This study was conducted in southeastern Spain (Fig. S1) in two monomictic reservoirs with contrasting morphometries. Cubillas (37.27° N, 3.68° W) is a small and shallow reservoir with a surface area of 1.94 km² and a total capacity of 19 hm³ (mean depth = 9.66 m). Iznájar (37.26° N, 4.33° W) is a big and deep reservoir with a surface area of 26 km² and a total capacity of 981 hm³ (mean depth = 37.55 m) (open database IDEAndalucia; https://www.ideandalucia.es/portal/, last access: 13 March 2024). Both reservoirs are impacted by large agricultural and urban areas in their watersheds, which results in large inputs of N and phosphorus (León-Palmero et al., 2020b, 2023). More information about the watersheds, morphometry, and water column characterization is provided in previous studies (e.g., León-Palmero et al., 2020a, b).

We sampled the water column of these reservoirs at the beginning (4 and 9 July) and the end (5 and 7 September) of the summer stratification in 2018. During the study period, intense human usage caused a decline in the volume and water level in both reservoirs, although this decline was more evident in the smaller reservoir (i.e., Cubillas). Cubillas reservoir decreased in volume from 17 hm³ in July to 11 hm³ in September and experienced a 3.4 m reduction in the water level. The hydraulic residence time during the study period was 83 d. Iznájar reservoir decreased in volume from 575 hm³ in July to 480 hm³ in September, with a 5.4 m reduction in the water level. The hydraulic residence time was 255 d during this period. The reservoir volumes and water levels on specific dates were obtained from the Confederación Hidrográfica del Guadalquivir open database (CHG; https://www.chguadalquivir.es/saih/, last access: 13 March 2024).

We sampled the water column near the dam, in the open water of the reservoir, at the same location during both the July and September campaigns. First, we conducted a vertical profile of the water column using a Sea-Bird 19plus CTD profiler, obtaining continuous measurements of temperature (°C), dissolved oxygen (DO, µmolL-1), and conductivity (µScm-1) in the reservoirs. Based on the temperature and DO profiles, we sampled three depths representing the epilimnion, oxycline, and hypolimnion or bottom waters. Water was collected at these three depths using a 5 L UWITEC bottle for further analyses and incubation experiments.

Samples for dissolved N2O analysis were taken in 250 mL air-tight Winkler bottles in duplicate, preserved with a solution of HgCl₂ (final concentration 1 mmolL-1) to inhibit biological activity, and sealed with Apiezon^® grease to prevent gas exchange. Samples were stored in the dark at a controlled temperature (25 °C) for less than six months until analysis at the University of Cádiz. Dissolved N2O concentration was measured using headspace equilibration in a 50 mL air-tight glass syringe in triplicate in each bottle from each sample. N2O concentration was quantified using a daily calibrated gas chromatograph (Bruker^® GC-450) as detailed in a previous study (León-Palmero et al., 2023).

Water samples for chemical and biological analysis were maintained at 4 °C until arrival at the laboratory. Particulate material from 500 to 1000 mL of water was filtered through pre-combusted (450 °C for 3 h) Whatman GF/F glass-fiber filters with a nominal pore size of 0.7 µm. Chlorophyll a (Chl a) was extracted from the filtered material and measured following the standard method (APHA, 1992). To obtain the cumulative Chl a (a proxy for fresh organic matter exported to the water column) in the whole water column (mg Chl a m⁻²) from the discrete depths, we summed the concentration of Chl a of each stratum using the trapezoidal rule (León-Palmero et al., 2020a). Dissolved organic carbon (DOC), NO3-, NO2-, and NH4+ were assayed in the filtered water. Samples for DOC determination were acidified with phosphoric acid (final pH <2) and measured by high–temperature catalytic oxidation using a Shimadzu total organic carbon analyzer (Model TOC-V CSH) (Álvarez-Salgado and Miller, 1998). NO3- concentration was assayed using the UV spectrophotometric method at the wavelength of 220 nm and correcting for DOC absorbance at 275 nm (APHA, 1992). NO2- and NH4+ concentrations were measured by Inductively Coupled Plasma Optical Emission Spectrometry at the Centro de Instrumentración Científica of the Universidad de Granada.

Two 60 mL glass serum bottles per depth were collected after overflow without headspace and poisoned with HgCl₂ to analyze the natural isotopic composition (δ15N) of the ambient pools of N2O, NO2-, and NO3-. Samples were maintained in darkness at room temperature for under six months before shipment to Princeton University for analysis. A 3 mL headspace was created with He before measuring the N2O, including standards with a known amount of N2O gas and internal standards for 15N-N2O. The total N2O in each bottle was extracted by purging with helium for 35 min at 38 mL min⁻¹. Then, N2O was trapped by liquid nitrogen and isolated from interference by gas chromatography (Frey et al., 2020; Ji et al., 2015). We detected the nitrogen masses 44 (i.e., ⁴⁴N2O representing ¹⁴N¹⁴N¹⁶O), 45 (i.e., ⁴⁵N2O representing ¹⁴N15N¹⁶O or ¹⁵N¹⁴N¹⁶O), and 46 (i.e., ⁴⁶N2O representing 15N15N¹⁶O), and the isotope ratios 45/44, 46/44 with a GC-IRMS system (Delta V Plus, Thermo). Standards in 20 mL glass vials with a known amount of N2O gas were measured every two to three samples to calibrate for the N2O concentration. The total N2O concentration and 45N2O/44N2O and 46N2O/44N2O ratios were converted to moles of ⁴⁴N2O, ⁴⁵N2O and ⁴⁶N2O. Both δ15N-N2O (‰) vs. Air-N₂ and δ18O-N2O (‰) vs. Vienna Standard Mean Ocean Water (VSMOW) were determined. Isotope measurements were linearity and offset corrected using an internal N2O reference gas with known isotopic composition. The N2O reference had the following isotopic composition: δ15N = -0.65±0.08 ‰ and δ18O = 37.37±0.27 ‰ present in ⁴⁵N2O and ⁴⁶N2O. Ideally, two known N2O reference gases would have been used for correction; however, due to this limitation, natural abundance isotope data were used to analyze trends in the sample dataset, rather than making comparison with previous studies.

The natural isotopic composition of the NO2-, and NO3- pools (i.e., δ15N-NO2- and δ15N-NO3-) were determined by converting these compounds to N2O and analyzing the isotopic composition of the resulting N2O. NO2- was converted to N2O by using the azide method (McIlvin and Altabet, 2005). Sample size was adjusted to contain 10 nmol of NO2-, transferred into 20 mL glass vials, and purged with He for 10 min. The NO2- was then converted to N2O using sodium azide in acetic acid. During this reaction, one N from azide is transferred into the N2O molecule; hence the resulting values were corrected by multiplying by 0.5. The 15N-N2O generated was measured on a Delta V Plus (Thermo) as described above.

We used the denitrifier method to convert NO3- to N2O (Granger and Sigman, 2009; Sigman et al., 2001; Weigand et al., 2016). The method is based on the isotopic analysis of the N2O generated from the NO3- by denitrifying bacteria that lack N2O-reductase activity (i.e., Pseudomonas chlororaphis). Sample size was adjusted to 20 nmol nitrate NO3-. The 15N-N2O generated was measured on a Delta V Plus (Thermo) as described above. We included known NO3- isotope international standards (USGS34 and IAEA N3) and converted them to N2O using the denitrifier method to correct δ15N-N2O values.

The abundance of unique functional genes involved in N2O cycling was quantified using quantitative PCR (qPCR), similarly to a previous study (León-Palmero et al., 2023). We pre-filtered water samples through 3 µm pore size filters, and concentrated the samples by centrifugation, then extracted DNA following Boström et al. (2004), and applied PCR and qPCR to assess presence, and abundance of target genes. We used standard reaction mix recipes, thermocycling conditions, and primer requirements specified by the manufacturer. Specific primers were selected from studies performed in natural freshwater samples when available. DNA from pure cultures was used as positive controls and for qPCR standard preparation. We targeted ammonia oxidizers using the archaeal amoA gene, as AOA dominated over AOB in these reservoirs (León-Palmero et al., 2023). Comammox amoA genes were targeted in PCR assays using degenerate PCR primers for clades A and B (Pjevac et al., 2017), but no positive control was available in this case. The nirS gene abundance was used as a proxy for denitrifiers, while nosZ gene (Clade I) abundance was assessed only at the deepest layer, assayed only bacteria reducing N2O to N₂. More details on the DNA extraction method, qPCR quantification, primers, specific conditions, standards, and positive controls are provided in the Supplement (Sect. S1).

Reservoir water from the three depths was drawn from the sampling bottle into 60 mL glass serum bottles after overflow. Once in the lab, samples from oxic water depths (refer to Table 1) were purged uncapped for 2 min to remove excess N2O, and a 3 mL headspace with ambient air was maintained after being exposed to ambient air for 30 min. Samples from anoxic waters were sealed with butyl rubber septa and crimped with aluminum seals immediately after filling. In these samples, a 3 mL helium headspace was retained after purging for 4 min. The serum bottles were weighed before and after filling them to account for the exact water volume in each sample. Table 1 compiles the incubation setup, conditions, and concentration of inorganic nitrogen added in each treatment. In the first treatment, we injected nine bottles from the same depth with 15N-NH4+ tracer (15NH₄Cl ≥ 98 atom % 15N, Sigma Aldrich) to a final concentration of 0.5 µmolL-1, obtaining a fraction labeled of the substrate pools between 0.1 and 1.0. In this treatment, we also added ¹⁴N-NO3-, equivalent to 0.10 of the NO3- pool. In the second treatment, 15N-NO3- tracer (K15NO₃, 98 atom % 15N, Sigma Aldrich) was injected to obtain a fraction labeled of the NO3- pool about 0.10. We also added ¹⁴N-NH4+ to a final concentration of 0.5 µmolL-1. Samples were incubated in the dark at the in situ temperatures from 13 to 26 °C (Table 1).

The first treatment (15N-NH4++14N-NO3-) was performed at all the depths (n=12), but the second treatment (15N-NO3- + 14N-NH4+) was performed only at the oxycline and hypolimnion (n=7, Table 1). Incubations were terminated by adding 0.1 mL saturated mercuric chloride (HgCl₂) to two bottles at t0 (≈0.25 h), two at t1 (≈1–3 h), two at t2 (≈12 h), and three at t3 (≈24 h). All samples were stored at room temperature in the dark for less than six months and shipped to the laboratory at Princeton University for further analysis.

The total N2O in each incubation bottle was extracted by purging with helium and measured with a GC-IRMS system (Delta V Plus, Thermo) as explained above. We included standards in 20 mL glass vials with a known amount of N2O gas every two to three samples to calibrate for the N2O concentration. The total N2O concentration and 45N2O/44N2O and 46N2O/44N2O ratios were converted to moles of ⁴⁴N2O, ⁴⁵N2O and ⁴⁶N2O. N2O production rates for each treatment were calculated from the slope of the increase in mass 45 and 46 during the linear phase over the four timepoints. The N2O production (R15N2O, nmolNL-1d-1) was calculated according to the following Eq. (1) (Santoro et al., 2020): 1R15N2O=FN-1Δ45N2OΔt+2Δ46N2OΔt×FN-1 where Δ45N2O and Δ46N2O represent the variation in the concentration of ⁴⁵N2O and ⁴⁶N2O over the incubation time (Δt), and the FN represents the fraction of 15N in the initial substrate pool (NH4+ or NO3-), which is assumed to be constant over the incubation time. The equation includes an extra factor of (FN)-1 to account for the probability of ⁴⁶N2O production, which is proportional to (FN)-2. Natural abundance 1000 ppm N2O carrier gas (50 µL in He) was injected before measurement to trap the produced labeled N2O and to ensure a sufficient mass for isotope analysis.

After N2O analysis, we analyzed the samples incubated with 15NH4+ and 15NO3- for 15NO2- production to determine the rates of NH4+ oxidation to NO2- (ammonia oxidation), and NO3- reduction to NO2- (first step of denitrification). The NO2- was converted to N2O by using the azide method (McIlvin and Altabet, 2005), and the 15N-N2O generated was measured on a Delta V Plus (Thermo) following the procedure and corrections described earlier. The rates of NH4+ oxidation to NO2- (RNO2-fromNH4+, nmolNL-1d-1) and first step in denitrification (RNO2fromNO3--, nmolNL-1d-1) were calculated following Eqs. (2), (3): 2RNO2-fromNH4+=FNH4+-1Δ15NO2-Δt3RNO2-fromNO3-=FNO3--1Δ15NO2-Δt where Δ[15NO2-] represents the variation in the concentration of 15NO2-, FNH4+ represents the fraction of 15NH4+ in the initial substrate pool, FNO3- represents the fraction of 15NO3- in the initial substrate pool, and Δt is the incubation time. Each rate was calculated from the first two time points, and two or three replicates per time point. Additionally, we also calculated the turnover time of NO2- (τNO2-, days), which represents the average time required to replace the nitrite pool given the measured production rate following Eq. (4): 4τNO2-=NO2-RNO2-fromNO3- where [NO2-] represents the concentration of NO2- (nmol N L⁻¹), and RNO2-fromNO3- represents the production rates of NO2- from NO3- (nmol N L⁻¹ d⁻¹).

15NO3- production rate was measured by the increase in 15NO3- in the samples incubated with 15NH4+. We converted NO3- to N2O using the denitrifier method (Granger and Sigman, 2009; Sigman et al., 2001; Weigand et al., 2016), and analyzed the resulting 15N-N2O following the previously outlined procedure and corrections. Net production of 15NO3- (RNO3-fromNH4+, nmolNL-1d-1) is referred to here as nitrification (i.e., it includes the two-step process of oxidizing ammonium to nitrite to nitrate) and was calculated following Eq. (5): 5RNO3-fromNH4+=FNH4+-1Δ15NO3-Δt where Δ[15NO3-] represents the variation in the concentration of 15NO3-, FNH4+ represents the fraction of 15NH4+ in the initial substrate pool, and Δt is the incubation time. Each rate was calculated from the first two time points, and two or three replicates per time point.

The N2O yield during NH4+ oxidation to NO2- (N2O-yieldAmox, %) was defined as the percent of the total N transformed to N2O during the incubation with 15N-NH4+ (Eq. 6): 6N2O-yieldAmox=R15N2OR15N2O+RNO2-×100 The N2O yield during nitrification (i.e., NH4+ oxidation to NO3-) (N2O-yieldNit, %) was defined as the percent of the total NH4+ transformed to N2O during the incubation with 15N-NH4+ (Eq. 7): 7N2O-yieldNit=R15N2OR15N2O+RNO3-×100 The N2O yield during denitrification (N2O-yieldDenit, %) was calculated as follows (Eq. 8): 8N2O-yieldDenit=R15N2ORNO2-+R15N2O×100

Statistical analyses were conducted in R (R Core Team, 2014) version 4.4.0. Data visualization was also performed in R, with final figure adjustments made using Inkscape (Inkscape Project, 2017). We assessed normality using the Shapiro-Wilk test of normality analysis and homogeneity of variances across groups using Levene's test. For normally distributed data with equal variances, we applied one-way ANOVA (F). When normality was met but variances were unequal, we used Welch's t test. For data that violated normality assumptions, we employed the Kruskal–Wallis rank-sum test (K–W) or the Wilcoxon test (W). Outliers were identified using the Grubbs test (G). Statistical significance was set at p<0.05. Linear regressions were used throughout the study to evaluate the rates and drivers of N2O concentration and production. Model assumptions were assessed, and the model performance evaluated using adjusted R2 values and predictor significance was determined using p values (α=0.05). Each sample was assigned a unique identifier (#1–12), which is shown in Table 1 and in the figures to facilitate data interpretation and highlight observed trends.

The water column of Cubillas reservoir was thermally stratified in July (16.5–25.9 °C), such that DO varied dramatically with depth, with a DO peak at the top of the thermocline (400 µmolL-1, 5.6 m) and decreasing concentrations until anoxia at 8 m (Fig. 1a). Dissolved N2O concentration increased from 0.11 in the epilimnion to 6.38 µmolNL-1 at the bottom of the reservoir. The decrease in the water level during the summer months due to human management presumably caused the mixing of the water column at the end of the summer, as evidenced in the homogenization of the temperature and DO profiles (Fig. 1a). Dissolved N2O distribution remained mostly homogeneous in September, ranging from 0.22 to 0.42 µmolNL-1 (Fig. 1a, Table S1 in the Supplement). The water column was always supersaturated in N2O. NO3- concentration decreased significantly from July to September (Fig. 1a, Table S1). The average NO3- concentration was reduced by half, from 321.2 µmolNL-1 in July to 162.4 µmolNL-1 in September. NO2- concentration varied from 13.8 to 33.0 µmolNL-1 (mean = 22.0 µmolNL-1). NH4+ concentration was below detection level at some depths, peaking at 4.3 and 6.9 µmolNL-1 in bottom waters. DOC concentrations varied from 217.6 to 247.7 µmolCL-1 (Table S1), and Chl a concentrations ranged from 5.4 to 18.1 µgL-1 (Fig. 1, Table S1).

Iznájar reservoir's water level decreased by over 5 m in summer, but thermal and oxygen stratification persisted due to its greater depth relative to Cubillas (Fig. 1b). The water column was always supersaturated in N2O (Table S1). Dissolved N2O increased with depth and over time, ranging from 0.05 to 0.26 µmolNL-1 in July, up to 3.60 µmolNL-1 in September, with the larger increase in the hypolimnion (Fig. 1b, Table S1). NO3- concentration also decreased from July to September, from 373.7 to 329.3 µmolNL-1 (average values, Fig. 1b), with the lowest values at the oxycline, where NO2- peaked. NH4+ was only detected in the oxycline in July and in the hypolimnion in September, with values of 5.7 and 8.7 µmolNL-1, respectively. The DOC concentrations varied from 186.0 to 228.0 µmolCL-1, and the Chl a concentrations from 3.8 to 12.4 µgL-1 (Fig. 1, Table S1).

Figure 1 and Table S2 illustrate depth distributions of DIN concentrations and isotopic compositions. Relationships between DIN concentrations and isotopic compositions are shown in Fig. 2. The natural abundance δ15N-N2O in the Cubillas reservoir ranged from -2.1 ‰ in the bottom waters in July to 3.6 ‰ in the epilimnion in September, while the δ15N-N2O in the Iznájar reservoir ranged from -8.7 ‰ in the hypolimnion in July to -2.3 ‰ in the hypolimnion in September (Figs. 1 and 2). The δ18O-N2O ranged from 41.6 ‰ in the bottom waters of the Cubillas reservoir in July to 64.4 ‰ in the bottom waters of the Cubillas reservoir in September (Fig. 2b, c). δ15N-NO3- was consistently positive (i.e., 15N enriched pool) in all the samples analyzed, and it varied from 8.9 ‰ to 13.4 ‰ (Fig. 2e). In the Iznájar reservoir, NO3- concentration also decreased from July to September, along with an increase in δ15N-NO3- (e.g., Fig. 2e, #7–9). In the study reservoirs, δ15N-NO2- varied more than δ15N-NO3-. In general, δ15N-NO2- increased with depth, showing changes in a few meters, from 15N-depleted to 15N-enriched values, except for the Iznájar reservoir in the July sampling (Fig. 1b).

N2O production from NH4+ ranged from 0.06 to 48.57 nmolNL-1d-1 in the Cubillas reservoir (Fig. 3), and from 0.02 to 3.72 nmolNL-1d-1 in the Iznájar reservoir (Fig. 4) (n=12, Table S3). Ammonia oxidation rates (i.e., NO2- production from NH4+, RNO2-fromNH4+) were only significant in Iznájar's hypolimnion in September, reaching 215.8±38.0 nmolNL-1d-1 (N2O-yieldAmox=0.041 %) (Table S3). In contrast, significant nitrification rates (i.e., NO3- production from NH4+, RNO3-fromNH4+) were detected at all study depths except in the hypolimnion of Iznájar in September (Figs. 3 and 4, Table S3). Nitrification rates varied from 6.1 to 56.1 µmolNL-1d-1 in Cubillas, and from 0.0 to 36.7 µmolNL-1d-1 in the Iznájar reservoir. The nitrification rates were significantly higher in July (mean ± SD = 24.6±19.4 µmolNL-1d-1) than in September (7.3±6.7 µmolNL-1d-1), and in Cubillas (mean ± SD = 22.2±17.9 µmolNL-1d-1), than in the Iznájar reservoir (9.6±13.6 µmolNL-1d-1) (p<0.05, in both cases). The N2O yields during nitrification (N2O-yieldNit) varied from 0.000 % to 0.086 %, with the maximum yield observed in the bottom waters of Cubillas in July (Table S3). The production of N2O from NH4+ was significantly correlated with the in situ NH4+ concentration except in the hypolimnion of both reservoirs in September (n=10, adj R2=0.44, p<0.05) (Fig. 5a). These two samples, which were excluded from this analysis, contained the highest NH4+ concentrations (>6 µmolL-1). The N2O production from NH4+ was an exponential function of the nitrification rates (Fig. 5b, adj R2=0.60, p<0.01).

N2O production from NO3- varied from 0.2 to 18.1 nmolNL-1d-1 in the Cubillas reservoir, and from 0.4 to 61.0 nmolNL-1d-1 in the Iznájar reservoir (Figs. 3 and 4, Table S3). The highest rates were detected in the oxyclines. NO3- reduction to NO2- (i.e., first step of denitrification, RNO2-fromNO3-) varied from 13.7 to 33.2 µmolNL-1d-1 in Cubillas, and from 10.1 to 28.6 µmolNL-1d-1 in the Iznájar reservoir. NO3- reduction rates were significantly higher in July (27.5±7.0 µmolNL-1d-1) than in September (12.2±1.9 µmolNL-1d-1) (p<0.05). This decrease in the NO3- reduction rates was accompanied by a decrease in the NO3- concentration from July to September in both reservoirs. Among all the samples, NO2- turnover varied from 0.2 d in the hypolimnion to 4.1 d in the oxycline of Iznájar in September (Table S3). The N2O yield of NO3- reduction (N2O-yieldDenit) varied from 0.001 % to 0.132 % in the Cubillas reservoir, and from 0.003 % to 0.603 % in the Iznájar reservoir. The maximum yields occurred in the oxycline of Iznájar reservoir in September and the oxycline-bottom waters of Cubillas in September. N2O production from NO3- was not significantly correlated to the in situ NO3- concentration (p>0.05).

The in situ abundance of the functional genes (archaeal amoA, nirS and nosZ) varied with depth, time, reservoirs, and with the N transformation rates (Figs. 3 and 4, Table S4). Archaeal amoA abundance ranged from 0 to 2.7×103 copies mL⁻¹ (n=12). In the Cubillas reservoir in July, the archaeal amoA gene was detected only in the oxycline, where NO2- concentration was maximal and NH4+ minimal. We detected the archaeal amoA gene at all three depths in September, and its abundance decreased with depth. In the Iznájar reservoir, the archaeal amoA gene was detected at all depths, with the minimum abundance in the oxycline in July. Archaeal amoA abundance wasn't correlated with the N2O concentration (p>0.05), the N2O production rates from NH4+ (p>0.05), or the nitrification rates (p>0.05).

The nirS abundance ranged from 4.5×104 to 5.3×105 copies mL⁻¹ in Cubillas, and from 8.1×104 to 4.7×106 copies mL⁻¹ in Iznájar (n=12). nirS was present in all the samples, and its abundance increased with depth and over time in Iznájar. The nosZ gene was only quantified in the deepest layers (n=4), where it ranged from 800 to 2.1×103 copies mL⁻¹ and was higher in September than in July in both reservoirs. N2O production from NO3- was not significantly correlated with the in situ nirS gene abundance (p>0.05).

In both reservoirs, the higher N2O concentrations were found in the deepest layers under suboxic conditions (i.e., DO <10 µmolL-1) (León-Palmero et al., 2023; Pinti, 2014), and coincided with the highest cumulative Chl a concentration (mg Chl a m⁻²), and the highest abundances of nirS gene (Figs. 1, 3 and 4). N2O concentration decreased exponentially as DO concentration increased (Fig. 6a), but it increased in a power function correlated with cumulative Chl a concentration (Fig. 6b). N2O concentration was also a power function of the nirS abundance (Fig. 6c). It is thus consistent that nirS abundance showed a negative correlation with DO concentration (Fig. 6d) and a positive correlation with cumulative Chl a concentration (Fig. 6e). Total production of N2O, calculated as the sum of the production from NH4+ and from NO3-, was significantly positively correlated with the nirS gene abundance (Fig. 6f, n=11). Sample #12 was excluded of this analysis.

Additionally, there was a positive correlation between δ15N-NO3- and the δ15N-N2O (Fig. 6g). We also detected a strong correlation between δ15N-NO2- and N2O concentration (Fig. 6h). The abundance of the archaeal amoA gene was not correlated to δ15N-NO2- (p>0.05). In contrast, δ15N-NO2- was significantly correlated with the nirS abundance (Fig. 6i, n=12, adj R2=0.28, p<0.05). Particularly, the nirS gene abundance explained up to 94 % of the variance in δ15N-NO2- in the Iznájar reservoir (Fig. 6i, n=6, adj R2=0.94, p<0.001).

We detected significant production of N2O from both NH4+ and NO3-. The rates of N2O production from NH4+ reported in this study are larger than those found in Lake Lugano (Frame et al., 2017) and closer to those detected in the Chesapeake Bay (Tang et al., 2022). These rates are also larger than the rates found in the eastern tropical South Pacific oxygen minimum zone (Frey et al., 2020; Ji et al., 2015). N2O production rates were significantly correlated with the availability of NH4+ and with nitrification rates, but not with archaeal amoA gene abundance. Despite the hypolimnion of Iznájar in September (#12) being apparently anoxic, we detected a significant production of N2O from NH4+, ammonia oxidation, and the presence of archaeal amoA genes. This combination of processes and gene detection suggests that trace amounts of oxygen may have been present at levels below the detection limit of our oxygen sensor. Similarly, the presence of trace levels of oxygen may explain the production of N2O from NH4+, and the nitrification rates in the anoxic waters of Cubillas, although in that case we did not detect the presence of archaeal amoA genes. The highest amoA abundance was measured in the oxycline of Cubillas in July (i.e., 2.7×103 copies mL⁻¹), but amoA was not detected in the surface and bottom waters within the same profile, precisely where the highest N2O production from NH4+ occurred. The absence of detectable archaeal amoA genes in samples with high N₂O production may reflect primer bias rather than true absence of ammonia-oxidizing archaea. Previous work in San Francisco Bay revealed that dominant AOA clades were not amplified by commonly used primers, including those employed in this study (Rasmussen and Francis, 2022). It is therefore possible that important AOA lineages present in these reservoirs were missed, leading to an underestimation of amoA abundance. We did not measure the bacterial amoA gene abundance, because AOA had previously been identified as the dominant ammonia-oxidizers in the study reservoirs (León-Palmero et al., 2023). Therefore, we cannot assess the potential contribution of AOB. We tested for Comammox using specific primers and did not detect them in any sample. Additionally, sample water was pre-filtered before DNA extraction (pore size = 3 µm), which may have excluded microbes attached to particles or suspended sediment, potentially including AOA or Comammox groups.

Significant nitrification rates were detected in 11 out of 12 samples, with values similar to those found in another eutrophic freshwater system, Lake Mendota (Hall, 1986), and several orders of magnitude higher than reported open ocean nitrification rates (e.g., 0.4–10 nmolNL-1d-1) (Small et al., 2013, and references therein). The detection of high nitrification rates, but no significant ammonia oxidation, might suggest that comammox is occurring at these depths. However, our PCR analysis showed no evidence of the presence of comammox bacteria (Fig. S2), although, because no positive control was available, we cannot completely exclude their presence. Therefore, we consider the possibility that complete ammonia oxidation could contribute to the observed nitrification rates. Alternatively, we hypothesize that the NO2- production by ammonia oxidation was tightly coupled to NO2- consumption by NO2- oxidizers, such that it could not be detected in the NO2- pool. NO2- production from ammonia oxidation was only detected in one sample in which we did not detect a significant nitrification rate (i.e., hypolimnion of Iznájar reservoir in September, #12), suggesting that NO2- could accumulate due to a decoupling of ammonia oxidation and nitrite oxidation in this sample. Ammonia oxidation is the rate-limiting step for nitrification in most systems, which is why NO2- rarely accumulates in the environment and could explain our observed mismatch between ammonia oxidation rates and total nitrification rates (Kowalchuk and Stephen, 2001). The rates of NO3- production detected here were often sufficient to account for a complete turnover of the NO2- pool during the incubation, consistent with the idea that NO2- did not accumulate, even though the in situ concentrations were substantial.

The production of N2O from NO3- was generally higher than from ammonium, suggesting that NO3- is the main substrate for N2O production. The highest rates occurred in oxycline samples, where NO3- concentration was often lowest, and the NO2- concentration peaked. However, the N2O production from NO3- was not significantly correlated with the in situ concentration of NO3-, probably because N2O production rates are not limited by NO3- availability. These rates were higher than the rates found in ocean waters (Ji et al., 2015), and in the Chesapeake Bay (Tang et al., 2022), but similar to those found in the eastern tropical South Pacific oxygen minimum zone (Frey et al., 2020). Similarly, these previous studies in oxygen minimum zones found the highest rates of N2O production close to the oxic-anoxic interface (Frey et al., 2020; Ji et al., 2015).

Denitrification is the main microbial process leading to NO3- removal in aquatic systems. Denitrifying bacteria (as represented by the nirS gene) were consistently found throughout the reservoir water columns and reached their highest abundances in the suboxic waters. Their abundance was not significantly correlated with the N2O production from NO3-, likely because of the small sample size (n=7). Frey et al. (2020) found that the nirS gene was not significantly correlated with N2O production from NO3-, but was correlated with NO2-. The total N2O production, calculated as the sum of the production from NH4+ and from NO3- (Table S3), was significantly correlated with nirS gene abundance (Fig. 6f), highlighting the importance of denitrification in the overall production of N2O. This is consistent with the higher production obtained from NO3- than from NH4+, and with the evidence from natural abundance isotopes, discussed below. The rates of NO3- reduction to NO2- in this study were up to 1000 times higher than those in the ocean (Füssel et al., 2012; Ji et al., 2015) and in the Chesapeake Bay (Tang et al., 2022). These eutrophic reservoirs exhibit high productivity, with elevated concentrations of NO3- and organic matter fueling intense denitrification and N2O production. This rapid processing activity may reflect a system-level response to external nutrient loading, whereby a portion of the nitrogen input is redirected toward atmospheric release (León-Palmero, 2023).

In general, N2O production by denitrification, nitrifier denitrification and bacterial nitrification produces a significant isotopic fractionation of 15N, meaning that the lighter ¹⁴N is preferentially used in N2O production, resulting in a N2O pool depleted in 15N relative to the respective substrate and a higher δ15N value in the substrate left behind (Wenk et al., 2013 and references therein). In contrast, AOA produce N2O that is enriched in 15N relative to the substrate, increasing δ15N-N2O, with an isotopic fractionation value of ≈-6 ‰ (Santoro et al., 2011; Stieglmeier et al., 2014). At the same time, the consumption of N2O by denitrifiers increases the proportion of 15N and ¹⁸O in the remaining N2O pool, increasing δ15N-N2O and δ18O-N2O values (Wenk et al., 2016).

To identify trends over depth or time, and interpret them in relation to the processes that leave their signatures in the isotopes, each sample is identified on the cross plots with a unique number (Table 1 and Figs. 2, 3, 4, 6). The trends that we observed in the natural isotopic composition of the N species suggested that denitrification was a significant process in the water column, in agreement with the rate data. In general, the increase in the N2O concentration with depth was coupled to the δ15N-N2O decrease (e.g., #1–3, #5–6 or #7–9 in Fig. 1 and black trend lines in Fig. 2a), which indicates net production of N2O by water column denitrification, nitrifier denitrification and/or bacterial nitrification. In contrast, the opposite trend occurred in Iznájar in September (#10–12, Fig. 1b and red trend line in Fig. 2a), which suggests that N2O may be a mix of consumption by denitrifiers and production by AOA in the hypolimnion at the end of the summer. There was also an increase in the δ18O-N2O with depth in each profile, accompanied by an increase in N2O concentration, which suggests a parallel production and consumption of N2O at the deeper layers. That trend was not observed in Cubillas reservoir in July, but rather a noticeable increase in the δ18O-N2O in bottom waters from July to September along with N2O concentration decrease (Fig. 2b, #3 and #6), indicating active N2O reduction. Besides, many samples are located along the slope δ18O:δ15N=2.5 in Fig. 2c, which is indicative of active N2O reduction (Ostrom et al., 2007). We detected the nosZ gene, which encodes the reduction of N2O during denitrification, in hypolimnetic waters with higher abundances in September. N2O consumption can occur in the anoxic hypolimnion of Mediterranean reservoirs and result in undersaturations up to 27 % in those with low N availability (León-Palmero et al., 2023). However, in the investigated reservoirs, the N2O reduction by nosZ-carrying denitrifiers did not cause an undersaturation of N2O in the investigated time frame, which is consistent with previous findings in eutrophic reservoirs with high N availability (León-Palmero et al., 2023).

In the Iznájar reservoir, the decrease in NO3- concentration coincided with the increase in δ15N-NO3-, suggesting that denitrification is consuming the lighter NO3- during these months (Fig. 2e, #7–9). We detected that δ15N-NO3- was correlated with δ15N-N2O (Fig. 6g), which is indicative of denitrification. Over time, as more N2O is produced from NO3-, the NO3- pool may get substantially enriched in 15N, and δ15N-N2O values may also increase, creating a trend line where higher δ15N-NO3- corresponds to higher δ15N-N2O values. In general, NO2- reduction enriches 15N in the remaining NO2- pool, while the production of NO2- may decrease its δ15N-NO2-. In the study reservoirs, the production of N2O by denitrification may have enriched 15N in the remaining NO2- pool, as evidenced by the tight coupling between N2O concentration and δ15N-NO2- (Fig. 6h) and the increase in the δ15N-NO2- was correlated with the abundance of denitrifying bacteria in the reservoirs (Fig. 6i). The gene used as a marker for denitrifying bacteria (i.e., nirS) encodes the NO2- reductase that catalyses the reduction of NO2- during denitrification. Thus, it acts directly on the NO2- pool. Furthermore, the abundance of the nirS gene in the water column was correlated with the dissolved N2O, as we also detected in a survey of twelve Mediterranean reservoirs (León-Palmero et al., 2023). These results suggest that denitrification was the main pathway of N2O production, and it resulted in a characteristic isotopic imprint in the remaining NO2- pool.

In addition, the cumulative Chl a concentration, which is a proxy for the vertical export of the autochthonous organic matter produced by primary producers in the whole water column, was significantly correlated with the abundance of the nirS gene and the dissolved N2O concentration (Fig. 6b, e). This is also consistent with our previous study in twelve reservoirs (León-Palmero et al., 2023), and may indicate that denitrification is enhanced by particulate material derived from the phytoplankton community in the water column. Several studies in marine waters have described that denitrification was affected by the quantity and quality of organic matter (Babbin et al., 2014; Ward et al., 2008). Dalsgaard et al. (2012) found that the higher denitrification rates were all found at marine stations with high Chl a levels in the overlying water, suggesting a subducted and potentially decaying algal bloom. In general, this organic matter export represents a high-quality carbon source, but also sinking particles with a surface for microbial colonization, an environment where both oxic and anoxic/low oxygen microenvironments coexist, and they even increase the probability of contact between bacteria and nitrogen (Liu et al., 2013; Xia et al., 2017).

The highest total N2O production in Cubillas coincided with the highest N2O concentration at the deepest depth in both months (Fig. 3). In the deeper reservoir, Iznájar, the highest production was measured at the oxycline, where there is a strong potential for N2O fluxes, while the highest N2O concentrations were detected in the hypolimnion (Fig. 4). In both reservoirs, the N2O turnover time at the oxycline was the lowest in the profile. In Iznájar, the N2O turnover time at the oxycline was as low as 13 d in July and 8 d in September (Table S3), suggesting that the N2O produced at this location does not accumulate there. Instead, an important fraction of the N2O produced at the top of the oxycline may be consumed or diffuse to the top layer. This diffusive flux, together with the N2O produced in situ in the epilimnion by microbial activity and photochemidenitrification (Leon-Palmero et al., 2025), determines the large N2O fluxes found previously in this reservoir, reaching up to 3.6 mg N N2O m⁻² d⁻¹, and even exceeding the CO₂ equivalent warming potential from CO₂ and CH₄ emissions combined (León-Palmero et al., 2020b).

An important feature observed in the water column of these reservoirs over the summer was the substantial decrease in the NO3- concentration, suggesting an active N filter for the high N loadings. Microbial activity in the water column and the sediments of reservoirs can reduce the excess of N through emissions of N₂, primarily produced during denitrification and anammox. In this study, N2O emissions also constitute an important loss of fixed N. Total DIN loss calculations from July to September showed that Cubillas lost 468 kg N d⁻¹, while Iznájar lost 5337 kg N d⁻¹, representing a 45 % and 11 % decrease, respectively (Table 2). The DIN loss rates (2.4 and 0.7 µmolNL-1d-1) were similar or even higher than those calculated in other lakes or in the Baltic Sea (Seitzinger, 1988). Normalized to reservoir surface area, the N loss was slightly higher in Cubillas. N2O production was two orders of magnitude higher in Iznájar than in Cubillas in terms of kg N per day, but production rates were more similar when normalized to area. In the water column of Iznájar, the percentage of the N2O production per DIN loss was higher than in Cubillas, at 1.9 % and 0.6 %, respectively. These percentages only refer to the biologically produced N2O in the water column and may increase if the N2O produced in the sediments, or the N2O produced abiotically by photochemodenitrification, which was initially described in the surface waters of these reservoirs (Leon-Palmero et al., 2025), are also incorporated in the calculation. These estimates represent a major seasonal N loss event rather than annual rates. They are based on DIN concentration differences between July and September, without considering whether the reservoirs received N inputs from their watersheds during that period. Since summer is the dry period, and drawdown of the reservoirs exceeded any input via rain or runoff, N inputs from the watersheds were likely minimal during the study period. Further details on the calculations and assumptions are provided in the Supplement (Sect. S2).

Zhou et al. (2019) described a decrease of 97 % in the NO3- concentration in the water column of Zhoucun reservoir during spring (2 months), and they related the N losses to aerobic denitrification occurring in the water column. Brezonik and Lee (1968) estimated that the hypolimnion of Lake Mendota lost 312 kg N d⁻¹. Beaulieu et al. (2011) found that <1 % of denitrified N was converted to N2O in streams. Thus, these reservoirs act as important sinks for fixed N during the summer at the landscape scale, particularly within agricultural and urban watersheds, and sources of N2O to the atmosphere. Denitrification significantly contributed to dissolved N loss and N2O production in the water column. Although N2O production per unit of DIN loss was less than 2 %, the absolute amount of N2O produced in the water column and likely emitted into the atmosphere is substantial.