both positive and negative—in large areas of the globe (22–24). For grasslands in many regions, including the European Alps, a predicted increase in extreme summer droughts and an increased autumn-winter precipitation will represent a critical disturbance for ecosystem functioning and nutrient cycles (25–28). Soil moisture has consistently been shown to be one of the most important parameters affecting soil oxygenation and thus determining N₂O production rates from nitrification, denitrification, and other pathways (13, 29–31). During drought, microbial activity and N₂O emissions are reduced (32–35). As soils are well oxygenated during drought, these low N₂O emissions are expected to result from nitrification, leading to nitrate accumulation; however, direct evidence for this assumption is missing. Rewetting after droughts usually causes large, fast pulses in N₂O fluxes (27, 34, 36)—“hot moments”—because of nitrogen substrate buildup; however, the processes driving these emissions are extremely challenging to characterize because of their highly dynamic nature (35, 37–39). These pulses are expected to result from denitrification due to increased soil moisture, thus leading to gaseous loss of reactive nitrogen. Similarly, high soil moisture during periods of increased rainfall is expected to increase denitrification and affect the balance between N₂O production and consumption; however, there are few direct observations of pathways under differing soil moisture regimes (40, 41). Soil properties such as bulk density and microaggregate physicochemical attributes will modulate the strength of coupling between precipitation changes and soil moisture, which will furthermore be dependent on microbial community composition and activity.

A detailed understanding of the impacts of climate change on coupling between soil macro- and microscale properties, microbial communities, and N₂O emission pathways is critical to drive accurate biogeochemical models (7) and to identify likely trajectories of N₂O emission dynamics from ecosystems under global change. However, there is a lack of studies integrating these parameters across spatiotemporal scales while providing sufficient time resolution to capture dynamic responses of N₂O emission pathways to experimental perturbations. The isotopic composition of N₂O can be particularly useful to distinguish between formation and consumption pathways (see fig. S2 and the “Describing N₂O isotopic composition” section) and to provide an integrated picture of N cycling.

The linear N₂O molecule has two distinct positions for isotopic substitution that can be measured accurately following recent developments in laser isotope spectroscopy (40, 42, 43). The difference in isotopic composition of these two positions—the isotopic “site preference” (SP)—is independent of the substrate isotopic composition and depends only on the reaction(s) forming and destroying N₂O (fig. S2). SP and δ¹⁵N are particularly useful to determine the partitioning between nitrification, denitrification, and N₂O consumption when considered in combination with other tools to obtain complementary insights into underlying processes. Investigations of microbial community structure and activity using molecular ecology techniques can confirm the identity of genes and/or microbes involved in N transformations, for example, by using quantitative polymerase chain reaction (qPCR) to measure the abundances of key functional genes in response to experimental treatments (44, 45). Spatial analysis techniques like NanoSIMS (Nanoscale Secondary Ion Mass Spectrometry) and STXM-NEXAFS (Scanning Transmission X-ray Microscopy - Near Edge X-ray Absorption Fine Structrure) can be used to show the distribution, speciation, and incorporation of elements in the soil matrix and within soil microaggregates (46–48). Measurements of soil substrate δ¹⁵N and monitoring of ancillary environmental and climate parameters also provide key data to understand drivers of observed isotopic changes and estimate isotopic mass balances. A combination of multiple techniques provides a more detailed and robust view of the N cycle and N₂O processes than would be possible with any single approach.

Here, we investigate N₂O emissions and isotopic composition from subalpine grassland monoliths subjected to an extreme drought and subsequent rewetting, in comparison to control and increased precipitation treatments throughout a growing season. In a climate manipulation experiment, N₂O isotopocules were measured online using automated chambers coupled to an N₂O isotopic spectrometer, giving a measurement frequency of <17 min for emitted (“source”) δ¹⁵N and SP. These results allowed an unprecedented level of detail for the analysis of variability in N₂O production and consumption pathways in response to environmental parameters. We aimed to quantify the net effect of drought and rewetting on emissions and pathways, which is currently unknown, as previous studies have not allowed for quantification of emission pathways with sufficient time resolution (13, 38, 49, 50). We tested the hypothesis that drought enhances nitrification and rewetting favors denitrification as the major source of N₂O emissions. Furthermore, we hypothesized that N₂O emissions during rewetting periods would be large and highly variable because of rapid utilization of N substrate built up during drought, and we expected to see a consistent increase in denitrification contribution with increasing soil moisture over all treatments. We expected that the impacts of precipitation on N₂O emission pathways would be mediated by soil microscale properties—investigated using NanoSIMS—as soil properties dictate water drainage through drought and during rainfall, and thus the nature of anoxic and oxic microsites in the soil, while the chemical environment will influence the availability of different N species through drought and rewetting.

RESULTS AND DISCUSSION

N₂O emission pathways in the pre-drought period

Sixteen intact monoliths were extracted from a lightly managed subalpine grassland and established at the University of Innsbruck for 5 weeks before the beginning of the drought experiment, during which time water-filled pore space (WFPS) was stabilized to 30 to 40% (Fig. 1 and fig. S8). Seven of the monoliths were subject to an extreme drought (D) (see section S1.4.2 for details on definition and strength of the drought treatment), while four control (C) monoliths received the mean precipitation and four wet (W) monoliths received 25% more than the mean precipitation (see the “Study design” section). Throughout the experiment, W monoliths produced significantly more N₂O than C monoliths, with a greater proportion from denitrification (section S1.4.4).

Fig. 1. Overview of N₂O fluxes from different pathways for control, wet (+25% precipitation), and drought treatments.

The results are the 12-hour mean of all monoliths in a treatment group (n = 4,4, and 7, respectively). Fluxes were partitioned into pathways using “correlation partitioning model” (CPM; see the “Estimation of microbial N₂O production and consumption pathways: The CPM” section): “Denitrification” and “nitrification” refer to net N₂O production by the respective pathways, whereas “reduction” indicates N₂O consumption and production of N₂ during denitrification. The time axis is normalized to the date of fertilization, as this produces the largest single change in fluxes, and time periods for drought, rewetting (RW), and recovery are indicated at the top of the figure. The y-axis scale changes at 1.4 nmol m⁻² s⁻¹ as indicated to better capture the range of observed fluxes. The scales on all y axes are equal, but the maxima differ.

All monoliths were fertilized with 200 kg-N ha⁻¹ as NH₄NO₃ 7 to 10 days before onset of the drought as described in the “Study design” section, which increased N₂O fluxes from 0.21 to 1.15 nmol m⁻² s⁻¹. An analogous unfertilized control watering showed no significant change in fluxes in the same period (Fig. 1 and fig. S9). Considering 0 to 90 days after fertilization—the main growing season (July to October)—1% of fertilizer N was released as N₂O, in good agreement with the Intergovernmental Panel on Climate Change (IPCC) default value of 1% (51). Over the first 2 weeks after fertilization, N₂O production from nitrification and denitrification for non-drought plots was relatively equal; however, 15 to 30 days after fertilization, N₂O from nitrification returned to prefertilization levels. Over the 90-day main growing season, 68% of fertilizer-induced N₂O was produced via denitrification. Similarly, soil ammonium concentration decreased in the weeks after fertilization, while soil nitrate concentration increased, as nitrate was produced from ammonium by nitrification soon after fertilization (labeled as “pre-drought” versus “peak drought” in fig. S10). δ¹⁵N of soil nitrate increased, reflecting production from isotopically heavier ammonium, as well as preferential consumption or leaching of isotopically light nitrate. Changes in δ¹⁵N of ammonium appear to primarily reflect strong isotopic fractionation favoring leaching of heavy ammonium. Our study reveals that N₂O source dynamics following fertilizer application lead to a shift from nitrification-dominated initially [as in (52); <10 days after fertilization] to denitrification-dominated when integrated over the entire growing season. These results complement the results of Castellano-Hinojosa et al. (53), who suggested that denitrification from deeper soil layers (15 to 20 cm) could dominate fertilizer-driven N₂O emissions after a year.

Impact of drought on N₂O emission pathways

Shortly after starting the drought treatment, N₂O fluxes dropped strongly, but then remained stable at 0.2 to 0.3 nmol m⁻² s⁻¹ below 30% WFPS (Figs. 1 and 2), with an average SP of <0‰ consistently measured for emissions below 30% WFPS. The lower source SP values measured for drought-stressed monoliths showed that the proportion of N₂O from denitrification increased under drought conditions. Denitrification produced 80 to 90% of N₂O between 20 and 30% WFPS, and around 70% of N₂O below 20% WFPS, despite little change in fluxes (Fig. 2). This pattern appeared stable across all monoliths, although isotope-based partitioning is relatively uncertain with low fluxes. N₂O reduction was not significantly affected by drought (P > 0.05). Although denitrification became somewhat more dominant during drought, drought did not induce a major shift in N₂O production pathways, in agreement with soil ${\text{NO}}_3^{-}$/${\text{NH}}_4^{+}$ (fig. S10), which showed slower N turnover for drought monoliths (i.e., lower magnitude) but no difference in the direction of concentration or isotopic changes compared to controls. Similarly, drought had no consistent impact on absolute or relative abundances of functional genes involved in N₂O production (fig. S11), in agreement with the results of a recent meta-analysis (35). Thus, it appears that drought primarily affects N₂O by restricting the overall rate of N₂O production, most likely through reduced solute availability for all pathways (54).

Fig. 2. Effect of reduced WFPS during the drought treatment (“drought,” n = 7) and 0 to 15 days after the drought (“rewetting,” n = 7) on N₂O fluxes and production pathways, compared to C/W monoliths (“control,” n = 8).

Large points show the mean value for all monoliths in each treatment for each 2% WFPS bin. Small paler points show individual 12-hour mean values.

Unexpectedly, very low WFPS did not reduce the proportion of N₂O from denitrification (low SP) pathways, although dry soils would be well oxygenated. This suggests that anoxic microsites remain present even in very dry soils, showing that reactive nitrogen loss to denitrification can continue throughout drought periods. NanoSIMS results at peak drought showed a strong enrichment of secondary ions indicative of nitrogen-bearing soil organic matter (N-SOM) (47, 48) at the surface of drought treatment microaggregates (Fig. 3 and section S1.4.3), likely formed from nitrosation (addition of NO group) of SOM (55). Co-occurring buildup of sulfur and N-SOM suggests that microbial cell death and subsequent release of organic components including S-containing amino acids could alternatively account for the changes in soil surface chemistry (56). Oxygen was not enhanced, thus suggesting that N-SOM was concentrated in an oxygen-deficient microenvironment under soil moisture deficit, providing a location for anoxic N₂O production to occur. During drought, soil microaggregates become increasingly hydrophobic (57, 58); thus, microsites can retain water strongly and remain hypoxic, supporting anaerobic mechanisms.

Fig. 3. NanoSIMS measurements comparing drought and control soil microaggregates, showing the enrichment in ¹²C¹⁴N⁻—indicative of organic N—relative to ¹⁶O⁻ on drought soil grains.

(A) Example NanoSIMS images of a typical control (monolith C1) and drought (D2) microaggregate soil grain showing ¹⁶O⁻, ¹²C¹⁴N⁻, and ratio images. Signal intensity is shown on the color bar. (B) Average depth profiles for control and drought grains at initial, peak drought, and post-rewetting time periods (see section S1.4.3 for details and statistics). “Relative depth” on the x axis refers to layers sputtered away sequentially in NanoSIMS analysis normalized by dwell time as described in the “Estimation of microbial N₂O production and consumption pathways: The CPM” section. The standard error for the drought treatment at peak drought is shown as a shaded area; other error ranges are not included for clarity.

The buildup of N-SOM leading to N₂O emissions during drought could be explained by the “codenitrification” or “chemodenitrification” pathways, whereby the N atom from ${\text{NO}}_2^{-}$ or NO combines with an N atom from another species, particularly organic N, to form N₂O or N₂ by chemical or biotic N-nitrosation (15–17). This pathway is usually studied using isotopic labeling, as it results in “hybrid” N₂O, where labels show the incorporation of N from two different substrates (59). The SP of N₂O from chemo/codenitrification is not well constrained and depends on the substrates involved; previous results suggest that a pathway involving hyponitrous acid can produce N₂O with SP <0‰ (60), in line with the results of this study.

N₂O emission pathways following rewetting of soils

At the end of the drought treatment, the seven D monoliths were rewet in two groups 1 week apart (D1 and D2 treatments) to allow high time resolution for N₂O flux and isotope measurements. For comparison to the post-drought rewetting, a summary of N₂O flux effects from triweekly watering of the non-drought monoliths is given in section S1.4.4. The drought monoliths were rewet with varying amounts of rainwater (20 to 90 mm; see the “Study design” section) to achieve a rewetting gradient and assess potential nonlinearity in rewetting effects (61). However, each monolith took up water to increase WFPS by 25 to 30% to 31 ±4%, and the additional water leached out of the monoliths. Therefore, the rewetting gradient did not produce a gradient in soil moisture and thus did not cause variability in rewetting response.

Rewetting of the drought treatment monoliths significantly increased the N₂O flux as well as CH₄ uptake and CO₂ emissions, showing the expected pulse response often termed “Birch effect” (62, 63) (fig. S12). N₂O emission peaks following rewetting were as high as 11.6 nmol m⁻² s⁻¹, of which 92± 12% was due to denitrification, and they occurred 51 ± 12 hours after rewetting (Fig. 2 and fig. S12). Control monoliths were watered with 5.7 mm of rainwater three times per week, including on the days when drought monoliths were rewet. On the first rewetting date, controls showed a strong watering response for both CO₂ and N₂O fluxes; however, this response was not seen for control monoliths on the second rewetting date (fig. S12). The rewetting peak was much larger than the average increase of 1.5 nmol m⁻² s⁻¹ seen following triweekly watering of control monoliths (fig. S7), and it also occurred much later (51 versus 12 hours), reflecting the significant stress of drought for the soil microbial ecosystem. Over the whole season, net emissions from control and drought plots were similar (0.124± 0.66 mmol m⁻² for control and 0.121 ± 0.76 mmol m⁻² for drought plots from mid-June to October); thus, reduced N₂O emissions during this extreme drought did not offset the effects of the rewetting peak.

N₂O emissions during rewetting were significantly higher for the same WFPS range than C or D emissions and highly nonlinear, illustrating hysteresis in the response of emissions to WFPS (Fig. 2). Hysteresis was also evident in the fraction of N₂O produced via denitrification, and the fraction of N₂O reduced: As WFPS increased during rewetting, more N₂O was emitted from denitrification than when WFPS decreased. Almost no N₂O reduction was evident in the correlation partitioning model (CPM) results during rewetting until WFPS was >30%, which suggests that initial rewetting emissions are from chemo- or codenitrification rather than from heterotrophic denitrification, as chemo- and codenitrification do not lead to N₂O reduction (16).

The drought-induced changes in soil microaggregate chemistry (Fig. 3) appear to be completely reversible, as no differences in elemental ratios were seen at the post-rewetting stage (table S3). Once water is available, N-SOM could be rapidly consumed via chemo- or codenitrification, thus accounting for some or all of the rewetting peak emissions and reversing the accumulation of N-SOM, after which heterotrophic or nitrifier denitrification once again dominates.

Abiotic versus biotic controls on N₂O emission pathways

WFPS was clearly the main driver of N₂O emissions throughout the drought-rewetting period (Fig. 2), displaying an exponential response of emissions to increasing WFPS despite minor changes in partitioning between production pathways. Triweekly watering of the control (C) and wet (W) monoliths provided further evidence of the nonlinear impact of WFPS on N₂O emissions (section S1.4.4). W monoliths (+25% precipitation; on average 3% higher WFPS) emitted 70% more N₂O from denitrification (1.2 versus 0.7 nmol m⁻² s⁻¹) and 40% more from nitrification (0.31 versus 0.22 nmol m⁻² s⁻¹), and thus a total average increase in N₂O fluxes of 55% for W compared to C monoliths (fig. S7).

Principal components analysis (PCA; Fig. 4) revealed additional controls on N₂O emissions. Before the drought, microbial functional gene abundances, bacterial biomass, and N₂O fluxes showed a strong positive relationship along PC1: Microbial functional gene abundances were clearly the main factor controlling N₂O emissions. Linear regressions confirmed that a strong relationship with N₂O fluxes was evident for nirK and nirS (nitrite reductase) abundances, which accounted for half of the variability in fluxes the week after fertilization (fig. S13). This supports the major contribution of denitrification evident from N₂O isotopocule measurements and CPM (see Fig. 1).

Fig. 4. PCA bi-plots describing the relationship between N²O emission pathways and fluxes and explanatory variables.

(A) Before the drought experiment began, and (B) through peak drought (0 to 7 days before rewetting) and rewetting (0 to 7 days after rewetting). Soil parameters are shown in green, features of N₂O emissions are shown in red, and microbial functional gene abundances are shown in blue. In (B), parameters measured during the rewetting (from the rewetting to 7 days after) are highlighted with open circles and italic font to distinguish from parameters measured during the drought. “N₂O-denit” and “N₂O-reduction” refer to the proportion of N₂O produced/consumed by denitrification/reduction, respectively. “N₂O max, delay” refers to the time between rewetting and maximum N₂O flux.

The PCA of factors controlling N₂O emissions through drought and rewetting showed clear differences to the initial period (Fig. 4). Microbial gene abundances were not a major factor controlling N₂O emissions during drought and rewetting, consistent with the results of (49). This was also evident in the pattern of microbial gene abundance throughout the experiment, which showed no clear drought-induced decrease (fig. S11), despite much lower N₂O fluxes. Much weaker correlations between measured gene abundances and fluxes were observed in the drought and post-drought period (fig. S13), supporting the role of codenitrification, which is not represented in the measured genes, or chemodenitrification, which is not microbially mediated.

N₂O fluxes at peak drought and during rewetting, as well as the rewetting peak maximum, were strongly correlated with soil temperature at rewetting and WFPS before rewetting, as well as with soil bulk density and ${\text{NO}}_3^{-}$ (PC1 in Fig. 4). PC2 revealed that a higher proportion of denitrification during rewetting was associated with higher fluxes, with a negative relationship to the delay between rewetting and peak fluxes. Overall, these results show that drought effects were more pronounced when bulk density was lower (57, 58), which was expressed in lower N₂O flux and WFPS at the peak of the drought, and larger delay until peak N₂O occurred after rewetting. This controls the mean and maximum flux experienced after rewetting: Less severe drought effects led to a larger rewetting peak, with a larger denitrification contribution.

Our results showed an unexpected dominance of chemo- or codenitrification for N₂O emissions during drought in managed grassland soils, driven by the buildup of organic nitrogen. In contrast to our initial hypothesis, drought did not promote nitrification: Although drought-affected soils are well oxygenated, substrate supply will limit N₂O production via the oxic nitrification pathway. As nitrite and N-SOM build up during drought, denitrifying pathways can continue, similar to previous observations in wetter soils (64, 65). Although recent studies have highlighted the potential impact of alternative denitrification pathways (15–17), drivers and rates are poorly known, and thus, these pathways are not included in most models. A summary of driving factors found in this study is shown in Fig. 5. Constraining the contribution of these pathways to N₂O emissions and nitrogen loss in a changing climate will be a critical focus for future studies, particularly combining isotopic labeling and NanoSIMS approaches, to facilitate the addition of these relatively unknown pathways to biogeochemical models.

Fig. 5. Schematic view of the effect of soil moisture, drought, and rewetting on N₂O emissions and pathways, shown by this study for a drought experiment using managed subalpine grassland monoliths.

(A) Linkages between key soil parameters regulating the emission of N₂O by different pathways are shown with gray arrows—thicker lines indicate a stronger relationship for some parameters. The impacts of drought are indicated in red and those of rewetting in green: + (++) signifies a (strong) increase, and − (− −) signifies a (strong) decrease. Inputs to the soil ecosystem are shown in purple and soil parameters in brown; microbial groups are shown with blue (nitrifiers), green (denitrifiers), and yellow (codenitrifiers). (B) Impact of WFPS on N₂O emission flux (solid lines) and the fraction of N₂O contributed by denitrification (dashed lines). The typical WFPS range is shown in purple, with the drought period highlighted in red and rewetting in green. The region where chemo-/codenitrification are important is highlighted in yellow. To illustrate the magnitude of the effects shown, the global mean emission density for croplands is shown in gray (7).

We found that drought and rewetting did not lead to an overall reduction in emissions across an entire season—for some drought monoliths, emissions were increased—despite strongly reduced fluxes in the drought period. This was due to hysteresis in the rewetting period causing emissions to overshoot the baseline WFPS-flux relationship (Figs. 2 and 5), in contrast to our initial hypothesis that emissions and pathways would exhibit a consistent response to WFPS. We observed that between monoliths in this experiment, increased drought severity reduced the size of the rewetting peak, contrary to expectations that continued fixation and nitrification during drought with reduced plant uptake would lead to an increased N₂O peak (35, 39). This suggests the possibility for a “tipping point” for overall N₂O emissions in response to drought-rewetting cycles, whereby moderate droughts may increase total emissions, while severe droughts reduce emissions.

Recent results have shown a strong acceleration in N₂O growth rate in the past decade (9). In addition, tropospheric background isotopic measurements have shown a decrease in SP of the total anthropogenic source over the past half-century, from 18‰ [1940 to 2008 average; (66)] to 11‰ [2018; (21)], reflecting an increase in the contribution of denitrification to total anthropogenic emissions with no identified cause. Fertilizer use accounts for around half the increase in N₂O emissions since the preindustrial era (7), with a strong acceleration after 1950 and an ever increasing proportion of synthetic fertilizer (67). Our results show that synthetic NH₄NO₃ fertilizer can strongly promote denitrification over a growing season. If this result is representative of impacts in different soil types, the increasing use of synthetic fertilizers may explain the increased global contribution of denitrification to anthropogenic N₂O emissions over the past half-century—a trend that will likely continue. However, this cannot explain the acceleration in N₂O growth rate in the past decade (9), which is much faster than the global increase in fertilizer application over the same period.

The observed acceleration in N₂O growth rate requires a large increase in N₂O emission factors (9) that is likely to be driven by climate change impacts on N₂O pathways. Our results show that the relationship between N₂O and WFPS is exponential in these grassland soils (Figs. 2 and 5), thus suggesting that climate change–induced autumn-winter precipitation increases could strongly increase total N₂O emissions, and further contribute to large gaseous losses of reactive nitrogen. In this study, N₂O emissions were enhanced by 55% for +25% precipitation increase. We can gain a first estimate of potential global impact by considering the average flux density from croplands (0.24 ± 0.1 nmol m⁻² s⁻¹) and from all ice-free land (0.04 ± 0.01 nmol m⁻² s⁻¹) [annual averages, (7)], and assuming a similar average sensitivity to increased precipitation across grasslands worldwide, which would need to be confirmed by further experimental studies of this type. In this case, just 2 months of +25% precipitation would increase N₂O emissions by 0.34 Tg N₂O-N a⁻¹, with equivalent losses of reactive nitrogen, which translates to an acceleration in growth rate of 0.04 nmol mol⁻¹ a⁻¹ (section S1.1). This shows the potential magnitude of climate feedback from a single mechanism [see also (11)]. Although numerous studies have shown the importance of soil moisture for N₂O emissions (13, 34, 68, 69), few measurements have been made at low WFPS, and the nonlinear response of N₂O to WFPS has not been clearly evident in studies that consider smaller soil moisture ranges (27). Our results show that climate change, in particular precipitation changes, is the most likely cause of the recent acceleration in N₂O growth rate, illustrating the clear need for more mechanistic experimental studies of N₂O production pathways across ecosystems globally to improve model predictions of N₂O emissions in the future.

We show that denitrification via various mechanisms is responsible for most of the N₂O emissions from the managed grassland investigated in this study, and perturbations such as fertilization, drought, rewetting, or increased precipitation do not act to enhance the proportion of emissions from nitrification. The role of “minor” denitrification pathways such as chemo- and codenitrification is clearly critical and requires further investigation. Incorporation of these mechanistic results into biogeochemical models will improve the capability to understand and predict impacts of climate change feedbacks on N₂O emissions, thus facilitating sustainable policy development through targeted mitigation.

MATERIALS AND METHODS

Study design

Sixteen intact soil monoliths were collected from the Long-Term Ecosystem Research Carbon, Water and Nitrogen (LTER-CWN) subalpine meadow site Kaserstatt Alm in Stubaital, Austria (47°7′N, 11°18′E) located 1820 m above sea level (www.lter-austria.at/en/cwn-sites-stubai-valley/). The location has a mean inclination of 20°, mean annual temperature of 3°C, mean annual precipitation of 1097 mm, and dystric cambisol loam-textured soil. Further detailed information regarding vegetation and soil properties has been published previously (70–73). All major dates throughout the sampling and experimental period are summarized in table S1.

The 28-cm-deep monoliths were installed in 25-cm-diameter steel cylinders, with a rooting mat on a steel grate and 12-cm open space below each monolith for drainage and leachate collection. Monoliths were collected on 8 May 2018 and transported to Innsbruck on 14 May 2018. The steel cylinders containing the monoliths were placed in 30- to 35-cm-deep holes in the ground in a common garden adjacent to the measurement laboratory, as shown in Fig. 6. Transportation to Innsbruck involves a minor degree of warming compared to the Kaserstatt Alm site. However, in high alpine sites, actual canopy temperatures can be much higher than air temperatures because of high irradiance; thus, the temperature difference experienced by plants is far less than expected from air temperature differences (74). The observed effects of temperature on ecosystem functioning and microbial ecology for alpine grasslands are relatively minor (74–76); thus, the effects observed in this experiment are expected to be robust, and the logistical benefit of transporting to Innsbruck was essential to the experiment’s success.

Fig. 6. A schematic view of the experimental setup showing the arrangement of the monoliths in the common garden, with the different treatment types indicated: C = Control (n = 4), W = Wet (n = 4), D1/D2 = Drought (n = 7).

Monolith D2-4 received rain through the side of the tent and is therefore not included in subsequent analyses. An example footprint for an open chamber is shown for monolith C3, and measurement groups A to D are indicated for each monolith. On the right-hand side, the measurement setup including the calibration manifold is shown.

Soil water content (SWC) and temperature probes were installed in eight of the monoliths and programmed to measure every 15 min (HOBO Micro Station, Onset Computer Corporation, MA, USA). The probes were moved every 1 to 8 days throughout the experimental period so that data were regularly available for all 16 monoliths. In addition, relative humidity and air temperature were monitored at the surface and at 1.5 m above ground level, and photosynthetically active radiation was measured at 1.5 m (S-LIA-M003 and S-THB-M002, Onset Computer Corporation, MA, USA). Microclimate data were gap-filled for the times each monolith was not measured by interpolating the nearest measurements and overlaying the mean variability for the treatment group. Based on the measured SWC, monoliths were watered 150 to 300 ml up to three times a week to bring the SWC of each monolith to approximately 30% (gravimetric) before beginning of the experiment with fertilization in early July (see below).

The monoliths were assigned to treatment groups control (C), wet (W), and drought (D1/D2) at random, as well as into four measurement groups designated A to D, where each had one monolith from each treatment. There were two drought treatment groups (D1 and D2) that were separately rewet so that emission dynamics following rewetting could be followed more closely (see the next section). For this study, drought is defined as a complete absence of precipitation, subsequently leading to stress and thus reduced biomass growth (77) discussed further in section S1.4.2. One monolith from treatment D2 (D2-4) was located at the edge of the rainout shelter and therefore received some rainfall during windy conditions—this monolith is not included in any subsequent analysis or discussion. On 2 July 2018, the eight monoliths in groups B and D were fertilized with 200 kg N ha⁻¹ as NH₄NO₃ (BioXtra ≥99.5%, Sigma-Aldrich Inc., Germany) in 300 ml of water each; groups A and C monoliths were also watered 300 ml as a control comparison for fertilization. Fertilization was carried out for the eight monoliths in groups A and C on 5 July 2018 in the same manner. Fertilization was carried out in two blocks so that the N₂O dynamics following fertilization could be monitored with higher time resolution, as described in the next section.

On 11 July 2018, a rainout shelter was set up over all the monoliths using transmissive plastic greenhouse foil (Lumitec Clear AF, folitec Agrarfolienvertriebs GmbH, Germany) open at the bottom (approximately 0.5 m) to allow air circulation. Throughout the drought treatment, D1/D2 monoliths received no water, while C monoliths received the average rainfall for the region [884 mm; (70)] distributed evenly across 3 days per week: 5.7 mm/280 ml per watering. W monoliths received 25% more than the average precipitation: 7.1 mm/350 ml per watering. Both C and W monoliths are within the normal range of variability and are therefore used to provide a wide WFPS range as a baseline to consider drought impacts. On 3 September 2018, the drought was ended for “D1” drought monoliths (groups B/C) with 20 to 80 mm of collected rainwater added slowly over 45 min. On 11 September 2018, the drought was ended for “D2” drought monoliths (groups A/C) with 30 to 90 mm of collected rainwater. Following rewetting, D1/D2 monoliths were watered according to the control watering schedule until the rainout shelter was removed on 18 September 2018, after which all monoliths were exposed to natural precipitation.

Measurements of trace gas fluxes and N₂O isotopic composition

Describing N₂O isotopic composition

Isotopic composition is expressed as a “δ” value, which describes the deviation from an international standard

$${\mathrm{\delta }}^{15}\mathrm{N}(‰)=(\frac{{\mathit{R}}_{\text{sample}}}{{\mathit{R}}_{\text{standard}}}-1)\times 1000$$ (1)

where R = [¹⁵N][¹⁴N] and R_standard is the isotopic ratio of atmospheric N₂ (AIR-N₂), or analogously for δ¹⁸O (e.g., in ${\text{NO}}_3^{-}$ or N₂O) using the O-isotope ratio of standard mean ocean water (V-SMOW). The linear N₂O molecule has two distinct positions for isotopic substitution—¹⁴N¹⁵N¹⁶O (α) and ¹⁵N¹⁴N¹⁶O (β)—that can be measured accurately following recent developments in laser isotope spectroscopy (40, 42, 43). The bulk ¹⁵N isotopic composition (hereafter δ¹⁵N) refers to the average δ¹⁵N at both positions. The N₂O SP refers to the difference in ¹⁵N isotopic composition between the central (α) position N and the terminal (β) position N (78)

$$\text{SP}={\mathrm{\delta }}^{15}{\mathrm{N}}^{\mathrm{\alpha }}-{\mathrm{\delta }}^{15}{\mathrm{N}}^{\mathrm{\beta }}$$ (2)

Coupling of automated chambers and spectroscopic measurements

Measurements of CO₂, CH₄, and N₂O fluxes and isotopic composition were started on 15 June 2018 using the setup shown in Fig. 6. Six LI-COR chambers (8100-104 opaque long-term chambers and 8100-104C clear long-term chambers, LI-COR Inc., USA) were connected using 15-m tubing to the LI-COR LI-8150 Multiplexer followed by an LI-8100A automated flux system to monitor CO₂ using a flow rate of 1 liter min⁻¹. A subsample of 260 sccm (standard cubic centimeters per minute) from the outflow of LI-8100A was directed to a Picarro G2301 cavity ring-down spectrometer (CRDS; Picarro Inc., USA) system for the measurement of CO₂ and CH₄ in the chambers. Outflow from Picarro G2301 was split using a T-connection. Most of the air was returned to the chambers via a Vacuubrand A2000 Picarro recirculation pump—30 sccm was subsampled to a Picarro G5131i N₂O CRDS isotope analyzer and replaced with ambient air to avoid a pressure drop in the chamber system. As shown in Fig. 6, a combination of ¹/₄″ Bev-A-Line tubing and ¹/₄″ and ¹/₈″ steel tubing connected with Swagelok fittings was used to interface the instrumentation. A chamber collar was installed on each of the 16 monoliths, and the chambers were moved between the monoliths every 1 to 5 days to obtain good data coverage on all monoliths.

For flux and isotope measurements, chambers were closed for 15 min, with 30-s flushing before and 45 s after each closure. In addition to chamber measurements, various calibration gases were measured:

1) AmbFS (ambient full system): Ambient air entering through LI-8150 and thus passing through the whole measurement system

2) AmbIso (ambient isotopic spectrometer): Ambient air entering directly into Picarro G5131i

3) MRDep (mixing ratio dependence): 2500 nmol mol⁻¹ N₂O in synthetic air (Air Liquide Austria GmbH) diluted to approximately 300, 400, and 500 nmol mol⁻¹ with synthetic air

4) Comp (compressed air): Medical compressed whole air (Air Liquide Austria GmbH)

The calibration setup used mass flow controllers (red-y SMART controller, GSC-A2SA-BB21, Vögtlin Instruments GmbH, Switzerland) and three-way solenoid valves (Burkert Austria GmbH) controlled with LabVIEW (National Instruments Inc., USA). A typical measurement sequence is shown in fig. S3. Including calibration as well as chamber and system flushing, every 6 hours and 40 min, all six chambers were measured four times, giving an average measurement frequency of 16.7 min.

Flux and isotope data processing and calibration

Measurement data were processed in blocks of 1 to 5 days, giving a balance between sufficient calibration points and required computing time. Data processing was carried out in R 3.5.1 (The R Foundation for Statistical Computing, 2018). Raw data from both Picarro instruments and a record of the calibration system (MFC flows, valve positions) from LabView were imported into R. CO₂ fluxes from LI-COR 8100A were first calculated from raw data using SoilFluxPro (LI-COR Inc., USA) using only the first 120 s of each chamber closure with a dead time of 20 s and imported into R to compare with data from the Picarro instrumentation.

The first step of data processing involved correction for MRDep of isotopic composition. The approximate time period corresponding to each (MRDep) measurement was found from the LabView record file. Within this time period, all MRDep data were first filtered to the target mixing ratio and then cut to include only the flattest part of the “peak.” These data were used to fit a linear model of N₂O mixing ratio versus δ¹⁵N and SP and δ¹⁸O, which was calculated using the R function lm. If the correlation was significant (P < 0.05), the MRDep slope was used to correct all of the measurement data for the 1- to 5-day block including calibration gases and chamber measurements. Direct measurement of MRDep was only made after July 31 when the full calibration manifold was installed; before this date, the mean MRDep for August 1 to 15 was used to correct data. The mean mixing ratio and isotopic composition for the flat peak for each calibration measurement (AmbSys, AmbIso, MRDep, Comp) was calculated and used to produce a dataset as shown in fig. S3.

The fluxes of CO₂, CH₄, and N₂O for each chamber measurement were then found. First, the approximate time period was identified from the LI-COR data file and the LabView record file. Large changes in CO₂ mixing ratio at the start and end of each chamber measurement due to flushing were then used to identify the exact time when closed chamber air was present in Picarro G2301. An exponential model (79) was fit to the measurements and used to calculate the CO₂ flux using a dead time of 20 s and a maximum of 120 s of data

$${\mathit{c}}_{\mathit{t}}-{\mathit{c}}_0=\frac{\mathit{Af}}{\mathit{V}\mathrm{\lambda }}(1-{\mathit{e}}^{-\mathrm{\lambda }\mathrm{t}})$$ (3)

where A is the chamber area, V is the system volume, t is the time, λ is a constant with units of time⁻¹, and c₀ and c_t are the CO₂ mixing ratios in the cell at initial time and at time t, respectively. When the exponential model failed to converge (<5% of measurements), a linear calculation was used. For both exponential and linear methods, fluxes were accepted at a confidence level of P < 0.05; measurements with no significant flux at P < 0.05 were designated as being below the detection limit. CH₄ and N₂O fluxes were calculated using the same method; however, all 15 min of chamber data was used. Convergence of the exponential model was worst for CH₄, because fluxes were lowest in magnitude relative to the CH₄ mixing ratio. The CO₂ fluxes from the Picarro G2301 data agreed very well with the fluxes from LI-COR 8100A calculated in SoilFluxPro, with R² of 0.75 to 0.95 and slopes of 0.8 to 1.1 for all data within each of the 1- to 5-day blocks. In all subsequent results and discussion, only CO₂ fluxes calculated from the Picarro G2301 data are shown.

The isotopic composition of source N₂O was calculated for each chamber closure using the Keeling plot method (80, 81), which uses a plot of inverse chamber mixing ratio against measured isotopic composition to find the source isotopic composition as the intercept

$${\mathrm{\delta }}_{\mathit{\text{meas}}}=\frac{1}{{\mathit{m}}_{{\mathrm{N}}_2\mathrm{O}}}+{\mathrm{\delta }}_{\mathit{\text{source}}}$$ (4)

where m is the mixing ratio and δ is the isotopic composition, e.g., δ¹⁵N. The 1-s raw isotopic data were very noisy; therefore, for the chamber closure, the data were first divided into six equal time blocks, and mean data were found for each block. The Keeling plot method was then applied to the six-block smoothed data to calculate the mean source isotopic composition. Error in the isotopic composition values (i.e., intercept) used the standard R algorithm for linear regression (lm). Isotopic composition values for the source were accepted if the Keeling plot regression was significant at P < 0.05 and the calculated error was less than 150‰. The constant 30-sccm dilution introduced into the system to balance the subsampling to Picarro G5131i (Fig. 6) was accounted for in flux and isotope data assuming that the added ambient air had the same mixing ratios and isotopic composition as measured ambient air; the total flow rate was 1 to 1.5 liters min⁻¹, so the effect of this correction was minor.

Following calculation of each 1- to 5-day block over the 5-month measurement period, all data were combined for postprocessing and calibration. Calibration to the international isotopic standard scale was accomplished by measuring two N₂O isotopic standards from Empa, Switzerland (Cal1 and Cal2; see table S2). These two standards were directly measured against the four calibration gas types in the “Measurements of trace gas fluxes and N₂O isotopic composition” section, as shown in table S2 (82). All chamber measurement data were corrected using a previously implemented approach described in detail in (83–85) and presented in brief here: AmbFS was first used as an anchor standard to correct drift in all measurement data by calculating deviations in AmbFS from its mean value and applying these to all data. Then, adjustment to the international standard scale was accomplished with a three-point calibration from AmbFS, Comp, and MRDep mean values over the entire measurement period. The final flux and isotope data were then filtered by removing outliers more than three SDs from the mean for each monolith. Of the total 9778 chamber measurements, 8202 (73%), 5341 (47%), and 4097 (37%) accepted values were obtained for N₂O flux and source δ¹⁵N and SP, respectively. Details of the measurement precision and performance are given in section S1.4.1.

Estimation of microbial N₂O production and consumption pathways: The CPM

The microbial pathways contributing to N₂O production pathways can be estimated on the basis of measured fluxes and isotopic composition. Current literature estimates show some range in SP fractionation depending on experimental conditions, microbes studied, and other parameters such as pH, substrate, and substrate availability; thus, best estimates for the SP endmembers of denitrifying pathways (production) and nitrification were taken as SP_D = −10‰ and SP_N = 30‰, respectively (86). Fractionation factors of α_SP = −5.4‰ and α_{15N − bulk} = −6.6‰ were used for SP and δ¹⁵N fractionation during N₂O reduction as the final step of heterotrophic denitrification (18, 86). These isotopic fractionation factors will hereafter be referred to as “base values” to distinguish them from different isotopic values used in sensitivity studies discussed below.

As N₂O is reduced in the final step of denitrification, isotopically light N₂O and N₂O − β are preferentially consumed; thus, the isotopic composition and SP of the remaining N₂O pool increase. It was therefore assumed that reduction activity would be evident as a negative correlation between SP and N₂O flux or a positive correlation between δ¹⁵N and SP (18, 87, 88). To quantify contributions from each pathway to the final N₂O flux, the data were considered for each monolith separately in 12-hour blocks beginning every hour throughout the entire measurement period. If more than three simultaneous measurements of flux, δ¹⁵N, and SP were available within the 12-hour block, the correlations between δ¹⁵N and SP, and flux and SP, were calculated using the R function lm. If either correlation was significant (P < 0.05) and positive/negative, respectively, it was taken as evidence for significant N₂O reduction. Several cases were considered:

1) If the mean SP within the block was lower than SP_D, denitrification was assumed to contribute 100% of N₂O with no N₂O reduction occurring.

2) If the mean SP within the block was between SP_D and SP_N but there was no evidence of reduction based on correlations between δ¹⁵N and SP, and flux, the relative proportions of nitrification and denitrification were calculated using a mixing model with the two endmembers, where f_D is the fraction of N₂O contributed by denitrification, f_N = 1 − f_D, and SP_mean is the mean SP in the 24-hour window

$${\mathit{f}}_{\mathrm{D}}=\frac{{\text{SP}}_{\text{mean}}-{\text{SP}}_{\mathrm{D}}}{{\text{SP}}_{\mathrm{N}}-{\text{SP}}_{\mathrm{D}}}$$ (5)

3) If the mean SP within the block was larger than SP_D and there was evidence for reduction, the measured slope between SP and δ¹⁵N was compared to α_SP/α_{15N − bulk} to determine the proportion of N₂O being reduced. For example, if a slope of 0.60 was measured for SP/δ¹⁵N, compared to α_SP/α_{15N − bulk} = 0.82, it was estimated that $\frac{{(\text{SP}/{\mathrm{\delta }}^{15}\mathrm{N})}_{\text{slope}}}{{\mathrm{\alpha }}_{\text{SP}}/{\mathrm{\alpha }}_{15\mathrm{N}-\text{bulk}}}=0.60/0.82=73\%$ of the variability in isotopic composition was due to reduction (f_red = 0.73), with the remaining variability resulting from mixing between nitrification and denitrification. f_red was used to find SP₀ according to Rayleigh fractionation (89)

$${\text{SP}}_0={\text{SP}}_{\text{mean}}-{\mathrm{\alpha }}_{\text{SP}}\times \text{ln}(1-{\mathit{f}}_{\text{red}})$$ (6)

where SP₀ is the SP of the directly emitted N₂O before any reduction took place. The proportion of the emitted N₂O contributed by nitrification and denitrification was then calculated from SP₀, according to Eq. 3, with SP₀ used in place of SP_mean.

For clarity, this approach to partitioning N₂O production pathways will hereafter be referred to as the CPM. Given the reported variability in endmember isotopic SP values for nitrification and denitrifying pathways (18, 86), CPM was tested using a simulated isotopic data series with a Monte Carlo approach (see section S1.3.3). The method was found to perform well, particularly when fluxes showed little variability in the modeled window (12 hours), and when reduction was low (usually <10% in this study). Uncertainty in partitioning N₂O pathways using CPM would also be higher in situations where denitrification and nitrification contributions are approximately equal, as numerous minor N₂O production pathways produce N₂O with an SP in this intermediate range (59).

CPM results in some “smearing” of the distinction between different pathways across time because of the 12-hour window used for each calculation. However, multiple measurements are needed to partition between three possible pathways; therefore, this is the best approach possible with the available data. CPM can perform poorly when pathways—particularly reduction—are highly variable within a 12-hour window. We found that N₂O fluxes did not vary strongly within most 12-hour data periods (see Results); therefore, we assume that pathways are not highly variable at this time scale. Future studies using higher measurement time resolution—for example, when instrumental precision improvements allow source isotopic composition calculation with shorter chamber closure—will allow shorter CPM windows and thus improve temporal resolution of pathway calculations.

Soil sampling and ancillary measurements

Biomass growth

In order for the automated chambers to open and close properly, the height of plant biomass was maintained at <10 cm above the chamber collars. Therefore, each week of the experiment, biomass was cut to the top of the chamber collar, dried for 4 to 5 days at 80^∘C, and weighed. Following the end of the measurement period, on 12 November 2018, all dead and living biomass was removed down to the soil surface and grouped according to dead grasses, dead forbs, live grasses, and live forbs—subsampled to the main alpine genera Cirsium sp., Plantago sp., Rumex sp., Geum sp., Leontodon sp., Trifolium sp., and other forbs. The weight wet and dry weight (4 to 5 days at 80^∘C) of each plant group were measured for each monolith.

Soil sampling during the experiment

A week after fertilization on 12 July 2018, as well as several days before and after each of the two rewetting events (on 31 August, 6 September, and 13 September 2018), soil sampling was conducted on the top 6 cm of soil within each collar using a 2.5-cm-diameter core for the analysis of soil N isotopic composition and microbial ecology (see table S1 for a summary of important dates). Holes created by sampling were filled with surrounding soil; no changes were seen in fluxes on soil sampling days, and all monoliths received the same treatment. Four grams of soil from each monolith was weighed, dried at 80^∘C for 3 days, and weighed again, before freezing at −18^∘C. Analyses for bulk δ¹⁵N, δ¹³C, and C and N were performed at the Department of Microbiology and Ecosystem Science, University of Vienna, using EA-IRMS (Elemental Analyzer Isotope Ratio Mass Spectrometry).

Soil sampling after the experiment

After the biomass was collected, the monoliths were destructively sampled on 19 and 20 November 2018 into three depth layers (0 to 6, 6 to 12, 12 to 18 cm depth) to gain detailed information on soil physical and chemical properties. A HYPROP sampling core was taken at the 0- to 6-cm and 6- to 12-cm depths and stored in the refrigerator at 4^∘C for several months before analysis of the soil moisture retention using HYPROP 2 instrumentation (METER Group Inc. USA). HYPROP instrumentation uses two high-precision tensiometers inserted into a soil core to monitor soil water potential as a saturated soil sample dries, thus allowing accurate and fast measurement of the moisture release curve.

A 6-cm-deep soil core with radius 14 mm (approximately 40 g) was taken at each depth in each monolith and dried for 24 hours at 105^∘C before weighing to determine the soil bulk density. A 1- to 2-g subsample of the dried soil was transferred to an Eppendorf Safe-Lock tube and ground for 5 min at 30 motions s⁻¹ in TissueLyser (QIAGEN Inc., USA). Subsamples (3 to 5 mg) were taken for measurements of total soil C and N using an elemental analyzer (“FlashEA 1112”, Thermo Fisher Scientific, MA, USA) with the standard NC Soils configuration.

Soil ${\text{NH}}_{\mathbf{4}}^{\mathit{+}}$ and ${\text{NO}}_{\mathbf{3}}^{\mathit{-}}$ nitrogen isotopic composition

For soil samples taken during the experiment (see the “Soil sampling during the experiment” section), 3 g of soil was weighed and immediately extracted in 1 M KCl (N-free, EMSURE, Merck KGgA) for 30 min while being shaken with 250 motions min⁻¹. The extracts were then filtered (Munktell folded filter paper, 80 g m⁻²) and frozen at −18^∘C. In addition, leachate was periodically collected from the base of the monolith steel containers and frozen at −18^∘C. For both extracts and leachates, ${\text{NH}}_4^{+}$ and ${\text{NO}}_3^{-}$ concentration and δ¹⁵N isotopic composition were measured at the Department of Microbiology and Ecosystem Science, University of Vienna. ${\text{NH}}_4^{+}$ and ${\text{NO}}_3^{-}$ concentration were determined colorimetrically, by measuring the spectrophotometric absorbance. For isotopic analysis, ${\text{NH}}_4^{+}$ was isolated from the extracts using microdiffusion and analyzed with EA-IRMS (90, 91). Isotopic analysis of ${\mathrm{\delta }}^{15}\mathrm{N}-{\text{NO}}_3^{-}$ was conducted with PT-IRMS (Purge-and-Trap Isotope Ratio Mass Spectrometry) (90, 91).

Functional gene analyses

For soil samples taken during the experiment (see the “Soil sampling during the experiment” section), 1 to 2 g of fresh soil were immediately frozen at −18^∘C in Eppendorf Safe-Lock PCR Clean tubes for microbial ecology analysis at the Helmholtz Zentrum München. DNA extraction was performed as described in (92) and (93) using a phenol/chloroform/isoamylalcohol mixture. qPCR was used to measure overall bacterial abundance [16S ribosomal RNA (rRNA) gene] and the abundances of key marker genes relating to the nitrogen cycle: amoA (AOA), amoA (AOB), nirK, nirS, nosZ, and nifH. qPCR protocols and primers are described in (94) for the N-cycle genes and in (95) for the 16S rRNA gene. To avoid PCR inhibition, the optimal DNA extract dilution (functional genes 1:256, 16S rRNA gene 1:4096) was determined by a dilution series qPCR. The R² of all standard curves was above 0.99, and the efficiencies of the amplification were as follows: 91.3% amoA (AOA), 81.6% amoA (AOB), 93.3% nirK, 81.7% nirS, 94.1% nosZ, 83.7% nifH, and 96.8% 16S rRNA gene.

NanoSIMS measurements

Soil samples collected before the drought, at peak drought, and a couple of days after rewetting (see the “Soil sampling during the experiment” section and table S1) were used for NanoSIMS analysis. Samples were held at −18^∘C until 8 months after collection, when they were prepared for NanoSIMS analysis. A tiny (<mg) amount of frozen soil was placed in a petri dish using tweezers, and 15 μl of isopropanol was added. Six microliters of isopropanol with suspended soil grains was picked up with a micropipette and dropped onto a 5 mm × 5 mm silicon wafer (Agar Scientific, UK), similarly to (46) and (47). The isopropanol was dried, and the soil grains remained adhered to the silicon wafer. A 10-nm platinum coating was added before measurements to improve sample conductivity and reduce lateral charging. Only a subset of samples were analyzed with the NanoSIMS (see table S3), as NanoSIMS measurements and data analysis are extremely time-consuming and expensive.

Measurements were performed with NanoSIMS 50L (CAMECA, France) at the Photon Science Institute, University of Manchester. This ion microprobe can provide extremely high lateral resolution as well as depth resolution for up to seven ion masses using a multicollector system to simultaneously achieve high mass resolution, sensitivity, and precision, as well as detection of the ion-induced secondary electrons (SE). In this study, a 16-keV Cs⁺ beam with a beam current of 0.7 to 0.8 pA (D1-4) and ∼0.3 pA (D1-5) was scanned over the surface to generate negative secondary ions (¹²C⁻, ¹⁶O⁻, ${}^{12}{\mathrm{C}}_2^{-}$, ¹²C¹⁴N⁻, ²⁸Si⁻, ³²S⁻, and ${}^{14}{\mathrm{N}}^{16}{\mathrm{O}}_2^{-}$) (48) that were analyzed in a double-focusing mass spectrometer. Before image collection, a 5 ×10¹⁵ Cs atom cm⁻² implantation dose was used to achieve steady state and sputter away the Pt coating. Large-area scans with a 30 × 30 μm raster over 256 × 256 pixels containing numerous particles were taken at several positions for each sample with 200 μs pixel⁻¹ dwell time and D1-4, ES-3, AS-2, and EnS-open. Smaller areas with a raster size of 3 to 10 μm were acquired to gain information in depth for single particles. The NanoSIMS tuning for depth analysis was 256 × 256 or 128 × 128 pixels, dwell time 200/500/1000 μs pixel⁻¹ according to D1 aperture position, D1-4, or D1-5; other slits were the same as for the large-area scans. ImageJ with the OpenMIMS plugin (Harvard, Cambridge, MA, USA) was used for data analysis. Regions of interest corresponding to each particle visible in larger raster images or single particles in small raster images (96) were defined using OpenMIMS and used to measure the ion intensity in each analysis layer, as well as ion ratios for ions of interest. A total of 218 individual grains were defined as regions of interest and analyzed with OpenMIMS to produce the results shown in Fig. 3 and table S3. The depth scale in NanoSIMS sputtering layers (Fig. 3) cannot be easily converted into nanometers without extensive further study of the soil matrix, which is beyond the scope of this paper (96); thus, the depth scale is shown as a relative depth normalized for differing dwell times.

Statistics

To compare overall differences within a treatment at different time steps, e.g., D monoliths pre-drought and peak drought, and to compare differences between treatments at the same time step, e.g., C and D at peak drought, analysis of variance (ANOVA) was used on data from the 16 monoliths with a cutoff significance level of P < 0.05. When comparing particular treatment times/groups, Wilcoxon signed-rank and rank-sum tests for nonparametric data were applied. Cutoffs for P values were included throughout this paper as a guide to the reader; however, when P values were close to cutoffs, they are reported directly and considered closely. Context of the results including previous research and consistent patterns across independent methods were used to interpret results and reach major conclusions as recommended by (97). A PCA was used to look at causes of variability in the data. This was applied to two periods of the dataset: The initial (pre-drought) period from 7 to 14 July 2018, comprising a total of 626 data points, and the combined peak drought (26 August 2018 to 2 or 10 September 2018 for D1 and D2, respectively) and rewetting periods (3 or 11 September 2018 to +5 days), comprising 1186 data points. Data were normalized to a range between 0 and 1 before performing the PCA using the R function prcomp (98). All statistical analyses were performed using R 3.5.1 (98).

The experimental design in this study accounts for the variability of a natural grassland by using four monoliths for C and W treatments and seven for D treatments (D1 = 4 and D2 = 3). Monoliths (>50 to 100) would ideally be needed to cover the true variability in N₂O fluxes [e.g., 100+ plot replicates (99, 100)], which is beyond the scope of this study. However, the high measurement frequency in this study can counterbalance the relatively low spatial measurement intensity (99), particularly considering the high temporal variability of N₂O emissions, as we report 1 to 2 orders of magnitude more data points for N₂O source isotopic composition than have been previously published [e.g., (87), (40)]. In addition, the use of two time points for both fertilization and rewetting accounts for the high variability of N₂O fluxes in response to current meterological and microclimatic conditions.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/7/6/eabb7118/DC1