The ATIC system used here was developed based on the automatic chamber system design introduced by Flechard et al. (2005), and the air sampling follows the system introduced by Ambus et al. (2010). The ATIC is operated as a non-steady-state flow-through chamber with a main loop that recirculates the headspace chamber air (Figure 1). The lid of the chamber closes automatically for 15 min. During this period, four headspace gas samples are collected (at 3.50, 7.25, 11.50, and 14.25 min after chamber closure) for 15 s and flushed into four different foil gas bags through a valve manifold. The use of the ATIC system allows flux measurements at relatively high frequency (like for online automatic chamber systems) but reduces the frequency of gas analysis and avoids the use of online trace gas analysis, which lowers the cost and energy consumption in the field. The ATIC runs with a battery (12 V) or power line, and the latter is used in our experiment. It consists of three parts: 1) a stainless steel chamber (L = l = 300 mm, H = 220 mm) with a pump (Thomas, Germany), CO₂ sensor (Senseair, Sweden), flow sensor (McMillan, New York City, NY, United States), lid inclinometer sensor (DIS Sensors, Netherland), and motor connected with the lid of the chamber through pulley and rope (Figure 1A, unit 2); 2) an associated controller system, including the main control module (Siemens, Germany), and the data logger (Onset, MA, United States) for the sensors attached with the chamber (Figure 1A, unit 1); and 3) 4 replaceable foil gas bags (Supelco, Germany; Figure 1A, unit 3). The controlling system opens and closes the chamber lid through the motor. Each chamber was placed on a PVC frame, inserted 5 cm into the soil. Flexible silicone attached below the chamber and the foam sealing above the chamber was used to achieve gas tightness of the chamber.

N₂O fluxes were measured quasi-continuously for two entire years from 28 February 2019 to 02 March 2021. Here, we designate the first sampling year period (from 28 February 2019 to 28 February 2020) as the first year and the second sampling year period (from 28 February 2020 to 02 March 2021) as the second year. In the study site, the ATIC systems (four on Cov, four on Ref) were installed on 13 February 2019 for testing. On 28 February 2019, eight ATIC systems (four on Cov and four on Ref) started to collect gas samples with a frequency (time between measurement cycles) of 3–9 h per individual chamber, differing between growing season and non-growing season. Here, we define a measurement cycle as a lid closing phase of 15 min with sequential gas sampling into each of the four foil gas bags.

The bags were filled at 3.50, 7.25, 11.50, and 14.25 min after chamber closure, respectively, and the gas samples from individual cycles were accumulated in these four foil gas bags. The final bag samples represented an average over a time period of 3–14 days (hereafter referred to as “sampling period”), depending on the sampling frequency and the total number of measurement cycles. Usually, gas samples accumulated in the bags over 30–40 cycles, which was limited by the volume of the foil gas bag (10 L) and the flow rate of the pump (1 L min⁻¹). After the sampling period, the foil gas bags were replaced with empty ones, and the filled gas bags were transferred to the lab to determine their gas concentration using the gas analyzer (G2308, Picarro, Santa Clara, CA, United States). Overall, each ATIC system was working for ∼3,200 cycles during the two entire years. The foil gas bags have been tested and proven to be suitable for long-term storage of N₂O near ambient concentrations in air (Supplementary Figure S2).

The accuracy of the ATIC system was accessed by the CO₂ and inclinometer sensors inside the chamber. The measurement of the real-time headspace chamber CO₂ concentration and the incline angle between the lid and chamber allowed us to detect any operational chamber problem (e.g., leakage and power failure). The CO₂ fluxes during each chamber closure time were evaluated by linear regression, and an overall R ² ≥ 0.9 was taken as an indicator for a fully functional ATIC within the sampling period. With the comparison between the average CO₂ fluxes during each chamber closure time and the CO₂ fluxes from the ATIC system within each sampling period, we determined whether the gas fluxes from the ATIC system represented the average gas fluxes within the sampling period (Supplementary Figure S3). N₂O fluxes calculated from the concentration gradient from the four foil gasbags were selected for post-processing after fulfilling certain quality criteria. First, R ² of CO₂ fluxes calculated from the regression lines of the four bags had to exceed 0.9, indicating that within the sampling days, the ATIC worked properly. R ² ≤ 0.9 indicated a failure of gas sampling within the sampling days, which led to a rejection of the N₂O flux data. Second, R ² > 0.9 of the N₂O regression lines was used as critical to accept the data for further analysis. However, low fluxes (±0.5 mg N m⁻² day⁻¹, calculated based on the detection limit of the Picarro) were accepted regardless of R ². With low fluxes, the random error of the measurement could be larger than the N₂O concentration difference between different sampling points, which could result in low R ², and therefore, rejection of the N₂O emission based on low R ² would lead to an underestimation of the overall fluxes. After two years of continuous field observation, the N₂O data could cover ∼86% of the sampling days, that is, a data gap of ∼14%.

For any missing N₂O emissions outside the fertilization event (background N₂O emissions), that is, values missing owing to a failure of ATIC systems or a rejection of data, a look-up table approach with two parameters (soil moisture and soil temperature) was used to fit the missing values (RMSE = 0.62 mg N m⁻² day⁻¹, R ² = 0.60), and tested by the available background N₂O values through cross-validation. For each chamber, background N₂O emissions were divided into 16 classes based on soil moisture (0–25th percentile, > 25th percentile–median, > median–75th percentile, and > 75th percentile), and soil temperature (0–25th percentile, > 25th percentile–median, > median–75th percentile, and > 75th percentile). With the assumption that without extra fertilizer input background N₂O should respond similar to similar soil temperature and moisture conditions at each site, the mean N₂O fluxes from each class were used to fit the missing value under the same soil temperature and moisture condition. To check the sensitivity of the N₂O gap-filling method for the background fluxes, two other methods were compared with the look-up table approach: 1) linear interpolation (RMSE = 0.65 mg N m⁻² day⁻¹, R ² = 0.51) to bridge the missing values and 2) taking mean values from the properly operating chambers for each site (RMSE = 1.3 mg N m⁻² day⁻¹, R ² = 0.36). For an N₂O gap caused by a power failure during the fertilization event on 30 August 2019 (site Ref), data were linearly interpolated to fill the data gap. For the failure of individual chambers (n = 6) during fertilization events, mean values from the properly operating chambers at each site were used to fill the data gap.

The air temperature was measured using a Vaisala Weather Transmitter (WXT520, Finland) and continuously logged every 10 min on a CR1000 data logger (Campbell Scientific, United Kingdom). Rainfall data and missing air temperature (27 December 2020 to 2 February 2021) were filled with data from a nearby meteorological station operated by MeteoSwiss (https://www.meteoswiss.admin.ch). Soil temperature and soil moisture (GS3 and 5 TE decagon devices, NE Hopkins Court, United States) were continuously recorded half-hourly at a depth of 5 cm for both Cov (n = 3) and Ref (n = 3). On 4 December 2019, 24 additional soil temperature sensors (UA-001-64 devices, Onset, MA, United States) were installed near each chamber at three depths (0, 2.5, and 5 cm) for recording surface soil temperature in winter. These sensors were taken out on 21 April 2020 for reading out the data, and the same process was followed for winter 2020/2021. Close to the chambers, the soil volumetric water contents at −5 cm depth were consistently recorded every 10 min with soil moisture sensor (EC-5, decagon devices, NE Hopkins Court, United States). Missing soil temperature data—due to a failure of a data logger between November to December 2019 and April to May 2020 (site Cov)—were fitted using linear regression between temperatures of the two sites with a similar temperature range (RMSE = 0.07°C, R ² = 0.98).

To determine the soil pore volume, on 12 April 2019, 72 undisturbed cylindrical soil samples (100 cm⁻³) were collected at three depths (3–8, 18–23, and 58–63 cm) with 12 replications at each site and transferred to the laboratory. In the laboratory, different pore diameters were measured following the approach by Keller et al. (2019). For this, samples were saturated from below and then drained to soil matric potentials of −30, −60, −100, −300, and −1,500 kPa.

Air porosity was calculated based on the difference between volumetric water content (VWC) and total pore volume. The relative gas diffusion coefficient (D_p/D₀) was calculated based on air-filled porosity by following the approach by Keller et al. (2019). A D_p/D₀ of 0.02 has been suggested as the critical threshold for adequate soil aeration, and the D_p/D₀ value lower than 0.02 indicates insufficient soil aeration (Schjønning et al., 2003). The water-filled pore space was determined by the ratio of volumetric water content and total pore volume, and field capacity was determined by water retention at −30 kPa. Here, we used the threshold of 80% of the field capacity for each site to roughly distinguish dry and wet conditions separately for Cov and Ref. Soil moisture of below 80% of the field capacity was designated as dry, and >80% of the field capacity as wet.

For soil organic carbon (C) and total nitrogen (N) content measurement, 22 soil samples were taken in 2018, with 11 replications at each site. Soil samples were dried at 105°C for 72 h, ground using mortar and pestle, and then milled in a ball mill (Retsch, MM 400, Germany) at 25-rotation s⁻¹ for 3 min. Samples containing carbonate (soil surface from Cov) were fumigated with hydrochloric acid overnight in a desiccator before being analyzed by elemental analysis (Hekatech, Germany). For soil pH, soil (unground) was suspended 10 g in 0.01 M calcium chloride (CaCl₂), shaken at 160 cycles min⁻¹ for 15 min, and left overnight before the soil pH was measured with a flat surface electrode (pH3310, WTW, Germany). For ammonium and nitrate measurements, 16 soil samples were taken in July 2021, with eight replications at each site. Soil N was extracted from 20 g field-moist soil with 0.01 M CaCl₂ solution and filtered. The filtrate was analyzed by segmented flow injection analysis (Skalar Analytical B.V., Breda, Netherlands). The C and N content in the slurry was determined in a central laboratory (Labor für Boden und Umweltanalytik, Eric Schweizer AG, Thun, Switzerland. For details about N application rate and frequency, see Table 2).

At both sites, high N₂O emission peaks were primarily triggered by fertilization events (F-peak) and lasted for 2–3 weeks before returning to background N₂O emissions (Figure 3). There were eight fertilization events during the experimental period, but we only observed six F-peaks during summer and autumn when the soil temperature was high (Table 2). To further explore the influence of N input on N₂O emissions, we defined a corresponding fraction of N loss as the ratio of the N₂O emissions during each F-peak and the corresponding fertilizer N input. We found the fraction of N loss to be higher at Ref than at Cov (p < 0.01) for each of the six individual F-peaks. In the first year, F-peaks contributed ∼78% to the annual N₂O emissions at Ref, corresponding to 25.56 ± 2.15 kg N ha⁻¹. At Cov, F-peaks contributed ∼64% to the annual N₂O emissions, corresponding to 2.02 ± 0.32 kg N ha⁻¹ (Figure 4A). For the second year, at both sites, F-peak fluxes only contributed ∼43% to the annual N₂O emissions, corresponding to 3.51 ± 0.84 kg N ha⁻¹ for Ref and 0.57 ± 0.11 kg N ha⁻¹ for Cov (Figure 4B). It needs to be noted that during the fertilizer event in August 2019, we observed a relatively high peak at both sites (Figure 3), which extended over 25 days (8 sampling periods). The N₂O emissions from this high peak contributed ∼78% to the annual N₂O emissions to Ref and ∼64% to Cov.

In order to identify the main driving factors for N₂O emissions, regression analyses were applied separately on fertilization-induced and background N₂O emissions. For fertilization-induced N₂O emissions, the regression analysis was performed on each individual fertilization-induced F-peak. We found that 50 to 60% of the variation in F-peak N₂O emissions could be explained by the multiple linear regression model with soil temperature, soil water-filled pore space, and amount of N input as explanatory variables (Table 3). At both sites, the variability of F-peak emission was mainly driven by soil temperature and N inputs (p < 0.05; p < 0.01), and the two parameters were positive drivers for F-peak N₂O emissions. In addition, the average WFPS during the fertilization events also contributed significantly to the variability of the F-peak N₂O emissions at both sites. For background N₂O emissions, the variation in N₂O emissions explained by the MLR model was low with R ² of 0.10 (Cov) and 0.16 (Ref). The impact of the two potentially driving parameters soil temperature and soil WFPS differed between Cov and Ref. Overall, soil temperature was a significant driver for background N₂O emissions at both sites (p < 0.01), but the effect of soil WFPS on background N₂O emissions was not significant (Table 3). A threshold of 80% of the field capacity was used to define dry and wet conditions in the field. At Ref, soil WFPS was a positive driver for background N₂O emissions during dry conditions in the field, whereas it was negative under wet conditions (p < 0.01). At Cov, soil WFPS only exerted limited influence on background fluxes.

In our study, soil temperature, soil water-filled pore space, and N input could explain more than half of the variance of fertilization-induced N₂O emissions (F-peak; Table 3). Higher F-peaks were found with warm temperatures and lower with cold temperatures (Table 2), most likely due to the reduced soil microbial activity (Holtan-Hartwig and Bakken, 2002). The fertilizer N inputs and the high WFPS did not compensate for the effect of cold temperatures (Table 2), indicating that in our field, high N₂O peaks only occur if all the driving variables (soil temperature, soil water content, and N availability) are supporting high N₂O production. Previous research also highlighted that peak N₂O emissions were not observed if one driving factor was below thresholds for soil temperature and moisture (Holtan-Hartwig and Bakken, 2002; Meng et al., 2005). In turn, if these driving factors were above the threshold, a fertilization event will lead to very high N₂O emissions. In the first year, we observed a very high F-peak after fertilization in August, which contributed more than half to the annual N₂O emissions for both sites and even 78% for Ref. This led to a significantly higher annual N₂O release in the first year than in the second year. One explanation might be the time line of the dry summer period followed by a wetting event together with the fertilizer application (Figure 3). This interpretation is supported by a large body of former research works that after dry and wet cycles for both mineral soils and organic soils, a greater amount of N₂O is emitted from grassland owing to the enhanced availability of C and N as related to soil organic matter mineralization (Priemé and Christensen, 2001; Beare et al., 2009; Harrison-Kirk et al., 2013). The differences in fertilization-induced N₂O emissions under different soil temperatures and soil WFPS suggest that as one N₂O mitigation option for the study site, fertilizer application should be avoided in hot summer periods or during frequent precipitation.

Despite receiving the same amount of N input and having similar soil temperatures and WFPS, as well as the same agricultural management, Ref had much higher N₂O emissions after fertilization. The different N₂O releases after fertilization might be related to various mechanisms. First, exogenous N inputs might prime the mineralization of SOM, thereby influencing N₂O production differently in Cov and Ref. Priming effects are defined as short-term changes of SOM mineralization in response to external stimuli, for example, exogenous N addition, and they alter the subsequent N₂O production from SOM-N (Kuzyakov et al., 2000; Daly and Hernandez-Ramirez, 2020; Thilakarathna and Hernandez-Ramirez, 2021). Priming may affect N₂O emissions positively or negatively, depending on soil moisture and SOM content (Roman-Perez and Hernandez-Ramirez, 2020). Former studies revealed positive N₂O priming effects to be associated with wetter soil conditions (WFPS >60%) and higher SOM content (Schleusner et al., 2018; Thilakarathna and Hernandez-Ramirez, 2021). In our study site, soil moisture for Cov and Ref were similar during the fertilization events (Table 2), but SOC content in surface soil of Ref was five times higher than that in surface soil of Cov (Table 1), which might have induced stronger priming by adding N, subsequently leading also to higher N₂O emissions for Ref (Perveen et al., 2019). For the background N₂O emissions, soil temperature was still a significant driver at both sites, but the influence of the soil water-filled pore space was limited (Table 3). Moreover, the variance of background N₂O emissions explained by the MLR model with soil temperature and soil water content was low, indicating that for both sites, the influence of soil temperature and soil moisture on background N₂O emission was limited. In our study, background N₂O emissions were also significantly reduced (p < 0.05) with mineral soil coverage, indicating that the higher N₂O emission from Ref was not only directly related to fertilization but also to the properties of the surface soil itself, for example, soil pH and soil N availability. Surface soil properties are of particular relevance for the amount of N₂O release as it has been shown that the soil depth from which emitted N₂O originates is only 0.7–2.8 cm (Neftel et al., 2000).

Second, surface soil pH increased from 5.2 to 7.3 (Table 1) with mineral soil coverage in our study site. It has been reported that the net production of N₂O from denitrification is strongly dependent on soil pH (Nadeem et al., 2020); N₂O emissions are negatively correlated with soil pH in organic soil, due to the decreased N mineralization with increasing soil pH under aerobic conditions of peat (Chapin et al., 2003; Weslien et al., 2009) and the possible enhancement of the synthesis of functional N₂O reductase from denitrification (Liu et al., 2014), resulting in a higher share of N₂. Hence, the relatively high soil pH at Cov may have contributed to the lower background N₂O emissions.

Third, the two topsoils differed in soil N content and soil N availability (Table 1). Soil available N from SOM mineralization is considered to be the main source for background N₂O production (Lampe et al., 2006). Our study site revealed a carbon loss of 3,100–6,300 kg C ha⁻¹ yr⁻¹ from peat oxidation after drainage based on a former study using a radiocarbon approach to estimate the soil carbon loss (Wang et al., 2021). Considering the C to N ratios of ∼25 (at a depth of 2 m) as representative for nitrogen stored in peat without fertilization, the study site has an N mineralization potential of 120–250 kg N ha⁻¹ yr⁻¹. This relatively high soil N supply at Ref, as also indicated by higher available soil N (Table 1), becomes available for microbial processing and consequently N₂O production. After mineral soil coverage with a thickness of ∼40 cm, N release from peat mineralization of topsoil SOM as an N source for N₂O formation is no longer available. Instead, mineralization of SOM from the mineral soil coverage, whose soil N content and the corresponding soil available N are much lower (Table 1), becomes a major N source. The decreased surface soil N availability due to mineral soil coverage for nitrification and denitrification may then restrain the N₂O production (Senbayram et al., 2012). As one consequence for management, the high soil N supply in drained peatland suggests that the fertilization demand for organic grassland soils might be substantially lower than that for mineral grassland soils.

Fourth, although the organic soil at site Cov still has the potential to produce N₂O, the mineral soil coverage might have pushed the organic soil underneath into a deeper zone with higher soil moisture and lower oxygen availability as compared to the organic topsoil at site Ref. Higher soil moisture could influence the exchange of N₂O between the site of production and the aerated pore space, thereby affecting the balance between N₂O production and consumption. With high soil moisture, a reduction of N₂O emission is expected owing to the higher consumption of N₂O when gas diffusion is slow (Kuang et al., 2019; Harris et al., 2021). Moreover, under strongly anaerobic conditions, N₂ as the end product of denitrification will be produced preferentially (Davidson et al., 2000). Thus, the amount of formed N₂O from the organic soil underneath might be reduced at site Cov via enhanced dissolution in soil water (Clough et al., 2006; Goldberg et al., 2008) or full denitrification before escaping into the atmosphere (Davidson et al., 2000). These effects are further pronounced by the lower gas diffusivity of the surface soil of Cov compared to Ref (Figure 2D). We suppose these effects in combination lead to lower N₂O emissions after mineral soil coverage.