Sampling was conducted in 72 distinct headwater streams located in Denmark, selected to obtain a wide range in pH, soil type, and land cover (Figs. 1 and 2). Many of these streams have been deepened and channelized, especially those located in agricultural landscapes. Due to regional geological variation, stream pH exhibits a pronounced east-west gradient (Thodsen et al., 2024). Eastern Denmark, including Zealand, Funen, and the easternmost part of Jutland, is underlain by calcareous glacial tills deposited during the Weichselian glaciation (Madsen, 1987). These sediments, originating from an ice advance from the Baltic Sea, are rich in chalk, which imparts a higher buffering capacity and contributes to elevated pH levels in surface waters (Fig. 1). In contrast, western Jylland is dominated by older, more leached glacial and meltwater deposits with lower carbonate content, resulting in reduced buffering capacity and generally lower stream pH. The catchments associated with each sampling location were delineated using a national digital elevation model (DEM) with a 10 m spatial resolution, obtained from the Danish Agency for Data Supply and Infrastructure. Delineation was carried out in ArcGIS Pro (version 3.5; ESRI Inc., Redlands, CA, USA) using the Fill, Flow Direction, and Watershed tools. Stream slope was quantified as the change in water-surface elevation over a 100 m reach, measured with a Leica GPS at the sampling site and a point 100 m upstream; where canopy impaired satellite reception, slope was derived from the DEM. Soil pH and soil type (JB number) for each catchment were extracted for five depths (0–15, 15–30, 30–60, 60–100, and 100–200 cm depth) from Adhikari (2013). Land use within each catchment was quantified by calculating the percentage cover of major land-use categories using the polygon-based national land-use map (scale 1:25000) from the Danish Area Information System (Nielsen et al., 2000). Daily precipitation data were obtained from the national 10 km grid, whereas air pressure data were from the corresponding 20 km grid covering the sampling sites, both provided by the Danish Meteorological Institute.

Water samples were taken from all streams seasonally during summer (19–26 August 2022), autumn (3–14 November 2022), winter (17–30 January 2023), and spring (19 April–4 May 2023) (Fig. 2). Of these, ten streams (Stream 1–10) were sampled monthly from April 2022 to April 2023. During the summer sampling campaign, 12 of the 72 streams were dry, with three remaining dry until the winter sampling period. Water samples were collected 20–30 cm from the stream bank using a 250 mL polypropylene bottle and stored in a transportable cooler. At each site, stream water velocity, width, and depth were measured. Stream water velocity and depth were measured across the stream with three points (right, middle and left). Stream velocity was measured using a vane wheel anemometer, positioned 0.4× depth from the streambed (Höntzsch GmbH & Co. KG, Germany). The discharge was calculated from average stream dimensions and average velocity. In-stream vegetation cover was visually estimated and recorded as a percentage, while the presence of iron ochre deposits in the stream was assessed on a scale from 0 to 3 (low to high abundance). To maintain consistency, the same observer conducted all assessments along a 2 m stream reach at each sampling location. Stream water temperature, dissolved oxygen, electrical conductivity, and pH were measured in situ using a multi-parameter probe (YSI Professional Plus, Xylem Analytics, USA). Samples for dissolved N₂O and carbon dioxide (CO₂) analysis were collected by drawing 50 mL of stream water into a syringe, followed by the addition of 10 mL of ambient air. The syringe was sealed using a stopcock and shaken vigorously for 1 min by holding the piston, while keeping the syringe in the shade to minimize changes in temperature. The 10 mL headspace gas was then transferred into a 5.9 mL pre-evacuated glass vial. Samples were stored in the dark until analysis. At each sampling location 10 mL ambient air collected above the water surface was injected into a 5.9 mL vial. Data on precipitation and atmospheric pressure were obtained from the Danish meteorological institute.

The samples for analysis of dissolved organic carbon (DOC), NO3-, NH4+, and soluble reactive phosphorus (SRP) were filtered within 24 h in the laboratory using Whatman GF/C filters (pore size 0.45 µm), which were rinsed with 300 mL demineralized water before use. Both filtered and unfiltered samples were stored cold (3–5 °C) and dark until analysis. The samples for NO3- and sulphate (SO42-) determination were analysed according to Danish standard DS/EN ISO 10304 (2009) by ion chromatography (Dionex ICS-1500 IC-system) with an anion Micro Membrane Suppressor (AMMS III 4 mm) as basic eluent. The system was equipped with a guard column (IonPac AG22) and a separator column (IonPac AS22). The eluent was a mixture of 4.5 mM Na₂CO₃ and 1.4 mM NaHCO₃. All samples for ion chromatography were filtered through a double-layered 0.22 µm glass fibre filter (SNY2225, Frisenette ApS, Knebel, Denmark). Ammonium (NH4+), total P (TP) and SRP were measured colorimetrically on a spectrophotometer (Shimadzu 1700, Shimadzu Corp., Kyoto, Japan) according to the Danish/European standard methods DS/EN ISO 11732 (2005) for NH4+ and DS/EN ISO 6878 (2004) for TP and SRP. Total organic carbon, DOC and TN were both measured on a TOC-L analyser equipped with a TNM-L module (Shimadzu, Kyoto, Japan) at a temperature of 720 °C following DS/EN ISO 1484 (1997) and DS/EN 12260 (2003), respectively.

The concentrations of N₂O and CO₂ in headspace samples and ambient air samples were determined using a dual-inlet Agilent 7890 GC system interfaced with a CTC CombiPal autosampler (Agilent, Denmark) configured and calibrated with standard gases according to Petersen et al. (2012), with detection limits of 0.15 ppm for N₂O. The gas chromatograph was equipped with an electron capture detector (ECD) for N₂O analysis and a thermal conductivity detector (TCD) for CO₂ measurements. The headspace concentration of N₂O and CO₂ (ppm) were converted to moles using the ideal gas law, accounting for the headspace volume, laboratory temperature, and site-specific atmospheric pressure. The concentrations of N₂O and CO₂ remaining in the water phase were calculated from the headspace gas concentrations according to Henry's law and using Henry's constant corrected for water temperature and atmospheric pressure at the sampling time (Weiss and Price, 1980). The total dissolved concentration originally present in the stream water was obtained as the sum of the gas transferred to the headspace and the remaining gas dissolved in the water. The expected headspace contribution from air was based on the ambient air sample, which was subtracted from the measured headspace values to correct for the ambient air introduced during the equilibration step. Stream water was considered undersaturated with respect to N₂O when dissolved concentrations were lower than the measured air concentration. The saturation levels were classified as undersaturated (<95 %), ∼ atmospheric equilibrium (95 %–105 %), and oversaturated (>105 %) following Aho et al. (2023). These threshold intervals were applied to explicitly incorporate the measurement uncertainty associated with the gas concentration determinations and the subsequent saturation calculations. Analytical precision of the GC measurements was quantified to be 2.6 % CV (SD = 0.0075 µg N L⁻¹, n=7, range = 0.3–2.0 µg N L⁻¹).

All statistical analyses were performed using R version 4.4.3 (R Core Team, 2025). The N₂O saturation was log transformed to ensure normality. Linear mixed-effects models (LMM) were used to identify environmental drivers of N₂O saturation using the lmer() function from the lme4 R package (Bates et al., 2015). To account for the hierarchical data structure and repeated measurements within the same stream, we included a random intercept for stream number. Prior to model development, correlations between predictor variables were checked, to ensure that parameters with a correlation coefficient above 0.5 were not included (Fig. A1). In addition, we calculated the variance inflation factor (VIF) to further evaluate potential collinearity among predictors (Table A1). All continuous predictor variables were standardized (z-transformed: centered and scaled to unit variance) to ensure comparability of effect estimates and facilitate interpretation of interaction terms. Model selection followed a backward elimination procedure, where non-significant predictors were sequentially removed. Model assumptions were evaluated through residual diagnostics, including visual inspection of Q-Q plots and residual vs. fitted plots (Zuur et al., 2009). Linear mixed-effects models were also applied to assess temporal and spatial differences in N₂O saturation, while accounting for repeated measures within streams. Pairwise comparisons between regions and seasons were conducted using estimated marginal means via the emmeans R package (Lenth, 2025). P-values were adjusted for multiple testing using the Bonferroni method to control the family-wise error rate (Haynes, 2013). The relationship between log-transformed N₂O saturation and stream pH was modeled using a generalized additive model (GAM) with a smooth pH term fitted by Restricted Maximum Likelihood (REML). Derivatives of the smooth term were used to identify pH ranges where N₂O changed significantly with pH. The first pH at which the derivative became significantly negative was defined as the onset of decline, and the pH with the most negative derivative as the point of steepest decline. Robustness of the onset threshold was evaluated using bootstrap resampling (n=500).

During the study period (April 2022 to March 2023) the mean (±SD) annual precipitation was 741±102 mm and mean air temperature was 8.9±0.4 °C (Table 1). Mean stream water N₂O concentration between all sampling sites was 2.4±2.8 µg N L⁻¹ ranging from 0.1 to 22.2 µg N L⁻¹. In the ten streams monitored monthly, N₂O saturation showed substantial variability both temporally and between streams. The stream with the lowest water pH (Stream 6, mean pH =5.8±0.2) generally exhibited higher N₂O saturation (Figs. 3, A2). For most streams N₂O saturation peaked during late winter and early spring coinciding with elevated NO3- concentrations (Figs. 3, A3). In contrast, NH4+ levels remained low and relatively stable throughout the study period (Fig. A4). Consistent with patterns observed for streams with monthly monitoring, mean N₂O saturation measurements across all distinct stream systems were significantly higher during winter, compared to summer (Tables A2 and A3). Summer N₂O saturation was also significantly lower compared to spring and autumn (Table A3). Stream-level differences explained the largest share of variance, with an intra-class correlation coefficient (ICC) of 0.67 according to our LMM (between-stream variance =0.119; residual =0.059). The remaining 33 % of the variance was attributable to within-stream variability. Regional variation accounted for a smaller proportion (ICC = 0.15; variance = 0.030; residual = 0.173), while seasonal differences were negligible (ICC = 0.016; variance = 0.0032; residual = 0.195).

The final model retained NO3-, pH, NH4+, DOC, land use, and stream depth as significant fixed effects (Table 2, Fig. A5). All retained predictors were positively correlated with N₂O concentrations, except pH. A significant negative interaction with pH (Estimate =-0.10, p<0.001) was included, indicating that N₂O increased with NO3- and the effect of NO3- was amplified under acidic conditions (Table 2). The model captured notable variation between stream locations (random intercept variance =0.041±0.020), although a considerable amount of unexplained variation remained (residual variance =0.041±0.020) (Fig. A6). Fixed effects alone explained approximately 47 % of the variation in N₂O (marginal R2), while the full model (including stream-level random effects) accounted for about 73 % (conditional R2) (Table A4). Initially SRP appeared as a significant predictor of N₂O saturation, but this relationship was driven by a single high observation and became non-significant upon its exclusion. Variables such as water temperature, oxygen in stream water, precipitation, sulphate and flow velocity were excluded from the final model due to a lack of statistical significance. To avoid the risk of introducing multicollinearity, CO₂ was excluded due to a strong correlation with pH (Fig. A1). Nevertheless, the effect of the interaction between pH and CO₂ on N₂O saturation was investigated separately to shed more light on the nature of this relationship (Fig. A5). A LMM analysis revealed a significant interaction between pH and CO₂ in relation to N₂O saturation (CO2× pH: Estimate =0.066±0.019, p<0.001). While N₂O saturation decreased with increasing pH overall, the positive relationship between CO₂ and N₂O became stronger at higher pH values. This was further investigated by another LMM showing that CO₂ was not significantly related to N₂O saturation at pH <6 (β=0.05±0.05, p=0.29), whereas at pH >6 N₂O saturation increased strongly with CO₂ (β=0.21±0.02, p<0.001).

The N₂O saturation increased with NO3- concentrations, but the magnitude of this response varied markedly with pH (Fig. 4). The highest N₂O saturations were observed at low pH, particularly below pH 6, where several observations exceeded 2000 % saturation even at moderate NO3- levels. Derivative analysis of the GAM indicated that N₂O saturation began to decline at pH ≈ 5.95, with the steepest decrease occurring at pH ≈ 6.37, defining a transition zone over which N₂O saturation dropped most sharply (Fig. A7). Bootstrap resampling confirmed the robustness of the start-of-decline estimate, with a median of 5.96 and a 95 % confidence interval of 5.83–6.19. The N₂O saturation was significantly higher for streams with pH < 6 (mean: 1275±927 %) compared to streams with pH > 6 (mean: 485±576 %) according to LMM analysis taking stream number into account (β=0.4±0.1, p<0.001).

Streams acted as net N₂O sinks in approximately 9 % of observations (n=35), as indicated by water column undersaturation relative to atmospheric equilibrium. Mean N₂O undersaturation was 69 %, with a range of 27 % to 94 %. Nitrous oxide undersaturation was observed primarily in streams with low NO3- concentrations, and there was no seasonal tendency. Prolonged periods of undersaturation were observed in three semi-natural streams whose catchments were characterized by a low proportion of agricultural land use. One stream (Stream 2) was undersaturated during all sampling campaigns, whereas Streams 3 and 4 were undersaturated in approximately 75 % of the campaigns. In addition, a single occurrence of undersaturation was observed in a stream receiving water from an agriculturally dominated catchment (stream 25).

Our observed N₂O concentrations (mean 2.4±2.8 µg N L⁻¹ or 608±722 %) were somewhat higher than levels from past studies of stream and rivers: median 1.0 µg N L⁻¹ in Sweden (Audet et al., 2020), mean 0.8 µg N L⁻¹ or 236 % in Michigan in United States (Beaulieu et al., 2008), 114±20 % in New Zealand (Clough et al., 2011) and 184±1081 % in United States (Aho et al., 2023). Stream water N₂O concentrations are the result of complex interactions among potential transport pathways (e.g. in-stream production, external input), within-stream physical factors (e.g. turbulence, slope) and biogeochemical conditions (e.g. pH, redox conditions). Here, we nonetheless assessed whether statistically robust relationships exist between N₂O saturation and selected in-stream environmental variables associated with N₂O production and consumption. Our finding that NO3- was a significant predictor of N₂O saturation is consistent with earlier studies (Baulch et al., 2012; Clough et al., 2011; Beaulieu et al., 2009). These findings may imply that heterotrophic denitrification contributes substantially to N₂O dynamics, whether through enhanced process rates or altered product ratios under elevated NO3- levels, though further evidence is needed to confirm this. Metagenomic analysis of the sediment from two agricultural streams from our stream network (Peterse et al., 2026) confirmed the presence of denitrification genes (nirS, norB, nosZ). Our results further indicate that the effect of NO3- on N₂O levels is stronger under acidic conditions. This implies that low pH in streams may promote elevated N₂O concentrations even under low N availability, consistent with Audet et al. (2020), who observed similar N₂O concentrations in forested (pH < 6) and agricultural streams despite much higher N concentrations in the latter.

Hydrological characteristics such as water residence time, slope, stream depth, flow velocity, and physical structure also influence N₂O concentration by affecting biogeochemical processes and gas exchange (Marzadri et al., 2021; Tonina et al., 2021; Mulholland et al., 2008; Mwanake et al., 2023). Our results show that deeper streams tend to exhibit higher N₂O saturation (Fig. A7). This pattern may be partially explained by the negative association between stream depth and lower pH levels. Physical constraints on gas exchange may also contribute, as deeper water bodies typically have lower surface area-to-volume ratios, potentially limiting the evasion of N₂O to the atmosphere, leading to higher N₂O concentrations in the water column. Several potentially important controls on N₂O concentrations were not accounted for in our study design, particularly those related to hydrological flow paths. Groundwater, often oversaturated with N₂O (Jurado et al., 2017; von der Heide et al., 2008), can significantly influence stream concentrations, particularly in headwater systems where it constitutes a major portion of baseflow (Bisson et al., 2025). Drain systems can deliver concentrated pulses of NO3- and N₂O directly into streams. Reay et al. (2003) reported N₂O concentrations at drain outlets up to five times higher than those measured 100 m downstream, suggesting rapid degassing upon discharge. Our study showed that seasonal patterns were weak, with slightly lower N₂O saturation in summer. This may reflect reduced substrate availability, enhanced N assimilation, and/or increased denitrification to N₂ under warmer conditions. Overall, our findings align with previous studies reporting inconsistent seasonality (Audet et al., 2020; Beaulieu et al., 2008; Hama-Aziz et al., 2017; Aho et al., 2022; Cole and Caraco, 2001; Rosamond et al., 2012), reinforcing the importance of local over temporal drivers.

In our study, N₂O saturation was negatively correlated with pH, and this effect was most pronounced at pH values below 6. Whether this relationship reflects a direct influence of pH or an indirect one mediated by other environmental variables remains unresolved, and it will most likely depend on the dominant N₂O production pathway. Only a small number of studies have investigated N₂O concentrations in streams with low pH (Audet et al., 2020). In contrast, the effect of pH on N₂O is well documented in soils, where numerous studies have demonstrated that low pH can impair N₂O reduction and shift end-product ratios toward N₂O rather than N₂ (Šimek and Cooper, 2002; Firestone et al., 1980; Liu et al., 2014; Qiu et al., 2024). This impairment is linked to reduced functionality of nosZ, the enzyme responsible for reducing N₂O to N₂ (Firestone et al., 1980). Metagenomic data from the sediment from two of our study streams (Peterse et al., 2026) revealed that nosZ was only slightly less abundant in an acidic stream compared with a neutral stream. However, previous studies have demonstrated that gene presence alone is a poor indicator of N₂O reduction capacity under low pH conditions, as acidic environments primarily impair nosZ enzyme functionality rather than gene abundance (Liu et al., 2014; Olaya-Abril et al., 2021). The limited number of studies in streams means that the generality of the pH–N₂O relationship across diverse landscapes is still poorly understood. One reason this effect has not been widely reported previously may be that earlier studies examined streams with relatively narrow and high pH ranges such as 7.2–9.0 (Clough et al., 2011), or 7.55–8.43 (Baulch et al., 2012). In other cases, pH was measured, but results were not shown (Beaulieu et al., 2009; Hinshaw and Dahlgren, 2013) or pH was not measured (Baulch et al., 2011a; Stow et al., 2005). More research is needed to determine whether similar pH–N₂O interactions operate under different climatic, hydrological, and biogeochemical conditions.

A range of environmental conditions and interactions influenced by pH may shape patterns of N₂O dynamics. In particular, dissolved CO₂ was strongly correlated with pH and showed a positive association with N₂O saturation overall, however this association strengthened progressively at higher pH, but was not significant at pH < 6. Overall, these results indicate that observed CO₂–N₂O relationships vary systematically across the pH gradient. Spatially, lower pH in streams draining western Jutland is consistent with sandy, low-alkalinity catchments with limited buffering capacity. Under such conditions, shifts in inorganic carbon speciation favor dissolved CO₂ (Wetzel, 2001). In addition, low buffering capacity further increases chemical pH sensitivity to dissolved CO₂. This strong physicochemical coupling complicates interpretation of CO₂–N₂O relationships in acidic systems. At higher, circumneutral pH, the significant CO₂–N₂O association may be explained by dissolved CO₂ acting as an integrated indicator of ecosystem metabolism and/or a direct control on N₂O production. Elevated CO₂ concentrations under circumneutral conditions may reflect increased respiratory activity and organic matter mineralization, processes that are often associated with enhanced nitrogen turnover and greater substrate availability for nitrification and denitrification (Hotchkiss et al., 2015; Mulholland et al., 2008). Increased respiratory oxygen demand could also promote localized low-oxygen conditions in sediments and biofilms, which are conducive to N₂O production. However, as oxygen availability was not measured directly, these interpretations remain inferential. Experimental evidence from recent sediment incubations indicates that elevated CO₂ can coincide with increased ammonia-oxidizing bacterial abundance and enhanced N₂O production, particularly under low DOC : NO3- ratios (Mwanake et al., 2025). While these experimental findings are consistent with the field-based patterns observed here, their applicability to heterogeneous stream environments is uncertain. In conclusion, because CO₂ and pH are inherently linked through carbonate system chemistry, their individual influences on N₂O dynamics cannot be disentangled in this study. As a result, neither the direction of causality nor the relative importance of biological versus physicochemical controls can be resolved. Under acidic conditions, nitrification intermediates such as hydroxylamine (NH₂OH) and nitric oxide (NO) can undergo chemical conversion to N₂O via non-enzymatic pathways (Zhu-Barker et al., 2015). A significant, but weak, positive correlation between NH4+ and N₂O indicates that nitrification, nitrifier denitrification, coupled nitrification-denitrification or nitrification coupled to abiotic processes may also contribute to N₂O production. The NH4+ concentrations in streams were generally low, but it is possible that NH4+ was released through the mineralization of organic matter in sediments, providing a substrate for nitrification (Arango and Tank, 2008). Supporting this, sediment data from ten streams with monthly monitoring showed that NH4+ concentrations were consistently higher in sediments than in the overlying water (Peterse et al., 2026). Microbial community profiling via 16S rRNA gene sequencing revealed nitrifiers in ten streams, with acidophilic Nitrosotaleaceae dominating acidic sites, indicating that pH might alter community structure (Peterse et al., 2026). Another potential N₂O source is chemodenitrification (Wankel et al., 2017), where oxidized N species are reduced by Fe²⁺, which may be important in iron-rich, acidic soils such as western Jutland, but its contribution could not be confirmed because Fe²⁺ and abiotic processes were not measured directly.

Freshwater ecosystems have been recognized as potential sinks for N₂O for some time (Kroeze et al., 2007), yet this role remains largely ignored in both global and national greenhouse gas inventories (Aho et al., 2023; Webb et al., 2019). Kroeze et al. (2007) proposed that inland waters can be N₂O undersaturated when N₂O consumption processes exceed in situ production and atmospheric invasion. In our study, undersaturated was observed for three streams located in areas with low impact from agriculture, and thus low stream water N concentrations. This finding is consistent with previous studies demonstrating that low N availability is a key factor influencing N₂O undersaturation in streams (Kroeze et al., 2007; Baulch et al., 2011b; Aho et al., 2023). Samples collected from streams across the United States, as part of the National Ecological Observatory Network, showed that 30 % of the samples (678 of 2288) were undersaturated. In our study undersaturation was observed across all seasons, with no statistically significant seasonal variation, implying that temporal factors such as temperature or hydrological conditions may play a lesser role than N status. Besides NO3-, N₂O undersaturation has also been linked to low oxygen and high DOC (Baulch et al., 2011b; Borges et al., 2019).