This study took place at FarmED, near Shipton-under Wychwood UK (51.869981,-1.581136, https://www.farm-ed.co.uk/). FarmED is an experimental agricultural station founded in 2018, consisting of strips of the same soil type under rotational herbal ley management (1–5 years). Prior to this layout, the soil (cambisol with an oolitic limestone bedrock) had been conventionally managed during 30 years for wheat and barley production. One experimental strip has been conserved under this conventional management (referred to as the arable strip). It includes the production of wheat and barley, minimum tillage and application of fertilizer, pesticides and herbicides. The herbal ley strips were all sown with the same mix, which includes white and red clovers (Trifolium repens and Trifolium pratense respectively), timothy (Phleum pratense), chicory (Cichorium intybus), lucerne (Medicago sativa) or sainfoin (Onobrychis viciifolia) among others (see Supplementary Table S1). For the present work, the 4-year-old herbal ley strip was studied and compared to the arable strip for a full year.

During our field campaign (April 2022–May 2023), the arable strip was sown with winter wheat in September 2021, which was harvested in August 2022, resown in October 2022 before being harvested again in August 2023. The herbal ley was mob-grazed by cows and sheep in July 2022 and April 2023. Synthetic liquid fertilizer was applied in the arable field with a rate of 179.5 kg N ha^-1 in three applications (Supplementary Table S2).

In this study, we selected eight locations in parallel within both fields (Supplementary Figure S1), where denitrification emissions from soil cores were measured (9 measurements during the study year) using our newly developed method (Micucci et al., 2024), field fluxes of N₂O and CO₂ were measured using conventional static chambers, and soil was sampled and brought back to the laboratory for quantification of N nutrients (nitrate and ammonium), dissolved organic carbon (DOC) and soil moisture.

During each measurement campaign, three soil samples were taken near core sampling locations (n = 8 per field) using a hand auger to a depth of 15 cm. These samples were then mixed to generate one composite sample per chamber location. After collection, they were transported to the laboratory, sieved to <2 mm, and stored at 4 °C in a refrigerator until further processing.

For mineral nitrogen analysis, 5 g of soil were extracted using a 0.5 M potassium sulfate (K₂SO₄) solution at a 1:8 ratio. The extraction mixture was shaken for 2 h at 200 rpm, followed by centrifugation at 3,000 rpm for 5 min. The supernatant was subsequently filtered through a 0.45 µm syringe filter and analyzed using a San++ continuous flow analyzer (Skalar, Breda, Netherlands). Nitrate (NO₃ ⁻) concentrations were determined via cadmium reduction, while ammonium (NH₄ ⁺) was measured using the modified Berthelot reaction. Detection limits were 0.02 mg N L^-1 for NO₃ ⁻ and 0.05 mg N L^-1 for NH₄ ⁺, with blank correction applied. For the quantification of dissolved organic carbon (DOC), 5 g of soil were extracted using 40 mL of deionized water, then shaken at 200 rpm on an orbital shaker for 2 h. The extracts were subsequently filtered through Whatman No. 42 paper (GE Healthcare). Filtrate analysis was performed using a Shimadzu TOC-L analyzer (Japan), with a detection limit of 1 ppm for DOC. Moisture content was measured gravimetrically based on mass loss after drying at 105 °C for 24 h and is expressed relative to dry mass. Water-filled pore space (WFPS) was calculated using bulk density and assuming a soil particle density of 2.65 g cm^-3. Other soil characteristics (such as texture, pH, organic matter content, etc.) were measured once at the start of the study and are presented in Supplementary Table S3.

We measured denitrification in situ using our newly developed method (Micucci et al., 2024), which combines the ¹⁵NGF with an artificial atmosphere depleted in N₂. Briefly, 10 cm-high intact soil cores (n = 8 per land-use type; 16 cores per month in total) were sampled inside 15 cm-long plastic liners (4.6 cm ID) and left to pre-incubate overnight under environmental conditions. The next day, ¹⁵N-NO₃ ^- tracer solutions were prepared by dilution of K-¹⁵NO₃ (98 at% ¹⁵N, Merck) and applied to reach a 50% ¹⁵N enrichment of the soil denitrifying pool (based on measured nitrate levels) and to induce an increase in gravimetric soil moisture <5% via 40 injections (at 0, 2.5, 5 and 10 cm depth). The liners were subsequently capped with gas tight lids with butyl rubber septa and flushed with a gas mixture containing 5% N₂, 20% O₂, 75% He and 0.11 ppm of N₂O (Supplementary Figure S1), and left to incubate in freshly dug holes in the soil under environmental conditions. Gas sampling occurred at times 0, 2 and 4 h where 15 mL of the liner’s headspace were sampled and transferred to pre-evacuated 12 mL Exetainer vials (Labco, UK). To equilibrate the loss of pressure, 15 mL of the artificial gas mixture were transferred back to the liners’ headspace. Errors resulting from this sampling procedure and diffusive fluxes due to incomplete liner sealing were corrected as detailed in Micucci et al. (2024). Upon return to the laboratory, 1 mL of the gas samples was analysed on an Agilent 7890A gas chromatograph to determine N₂O and CO₂ concentrations, with repeatability (1σ) of 3 ppb and 11 ppm, respectively. The remaining 14 mL of the samples were analysed over a continuous flow IRMS (Elementar Isoprime PrecisION; Elementar Analysensysteme GmbH, Hanau, Germany) for the determination of ratios R29 (²⁹N₂/²⁸N₂) and R30 (³⁰N₂/²⁸N₂) for N₂ as well as ratios R45 (⁴⁵N₂O/⁴⁴N₂O) and R46 (⁴⁶N₂O/⁴⁴N₂O) for N₂O. The repeatability (1σ) for these measurements were (1.5 × 10⁻⁶) for R29, (9.3 × 10⁻⁶) for R30, (3.1 × 10⁻⁵) for R45 and (8.2 × 10⁻⁵) for R46. The full description of these analyses can be found in Section 2 of the Supplementary File. Denitrified N₂ and N₂O fluxes were calculated using the approach described in Micucci et al. (2024).

From the liner measurements, we calculated two denitrification parameters, the source partitioning coefficient (SPC) and the macroscopic product ratio (R_N2O, see Micucci et al., 2025) defined as per Equations 1, 2: S P C = f N 2 O   d e n i t r i f i e d f N 2 O   t o t a l (1) R N 2 O = f N 2 O   d e n i t r i f i e d f N 2 O   d e n i t r i f i e d + f N 2   d e n i t r i f i e d (2)

Where f is a flux of nitrogen (either as N-N₂O or N-N₂; in kg N ha^-1 month^-1). The SPC represents the part of soil N₂O emissions attributable to denitrification while the R_N2O represents the proportion of denitrification product emitted as N₂O rather than N₂.

Two weeks before the measurement campaign of April 2022, eight collars (20 cm OD, 15 cm height) were inserted 10 cm deep into the soil for installing conventional static chambers (20 cm OD, 20 cm total height) in both the herbal ley and the arable field (Supplementary Figure S1). Because of the distance between chambers and the fact that the soil core incubation was running at the same time, the sampling of the chambers’ headspace occurred every hour for 2 h. Since this does not respect the criterion of chamber height to deployment time ratio ≥40 cm h⁻¹ recommended by Zaman et al. (2021), we corrected for diffusion issues using the HMR model (Pedersen et al., 2010), with the associated R script (R Core Team 2023) from CRAN (Comprehensive R Archive Network). To illustrate the magnitude of the diffusion correction, in May 2022, 14 out of 16 chambers required this correction, leading to an average 47% increase in N₂O fluxes. This is consistent with the results of Anthony et al. (1995) and Pedersen et al. (2010), who reported values of 54% and 52% respectively. Gas samples were collected and analysed using the same approach as for the cores, but without the IRMS analysis.

While our nine field campaigns represent a limited temporal resolution at the annual scale, they still allow us to derive a first-order denitrification budget and estimate field-scale denitrification product ratios. Chamber and core measurements both have strengths and limitations. Open-bottom static chamber measurements do not require soil disturbance (after initial insertion of the chamber collar) and capture the integrated soil profile emissions at the study location. They are however sensitive to downward or lateral diffusive losses, especially for prolonged measurement times (Hutchinson and Mosier, 1981), which can lead to underestimations of total denitrification fluxes by up to 37% (Well et al., 2019b). These diffusive exchanges with the ambient air also prevent the use of an artificial atmosphere (Micucci et al., 2024). On the other hand, the seal of the liners enables them to retain the artificial atmosphere for prolonged periods of time, allowing for N₂ detection, and their bottom-closed structure prevents diffusive losses. However, the magnitude of flux measured through these liners is probably not representative of field rates, given the small height (10 cm) and surface area (ID 5 cm) of the cores. Nonetheless, since most of the denitrification activity occurs in that zone (more generally in the tillage zone; Groffman et al., 2009; Well et al., 2019a), we hypothesized that the dynamics of denitrification and in particular the SPC and R_N2O coefficients measured through these cores would be representative of field dynamics. Combining these coefficients with the N₂O fluxes measured from static chambers we can derive an alternative denitrification budget, as follows: f N 2 O   d e n i t r i f i e d = S P C × f N 2 O   c h a m b e r (3) f N 2 = f N 2 O   d e n i t r i f i e d × 1 − R N 2 O R N 2 O (4)

Where f_{N2O chamber} is the flux of N-N₂O derived from the static chamber measurements. Given that the soil beneath the flux chambers remained isolated throughout the entire campaign (due to the collars inserted 10 cm deep, see Section 2.4), while new soil cores had to be sampled near the chamber locations every month, and given that denitrification exhibits high spatial variability at the micro-scale, monthly averages were used for chamber-based N₂O flux magnitudes as well as denitrification parameters. The propagation of error was determined as per Equations 5, 6: S E f N 2 O   d e n i t r i f i e d = S P C × S E f N 2 O   c h a m b e r 2 + S E S P C × f N 2 O   c h a m b e r 2 (5) S E f N 2 = S E f N 2 O   d e n i t r i f i e d × 1 − PR PR 2 + S E PR × f N 2 O   d e n i t r i f i e d PR 2 2 (6)

Where SE is the standard error.

To take into account our experimental design (measurements repeated in time and space), we used linear mixed models with the lme4 package of R (R Core Team 2023). The unique combination of “Month” and “Land Use Type” was used as a random factor named “unique_sample”.

Models were validated by assessing the normality and homoscedasticity of residuals, and the best-fitting model was selected using the Akaike Information Criterion (AIC). Type III ANOVA, using the Satterthwaite method was performed to assess the effect and significance of the different predictors. To characterize the temporal patterns of the study variables, we used linear interpolation between monthly means; yearly fluxes were derived using the trapezoidal rule. All results are given ±standard error.

The evolution of total N₂O emissions (Figure 1a) and denitrified N₂O emissions (Figure 1c) measured with the liners showed a clear peak in May 2022 in the arable field (respectively 0.84 ± 0.16 and 0.32 ± 0.15 kg N ha^-1 month^-1), during the fertilization period (March to May 2023). At the same time, N₂ emissions were relatively low 0.92 ± 0.31 kg N ha^-1 month^-1 (Figure 1d), which resulted in the highest product ratio recorded during our monitoring period, with a value of 32.0% ± 7.8%. Interestingly, in the prior month (April 2022), the opposite trend is observed, with a high peak of N₂ production at 4.13 ± 0.81 kg N ha^-1 month^-1 and modest emissions of denitrified N₂O at 0.09 ± 0.02 kg N ha^-1 month^-1, resulting in a product ratio of 3.7% ± 1.3%. Other than that, emissions were relatively stable with denitrified N₂ emissions around 1 kg N ha^-1 month^-1 in both fields and denitrified N₂O emissions around 0.1 kg N ha^-1 month^-1. There was a second peak of N₂O emissions in October, observed both in the liners (total and denitrified) and the chambers. The SPC coefficients for both fields were characterized by an almost Gaussian evolution between the months of March and September (Figure 1e), with some anomalies for the herbal ley, having a peak at 62% ± 10% in March and another peak in July with 75% ± 12%. Similar to N₂O emissions, a second smaller peak event was observed around the month of October for both fields. On the other hand, the denitrification product ratios (Figure 1f) were mostly under 15%, at the exception of the month of May 2022 as mentioned previously. We can note that our newly developed method was successful at measuring denitrified N₂ emissions 90% of the time (n_total = 144).

The total N₂O emissions from cores and chambers were in the same order of magnitude (∼1 kg N ha^-1 month^-1 Figures 1a,b). However, the emissions from the herbal ley tended to be higher in the liners, especially from May 2022 onwards, where they were on average 0.12 kg N ha^-1 month^-1 compared to 0.02 kg N ha^-1 month^-1 in the chambers. A peak of emissions (0.26 kg N ha^-1 month^-1) was observed in the chambers in April 2022 but not in the liners. The same thing is observed in the arable field, where peak of almost 2 kg N ha^-1 month^-1 is observed in the chambers in April 2022 but not in the liners.

Finally, the fluxes of CO₂ followed the same pattern whether they were measured with chambers or liners. They were at an all-time high in April 2022 and decreased gradually until August 2022 before a local spike in October 2022 and decreasing to reach all time low levels in February 2023. However, the fluxes measured through the chambers were about 6 times higher than in the liners.

The ammonium content of the two land uses stayed relatively constant (below 3 kg N ha^-1; Figure 2) despite fertilizer application, which contained approximately 75% of ammonia and urea (Supplementary Table S2). On the other hand, fertilizer application in 2022 resulted in a clear increase of the observed nitrate concentration in the arable field (Figure 2). The three applications of fertilizer in 2022 (∼41.5 kg N ha^-1 of N-NO₃, Supplementary Table S2) led to significantly elevated nitrate levels, reaching 56.8 ± 9.8 kg N ha^-1 in May. Fertilizer-derived nitrate accumulated and persisted in the soil until October, at which point the nitrate levels reached their pre-fertilization levels (∼2 kg N ha^-1). On the other hand, the herbal ley had low and constant amounts of available nitrate (<5 kg N ha^-1), except for a moderate increase from July to September 2022.

For both fields, moisture decreased from March 2022 (WFPS ∼75%) to August 2022 (∼20%) at the exception of June 2022, which saw a local maximum peak (70%) due to extremely wet conditions (Figure 2). From there, the moisture levels rose again up to the end of October 2022, where they stabilized until the end of the field campaign in February 2023.

On average, the herbal ley retained more moisture than the arable field. Finally, DOC levels in both fields showed peak events in August 2022 (∼100 kgC ha^-1) which coincide with the grazing period in the herbal ley and the presence of crop residues following harvesting in the arable field. There were also peaks of DOC levels in April 2022 and 2023 (∼70 kgC ha^-1, except in the herbal ley in 2023) aligning with the early spring surge in plant and microbial activity. Otherwise, the levels remained relatively steady (between 25 and 50 kgC ha^-1).

To analyse the dependence of denitrification with environmental variables, we performed linear mixed model analyses as described in Section 2.6. Our four dependent variables were the flux of denitrified N₂O, flux of denitrified N₂, source partitioning coefficient as well as product ratio. Since none of these variables followed a normal distribution, we had to use non-linear transformations. The fluxes of denitrified N₂O and N₂ were log-transformed (Supplementary Figure S2). The SPC and R_N2O parameters varied between 0 and one and therefore could not be long transformed. A Box-Cox test determined that a square root transformation was the most appropriate (Supplementary Figure S3), resulting in these ratios approximating a normal distribution (Supplementary Figure S4). While other studies have used a Beta transformation (e.g., Eckei et al., 2024), our data were more normally distributed using a square root transformation, as indicated by a significantly lower Akaike Information Criterion compared to the Beta transformation for the final model (AIC: 182 vs. 314). We followed the same procedure for the SPC coefficient and found that a power transformation with an exponent of 0.75 was the best transformation (Supplementary Figure S5). This initial model was used for the four dependent variables: s t u d y   v a r i a b l e   ∼   L a n d   u s e + N   e n r i c h m e n t + N i t r a t e + M o i s t u r e + A m m o n i u m + C 15 O 2 + D O C + 1 u n i q u e _ s a m p l e (7)

Although the ¹⁵N enrichment of the denitrifying pool appears collinear with Nitrate, these two variables represent distinct processes (correlation coefficient = - 0.29). Soil nitrate levels were measured before the incubation experiments, whereas ¹⁵N enrichment resulted from the application of ¹⁵N-labeled nitrate during the experiments. Therefore, although both variables assess the dependence of R_N2O on nitrate, the variable Nitrate reflects the influence of native nitrate levels, while ¹⁵N enrichment captures the potential stimulation effect induced by nitrate addition. The models were then adjusted to maximize the explained variance and minimize the AIC.

The results show that CO₂ emissions measured from the cores were significant predictors of all four response variables. Denitrified N₂O emissions also depended on nitrate levels (p < 0.1) and moisture (p < 0.01) while N₂ emissions depended on the ¹⁵N enrichment of the soil denitrifying pool. The SPC was dependent on DOC levels (p < 0.1), moisture (p < 0.001) and ammonium (p < 0.1), while R_N2O depended on nitrate levels (p < 0.01) and moisture (p < 0.05). Interestingly, neither the SPC nor R_N2O depended on the ¹⁵N enrichment of the soil denitrifying pool. Plotting these evolutions (Figures 3a,b) confirms that the influence of the ¹⁵N enrichment on the observed values was minimal.

Since the SPC and R_N2O were not dependent on the quantity of added tracer, we could combine liner and chamber measurements (see Section 2.5) to derive field scale fluxes of denitrification as per Equations 3, 4, (Figure 4). According to this approach, the denitrification magnitude was high in the arable field in April 2022 (∼14 kg N month^-1 ha^-1) following fertilizer application, although it was associated with a high uncertainty (7.4 kg N month^-1 ha^-1). The rest of the time, emissions were below 3 kg N month^-1 ha^-1 in this field. On the other hand, the herbal ley consistently emitted denitrification fluxes below 0.2 kg N month^-1 ha^-1, at the exception of April 2022, where fluxes reached 1.2 ± 0.68 kg N month^-1 ha^-1. Using an initial statistical model similar as in Equation 7, denitrification fluxes were log-transformed. The final statistical model used for the total denitrification fluxes was: log t o t a l   d e n i t r i f i c a t i o n ∼   L a n d   u s e + N i t r a t e + M o i s t u r e + D O C + 1 u n i q u e _ s a m p l e (8)

In this model (R² _m = 46%, R² _c = 70%), Land use was the predominant fixed effect (F = 69.5, p < 0.05) while moisture played a marginal role (F = 4.12, p < 0.1). The levels of DOC were also close to a marginal effect (F = 2.47, p = 0.13).

We derived an annual denitrification budget for the period from April 2022 to April 2023 to represent a full year of emission using either core measurement or the combination of core and chambers (Table 2).

The annual denitrification budgets based on either cores or cores and chambers gave similar results for the arable field, with estimated total denitrification losses of 17.97 and 22.12 kg N ha^-1 respectively. This trend was not observed for the herbal ley where the measurements were significantly different between the two methods (14.62 and 2.41 kg N ha^-1 respectively). Based on the core and chamber approach, and accounting for a fertilizer application rate of 179.5 kg N ha^-1 in the arable field, we estimate a denitrification emission factor of 11% (19.71 kg N ha^-1 losses) after subtracting background emissions measured in the herbal ley. Similarly, we can derive a N₂ emission factor of 10.2%.

Our newly developed liner method was characterized by a high success rate of N₂ flux detection (90% of measurements). Data gaps in denitrification dataset are frequent due to the very high sensitivity required to discriminate soil-N₂ fluxes (Mulvaney and Kurtz, 1984; Kulkarni et al., 2014; Buchen et al., 2016). Although no existing number on the proportion of dataset gaps could be found, a 90% success rate can be considered very high for in situ denitrification measurements. Deriving a denitrification annual budget using either the core or the combination of cores and chambers results gave consistent results for the arable field, with 17.97 and 22.12 kg N ha^-1 year^-1 respectively (Table 1). However, this trend is not observed for the herbal ley, where the two methods yielded 14.63 and 2.41 kg N ha^-1 year^-1 respectively. Even when intact, soil cores do not fully reproduce field conditions, as they are laterally isolated, disconnected from the surrounding soil matrix, and subject to artificial boundary conditions (liner and headspace), which can slightly reduce oxygen diffusion, facilitate gas accumulation, and stabilize anoxic microsites. Nonetheless, the observed differences between cores and chambers for the herbal ley might be better explained by a stimulation effect due to soil disturbance, release of organic carbon from the legume roots and tracer injection in the herbal ley, where consistently low nitrate levels were observed (Figure 2). In particular, the statistical analysis revealed a marginal dependence of denitrified N₂ fluxes on the ¹⁵N enrichment of the nitrate pool (p < 0.1), reflecting a direct effect of the tracer dose applied. Another evidence for this effect is in the CO₂ fluxes, which are usually considered consistent and proportional to the amount of soil incubated (Cueva et al., 2019). Here, CO₂ fluxes followed a similar trend between cores and chambers. However, the herbal ley emitted relatively more CO₂ than the arable field in the cores compared to the chambers, resulting in a more pronounced difference between land uses under core conditions. Emissions of CO₂ were a strong indicator of denitrification activity in this study, further pointing to a stimulation effect. Although denitrification—and to some extent, nitrification—exhibits high spatial variability (Ambus and Christensen, 1993), leading to significant differences in N₂O emission magnitudes between closely located sampling points (i.e., core and chamber locations here), a stimulation effect remains the more plausible explanation here. Such effects have been previously reported in the literature (Dodds et al., 2017; Warner et al., 2019; Lewicka-Szczebak and Well, 2020).

If addition of tracer can impact the magnitude of emissions, can it influence the SPC and R_N2O coefficients? This was one of the key questions raised in Micucci et al. (2024), where a significant correlation between nitrate addition and R_N2O was observed in the laboratory but not necessarily in the field. The available literature reports several instances of product ratio stimulation under high amendment of nitrate (>75 kg N ha^-1; Jarvis et al., 1991; Loick et al., 2017). Here, using 144 field measurements, we show that adding tracer-level amounts of nitrate had no significant impact on these coefficients, although the p-value for the product ratio approached marginal significance (p = 0.13). Furthermore, field nitrate levels were a significant predictor of the R_N2O (p < 0.01), suggesting that a potential influence of the ¹⁵N–NO₃ ^- tracer cannot be completely ruled out. However, in the context of our study, there is enough evidence to consider the R_N2O and SPC as independent from the addition of nitrate tracer (Figure 2) and use them in combination with N₂O fluxes measured from undisturbed soil under static chambers. The denitrification budgets for the herbal ley and arable field using this approach (2.41 and 22.12 kg N ha^-1 year^-1 respectively, Table 1) are much more realistic and consistent with the nitrate levels observed during the study period in these fields than the budgets derived from cores alone. Indeed, the nitrate levels were mostly below 5 kg N ha^-1 in the herbal ley but reached almost 60 kg N ha^-1 in the arable field (Figure 2). Therefore, the combination of core and chamber measurements was considered the most appropriate approach for estimating the denitrification budget in our study.

While the ¹⁵NGF approach traces the transformation of nitrate into N₂, unlabelled ammonium can be converted to nitrate via nitrification and thereby indirectly contribute to N₂ emissions. This is relevant in the present study, as the application of ammonium and urea fertilizers (Supplementary Table S2) did not result in increased field ammonium concentrations (Figure 2), indicating that a fraction of the applied NH₄ ⁺ was likely nitrified. The ¹⁵NGF calculations explicitly recalculate soil ¹⁵N enrichment and therefore partially account for this potential dilution of the tracer signal, although the complete nitrification-coupled pathway cannot be resolved explicitly without additional tracers.

To further improve our methodology in future studies, we suggest: i) Applying the ¹⁵N tracer at the same time as the sampling of cores (i.e. 24 h before measurement), which would allow for the tracer to diffuse in the soil matrix and could potentially limit stimulation effects. In our case, the measurements were made after tracer injection to ensure sensitivity and because samplings at times 0 already shown enrichment in the ¹⁵N signature of N₂O, indicating the start of tracer denitrification. However, our high success rate during this 1-year field campaign shows that sensitivity is no longer an important issue with our new method. ii) We also suggest using longer liners which will enable to sample the full soil A horizon and introduce larger headspace, where gas sampling will be less disruptive. iii) Increasing the frequency of denitrification measurements and coupling them with automated chamber and continuous analyser will allow a more thorough characterization. iv) Finally, a¹⁵N enrichment of 50% has been shown to be ideal for optimizing the sensitivity of the ¹⁵N gas flux method (Micucci et al., 2023) and should be maintained as much as possible. However, this remains particularly challenging in natural environments, where nitrate levels exhibit high spatial variability.

Overall, the combined use of core and chamber measurements aligns with recent methodologies, such as those employed by Wang et al. (2020) or Bizimana et al. (2022), where laboratory-derived product ratios were applied to field chamber data. However, in our study, the product ratio is determined under in situ conditions, likely providing a more accurate representation. Such an approach can help unravel the complexities of denitrification at a relatively low cost. Indeed, our method relies on the use of a small volume of artificial atmosphere to replace the headspace at the beginning of the incubation experiment and does not require a continuous flow-through system in contrast to other approaches (e.g., Cárdenas et al., 2003; Well et al., 2019a), resulting in reduced consumption of the expensive artificial atmosphere mixture. The approach nevertheless remains costly due to the use of ¹⁵N tracers. Combining this method with automated flux chambers would therefore be particularly advantageous, as it would substantially increase measurement frequency—one of the main limitations of the present study—while maximizing the information gained per incubation.

In our study, nitrate and moisture impacted denitrified N₂O emissions, but more importantly the denitrification product ratio (Table 1), which directly affects the efficiency of denitrification as a sink for nitrous oxide. The denitrification product ratios in both fields stayed below 15% during the measurement campaign, at the exception of May 2022 where it reached 32% in the arable field. This coincides with the highest levels of nitrate recorded during our campaign (∼55  kg N/ha, following fertilization, Figure 2) and with a local minimum in soil WFPS (50%). These results are consistent with the conditions that favour incomplete denitrification, such as a relatively low moisture (<75% WFPS, Davidson, 1992; Saggar et al., 2013; Weier et al., 1993) and fertilizer-levels of nitrate amendments (Clayton et al., 1997; Loick et al., 2017). Another parameter that influences denitrification is temperature, which has rarely been characterized in the field due to lack of measurements. Since denitrification is a microbially mediated process, it tends to be minimal around freezing temperatures (Dorland and Beauchamp, 1991), however the effect on the denitrification product ratio remains uncertain under field conditions. Laboratory studies have shown that increasing temperature can increase this ratio, favouring N₂O accumulation over complete reduction to N₂ (Phillips et al., 2015). Here, we characterized exceptionally low product ratios during winter, with values of 0.39% and 0.22% for the herbal ley and arable field, respectively under a temperature of 4 °C. This seems to indicate that the product ratio decreases at low temperatures, consistent with the results of Phillips et al. (2015). To our knowledge, this aspect has rarely been characterized, and more generally, the temporal variations of the product ratio under in situ conditions remain largely unexplored. Indeed, in a meta-analysis compiling 236 studies, it was found that 74% of the reported denitrification fluxes were measured instantaneously rather than over a time period (Almaraz et al., 2020).

Multiple microbial processes contribute to N₂O production, denitrification and nitrification are the dominant pathways (Bremner, J. M., 1997; Barnard et al., 2005; Zou et al., 2014). Therefore, the SPC essentially reflects the balance between nitrification and denitrification. In our study, it was strongly influenced (p < 0.001) by moisture and marginally (p < 0.1) by DOC content and ammonium levels. The availability of carbon substrate favors denitrification (Seitzinger et al., 2006) while ammonium promotes nitrification as it serves as its primary substrate (Ayiti and Babalola, 2022). Moisture has been shown to regulate the balance between denitrification and nitrification, in particular with a shift towards denitrification around 60%–70% WFPS (Ciarlo et al., 2007; Congreves et al., 2019; Wang et al., 2023). However, we observe an opposite trend here where the SPC is maximal during the summertime where the moisture is at a minimum. This can be explained by the very low magnitude of total N₂O fluxes during this period (<0.1 kg N ha^-1 month^-1 for July and August, as per measured in the static chambers), which likely makes SPC values more sensitive to small absolute variations in isotopic composition. The correlation between total denitrification and moisture (p < 0.1) is clearer, with higher rates in the period between April and June 2022 and in October 2023, when moisture levels were between 50% and 75% WFPS (Figure 2).

Total denitrification was mainly explained by land uses (p < 0.05), which are principally differentiated by N fertilizer application. Finally, DOC levels showed a potential influence but was not statistically significant (p = 0.13), in consistence with the controls of denitrification mentioned previously.

Although our annual denitrification budgets remain inherently sensitive to our sampling resolution, they show as expected that the arable field was a much bigger source of denitrified N₂O and N₂ than the herbal ley (Table 2), with a yearly loss of 22.12 kg N ha^-1 compared to 2.41 kg N ha^-1 (using the combination of chamber and core measurements). This loss occurred in two phases. The first phase occurred during the fertilization and growing periods (April to July 2022), with the highest fluxes recorded in April, reaching 14 kg N ha^-1 month^-1. The static chambers indicated that around 2 kg N ha^-1 month^-1 of N₂O were lost during April and May. However, the SPC coefficient was respectively at 35% and 56%, indicating that nitrification was the dominant process in these fertilizer-derived N₂O emissions. The second phase occurred around October 2022 (∼2.5 kg N ha^-1 month^-1), when nitrate became available again upon rewetting. Indeed, application of fertilizer resulted in the accumulation of nitrate, which started in April 2022 (Figure 2). A large portion of this nitrate was either taken up by plants or lost to the environment by June 2022, but nitrate levels did not return to their pre-fertilization values before October. These dynamics coincided with high N fertilizer application and drought conditions (particularly in July and August, during the second most severe British drought; Barker et al., 2024), under which nitrate accumulation was observed in the soil. The return of nitrate levels to pre-fertilization levels in October coincided with milder and wetter climatic conditions following the drought and resulted in a denitrification peak of 2.5 kg N month^-1 (Figure 4) while nitrate levels dropped by 11.2 kg N ha^-1 (Figure 2). This nitrate build-up was also probably responsible for the high denitrification activity recorded in April/May. Under lower fertilization rates (120.5 vs. 179.5 kg N ha^-1) and milder climatic conditions, nitrate levels did not accumulate in 2023, likely due to more efficient plant uptake. This resulted in total N₂O emissions (as measured using static chambers) being 22 times lower. These findings suggest that denitrification was also drastically reduced in the absence of nitrate accumulation, further emphasizing the critical role of fertilizer management, including not only total nitrogen inputs but also fertilizer form and application timing. In this study, fertilizer was applied in three split applications (Supplementary Table S2); however, more refined management strategies that further synchronize nitrogen supply with crop demand may be required to limit transient nitrate accumulation and associated denitrification losses (Cassman et al., 2002). In the herbal ley, only 2.41 kg N ha^-1 were emitted through denitrification annually, with the highest flux of 1.21 kg N ha^-1 month^-1 recorded in April 2022 and emissions below 0.2 kg N ha^-1 month^-1 the rest of the time. This is consistent with the levels of nitrate observed in this field (3 kg N ha^-1 on average). If we assume that the difference in denitrification between the arable field and the herbal ley is primarily driven by fertilizer application, we can estimate that approximately 11% of the applied N-fertilizer was denitrified, with 7.3% of these losses being emitted as N₂O rather than N₂. This estimate remains approximate, as the herbal ley is expected to influence internal nitrogen cycling compared to an unfertilized arable field. However, given that both systems share very similar baseline soil physical and chemical properties (Supplementary Table S3)—including texture, bulk density, and pH, which strongly control oxygen diffusion, water retention, and nitrate residence time—this influence alone is unlikely to explain the large contrast in denitrification observed here. Rather, differences in fertilizer inputs and nitrate availability are expected to be the dominant drivers, suggesting that our estimates may be slightly conservative.

Overall, these results agree particularly well with the literature. Indeed, in a meta-analysis compiling 336 denitrification measurements (using chambers, cores or N balance techniques), denitrification annual budgets of agricultural soils were mostly found to be in the range of 20–30 kg N ha^-1 year^-1 (Hofstra and Bouwman, 2005). The results of the arable field in the present study (22.12 kg N ha^-1 year^-1) are thus in good adequacy with this range. Our N₂ emission factor (10.2%) also in good agreement with the value recently reported by Kleineidam et al. (2025), who estimated it at (8.6 + 1.9) % for croplands. In their study on denitrification measurements in a German winter wheat field, Eckei et al. (2024) reported lower total denitrification losses (8.60 ± 2.21 kg N ha^-1 over 189 days) despite a higher fertilizer application rate than in our study (210 vs. 179.5 kg N ha^-1). Their methodology, based on chamber-scale denitrification measurements (compared to core-scale measurements in our study), employed a more precise approach—using a continuous flush of artificial atmosphere—and more frequent measurements. This ultimately led to a more accurate quantification of denitrification. However, differences in geographic regions and management practices may also contribute to the observed discrepancies, although our results remain within the same order of magnitude. Reporting field-based experimental denitrification rates remains critical to constrain global nitrogen budgets, as denitrification is often the least quantified component—frequently missing or underestimated in global models (Bai et al., 2013).

We derived denitrification product ratios of 5.81 and 7.05 for the herbal ley and the arable field respectively (Table 1). These ratios are also in good agreement with the literature. In a meta-analysis on terrestrial denitrification N fluxes, Scheer et al. (2020) reported a mean denitrification product ratio of 8% (6%–11%); and more specifically a ratio of (12.40 ± 3.1) % for soils under natural vegetation and (10.90 ± 2.0) % for agricultural soils. More recently, Kleineidam et al. (2025) compiled 18 studies and found mean product ratios of 12% for grassland and 16% for croplands (and medians of 4% and 8% respectively). Eckei et al. (2024), who also followed temporal dynamics of denitrification (although not in winter) found 12% in the German winter wheat crop amended with 210 kg N ha^-1 of fertilizer. Our results are thus in the lower range of the reported values. This could potentially be explained by the slightly alkaline pH of the FarmED soil (Supplementary Table S3), which tends to favour the reduction of N₂O to N₂, thereby decreasing the denitrification product ratio (ŠImek and Cooper, 2002). Furthermore, if we look at monthly reported values (Figure 1), product ratios around 10% are frequently encountered (especially between June and August). If most of the studies analysed by Kleineidam et al. (2025) and Scheer et al. (2020) measured this ratio in one-time experiments—as was the case in 74% of studies according to Almaraz et al. (2020)—this could explain why our annual mean ratios are slightly lower than the reported values.