Lakes as nitrous oxide sources in the boreal landscape

Abstract Estimates of regional and global freshwater N2O emissions have remained inaccurate due to scarce data and complexity of the multiple processes driving N2O fluxes the focus predominantly being on summer time measurements from emission hot spots, agricultural streams. Here, we present four‐season data of N2O concentrations in the water columns of randomly selected boreal lakes covering a large variation in latitude, lake type, area, depth, water chemistry, and land use cover. Nitrate was the key driver for N2O dynamics, explaining as much as 78% of the variation of the seasonal mean N2O concentrations across all lakes. Nitrate concentrations varied among seasons being highest in winter and lowest in summer. Of the surface water samples, 71% were oversaturated with N2O relative to the atmosphere. Largest oversaturation was measured in winter and lowest in summer stressing the importance to include full year N2O measurements in annual emission estimates. Including winter data resulted in fourfold annual N2O emission estimates compared to summer only measurements. Nutrient‐rich calcareous and large humic lakes had the highest annual N2O emissions. Our emission estimates for Finnish and boreal lakes are 0.6 and 29 Gg N2O‐N/year, respectively. The global warming potential of N2O from lakes cannot be neglected in the boreal landscape, being 35% of that of diffusive CH4 emission in Finnish lakes.


| INTRODUC TI ON
Lakes and streams acting as recipients of carbon, nitrogen, and other nutrients transported from terrestrial ecosystems contribute to landscape greenhouse gas (GHG) balances emitting carbon dioxide (CO 2 ), methane (CH 4 ), and nitrous oxide (N 2 O). Freshwater N 2 O has received minor interest compared to carbon gases, CO 2 and CH 4 .
Consequently, both the variability of N 2 O concentrations and factors regulating N 2 O fluxes from lakes at regional and global scales have remained poorly constrained and estimates of freshwater N 2 O emissions are still uncertain due to sparse data (Deemer et al., 2016;DelSontro, Beaulieu, & Downing, 2018;Soued, del Giorgio, & Maranger, 2015). Majority of freshwater studies have focused on rivers and streams in N-rich agricultural environments (Beaulieu et al., 2011;Hu, Chen, & Dahlgren, 2016;Mulholland et al., 2008) excluding landscapes dominated by forests and peatlands, the most widely distributed ecosystems in the boreal zone.
The main processes involved in N 2 O cycling are aerobic nitrification and anaerobic denitrification, which are regulated by several environmental factors like oxygen and organic matter content, pH, temperature, and the availability of ammonium and nitrate (Butterbach-Bahl, Baggs, Dannenmann, Kiese, & Zechmeister-Boltenstern, 2013). In contrast to most previous studies, which have predominantly focused either on a few lakes and/or summer time measurements, we measured N 2 O concentrations for the four seasons in the water columns of 112 lakes in Finland covering different lake types, locating between the latitudes 60°N and 67°N ( Figure 1; Table 1). We examined how seasonal and spatial variation of N 2 O concentrations in lakes was associated with the characteristics of lakes (area, maximum depth, water chemistry, temperature, oxygen content) and catchments (area, elevation, and land use cover) and compared spatiotemporal variation in N 2 O concentration with that of CO 2 and CH 4 measured simultaneously with the N 2 O (Juutinen et al., 2009;Kortelainen et al., 2006).
Besides N 2 O lake chemistry and morphometry (lake area [LA], max depth), catchment characteristics (catchment area [CA], elevation), land use cover (agricultural land %, peat %, forest %, urban %), and climate-related variables (latitude, water temperature) were determined for each lake in order to identify key drivers contributing to seasonal variation of N 2 O concentrations in boreal lakes. The F I G U R E 1 Location of the study lakes (n = 112)  [1998][1999] and for Nordic Lake Survey (NLS) lakes (sampled in 1995) lakes were sampled once in winter, spring, summer, and autumn during 1998-1999 from four depths at the deepest point of the lake for N 2 O concentrations and physical and chemical characteristics. For CO 2 and CH 4 analyses, a subpopulation of 177 lakes was randomly selected from the NLS data (Juutinen et al., 2009;Kortelainen et al., 2006). Gas analyses were carried out in the lab-

| Calculation of gas fluxes
N 2 O measurements were carried out by the headspace equilibration technique (McAuliffe, 1971). Ultra pure N 2 gas (30 ml) was added to 60 ml syringes containing 30 ml water, the syringes were then shaken vigorously for 3 min. The headspace gas concentration was quantified with a gas chromatograph (Hewlett Packard Series II and Shimadzu GC-14-A) equipped with an FI-detector.
Lake-atmosphere gas fluxes were estimated from the surface water gas concentrations according to the First Fick's law of diffusion (Fick, 1855;Wanninkhof & Knox, 1996): where F gas is the lake-atmosphere flux of N 2 O, kN 2 O is the gas transfer velocity (m/day), C gas is the concentration of the gas in the surface water (μmol/L), C eq is the concentration of the gas (μmol/L) in equilibrium with the atmosphere. Since the data behind measured gas transfer coefficients (k values) are limited and the relationship between lake size and k values varies, we used three approaches to estimate lake-atmosphere gas exchange (Heiskanen et al., 2014;Holgerson, Farr, & Raymond, 2017;Vachon & Prairie, 2013). We calculated the gas fluxes for the 94 lakes that had N 2 O concentrations measured in all four seasons.
Some lakes are located in remote areas and were difficult to sample before and/or after ice melt.
First, following the approach by Heiskanen et al. (2014), kN 2 O value was calculated, using the average values of kCO 2 from a small Finnish lake during a 4 month period, which we transformed into the kN 2 O: where Sc gas is the Schmidt number for a given gas (Jähne, Heinz, & Dietrich, 1987) and n is 1/2.
The wind speed was assumed to be 3 m/s, which is an average open water period wind speed at the height of 10 m for the inland measurement stations in Finland (Leinonen, 2000).

| Estimation of annual fluxes and upscaling
First, following similar approach as for CO 2 (Kortelainen et al., 2006), Secondly, we estimated the annual median N 2 O evasion both for each lake size class separately (lake size-specific evasion) and for all size classes combined (the mean, median, and summer median of individual lakes) for Finnish and Boreal lakes. The randomly selected lakes, which could be sampled during all four seasons (1) F gas = k gas C gas − C eq , (3) k 600 = 2.51 + 1.48 U 10 + 0.39 U 10 log 10 LA.
(n = 71), were divided into the different size classes and the median annual evasion of each size class was multiplied with the area of the Finnish and boreal lakes belonging to the respective size class.
Summing up the size class-specific estimates results in an estimate for Finland and boreal region, respectively. The rationale was that when upscaling is based on the evasion estimated specific to each of the four lake size classes, the impact of the numerous small lakes in our data on the regional estimate is not disproportionally large.
Furthermore, we estimated the contribution of N 2 O emission from lakes to that of forests both in Finland and in the boreal zone. For Finland, we used the lake surface area of 32,663 km 2 (Raatikainen & Kuusisto, 1990) and the forest area of 0.203 × 10 6 km 2 (Vaahtera et al., 2018). For the boreal region, the lake surface area was estimated from MODIS data (1,422,448 km 2 ), the forest area of 12.1 × 10 6 km 2 was derived from Potter, Matson, Vitousek, and Davidson (1996).

| Water chemistry
The lakes were sampled in each sampling occasion for dissolved oxygen, alkalinity, conductivity, pH, color, total nitrogen (N tot ), nitrate (+ nitrite) nitrogen defined as (NO 3 -N, nitrite had negligible contribution to the total amount of nitrate and nitrite), ammonium nitrogen (NH 4 -N), total phosphorus (P tot ), phosphate phosphorus (PO 4 -P), total organic carbon (TOC), and total iron (Fe tot ). Water chemistry was analyzed from unfiltered samples in the accredited laboratories of the Regional Environment Centers. N tot was determined by oxidation with K 2 S 2 O 8 , reduction of NO 3 -N to NO 2 -N in Hg-Cd (Cu-Cd) column and colorimetric determination of azo-color. The sum of NO 3 -N and NO 2 -N was measured by reduction of NO 3 -N to NO 2 -N in Hg-Cd (Cu-Cd) column, followed by colorimetric determination of azo-color. NH 4 -N was measured colorimetrically with hypochlorite and phenol. P tot and PO 4 -P were measured colorimetrically. TOC was determined by oxidizing the sample by combustion and measuring inorganic C by IR-spectrophotometry (National Board of Waters, 1981).

| Catchment characteristics
The CAs of the NLS lakes were determined from the topographic maps, and the catchment boundaries were digitalized and combined with land use data based on satellite images using the Arc View georeferencing software. Lake area, CA, catchment to LA ratio, latitude, and the proportion of peatland, forest on mineral soil, agricultural land, water (consisting of the upstream water bodies and the lake itself), and built-up area in the catchments were determined.

| Statistical analysis
The relationships between the N 2 O concentrations and lake chemistry, morphometry, latitude, and catchment characteristics were examined using Pearson's correlation coefficients using SAS 9.4 for Windows software. The variables were log e or square root transformed in order to normalize their distribution.
Stepwise multiple linear regression models predicting N 2 O concentrations were carried out using lake chemical, physical, and morphometric variables, climatic variables (temperature, latitude), and catchment properties as predictors. The cases with an absolute value of the studentized residual exceeding 3 were excluded and only the independent variables with p < .05 were included in the models.
Linear mixed models were used to take into account that each lake was sampled four times and from four depths (interdependence of within-lake sampling). We run a linear mixed model with NO 3 -N, depth, season as fixed factors, and lake as random factor. That is, the

| Seasonal and spatial variation in N 2 O
During the open water season, 71% of the surface water samples were saturated with respect to the atmospheric equilibrium value of N 2 O, that is, lakes were mostly sources of N 2 O. Nitrous oxide, similar to CO 2 , peaked in winter. In contrast, CH 4 concen-  (Juutinen et al., 2009;Kortelainen et al., 2006).
There was no district correlation between N 2 O and water pH.
Both the highest and the lowest N 2 O concentrations occurred around the median pH of 6.5 ( Figure 6). Even though there was a weak correlation between oxygen and N 2 O (Table 2) F I G U R E 2 Seasonal distribution (median, first and third quartile) of the concentrations of N 2 O (n = 87) (a), CO 2 (n = 177) (b), and CH 4 (n = 177) (c), in randomly selected lakes, all depths. Minimum N 2 O concentrations were measured in summer in contrast to CO 2 and CH 4 distributions. Concentrations of CO 2 and CH 4 are based on the data from Kortelainen et al. (2006) and Juutinen et al. (2009). Note that y-axis is on a log scale Lake chemistry predicted N 2 O better than catchment land use cover.
In linear multiple regression models, electron acceptors, nitrate and oxygen, and lake water temperature as independent variables predicted best (r 2 = .55, stepwise procedure) the N 2 O in the entire data (n = 1,396, all seasons and depths). Nitrous oxide in bottom water (all seasons) was best predicted by nitrate and oxygen concentration (r 2 = .54). The surface water model (all seasons) had nitrate, temperature, and the percentage of agricultural land in the catchment as the independent variables explaining 58% of the variation in N 2 O. The best model for winter (all depths) explained 58% of the variability in N 2 O selecting nitrate, latitude, and pH as the independent variables (Table 3). The linear mixed model results demonstrated that the significant relationship between nitrate and N 2 O remained (p < .001) even after the influence of depth (p = .096) and season (p < .001) had been taken into account.

| N 2 O evasion
We estimated the median N 2 O evasion based on 71 lakes (<100 km 2 ) as 0.009 g N m −2 LA year −1 . The seasonal median fluxes were 0.002 g N/m 2 LA at the thaw (during 0.5 months), 0.002 g N/m 2 LA in spring (1.5 months), 0.001 g N/m 2 LA in summer (3 months), and 0.003 g N/m 2 LA in autumn (2 months). For the largest lakes (>100 km 2 ), we used the evasion estimate from the 10-100 km 2 lake size class.
The median evasion for the <0.1, 0.1-1, 1-10, 10-100 km 2 lake size classes was estimated as 0.0047, 0.007, 0.018, and 0.02 g N/m 2 LA, respectively (using k values from Holgerson et al., 2017). Nitrous oxide data were not available for lakes larger than 100 km 2 ; for this lake size class, we used the median evasion estimate from the 10-100 km 2 size class. N 2 O evasion per surface area unit was highest in the largest lakes reflecting the distribution of nitrate concentrations. In contrast, CO 2 evasion estimates per surface area unit (Kortelainen et al., 2006) were largest in small lakes.
Total annual N 2 O flux from Finnish lakes (total LA 32,663 km 2 ) was estimated as 0.6-0.8 Gg N 2 O-N/year, based on the areas of the lake size distribution by Raatikainen and Kuusisto (1990) (Table 4; Tables S1 and S2).
Freshwater N cycling integrates numerous simultaneous temperature-dependent microbiological processes. Also, our data demonstrated  (Table 5; Table S2)  driver for CO 2 in our data, with similar explanation power, 78%, of the variation across all lakes and depths (Kortelainen et al., 2006).
Using Equation (4) Large lakes turned out to be disproportionately important N 2 O sources among the lake population on the landscape level.
Lakes larger than 10 km 2 were estimated to contribute 77% of the total N 2 O emission from Finnish lakes. In contrast, CO 2 evasion estimates demonstrated that lakes smaller than 10 km 2 dominated landscape CO 2 evasion among the lake population while lakes larger than 10 km 2 (representing 65% of the total LA distribution) represented only 45% of the estimated CO 2 evasion (Kortelainen et al., 2006). Our estimated N 2 O-N emissions from lakes represent 17% of the N 2 O-N emissions from boreal forests, the dominating ecosystem in Finland where lakes cover 10% of the total land area (Table 6).  (Juutinen et al., 2009) Figure 6). The variability in N 2 O concentration was the largest around the median pH 6.5 (Figure 6), which may reflect optimal pH of the accumulation of N 2 O, that is, net production of N 2 O from nitrification and denitrification. Low pH inhibits the N 2 O reductase which increases the N 2 O to N 2 ratio in denitrification (Richardson, Felgate, Watmough, Thomson, & Baggs, 2009). Nitrification is further inhibited at high C:N ratios (Her & Huang, 1995), typical for boreal Finnish lakes and often accompanied by low pH (Kortelainen et al., 2013). Supportingly, Humic large lakes, including only seven lakes in our dataset, also had high N 2 O concentrations (Figures 3 and 10).

| Seasonal and spatial variation in N 2 O concentrations and fluxes
Nitrous oxide peaked in winter similar to CO 2 , while CH 4 con- Generally, N 2 O production is limited by low N turnover and low N mineralization in the high latitude N limited ecosystems (Potter et al., 1996). In N limited boreal terrain, N turnover is rapid and internal N cycling across forested ecosystems dominates the spatial nitrate distribution, which often reflects more closely catchment land use cover and topography than N deposition-in spite of N deposition being  (Figure 7). In summer, low nitrate content and high N 2 O/NO 3 -N ratio resulted from nitrate being consumed in primary production, denitrification, and other microbial processes. In freshwaters, 0%-4% of N is generally released as N 2 O in denitrification (Mulholland et al., 2008;Seitzinger, 1988;Silvennoinen, Liikanen, Torssonen, Stange, & Martikainen, 2008).
The explanation power of our statistical models for N 2 O concentrations in lakes (Table 3) is comparable to the power of the models developed for terrestrial N 2 O emissions (Leppelt et al., 2014;Pärn et al., 2018). In the lake dataset from boreal southern Norway and Sweden, N 2 O concentrations correlated positively with nitrate in summer (Yang et al., 2015). In

| Landscape scale patterns
Key processes and feedbacks of landscape scale GHG fluxes have remained poorly quantified. Dynamics of N 2 O in our lakes at landscape scale did not follow those of CO 2 (Kortelainen et al., 2006) and CH 4 (Juutinen et al., 2009; Figure 1). Large lakes dominate the lake surface area distribution in Finland (Raatikainen & Kuusisto, 1990). Furthermore, estimated N 2 O emissions per surface area unit were largest from large lakes (Figure 9) reflecting the distribution of nitrate concentrations. In contrast, concentrations and estimated emissions of CO 2 and CH 4 decrease with decrease in LA (Denfeld, Kortelainen, Rantakari, Sobek, & Weyhenmeyer, 2016;Juutinen et al., 2009;Kortelainen et al., 2006 Annual mean and median N 2 O flux estimates from our lakes including seasonal data were surprisingly close to each other (Table 5) while estimates based only on summer measurements underestimated annual N 2 O emissions (Table 4). Furthermore, the different annual N 2 O flux estimates resulting from the three approaches for gas transfer coefficients (k values) and their dependence on lake size (Heiskanen et al., 2014;Holgerson et al., 2017;Vachon & Prairie, 2013) demonstrate that there is uncertainty due to limited mea-  (Cui et al., 2016). Rising temperature has further been shown to result in earlier spring snow melt floods throughout northeastern Europe (Blöschl et al., 2017) which contributes to seasonal distribution of nitrate transport from land to lakes and further to the overall role of lakes as N 2 O sources in the boreal landscape.

ACK N OWLED G EM ENTS
Sincere thanks to our dear colleague Jari Huttunen, whose enthusiasm on N 2 O inspired us to finish this paper after his passing away, to two anonymous reviewers for the comments, which significantly improved the MS, to the personnel of the Water and Environment Districts for sampling and analyzing the water samples, to Antti Räike for preparing Figure 1 and

CO N FLI C T O F I NTE R E S T S
The authors declare no competing financial interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.