Forest streams are important sources for nitrous oxide emissions.

Streams and river networks are increasingly recognized as significant sources for the greenhouse gas nitrous oxide (N2 O). Nitrous oxide is a transformation product of nitrogenous compounds in soil, sediment and water. Agricultural areas are considered a particular hotspot for emissions because of the large input of nitrogen fertilizers applied on arable land. However, there is little information on N2 O emissions from forest streams although they constitute a major part of the total stream network globally. Here, we compiled N2 O concentration data from low order streams (~1000 observations from 172 stream sites) covering a large geographical gradient in Sweden from the temperate to the boreal zone and representing catchments with various degrees of agriculture and forest coverage. Our results showed that agricultural and forest streams had comparable N2 O concentrations of 1.6±2.1 and 1.3±1.8 µg N L-1 , respectively (mean±SD) despite higher total nitrogen (TN) concentrations in agricultural streams (1520±1640 vs 780±600 µg N L-1 ). Although clear patterns linking N2 O concentrations and environmental variables were difficult to discern, the percent saturation of N2 O in the streams was positively correlated with stream concentration of TN and negatively correlated with pH. We speculate that the apparent contradiction between lower TN concentration but similar N2 O concentrations in forest streams than in agricultural streams is due to the low pH (<6) in forest soils and streams which affects denitrification and yields higher N2 O emissions. An estimate of the N2 O emission from low-order streams at the national scale revealed that ~1.8×109 g N2 O-N are emitted annually in Sweden, with forest streams contributing about 80% of the total stream emission. Hence, our results provide evidence that forest streams can act as substantial N2 O sources in the landscape with 800×109 g CO2 -eq emitted annually in Sweden, equivalent to 25% of the total N2 O emissions from the Swedish agricultural sector.

hotspot for emissions because of the large input of nitrogen (N) fertilizers applied on arable land. However, there is little information on N 2 O emissions from forest streams although they constitute a major part of the total stream network globally.
Here, we compiled N 2 O concentration data from low-order streams (~1,000 observations from 172 stream sites) covering a large geographical gradient in Sweden from the temperate to the boreal zone and representing catchments with various degrees of agriculture and forest coverage. Our results showed that agricultural and forest streams had comparable N 2 O concentrations of 1.6 ± 2.1 and 1.3 ± 1.8 µg N/L, respectively (mean ± SD) despite higher total N (TN) concentrations in agricultural streams (1,520 ± 1,640 vs. 780 ± 600 µg N/L). Although clear patterns linking N 2 O concentrations and environmental variables were difficult to discern, the percent saturation of N 2 O in the streams was positively correlated with stream concentration of TN and negatively correlated with pH. We speculate that the apparent contradiction between lower TN concentration but similar N 2 O concentrations in forest streams than in agricultural streams is due to the low pH (<6) in forest soils and streams which affects denitrification and yields higher N 2 O emissions. An estimate of the N 2 O emission from low-order streams at the national scale revealed that ~1.8 × 10 9 g N 2 O-N are emitted annually in Sweden, with forest streams contributing about 80% of the total stream emission. Hence, our results provide evidence that forest streams can act as substantial N 2 O sources in the landscape with 800 × 10 9 g CO 2 -eq emitted annually in Sweden, equivalent to 25% of the total N 2 O emissions from the Swedish agricultural sector.

K E Y W O R D S
agriculture, forest, greenhouse gas, nitrogen, nitrous oxide, river, stream

| INTRODUC TI ON
Nitrous oxide (N 2 O) is a potent greenhouse gas with a global warming potential (GWP) about 300 times that of carbon dioxide (CO 2 ) over a 100-year timeframe (IPCC, 2013). N 2 O is also the current dominant ozone-depleting substance, and N 2 O emissions thus have a negative impact on the recovery rate of the ozone hole (Ravishankara, Daniel, & Portmann, 2009). At a global scale, agriculture is the largest anthropogenic source of N 2 O, contributing 4.1 Tg N/year, that is, ~60% of all anthropogenic N 2 O emissions (Ciais et al., 2014).
Nitrification is the microbial oxidation of ammonia (NH 3 ) or ammonium (NH + 4 ) to nitrate (NO − 3 ); during the first step of this oxidation, namely the oxidation of NH 3 or NH + 4 into nitrite (NO − 2 ), N 2 O can be formed as an intermediate product (Prosser & Nicol, 2012). Denitrification is the sequential reduction of nitrogenous oxides (NO − 3 or NO − 2 ) to gaseous forms (NO, N 2 O and N 2 ;Tiedje, 1988;Wrage et al., 2001). The production of N 2 O is largely dependent on environmental conditions, and the major regulators are carbon and nitrogen (N) availability, temperature, pH and moisture (Mosier et al., 1998).
Soils and livestock management are the main anthropogenic sources of N 2 O in agricultural landscapes (Ciais et al., 2014).
However, a fraction of N fertilizers applied onto fields can be leached to ground-and surface waters. During leaching and transport in ground-and surface waters, transformation processes (e.g., denitrification) result in the production of N 2 O, which is water-soluble (Baggs & Philippot, 2011;Wrage et al., 2001). Hence, drainage networks (i.e., ditches and streams) are hotspots for N 2 O emissions (Reay et al., 2012;Rees et al., 2013). Studies on streams in the United States, France and Sweden have demonstrated that, although streams constitute only a small fraction of the total area in the landscape (~0.1%), they can have a disproportionately large impact on total N 2 O emissions from agriculture (3%-6%; Audet, Wallin, Kyllmar, Andersson, & Bishop, 2017;Beaulieu, Arango, Hamilton, & Tank, 2008;Grossel et al., 2016). Considering that the consumption and use of agricultural N fertilizer is increasing to meet the food demand of the growing global population (Bodirsky et al., 2014), it is likely that agricultural N 2 O emissions will continue to increase in the future and contribute to climate forcing and ozone depletion (Ravishankara et al., 2009;Reay et al., 2012).
Consequently, many N 2 O studies have focused on streams draining agricultural areas, resulting in a lack of data on N 2 O emissions from streams within other types of land use, especially forest streams (Davidson & Swank, 1990;Holl, Jungkunst, Fiedler, & Stahr, 2005;Vidon & Serchan, 2016), despite the potential of forested catchments to process and transform N (e.g., Brookshire, Valett, Thomas, & Webster, 2005;Kortelainen et al., 2006;Sponseller et al., 2016). Estimation of N 2 O emissions from forest streams would be especially relevant in countries where forest covers large proportions of the total land mass such as Finland (73%), Sweden (69%), Russia (50%) and Canada (38%;FAO, 2015). Hence, even if it is likely that forest streams have much lower N availability and less N 2 O emissions per unit area than agricultural streams, the former might still be a larger N 2 O source at the national and global scale. Such information is crucial for developing targeted and effective mitigation schemes aiming at reducing N 2 O emissions.
To fill the knowledge gap, we assembled a unique data set comprising approximately 1,000 stream N 2 O concentration measurements from agricultural and forest streams in Sweden. We focused especially on low-order streams (Strahler order ≤ 4) because of their strong hydrological and hydrochemical connectivity with surrounding soils and the fact that they often constitute the majority of the total stream length (Bishop et al., 2008). We hypothesized that (a) streams in forested catchments will have lower N 2 O concentrations than streams draining agricultural catchments because of lower N availability; (b) when scaled to the national level, Swedish forest streams will emit more N 2 O than agricultural streams due to their greater length and surface area.

| Data set and site descriptions
The data set of the present study comprises direct concentration measurements of N 2 O from Swedish streams. The data set is a combination of catchment and regional surveys performed during 2004-2017 in six catchments or regions: Krycklan (KRY), South-East Sweden (SES), Skogaryd Research Catchment (SRC), Scania (SCA), and Uppsala 1 and 2 (UPP1 and UPP2). The sites spanned a large geographical range of Sweden from approximately 55°N to 64°N, thereby covering most climatic zones with the exception of the sub-Arctic ( Figure 1). All data were collected from low-order streams (Strahler order ≤ 4), except for two sampling sites at UPP2 where the Strahler order was 5.
The mean annual precipitation and temperature at the sites ranged from 550 to 900 mm and from 2 to 7°C, respectively (Table 1).
The area of the subcatchments at the sampling sites ranged from 0.03 to 834 km 2 . The catchments at KRY, SES and SRC were dominated by forest land use (average >80%), while the streams at SCA, UPP1 and UPP2 had an average of 69%, 49% and 36% of agricultural land use in their respective catchments. Wetlands were also present at some of the sites, especially at KRY (mean cover 17%; Table 1).
The KRY data were collected between January and December 2004 (~28 sample collections) at 15 stream sampling sites within the boreal KRY catchment as part of the Krycklan Catchment Study . The sites at SES represent first-order streams that were part of a seasonal survey in late summer and autumn 2016 as well as spring 2017 (Hawkes et al., 2018;Wallin et al., 2018). Approximately 100 sites were included in each sea-

| Water chemistry
Grab samples of stream water were taken for nutrient analysis at all sites at every visit. Stream water pH was recorded at the majority of the sites. At KRY, pH was measured at room temperature after returning to the laboratory using a Ross 8102 low-conductivity combination electrode (ThermoOrion; Buffam, Laudon, Temnerud, Mörth, & Bishop, 2007). At SCA, pH was measured directly in the field using a WTW ProfiLine Multi 3320. At SES, pH was measured at room temperature upon arrival at the laboratory using a titrosampler Metrohm 855 with a built-in pH probe. At SRC, pH was measured in situ using a Hach HQ40D-PHC10105 pH electrode at eight of the 16 sampling sites. At UPP1-2, pH was measured directly in the field using a Multiparameter Meter Hi 9829 from Hannah Instruments. Total organic carbon (TOC) was measured in the water samples from KRY, SCA and SES, while dissolved organic carbon (DOC) was determined at SRC and UPP1-2. TOC is generally equivalent to DOC in Swedish forest streams (Laudon et al., 2011;Laudon, Köhler, & Buffam, 2004) and will, therefore, hereafter be referred to as DOC. Total N (TN) was measured in samples from KRY, SCA, SES and SRC, whereas only NO − 3 was measured at UPP1-2. However, the NO − 3 fraction generally constitutes most of TN in sites dominated by agricultural land use (Kyllmar et al., 2014) and will, therefore, for simplicity, be referred to as TN hereafter. All chemical analyses were performed according to Swedish standard methods (Fölster, Johnson, Futter, & Wilander, 2014). Stream water temperature was recorded upon sampling at all sites except at SRC where the temperature was recorded at the most downstream sampling site in the catchment.

| In-stream concentrations of N 2 O
The data set of in-stream concentrations of N 2 O was formed by combining results from several sampling campaigns which used different protocols but all relied on headspace equilibration method (McAuliffe, 1971) and gas chromatography (GC) analyses.
At KRY, water samples were collected in N 2 -filled 60 ml glass vials sealed with a bromobutyl rubber septa. For each sample, a 15 ml aliquot of bubble-free water was injected into the glass vial, subsequently acidified to pH 2-3 with one drop of 30% ultrapure HCl (0.5% v/v) and stored cold at ~2°C. At SCA, SES, UPP1 and UPP2, 10 ml of stream water was collected in a 22.5 ml gas-tight glass vial preflushed with N 2 ; the vials also contained 0.2 ml of ZnCl 50% (w:v) for sample preservation. At SRC, 5 ml stream water was added to 20 ml vials preflushed with N 2 and containing 100 µl H 3 PO 4 for sample preservation. The samples were stored in the dark until analysis generally within a week and up to a maximum of 1 month. The headspace N 2 O concentrations in the vials from all sites were directly analyzed by GC with electron capture detector (GC-ECD). The GC brands varied among laboratories, but certified N 2 O standards were used in all cases for calibration and validation.
Headspace N 2 O concentrations obtained after GC analysis were converted into dissolved N 2 O concentrations (C obs ) using the N 2 O solubility function by Weiss and Price (1980) and taking into account the stream water temperature and atmospheric pressure at the sampling time. Data on atmospheric pressure were obtained from the closest monitoring station from the Swedish Meteorological and TA B L E 1 Characteristics and sampling information on the streams sampled in six regions or catchments Wetland

| Estimate of total N 2 O emissions from loworder streams in Sweden
The total N 2 O emissions from low-order streams in Sweden were estimated using the same approach as in Wallin et al. (2018), where a national estimate of CO 2 and CH 4 emissions from low-order streams was derived. Wallin et al. (2018) provided estimates of gas transfer velocities for CO 2 (k 600 ) for every combination of stream order (1-4) and land-use class (i.e., agriculture or forest). The gas transfer velocities were modeled based on slope, catchment area and daily specific discharge for more than 400,000 stream segments. The mean values of gas transfer velocities for CO 2 (k 600 ) specified in Wallin et al. (2018) were converted to k N 2 O following Wanninkhof (1992): where Sc N 2 O is the Schmidt number calculated as described in Wanninkhof (1992), accounting for changes in water temperature.  for alpine regions, which represent only 6.5% of the total stream surface area in Sweden, these were not included in our assessment.

| Statistics
The statistical analyses were performed using the open source statistical software R version 3.4.4 for Windows (R Development (1) %sat = (C obs ∕C eq ) × 100, Core Team, 2018), with the package 'nlme' and the function 'lme' therein (Pinheiro, Bates, DebRoy, Sarkar, & R Development Core Team, 2012). Linear mixed effect models were used to explore linkages between N 2 O %sat and selected environmental variables, as these models are particularly suitable to examine the patterns in time series datasets from different sites (Zuur, Ieno, Walker, Saveliev, & Smith, 2009). The mixed models were checked for normality and homogeneity of variance by visual inspection of plots of residuals against fitted values (Zuur et al., 2009). The significance of the models was assessed by comparison with a nullmodel using the likelihood ratio. The potential predictor variables were checked for multicollinearity using the variance inflation factor (VIF) values (VIF < 10 indicating low risk of multicollinearity).
We used spatial correlograms (function spline.correlog in the R package 'ncf'; Bjornstad, 2018) to verify the absence of spatial autocorrelation in the residuals of the models. Finally, we tested the presence of temporal autocorrelation in the mixed models by adding the correlation structure 'corAR1' from the package 'nlme' and examining the residuals (Pinheiro & Bates, 2000;Pinheiro et al., 2012). All N 2 O observations and corresponding ancillary variables were included in the following models.
The aim of the first analysis was to test whether N 2 O %sat and potential regulators of N 2 O production (TN, pH and DOC) differ between forest and agricultural streams. Hence, N 2 O %sat, TN, pH and DOC were individually tested for significant differences between forest (KRY, SES and SRC) and agricultural (UPP1, UPP2 and SCA) streams (Table S1, models 1-4). To reduce variance heterogeneity in the data and to meet the assumptions of linear mixed effect models, N 2 O %sat was transformed using natural logarithm before inclusion in the models. The regions or catchments were added as a random effect. Using the same approach, the differences in N 2 O %sat in forest and agricultural streams across seasons and stream order were also tested (Table S1, models 5-8). When season or stream order was found significant in the models, the variations among the different seasons or stream orders were tested using Tukey's posthoc test. In a second analysis, the aim was to test the effect of selected We used Monte Carlo simulations (mean of 10,000 repetitions of a Monte Carlo simulation with 10,000 iterations) in R to estimate the uncertainty of the total N 2 O emissions from low-order streams. The level for significance of all analyses was set at p < .05. (1,620 ± 1,490 µg N/L) and UPP2 (1,050 ± 1,200 µg N/L) than at the rest of the sites, although SES also had a few higher values

| D ISCUSS I ON
The results of the present study reveal that Swedish streams are This proportion is in reasonable agreement with previous research at KRY and other boreal catchments (Kortelainen et al., 2006;Sponseller, Blackburn, Nilsson, & Laudon, 2018) and if we consider that the same proportion holds true at the other forest regions, this would confirm that the ratio N 2 O:NO  (Eriksson, Karltun, & Lundmark, 1992). The current recovery from acidification observed in many streams in Northern Europe and North America (Garmo et al., 2014;Kothawala, Watmough, Futter, Zhang, & Dillon, 2011) opens the question of whether stream N 2 O emissions from acidified forested areas are experiencing a decreasing trend.
Another plausible explanation for the relatively high N 2 O concentration in forest streams could be that N 2 O is produced by chemodenitrification, which is the abiotic reduction of oxidized N species (i.e., NO  Hansel, Wankel et al., 2017). Boreal forested catchments, typically on podzols, generally have a high iron export to surface water (Ekström et al., 2016;Kortelainen et al., 2006) and thus are likely to offer conditions suitable for chemodenitrification and potentially high yields of N 2 O (Kulkarni, Yavitt, & Groffman, 2017 most N transported to surface waters will primarily reach low-order streams and is assumed to be rapidly processed before being transported downstream (Alexander, Boyer, Smith, Schwarz, & Moore, 2007;Marzadri, Dee, Tonina, Bellin, & Tank, 2017;Peterson et al., 2001). For example, a decrease in N 2 O emissions with increasing stream order was observed in a study from Minnesota, USA (Turner et al., 2015). However, we did not observe a similar pattern in our agricultural streams, perhaps because our data set comprised only low-order streams from several regions, whereas Turner et al. (2015) investigated streams and rivers ranging in order from 1 to 10, located in the same region and with similar crop cover (mainly corn production).
It is unclear whether most N 2 O in streams is produced in upland or riparian soils before being transported to surface waters or whether it is produced in situ. Upland forest soils are generally believed to act as weak sources or sinks of atmospheric N 2 O, and production of N 2 O can proceed through both nitrification and denitrification (Laverman, Zoomer, & Verhoef, 2001;Peichl, Arain, Ullah, & Moore, 2009;Skiba, Pitcairn, Sheppard, Kennedy, & Fowler, 2005). Increased N 2 O production has been observed after both increasing moisture content and increased N load (Sitaula & Bakken, 1993;Ullah, Frasier, King, Picotte-Anderson, & Moore, 2008), and this N 2 O could then be transferred from soils to streams.
Additionally, the role of the riparian zone as source of N 2 O production needs to be clarified, considering the strong controls that it exerts on a wide range of biogeochemical processes in forested and agricultural catchments (Blackburn, Ledesma, Näsholm, Laudon, & Sponseller, 2017;Ledesma et al., 2018;Ranalli & Macalady, 2010).
The proportion of wetlands in the catchment also strongly alters TN, DOC and iron dynamics in headwater forest streams (Löfgren, Fröberg, Yu, Nisell, & Ranneby, 2014;Sponseller et al., 2018) and thus can affect N 2 O production processes. In situ production of N 2 O in the hyporheic and benthic zones of the stream is suggested to be a major source of stream N 2 O (Marzadri et al., 2017). However, a study of 72 headwater streams determined that in-stream denitrification contributed, on average, only 26% of the total N 2 O emissions (Beaulieu et al., 2011), whereas the contribution by other processes (e.g., nitrification or chemodenitrification) remains largely unknown.  (Beaulieu et al., 2008) or no seasonal trend (Baulch, Schiff, Maranger, & Dillon, 2011 (Gauthier, Bernier, Kuuluvainen, Shvidenko, & Schepaschenko, 2015;Hansen et al., 2013).
This study suggests that even relatively low N levels processed and leached to surface waters via acidic soils in forested catchments can yield significant amounts of N 2 O emitted to the atmosphere. Consequently, N deposition and fertilization of forest soils might lead to higher N 2 O emissions than anticipated if N is leached to surface water. Atmospheric N deposition reaches about 7 kg N/ha year in southern Sweden, 3.9 kg N/ha year in central Sweden and 1.2 kg N/ha year in northern Sweden (Lucas, Sponseller, & Laudon, 2013 Turner et al., 2015). On the other hand, a recent paper, using a modeling approach, concluded that N 2 O emissions from inland waters might be overestimated by an order of magnitude (Maavara et al., 2019). However, a review concluded that the IPCC factor for N 2 O riverine emission was actually very similar to the factor calculated from a global data set of N 2 O stream emissions (Tian, Cai, & Akiyama, 2019). In our study, we showed that the ratio N 2 O:NO − 3 at the agricultural streams (0.0015) compared well with the EF 5r (0.0025) while the forest streams seemingly had a higher emission factor (0.011). Hence, there is still a great need to better constrain estimates of riverine N 2 O emissions, especially in forest streams.

ACK N OWLED G EM ENTS
The authors wish to thank Maud Oger and My Osterman for field assistance, Audrey Campeau for her GIS map design and Tinna Christensen for her assistance with the graphical abstract. We are grateful for the financial support for each of the original studies that together have enabled the large N 2 O concentration data set.
The authors thank all the samplers from the different catchments,