Water Flow Controls the Spatial Variability of Methane Emissions in a Northern Valley Fen Ecosystem

Hui Zhang1,2*, Eeva-Stiina Tuittila3, Aino Korrensalo3, Aleksi Räsänen2,4, Tarmo Virtanen2,4, Mika Aurela5, Timo Penttilä6, Tuomas Laurila5, Stephanie Gerin5, Viivi Lindholm4, Annalea Lohila1,5 1Institute for Atmospheric and Earth System Research (INAR), Department of Physics, P.O. Box 68 (Pietari Kalmin katu 5), 5 University of Helsinki, 00014 Helsinki, Finland 2Helsinki Institute of Sustainability Science (HELSUS), 00014 Helsinki, Finland 3Peatland and soil ecology research group, School of Forest Sciences, University of Eastern Finland, 8010 Joensuu, Finland 4Ecosystems and Environment Research Programme, Faculty of Biological and Environmental Sciences, University of Helsinki, 00014 Helsinki, Finland 10 5Climate System Research, Finnish Meteorological Institute, PL 503, 00101, Helsinki, Finland 6Natural Resources Institute Finland, 00790 Helsinki, Finland *Correspondence to: Hui Zhang (hui.zhang@helsinki.fi)


Introduction
Northern peatlands, which cover approximately 15 % of the boreal and arctic regions, are long-term sources of the greenhouse gas methane (CH4) (Korhola et al., 2010;MacDonald et al., 2006), partly counteracting the cooling impact of 35 related long-term carbon dioxide (CO2) uptake. The response of northern peatlands to global warming has partly contributed to the recent increase in atmospheric CH4 concentrations (Bousquet et al., 2011;Ciais et al., 2014;Kirschke et al., 2013), and modelling projections have suggested that, globally, wetland CH4 emissions will continue to increase during the 21 st century and have a positive feedback on global warming (Zhang et al., 2017). However, large uncertainties remain in the global CH4 budget models due to limited knowledge of the relative contribution of the various environmental drivers that control CH4 40 fluxes (Riley et al., 2011). To upscale observed CH4 fluxes and produce realistic scenarios for future projections of atmospheric CH4 concentrations, it is crucial to understand and quantify the correlations between peatland CH4 emissions and their environmental drivers.
In peatlands, CH4 is produced in wet and anoxic conditions below the water level by methanogens, and then released from the peat to the atmosphere. During the transport process, part of the produced CH4 is consumed/oxidised by methanotrophs. 45 The processes of CH4 production, consumption, transport and final release to the atmosphere are affected by several environmental factors, such as water level, organic substrates, and temperature (Abdalla et al., 2016;Bellisario et al., 1999;Larmola et al., 2010). There is also evidence that peatland vascular plant functional types can affect CH4 emissions by altering microbial community structure (Robroek et al., 2015). Sedge-dominated fens are CH4 emission hotspots due to greater methanogenic activity (Juottonen et al., 2005) and litter degradation rates (Aerts et al., 1999). Also, the greater 50 abundance of sedges (Carex spp.) in fens provides both a direct route for CH4 movement to the atmosphere through aerenchyma tissue, thereby avoiding the oxidation of CH4, and also provides high-quality litter into the soil, which promotes CH4 production (Noyce et al., 2014).
Fens, unlike bogs, are fed by mineral-rich water as seepage from the mineral soil below (soligenous fens) or from surface water flow from the catchment (topogenous fens) (Wheeler and Proctor, 2000). Valley fens that are located in water 55 collecting depressions can receive water from both sources. The spatial variation in the quantity and quality of incoming water creates spatial patterns in vegetation and microbial communities (e.g., methanogens and methanotrophs), and thus CH4 production and oxidation, transportation and ultimately emissions to the atmosphere (Andersen et al., 2011;Juottonen et al., 2015;Kokkonen et al., 2019;Robroek et al., 2015). Several studies have focused on the interactions of CH4 with vertical water level fluctuations. For example, long-term lowering of the water level has been associated with a decreased abundance 60 of Sphagnum mosses and aerenchymous plants, decreased CH4 emissions and CH4 production potential (Yrjälä et al., 2011).
However, due to the heterogeneity of peatlands, inconsistent patterns can also be found. For instance, several studies have indicated that greater CH4 emissions occur when the water level is close to the surface of the peatland (Bubier et al., 2005;Pelletier et al., 2007), while other studies have found maximum fluxes occurred at intermediate water levels (Turetsky et al., 2014), or found no connection between CH4 emissions and water level (Euskirchen et al., 2019, Korrensalo et al., 2018 Nevertheless, water level has been suggested as a more important forcing factor on CH4 cycling in fens than either https://doi.org/10.5194/bg-2020-268 Preprint. Discussion started: 14 August 2020 c Author(s) 2020. CC BY 4.0 License. temperature or vegetation composition alone Mäkiranta et al., 2018;Riutta et al., 2020). In addition to vertical water level changes, the lateral flow of water in fens can be even more important in driving the processes that underpin CH4 emissions, because flowing water not only ensures a water supply for the vegetation, but also transports nutrients, which benefits vegetation and microbial communities (Laitinen et al., 2007). At the same time, flowing water is 70 likely to transport more oxygen (Ingram, 1983), thus enhancing CH4 oxidation and suppressing production. While fens are typically the highest CH4 emitters of all peatlands (Turetsky et al., 2014), the influence of lateral water flow on fen CH4 emissions has not been studied to date.
At a global scale, climate warming is projected to continue in the decades ahead, while changes in precipitation patterns are projected to be more regional (Collins et al., 2013). The Coupled Model Intercomparison Project (CMIP5), under a RCP8.5 75 scenario, predicts a warmer and wetter climate for Fennoscandia (Collins et al., 2013). As peatland hydrology is driven by several processes, such as precipitation, lateral water fluxes, transpiration and evaporation, climate model predictions cannot be directly applied to infer peatland hydrological conditions (Helbig et al., 2020;Tuittila et al., 2007;Wu et al., 2010;Zhang et al., 2018), especially in minerotrophic fens. Nevertheless, peatland habitats can be impacted under both warming-dry and warming-wet scenarios (Bjorkman et al., 2018;. In addition, fens may be more sensitive to water level 80 changes than bogs; in particular, their plant communities have been shown to experience clear species turnover under drier conditions (Kokkonen et al., 2019). Aside from the vertical fluctuations in the water level, climate change is also likely to affect the water that enters fens as it will control the hydrological conditions within the catchments, e.g., the temperature sum in spring strongly controls the timing and amount of snowmelt water that enters the fen. This type of change in catchment conditions is likely to impact, for example, plant phenology and biomass production (Mäkiranta et al., 2018). This will, in 85 turn, impact on C cycling between the peatland and the atmosphere due to different photosynthesis, decomposition and gas transportation rates, and on other factors at the plant functional type and even at the species levels (Hajek et al., 2009;Laine et al., 2011;. Hence, a full insight into the complex climate-peatland-ecohydrology-CH4 relationship is needed to predict the impact of changing catchment hydrology on fen CH4 emissions under climate change scenarios. Prior to importing peatland-scale CH4 emissions into global circulation models, we first need to bridge the gap of understanding as 90 to how water flows control fen microhabitats and CH4 emissions. In this study, we aimed to assess the role of flowing water in regulating spatial variations in valley fen vegetation and CH4 emissions. More specifically, we asked the following research questions: 1) How does a flowing stream within a valley fen impact microhabitat conditions, vegetation composition and biomass production? 2) Does the distance to a stream modify CH4 fluxes? 3) How does vegetation composition and stream-related variables control CH4 emissions? We hypothesised that: 95 (H1) water table, temperature, oxygen concentration, vegetation structure and biomass are related to the proximity of the stream; (H2) spatial variation in CH4 fluxes is related to the distance from the stream; (H3) regulation of CH4 fluxes by the stream is mediated by the vegetation and by environmental variables, such as oxygen concentration. It is an open mesotrophic sedge fen that is located in a valley in the hilly Pallas region of northern Finland (Figure 1).
Based on the 30-year average (1981-2010; Kittilä Pokka meteorological station), the annual average temperature and total precipitation are -1.3 °C and 547 mm, respectively (Pirinen et al., 2012).   During the three flux measurement years (2017-2019), the summer of 2018 was exceptionally warm, up to 5 °C warmer than the long-term average (Figure 1b). Based on the ICOS continuous peat profile temperature measurements (at 5, 10, 20, 30, 115 50 and 100 cm; Figure 1c) in 2018, peat temperature at Lompolojänkkä varied along depth and also for different habitats ( Figure A1). During summer, peat temperature decreased from the surface to the deeper layers and the pattern was reversed for the other seasons. Peat temperatures were greater in the drier parts of the study site compared to the wetter parts (closer to the stream) at all measured depths, and there were also larger temperature variations between the different depths.
Peat accumulation at Lompolojänkkä initiated around 10,000 cal. yr BP (calibrated years before the present 1950 AD) and 120 the deepest peat depth is approximately 2.5 m (Mathijssen et al., 2014). Almost the whole peatland is water saturated throughout the year. The relatively dense vegetation layer is dominated by different sedges (e.g., C. rostrata, C. chordorrhiza) in the wet areas and various deciduous shrubs (e.g., Betula nana, Salix phylicifolia) in the relatively dry areas.
Moss cover (e.g., Sphagnum spp.) is patchy with c. 57 %-cover (Aurela et al., 2009). A small stream flows through the long and narrow valley fen (outlined in Figure 1a) and empties into the nearby Pallasjärvi lake. The flow and size of the stream 125 varies seasonally; being largest in spring after snow melt in the catchment. During summer, the stream water level in many locations is below the vegetation surface and may not be visible (Figure 1d). For more detailed descriptions of Lompolojänkkä, see Aurela et al. (2009) and Lohila et al. (2010).

Sampling / sample plot set up 130
To quantify the spatial variability in CH4 fluxes in the valley fen, we installed 15 permanent sample plots 60 cm x 60 cm (W x L) at varying distances from the stream in 2017 (Table 1). The sample plots were set up as clusters of three to six plots that were typically located within a metre of each other. Initially, the closest cluster from the stream was located within a 10-m distance, and the furthermost at a 40-m distance. In 2019, we sought a better mechanistic understanding of the controls on CH4 fluxes and so we added nine more sample plots, located in three clusters at 50, 60 and 90 m distance from the stream 135 (Table 1). In total, 24 permanent gas flux measurement plots were established (Table 1, Figure 1c). The sample plots are coded according to their distance to the stream/visible flowing water as 10a-c (a cluster within 10 m to the stream with three replicates a-c), 20a-c, 30a-f, 40a-c, 50a-c, 60a-c and 90a-c. The location of each plot was measured with a Trimble R8 GPS device with ±5 cm accuracy and the distance to the stream from each sample plot was calculated based on the National Land 145 Survey of Finland topographic database. For determining fluxes, the closed chamber method with fixed collars was used for clusters 30-90, and a floating chamber method without collars was employed for clusters 10 and 20 (Alm et al., 2007). The size of the opaque aluminium chamber was 60 cm x 60 cm x 40 cm (W x L x H) and each chamber was equipped with a fan. The sample gas was sucked from the 155 chamber at a flow rate of 200-200 ml min -1 using 50-m long tubing (d=6 mm) into a LGR gas analyser (LGR GCA-24p-EP, model 911-0011-0004, Los Gatos Research Inc., Ca, USA) located in a temperature-controlled cabin. The duration of one measurement was approximately 5 mins. The floating chamber (60 cm x 60 cm x 30 cm) was used at plots with permanently high, flowing water. In addition, gross primary production (GPP) was measured at clusters 10-40 using a transparent chamber on 24-25 July 2019 at the time of peak growing season. Same gas analyser as described above was used. 160

CH4 and CO2 flux measurements
Photosynthetically active radiation in the chamber was measured using a Kipp&Zonen PQS1 PAR Quantum Sensor (Kipp & Zonen B.V., Delft, the Netherlands). In order to fit a light-response curve to the net CO2 exchange (NEE) data, NEE was first measured in full light, after which the chamber was covered with fabrics to create four different light levels (white shade, black shade, double black shade, and double black with green shade). In addition, one measurement with full shading to capture dark respiration was performed. 165 The CH4 and CO2 fluxes from each measurement were calculated from the linear slope (R 2 > 99 % for over 90 % measurements and R 2 > 90 % for other measurements) in gas concentration over time, taking into account chamber volume, chamber air temperature and air pressure at the measuring point. The volume in the chamber during each measurement was specified according to the instant ambient water level. The air temperature and air pressure data were derived from the nearest meteorological station, and air pressure was calibrated for each chamber, taking into account the altitude of the plot. 170 We determined the GPP-light response curve for each sample plot (based on the NEE measurements with the transparent chamber), and derived sample plot specific GPPmax values at a photosynthetic photon flux density level of 800 μmol m -2 s -1 .

Environmental data collection
To reach a mechanistic understanding of the spatial pattern of CH4 fluxes, we collected data on the potential environmental factors that control emissions in combination with each flux measurement conducted in 2019. These factors were air and 175 peat temperature, water table, dissolved oxygen concentration, leaf area index, and plant community cluster (Table 1).
Air temperature was either measured using a temperature sensor fixed inside the chamber or measured at 2-m height at the site (Lompolonvuoma meteorological station of Finnish Meteorological Institute (FMI)). Peat temperature was measured at 5 cm below the moss surface (T5) using a Pt100 thermometer (Omega HH376, Omega Engineering Inc., CT, USA). Water table relative to the moss surface (WT) was measured from a plastic tube installed in the peat next to each sample plot. 180 Dissolved oxygen concentrations at 20 (DO20) and 40 cm (DO40) below the surface (except cluster series 60) were measured using a YSI Professional Series Digital handheld meter.
The leaf area index (LAI) of four vascular plant functional types (PFTs; deciduous shrub, evergreen shrub, forb and graminoid), and moss cover were estimated. The estimation of LAI followed Juutinen et al. (2017). First, we selected 31 square plots (50 cm × 50 cm) located within the fen and surrounding areas in July-August 2019, and estimated green 185 projection cover (%) and measured mean height for each PFT in the plots. Second, to measure LAI of the samples, we harvested the aboveground parts of the vascular plant species, scanned them with an A4 scanner and calculated the proportion of green pixels in GIMP 2.8 (The GIMP Team, www.gimp.org). Third, we constructed empirical relationships between cover or plant volume (cover × height) and LAI with ordinary least squares (OLS) regressions for four PFTs found in the site. We chose the optimal predictor (cover or volume) by minimising the root mean square error value, and in the 190 final models, the adjusted coefficient of determination (Adj.R 2 ) varied between 0.73 and 0.89 (Table A1). Fourth, we used the equations from the OLS regressions to model seasonal LAI development curves for each CH4 sample plot in which we had measured green projection cover and height for the four PFTs throughout the summer of 2019. Finally, we derived LAI values for each flux measurement time from the seasonal LAI development curves. We also calculated LAI values for the aerenchymous plants in each plot, which included C. aquatilis, C. canescens, C. chordorrhiza, C. lasiocarpa, C. limosa, C. 195 rostrata, Comarum palustre, Equisetum fluviatile, Eriophorum vaginatum, and Menyanthes trifoliata. The calculation of aerenchymous LAI was carried out by applying the same OLS regression equations used for forb and graminoid PFTs to datasets that included only aerenchymous plant species.
In addition, we delineated four plant community types/clusters for the CH4 sample plots as follows. First, we calculated the Bray-Curtis distance matrix of the plant species projection cover data from the sample plots and, in addition, 200 200 systematically sampled vegetation plots that were inventoried in the fen in 2018. Second, we derived four non-metric multidimensional scaling ordination (NMDS) axes from the distance matrix. Third, we delineated four plant community clusters from the NMDS axes with the Partitioning Around Medoids (PAM) method. The clustering was conducted in R with packages 'vegan' (Oksanen et al., 2019) and 'cluster' (Maechler et al., 2019). A map showing the location of the vegetation community clusters in the study site can be found in Figure A2. 205

Data analysis
NMDS was used to explore the linkages between peak season vegetation composition, distance to the stream, biomass production and flowing water. Peak season total LAI was used as a proxy for biomass production, and early summer DO20 and DO40 were used as proxies for flowing water/nutrient availability. For a robust analysis, plant species with occurrence lower than 3 % were excluded from the analysis. 210 Linear mixed-effect models were applied to the CH4 flux and environmental data to identify the potential drivers of CH4 flux using two different approaches. First, we explored the spatial variation in CH4 fluxes by constructing a model with CH4 data from all three measured years. Here, potential fixed predictors for CH4 flux were distance to the stream, air temperature and the factorial plant community cluster. To account for repeated measurements, we included the nested random effects of year, month and measurement plot. Second, to gain a mechanistic understanding of the controls on CH4 fluxes, we used a dataset 215 with additional variables gathered during 2019. Here, potential fixed predictors were DO20, DO40, T5, air temperature, WT, GPPmax, LAI of all vascular, aerenchymous and ericoid plants, moss cover (% coverage), CO2 dark respiration, distance to the stream, and the factorial plant community cluster. To account for repeated measurements from the plots over the growing season, we included the crossed random effects of measurement day and plot.
In building the models, we manually added the potential fixed predictors one by one and tested whether the resultant, more 220 complex model was significantly better than the model without the added predictor, using conditional F-test and Akaike information criterion (AIC). To account for the nonlinear relationship between CH4 flux and some environmental variables (such as temperature), we tested several response shapes for the fixed predictors: i) linear response, ii) quadratic response, iii) linear response above or below a certain threshold value, but constant otherwise, and iv) quadratic response above or below a certain threshold value, but constant otherwise. In cases iii) and iv), the response type and threshold value were determined 225 visually by plotting the residuals of the previous model against the fixed predictor to be added. The final response shape and threshold value were selected based on the conditional F-test and AIC values. Furthermore, we tested the interactions between all fixed predictors in the final models and only included those predictors that led to a significant improvement in model performance. The first explorative model was fitted with function lme of the package 'nlme', and the second more complex mechanistical model was fitted with function lmer of the package 'lme4' in R. 230

Variations in vegetation and environmental factors
The studied valley fen exhibited clear but distinctive patterns in vegetation composition, WT, LAI, and DO concentrations related to distance from the stream (Figures 2 and A2-5). Moreover, the temporal patterns in WT and DO concentration 235 showed distinct variations at locations further away and closer to the stream, respectively ( Figure A4).
In total, four plant community types were identified (Figure 3, Table A2). Community type (1) fluvial, which was found in the wetter parts of the fen, was dominated by E. fluviatile and C. limosa. Community type (2) riparian represented riparian vegetation that were taller, such as C. aquatilis, S. lapponum, S. phylicifolica and Comarum palustre. Community type (3) lawn, and community type (4) hummock contained vegetation typical of drier fen conditions, with the hummock type found 240 in the driest areas. The dominant species in these community types included S. riparium, Vaccinium oxycoccos and C. livida (lawn), and S. russowii, V. uliginosum, Betula nana and Rubus chamaemorus (hummock). The overriding pattern was related to the distance to the stream (Figure 2a and A2), i.e. fluvial and riparian community types were recorded in the locations closest to the stream, while lawn and hummock types were located at the plots furthest from the stream. In addition, the plant communities in the sample plots were suggestive of a spatially heterogeneous structure in the fen, i.e. different types were 245 recorded within a short distance (Figures 2a and 3). The NMDS ordination ( Figure 3) revealed that the main pattern in vegetation structure related to the distance to the stream was correlated strongly with, and was better explained by, peak season oxygen concentration. Total LAI increased with peak season oxygen concentration, which was negatively correlated with distance. Aquatic species, such as C. aquatilis and species that typically benefit from moving water, such as S. lapponum, C. palustre and M. trifoliata, exhibited relatively high positive values on the first NMDS axis, revealing a strong 250 relationship between the stream and some specific plant species. Species adapted to drier surfaces, such as mosses Rhizomnium sp. and S. warnsdorfii, and the sedge E. vaginatum, were located at the other end of the axis. As peak season GPP data were only available for clusters 10-40, they were not included in the NMDS analysis, but were analysed separately against oxygen concentration and total vascular LAI data ( Figure A6). GPP was clearly higher closer to the stream (> 0.45 mg CO2 m -2 s -1 ) than further from the stream (< 0.35 mg CO2 m -2 s -1 ). In addition, GPP was strongly related to total vascular 255 LAI, at least when LAI < 2. In the only sampling point with a LAI value > 2, GPP did not increase any further.
The WT pattern at the sample plots was strongly linked to their distance to the stream, i.e. WT was higher closer to the stream (Figure 2b). At clusters 10 and 20, close to the stream, there was approximately 10 cm of water above the peat surface, while at cluster 90, furthest from the stream, the WT was approximately 10 cm below the surface. The other clusters displayed intermediate WT values. In general, the lowest (deepest) WT levels were measured at all sample plots during late 260 July, when precipitation was low and air temperature had reached the seasonal peak ( Figures A4 and A7).
The sample plots located next to the stream (cluster 10) showed significantly larger mean seasonal vascular LAI values (mean 1.5) but were similar to cluster 60 (with lawn vegetation) (Figure 2ac). Clusters 10 and 60 both showed significantly higher aerenchymous LAI values than the other clusters (~0.5), although cluster 10 (mean 1.4) had a significantly higher value than cluster 60 (mean 1.1) (Figure 2c). Plot 10a appeared to be an outlier with higher total and aerenchymous LAI 265 values (~4) than the other plots (< 2), which was attributed to the presence of the abundant forb C. palustre in that plot Dissolved oxygen concentrations at both 20 and 40 cm depths showed a similar spatial pattern, with higher concentrations 270 recorded close to the stream (in clusters 10 and 20) (Figure 2d). However, large temporal variations existed in DO values at both the 20 and 40 cm depths, which generally peaked in early summer during a high flow of water ( Figure A4). Also, DO concentrations showed a similar temporal pattern to precipitation, with higher concentrations recorded during periods with higher precipitation (Figures A4 and A7).
The proximity of the stream reduced the temporal variation in the peat temperature measured at 5 cm depth in 2019 (Figure  275 A4); while the temperature at sample plots further away (clusters 50-90) varied between 3 and 23 °C, and the temperature at sample plots close to the stream (clusters 10-40) stayed between 7 and 15 °C.

Response of CH4 fluxes to environmental forcing
In the mixed-effect model (three-year dataset), which was constructed to examine spatial variability, CH4 fluxes were controlled by the distance to the stream and by air temperature (fixed predictors), while plant community type was not a significant predictor when distance to the stream was included (Table A3a). There was a quadratic relationship between CH4 320 fluxes and distance to the stream, with the highest fluxes observed at an intermediate distance (Figure 5a). There was a positive linear correlation between air temperature and CH4 fluxes only at temperatures above a threshold value of 18 °C.
Below that threshold, CH4 fluxes remained unaffected ( Figure 5). There was a significant interaction between distance to the stream and air temperature (p = 0.03), with the greater impact of temperature on CH4 flux observed at plots closer to the stream (Figure 5b). 325  In the second model (2019 dataset), which was constructed to provide a robust mechanistic understanding of CH4 dynamics in the fen, temporal and spatial variation in CH4 flux were found to be best explained by peat temperature at 5 cm (T5), WT, DO concentration at 20 cm below the surface (DO20), graminoid LAI and vascular LAI as fixed predictors ( Figure 6, Table   A3b). When DO20 was included, the distance to the stream and plant community type were not significant predictors. Of 335 these predictors, DO20 linearly decreased the flux until a threshold value of 40 % was reached, above which it remained constant, while there was a linear relationship between CH4 fluxes and the other predictors ( Figure 6). Both T5 and graminoid LAI were observed to linearly increase CH4 fluxes, while fluxes were negatively correlated with WT and vascular LAI (i.e., fluxes were lower at higher water levels and greater vascular LAI values). DO20 interacted with T5 and WT ( Figure 6e, f), so that DO20 decreased CH4 flux more steeply at lower T5 and WT values. Also, T5 and WT responses were 340 steeper at low DO20 values. Furthermore, vascular LAI had less impact on CH4 flux at high WT levels.

The role of the stream in driving fen vegetation and biomass production
As hypothesised, the spatial heterogeneity in environmental variables in this valley fen site was highly related to the distance from the stream. Peatlands typically have spatially heterogeneous microhabitats due to wide variations in hydrology and 355 nutrient availability (Rydin and Jeglum, 2013), which impact microbial activities and subsequent CH4 emissions (Juottonen et al., 2005;Juottonen et al., 2015;Noyce et al., 2014;Ström et al., 2003). Water table level is one of the most important influences on plant occurrence and growth in peatlands (Rydin and Jeglum, 2013), and in this study, was highest closer to the stream. As a result, hydrophilic species, such as C. aquatilis, S. lapponum, C. palustre, and M. trifoliata, were abundant in places close to the stream. Even though we did not measure the chemical composition of the water, the abundance of these 360 species implies a minerogenic supply established by water flow (Wassen et al., 1990).
The observed positive link between early summer oxygen concentrations (a proxy for flowing water) and total LAI further confirmed that flowing water likely delivers more nutrients and better supply plant growth and photosynthesis, and therefore provides more C substrates for microbial activities (Bellisario et al., 1999). In addition, GPP, the key driver of the peatland C cycle (Whiting and Chanton, 1993) and influences peatland vegetation composition and abiotic factors, such as air 365 temperature and water level (Peichl et al., 2018), was consistently higher in plots located nearer the stream. Similarly, dissolved oxygen concentrations that acted as a proxy of the mineral-nutrient rich water were also higher in those plots. It has been shown that increased water supply alone can cause substantial increases in biomass on nutrient-rich soils, while fertilisation/nutrient addition has little effect (Eskelinen and Harrison, 2015). As such, the forbs and mosses that dominate such wet fens might benefit from higher water tables for biomass production (Mäkiranta et al., 2018). In this study, as the 370 stream can bring both water and nutrients to the site at the same time, we are not able to distinguish whether the impact of the stream on the vegetation at our site was caused by the water or by nutrient supply, or both. Nevertheless, our results suggest that flowing water acted as a decisive factor on peatland vegetation composition and biomass production, through the addition of either water or nutrients. Therefore, the stream is likely to play a key role in regulating peatland CH4 emission patterns. 375

Role of stream-induced microhabitats in driving CH4 emissions
Consistent with our second hypothesis, the overall pattern of CH4 fluxes showed clear spatial variations in relation to the distance from the stream. The impact of the stream was greater than the influence of vegetation community types, which represent general microform conditions and have been commonly reported to regulate CH4 emissions (e.g., Riutta et al., 2007). Specifically, as expected in the third hypothesis, factors such as peat temperature at 5 cm depth (T5), WT, DO20 and 380 LAI, which were to some extent shaped by the stream, were all significant factors in driving CH4 emissions at this site.
Our data suggest that CH4 emissions increased with higher T5 values, similarly to many previous studies (e.g., Korrensalo et al., 2018;Rinne et al., 2018). Rising temperature is known to increase the activity of CH4 producing microbes, and also enhance CH4 transport through aerenchymous plants (Dunfield et al., 1993;Grosse, 1996;Kolton et al., 2019). Moreover, https://doi.org/10.5194/bg-2020-268 Preprint. Discussion started: 14 August 2020 c Author(s) 2020. CC BY 4.0 License. the temperature sensitivity in our study site was stronger closer to the stream, which is possibly due to a higher dissolution 385 rate in cold water. The pattern implies that in the fertile conditions next to the stream, higher oxygen concentrations in the cool water limits emissions by supressing CH4 production or by enhancing oxidation, and that warming of the water removes this limitation. In support of this finding, higher DO20 values were found to decrease temperature sensitivity. Similar to the CH4 response to T5, higher DO20 values also reduced the impact of WT position on CH4 emissions. Both responses highlight the importance of oxidation when considering how CH4 emissions respond to environmental changes (Song et al., 390 2020). The patterns might also indicate higher CH4 production under warmer conditions within the catchment and, consequently, on higher CH4 concentrations in the flowing water (Juutinen et al., 2013). However, in this study we were not able to determine the origin of the emitted CH4.
In our sampling campaign, WT levels were observed both above and below the soil surface, and CH4 emissions were found to generally decrease with rising WT levels. This decrease is in contradiction with many other studies that mainly cover sites 395 with WT levels below the soil surface (Bubier et al., 2005;Pelletier et al., 2007;. However, low emissions were also observed in the drier parts of the fen, which is in agreement with previous studies, in addition to very low emissions observed close to the stream. The lower emissions and a generally unimodal response to WT level were overridden in the whole dataset by the much stronger pattern captured close to the stream. Two plausible explanations for the observed WT-CH4 emission pattern are, 1) the potential CH4 production zone is smaller and the potential CH4 oxidation 400 zone is greater in drier conditions (Lai, 2009), and 2) in wet conditions, where there is moving water, lower CH4 emissions can be attributed to enhanced CH4 oxidation in the oxygen-rich water column, and a lower CH4 production rate due to the presence of oxygen (Bubier, 1995). Also, lower peat temperatures due to the higher water table and flowing water likely contribute to a lower CH4 production rate.
In our study, vascular LAI was found to have a negative linear correlation with CH4 emissions. Plots nearest the stream had 405 the highest vascular LAI values but the lowest CH4 fluxes, i.e. the impact of the stream was again predominant over the impact of LAI. Studies have shown that shrubs can hinder CH4 production because of their poor-quality substrate for methanogenesis (Riutta et al., 2020, Yavitt et al., 2019, although the cover of shrubs at our study site was very small. Therefore, low CH4 emissions at high vascular LAI values is likely due to in situ lower peat temperature and the higher oxygen concentrations caused by the moving water. As aerenchymous LAI showed a very similar pattern to vascular LAI, it 410 was not included in the mechanistic model. Instead, graminoid LAI, which was not impacted by the stream, showed a positive link with CH4 emissions, which is in line with several previous studies (e.g. Bhullar et al., 2013ab). The exceptionally high CH4 fluxes measured at cluster 50 where the graminoid LAI was low is potentially linked to one aerenchymous species growing in the cluster, i.e. Eriophorum vaginatum, which can enhance CH4 release and increase C substrate input to methanogens (Greenup et al., 2000). 415 In general, CH4 emissions in stream-dominated fens are likely to show a quadratic response pattern in regard to their distance to the stream, with low emissions occurring at both the closest and farthest locations from the stream, mainly due to high oxygen concentrations and water depletion, respectively. The highest CH4 emissions were found in places at intermediate distances to the stream, which benefit from both sufficient water and nutrient supply but have inherently low soil oxygen concentrations. However, we acknowledge the challenge of defining the stream at our site due to the seasonal variation in 420 catchment hydrological conditions. Hence, this study only demonstrates the stream-CH4 emission pattern, rather than providing quantitative information for projections.

Future peatland CH4 emission trajectories under climate change
Projection of global peatland CH4 emissions under different climate change scenarios is a major challenge due to the reported variabilities in emissions, and also because of the interactions between the various environmental predictors (Strack 425 and Waddington, 2007;Strack et al., 2004;Weltzin et al., 2000;Zhang et al., 2002). Our study further highlights that the impacts of climate change on CH4 emissions in flow-through peatland systems are even more complicated due to the additional effects of the flowing water, which poses a challenge for accurate predictions of the global CH4 budget.   respectively (e.g., Mäkiranta et al., 2018;Roulet et al., 1992;Yavitt et al., 2019).
The majority of peatlands are located in northern high latitudes where the climate is currently experiencing a greater rate of change than in other regions (Collins et al., 2013). Climate warming is expected to promote microbial activity, and therefore faster C cycling. However, warming in tandem with drying has been shown to decrease the methanogenic archaea 440 populations (Peltoniemi et al., 2015). In our study, vegetation composition, as such, was not a significant controller of CH4 emissions, although biomass production (GPP and LAI), influenced by the stream, was a very important controller as it likely provides substrates for methanogens. However, this is in contradiction with the suggestion that vegetation mainly influences CH4 emissions at minerotrophic sites by facilitating transportation, while at ombrotrophic sites it is through substrate-based interactions regulated by plant photosynthetic activity (Oquist and Svensson, 2002). Climate warming and/or 445 peat surface drying can alter vegetation composition and affect the contribution of the biomass that is produced. For example, shrubs can benefit from these environmental changes, while forbs and mosses may suffer (Kokkonen et al. 2019, Mäkiranta et al., 2018. Even though such hydroclimatic impacts on vegetation might be modified by nitrogen availability (Luan et al., 2019), high latitudes generally experience little nitrogen deposition (Du et al., 2020). The abundance and functional types of the plants, especially graminoid plants, regulate CH4 fluxes, but such impacts might be overruled if 450 the water table level drops substantially (Riutta et al., 2020). In addition, there is some evidence of microtopographic differences in peatland nutrient dynamics in response to drying (Macrae et al., 2013), whereas flowing water will benefit the nutrient supply at a specific site. Furthermore, the expansion of shrubs, in response to drying, might potentially decrease peat temperatures due to increased shading and the evaporative cooling effect (Strakova et al., 2012), and thereby reduce CH4

emissions. 455
Flowing water also tends to keep the peat temperature lower, even though fens with moving water warm up earlier than other peatlands in the spring and early summer (Rydin and Jeglum, 2013). In contrast to temperature predictions, predicting precipitation remains more uncertain, although in general, it is expected to increase in several regions (Collins et al., 2013).
Although peatland hydrological processes are not directly impacted by precipitation due to, for example, evapotranspiration and runoff, it has been shown that precipitation can decrease CO2 uptake and GPP due to cloudiness and associated reduced 460 light availability (Nijp et al., 2015), thus influencing CH4 emissions. Precipitation can also cause more dynamics of water and decrease CH4 emissions by providing more oxygen for CH4 oxidation (Mitchell andBranfireun 2005, Radu andDuval 2018), which can be further accelerated under a warmer and drier peat surface scenario.

Conclusions 465
Our data from a flow-through valley fen demonstrates that hydrology in northern fen systems has a dual role in controlling CH4 emissions, depending on the presence or absence of a stream. Flowing water not only enhances nutrient transportation and oxygen availability, but also decreases peat temperature, all of which are significant direct or indirect controllers of CH4 emissions. At places close to the stream there were higher water levels, lower peat temperatures, and greater oxygen concentrations; these supported the highest total leaf area and gross primary production rates but resulted in the lowest CH4 470 emissions. Further from the stream, the conditions were drier and CH4 emissions were also low. CH4 emissions were highest in the intermediate distance from the stream where oxygen concentration in the surface peat was low but gross primary production was still high. Our results show how a stream controls CH4 emissions in a flow-through fen, which is a common peatland ecosystem type from the arctic to the temperate zones. Therefore, future projections of the global CH4 budget need to take into account flowing water features in fen systems. 475 Figure

Data availability
The data used in this study will be made available on the Figshare repository after the article is accepted for publication.