Dynamic C and N stocks – key factors controlling the C gas exchange of maize in heterogenous peatland

. The drainage and cultivation of fen peatlands cre-ate complex small-scale mosaics of soils with extremely variable soil organic carbon (SOC) stocks and groundwater levels (GWLs). To date, the signiﬁcance of such sites as sources or sinks for greenhouse gases such as CO 2 and CH 4 is still unclear, especially if the sites are used for cropland. As individual control factors such as GWL fail to account for this complexity, holistic approaches combining gas ﬂuxes with the underlying processes are required to understand the carbon (C) gas exchange of drained fens. It can be assumed that the stocks of SOC and N located above the variable GWL – deﬁned as dynamic C and N stocks – play a key role in the regulation of the plant- and microbially mediated CO 2 ﬂuxes

−127 ± 212 g C m −2 yr −1 in HS. While static SOC and N stocks showed no significant effect on C fluxes, SOC dyn and N dyn and their interaction with GWL strongly influenced the C gas exchange, particularly NEE and the GPP : R eco ratio. Moreover, based on nonlinear regression analysis, 86 % of NEE variability was explained by GWL and SOC dyn . The observed high relevance of dynamic SOC and N stocks in the aerobic zone for plant and soil gas exchange likely originates from the effects of GWL-dependent N availability on C formation and transformation processes in the plant-soil system, which promote CO 2 input via GPP more than CO 2 emission via R eco .
The process-oriented approach of dynamic C and N stocks is a promising, potentially generalisable method for system-oriented investigations of the C gas exchange of groundwater-influenced soils and could be expanded to other nutrients and soil characteristics. However, in order to assess the climate impact of arable sites on drained peatlands, it is always necessary to consider the entire range of groundwater-influenced mineral and organic soils and their respective areal extent within the soil landscape. soil organic carbon (SOC) mineralisation rates and the associated CO 2 emissions (Couwenberg et al., 2010;Kasimir-Klemedtsson et al., 1997;Nykänen et al., 1995), resulting in the creation of small-scale mosaics of soil types with extremely variable SOC stocks, especially in the case of fens. The respective soil types range from deep peat soils to humus-rich sandy soils, which are not classified as peat soils due to an SOC content of < 12 % (IUSS Working Group WRB, 2007). These individual soil types are typically found at similar relative elevations within an increasingly undulating landscape, and the groundwater level (GWL) is often subject to considerable short-term fluctuations. As a result of the tight coupling between soil types and elevation, mean GWL may differ considerably between individual soil types (Aich et al., 2013;Heller and Zeitz, 2012;Dawson et al., 2010;Teh et al., 2011;Dexler et al., 2009;Müller et al., 2007;Schindler et al., 2003). These sites are typically used as grassland or cropland (Joosten and Clark, 2002;Byrne et al., 2004).
The relevance of these soil type mosaics originating from drained fen peatlands as a source or sink for greenhouse gases such as CO 2 and CH 4 , especially if the land is used for cropland, still cannot be exactly determined. In particular, knowledge about the influence of variable soil C stocks on the C gas exchange is still limited. In light of the extreme complexity of site conditions, it seems unlikely that the common focus on interactions between C stocks and particularly relevant control parameters such as groundwater and temperature (Adkinson et al., 2011;Berglund et al., 2010;Kluge at al., 2008;Jungkunst and Fiedler, 2007;Daulat et al., 1998) will result in reliable and generalisable conclusions about the C gas fluxes in degraded fens, mainly because this approach fails to account for the plant-induced C gas input counteracting the C gas emissions determined by soil characteristics and microorganisms.
Therefore, new insights are much more likely to be derived from system-oriented studies analysing all interrelated C gas fluxes, e.g. CH 4 exchange, CO 2 uptake during photosynthesis and CO 2 emission via respiration, together with the underlying processes and control mechanisms (Chapin III et al., 2009;Schmidt et al., 2011). Indeed, there are numerous indications suggesting that this approach may also be promising for the C gas exchange of drained fen sites.
Short-and long-term fluctuations in GWL and its interactions with soil and plants very likely also play a key role in the C cycle of other groundwater-influenced soil types similar to true peat soils (Couwenberg et al., 2011;Berglund and Berglund, 2011;Flanagan et al., 2002;Augustin et al., 1998;Martikainen et al., 1995;Nykänen et al., 1995). For peat soils, many studies have documented the impact of GWL on the interactions between soil C dynamics and gaseous C emissions in the form of CH 4 and CO 2 , the latter originating from autotrophic root respiration and heterotrophic microbial respiration. Ultimately, these GWL effects are a result of the ratio between SOC stocks located in the aerobic, i.e. above-GWL, and the anaerobic, i.e. below-GWL, zone (Laine et al., 1996). However, very few (Leiber-Sauheitl et al., 2014;Jans et al., 2010;Jungkunst et al., 2008;Jungkunst and Fiedler, 2007) studies have investigated Gleysols and groundwaterinfluenced sandy soils, which make up a significant portion of fen landscapes. It also remains unclear whether the impact of GWL on the gas exchange is modified by the highly variable density typical of the SOC-rich soil horizons of drained peatlands.
Knowledge gaps also limit the quantification of direct GWL effects on plant-mediated CO 2 uptake via photosynthesis. Site-adapted plants growing on undisturbed peat soils and perennial grasses cultivated on groundwater-influenced soils can tolerate changing GWL without considerable deterioration of photosynthetic performance (Farnsworth and Meyerson, 2003;Crawford and Braendle, 1996). In contrast, GWL fluctuations likely have a particularly strong impact on annual crops cultivated on drained peatlands, as most crops typically react to waterlogging, i.e. anoxic soil conditions as a result of high GWL, with reduced photosynthesis, plant respiration and growth (Zaidi et al., 2003;Asharf, 1999;Singh, 1984;Wenkert et al., 1981). Other studies indicate that crops cultivated on groundwater-influenced soils feature better growth when GWLs are low (Glaz et al., 2008), but it is unclear whether this is a direct result of improved aeration or an indirect effect of increased soil volume, allowing for better root development and thus increased nutrient uptake (Glaz et al., 2008;Livesley et al., 1999).
Despite the system-orientated approach mentioned above, it can therefore be assumed that the amounts of soil C and N located above the temporally variable GWL -hereafter referred to as dynamic C and N stocks -are of essential relevance to plant-and microbially mediated C gas fluxes on drained peatland soils. Moreover, investigations into the effects of dynamic C and N stocks may yield new insights into the mechanisms controlling the C dynamics at these sites. This would be a significant advancement with respect to a comprehensive and generalisable understanding of the CO 2 and CH 4 source and sink capacity of drained arable fen peatlands.
The present study tests the above-mentioned assumption by means of multiyear manual-chamber measurements, subsequent modelling and complex statistical analysis of all relevant C gas fluxes in maize cultivated on different groundwater-dependent soil types representing a steep SOC gradient; the gas fluxes are the net CO 2 exchange resulting from gross primary production (plant photosynthesis) and ecosystem respiration (sum of plant and soil respiration) and the CH 4 exchange. In particular, the study focuses on answering the following research questions: 1. Are there differences among soil types regarding the dynamics and the intensity of the C (CO 2 and CH 4 ) gas exchange of drained arable peatland soils?
2. (a) Which factors and factor interactions influence the C gas exchange of drained arable peatland soils? (b) In particular, what is the influence of the amount and the dynamics of soil C and N stocks located in the aerobic zone above GWL on the C gas exchange of drained arable peatland soils?
2 Materials and methods

Site description and land use history
The study sites are located near the village of Paulinenaue, in the shallow and drained peatland complex "Havelländisches Luch" of NE Germany (51 km west of Berlin; 52 • 41 N, 12 • 43 E). This peatland was first drained at the beginning of the 14th century (Behrendt, 1988). A systematic amelioration for the entire Luch took place from 1718 until 1724 and included the construction of ditches and dams to drain the formerly swampy terrain and to provide access to the land. Grasslands with hay production dominated the Luch at that time. In order to prevent repeated flooding and to increase grassland productivity, a second amelioration with deeper drainage ditches was implemented between 1907 and 1925. A substantial increase in total ditch length occurred between 1958 and 1961, when approx. 1000 km of new ditches were established in the area (Behrendt, 1988). The next huge effort to increase productivity was started in the early 1970s by the so-called "Komplexmelioration", which lasted until the late 1980s. The basic idea was to establish a system of pumping stations and related ditches in order to increase and lower the groundwater table dynamically throughout the vegetation period depending on the actual plant water demand. In addition, fertiliser application rates, including organic manure, increased and the acreage of arable land doubled at the expense of grassland. After the reunification of Germany in 1989, a substantial de-intensification took place, resulting in the reconversion of arable land to grassland, the reduction of fertiliser input and the abandonment of hydraulic technical devices for economic reasons. The region is characterised by a continental climate with a mean annual air temperature of 9.2 • C and a mean annual precipitation of 530 mm .
The study sites are located along a representative and steep landscape gradient in terms of soil organic carbon stocks (SOC stocks ; 0-1 m), which is related to topographic position (Table 1): AR, a Haplic Arenosol developed from aeolian sands with low SOC stocks (8 kg C m −2 m −1 ) in an elevated topographic position (29.6 m a.s.l.); GL, a Mollic Gleysol developed from peat overlying fluvial sands with medium SOC stocks (38 kg C m −2 m −1 ) at 29.0 m a.s.l.; and HS, a Hemic Histosol developed from peat featuring high SOC stocks (87 kg C m −2 m −1 ) at the edge of a local depression (28.8 m a.s.l.). Moreover, the vertical distribution of C and N differs between sites: at AR almost all SOC and N is concentrated in the plough layer (Ap horizon), whereas GL and HS show larger portions of SOC and N in subsoil horizons ( Fig. S1 in Supplement).
All sites were identically managed during the study period (Table S1 in Supplement), i.e. cultivated with a monoculture of grain maize with annually changing varieties. The AR and HS sites are located 150 m apart within the same managed field, while GL is located 1.5 km from AR and HS. However, field operations such as tillage, sowing, fertilisation and harvest were conducted almost concurrently at all sites. Maize was fertilised with diammonium phosphate (DAP) containing 22 kg N ha −1 and 24 kg P ha −1 in the course of sowing, followed approx. 2 weeks later by fertilisation with calcium ammonium nitrate (CAN) containing 100 kg N ha −1 . During harvest, total plant biomass within the measurement plots was collected, chipped, dried at 60 • C to constant weight and weighed. Grain yield was not recorded due to technical complications. Total plant biomass subsamples were analysed for C content at the ZALF Central Laboratory. After harvesting, all sites were mulched and ploughed.

Environmental controls
Half-hourly values of air temperature (20 cm height), soil temperatures (2, 5 and 10 cm depth), PAR (photosynthetically active radiation) and precipitation were continuously recorded by a climate station installed within 1 km of the sites. Site-specific air and soil temperatures were manually measured simultaneously with CO 2 and CH 4 flux measurements. Site-specific half-hourly air and soil temperature models were derived from correlations between the respective climate station temperature records and site-specific manual temperature data. Sunshine hours and long-term climate data originate from the Potsdam station of the German Weather Service (DWD). GWL at GL and HS was measured manually every 2 weeks using short 1.5 m dip wells. The measured piezometric heads are considered representative of the phreatic water levels in the peat layer because the organic soil layer directly overlies a sand aquifer without any major lowconductance soil horizons in between. At HS, GWL was additionally recorded every 15 min by a data logger (Minidiver, Schlumberger). Time series modelling was used to fill several small data gaps and to obtain continuous daily GWL data for the entire study period. The applied PIR-FICT (predefined impulse response function in continuous time) approach (von Asmuth et al., 2008) implemented in the Menyanthes software (von Asmuth et al., 2012a) is a physically based statistical time series model specifically developed to model hydrologic time series, including shallow GWL fluctuations. As input, the model requires continuous precipitation (DWD station Kleßen) and evapotranspiration data (FAO56 Penman-Monteith; DWD station Kyritz) and optional control parameters, e.g., in our case, deep GWL data recorded from a local dip well (Ministry of Rural Development, Environment and Agriculture -LUGV, Brandenburg). The calibrated model explained 80-87 % of the data variance, a good result for this data and model type (von Asmuth et al., 2012b). Confidence intervals of GWL time series predictions were obtained by means of stochastic simulation (see von Asmuth et al., 2012a). Due to the short distance between AR and HS and the highly significant correlation of GWL at these sites (R 2 = 0.836), daily GWL values for AR were calculated by shifting the modelled time series of HS with a constant offset of 0.9 m.

Concept and calculation of dynamic C and N stocks
The concept of "dynamic" groundwater-dependent C and N stocks was developed to account for the interaction of the most important drivers of the C gas fluxes in peatlands, namely GWL and soil C and N stocks. The underlying idea is to derive a quantitative, dynamic proxy for the aerated, unsaturated zone which determines the actual nutrient and O 2 availability and is therefore highly relevant for root and shoot growth, microbial activity and, consequently, all C gas fluxes. Using daily GWL data, it was determined for each 1 cm soil layer up to a depth of 1 m whether the respective layer was saturated with groundwater or not. In daily time steps, SOC and N stocks were then calculated for all non-saturated 1 cm layers and cumulated over the entire nonsaturated soil profile, i.e. above GWL, to generate daily dynamic SOC (SOC dyn ) and N (N dyn ) stocks. For further analysis, daily SOC dyn and N dyn values were averaged monthly and annually.

Gas flux measurements
Periodic trace gas measurements were carried out at three permanently installed soil collars (0.75 × 0.75 m) at each site. In the summer of 2007, due to flooding, soil collars at the HS site had to be relocated within a radius of 10 m to (i) technically allow for gas flux measurements and (ii) ensure that all soil collars contained flood-affected but viable plants in order to maintain comparability with the GL and AR sites, where maize mortality was not increased by flooding.
Throughout the entire study period, CH 4 measurements were conducted one to two times per month using static non-flow-through non-steady-state opaque chambers (vol. 0.296 m 3 ; Livingston and Hutchinson, 1995;Drösler, 2005) for a total of 51-60 campaigns per site. At HS, CH 4 measurements were terminated already in October 2010 due to management constraints. The exchange of CH 4 was measured by taking four consecutive 100 mL gas samples from the chamber headspace at 20 min intervals (closure time 60 min), subsequently analysed using a gas chromatograph (Shimadzu GC 14B, Loftfield, Göttingen, Germany) equipped with a flame ionisation detector.
CO 2 exchange was measured using dynamic flow-through non-steady-state transparent (net ecosystem exchange -NEE; light transmission of 86 %) and opaque (ecosystem respiration -R eco ) chambers (Livingston andHutchinson 1995, Drösler 2005) attached to an infrared gas analyzer (Li-820, Lincoln, NE, USA). Full-day CO 2 measurement campaigns with repeated (30-50) individual chamber measurements (closure time 3-5 min) were conducted regularly every 4-6 weeks from May 2007 to April 2011, for a total of 29-37 full campaigns per site. Further details on the CO 2 measurement methodology are given in Hoffmann et al. (2015).

Flux calculation and gap filling
Flux calculation for CO 2 and CH 4 was based on the ideal gas equation accounting for chamber volume and area, air pressure, and average air temperature during the measurement. CH 4 fluxes were calculated with the R package "flux 0.2-2" (Jurasinski et al., 2012), using linear regression analysis with stepwise backward elimination of outliers based on the normalised root mean square error (NRMSE ≥ 0.2) up to a minimum of three data points. Fluxes with NRMSE > 0.4 were rejected. The calculated flux rates were then averaged for the respective measurement day and linearly interpolated to determine annual CH 4 exchange.
For CO 2 , the R script of Hoffmann et al. (2015) was used for flux calculation as well as the subsequent separation into and modelling of R eco , gross primary production (GPP) and NEE. Measurements <30 s were rejected and measurements > 1 min were shortened by a death band of 10 % at the beginning and end, respectively (Kutzbach et al., 2007). For each measurement, the final flux rate was selected from all potential flux rates generated by a moving window approach using a stepwise algorithm, numerous quality criteria and the Akaike information criterion (AIC; for details see Hoffmann et al., 2015). For R eco , gap filling between measurement campaigns was performed using campaign-specific temperature-dependent Arrhenius-type models by Lloyd and Taylor (1994). GPP fluxes were calculated by subtracting modelled R eco fluxes from measured NEE fluxes, and then modelled using campaign-specific hyperbolic PARdependent models (Wang et al., 2013;Elsgaard et al., 2012;Michaelis-Menten, 1913). Average measured flux rates were used if no significant fit was achieved for campaign-specific R eco or GPP models (Hoffmann et al., 2015). Half-hourly NEE values were calculated from modelled R eco and GPP fluxes (Hoffmann et al., 2015;Drösler, 2005) and cumulated from 1 May to 30 April of the following year (Table S1), resulting in four consecutive annual CO 2 balances. Negative values represent a C gas flux from the atmosphere to the ecosystem; positive values a flux from the ecosystem to the atmosphere. The uncertainty of the annual CH 4 and CO 2 exchange was quantified using a comprehensive error prediction algorithm described in detail by Hoffmann et al. (2015).

Data analysis
Daily values for CH 4 efflux, GPP, R eco and NEE were cumulated monthly for a total of 48 monthly data sets per site to reduce the effects of temporal autocorrelation. The respective environmental controls were cumulated (sunshine hours, precipitation and linear modelled biomass) or averaged (for GWL, SOC dyn , N dyn , and air and soil temperature) for each month. Gas flux balances for longer time periods may vary considerably depending on the duration of the respective cumulation period. As the wavelet analysis of daily NEE data for inherent signals revealed strong annual dynamics (Stoy et al., 2013; Fig. S2), a 365-day cumulation period was used to calculate gas flux balances. Additional variability in annual balances can result from arbitrarily chosen starting dates of the cumulation period. To account for this uncertainty in the calculation of annual balances, a 365-day moving window was shifted in monthly time steps through the entire study period, resulting in a total of 111 data sets (37 per site) for annual NEE, GPP, R eco and CH 4 efflux and the respective environmental control parameters.
Subsequently, generalised linear model (GLM) analyses (SPSS generalised linear -GENLIN -procedure) were performed to determine the influence of environmental controls and their interactions on the cumulated annual CH 4 , R eco , GPP and NEE balances as well as the GPP : R eco ratio. Models were defined using a gamma probability distribution and a log link function and calculated in a stepwise backward elimination procedure, dropping non-significant variables until no further improvement of the AIC was achieved (correction for finite sample sizes: AIC c ). Parameter and interaction effects were evaluated based on the Wald χ 2 statistic, appropriate for non-normally distributed continuous variables. Prior to analysis, CH 4 data were log-transformed after adding the minimum CH 4 value to each data value in order to allow for the application of the GLM log link function. Analogously, absolute values of GPP were used for the analysis and NEE data were transformed to positive values by adding the minimum NEE value to each data value.
Multiple nonlinear regression analyses were performed to derive a model for NEE based on GWL and SOC dyn , N dyn , SOC dyn : N dyn ratio and biomass, representing the main GLM parameter groups. For model calculation, data were averaged for 12 site-specific GWL classes to account for uncertainty from GWL model data. The class number was determined using Sturges' rule, appropriate for n < 200 (Scott, 2009). All data analyses were performed using the R (R 3.0.3) and SPSS (SPSS 19.0.1, SPSS Inc.) software.

Environmental controls
During the study period (May 2007-April 2011), weather conditions were somewhat cooler (8.7 • C) and wetter (634 mm) compared to the long-term average (1982-2012; 9.2 • C; 530 mm). Particularly the 2010/11 measurement year considerably deviated from the long-term temperature average, with an annual air temperature that was 1.5 • C below the long-term average -1 SD (data not shown). While PAR and air temperature showed high daily and seasonal dynamics (Fig. 2a), no pronounced seasonal patterns were observed for precipitation (Fig. 1). Instead, precipitation featured an extremely high interannual variability with particularly heavy rainfalls during the summer months of 2007 (May-July; Fig. 1). The precipitation sum during this period (507 mm) exceeded the long-term average (179 mm) by > 180 % (data not shown). Reflecting the precipitation dynamics, GWL showed similar temporal dynamics at the three sites, but at different levels. In summer, GWL remained generally low, with the exception of July-August 2007. The HS site, which consistently featured the highest average GWL (−0.5 m; Fig. 1, Table S2), was flooded during this period (GWL + 0.2 m; data not shown).
The SOC dyn and N dyn stocks calculated based on the modelled GWL showed the highest fluctuations at the HS site (Fig. 1). During times of high GWL, such as in summer 2007, the HS and GL site featured drastically lowered SOC dyn and N dyn values, amounting to only 6.2 kg C m −2 and 0.5 kg N m −2 , respectively, with SOC dyn and N dyn reduced to 0 during flooded periods. In contrast, pronounced peak values at HS were calculated for the low-GWL summer months during the rest of the study period, with monthly averages of 21-86 kg C m −2 and of 2-5 kg N m −2 . The HS site always featured the highest annual SOC dyn (52 kg C m −2 ) and N dyn (4 kg N m −2 ) stocks, except in 2007/08 ( Fig. 1; Table S2).

Daily and annual carbon gas exchange
All sites generally featured very low daily CH 4 fluxes (−0.01 to 0.01 g CH 4 -C m −2 d −1 ) throughout the study pe- Figure 1. Seasonal dynamics of (from top to bottom) daily precipitation, average daily GWL including 95 % confidence intervals (dotted lines), and daily dynamic SOC dyn and N dyn stocks by site (for 0-1 m depth). riod (Fig. S3). However, considerable CH 4 emission peaks were observed at the HS and GL sites during times of flooding or high GWL, e.g. during summer 2007 and spring 2008. At HS, this resulted in a maximum CH 4 flux of 1.2 g CH 4 -C m −2 d −1 on 1 August 2007, which is approx. 60 times higher than the median flux (0.02 g CH 4 -C m −2 d −1 ) at this site. As a result of the flooding, annual CH 4 emissions in 2007/08 at HS amounted to 28 ± 4 g CH 4 -C m −2 yr −1 and were thus nearly 100 times higher than observed for HS in the following years (0.3 ± 0.5 and ± 0.2 g CH 4 -C m −2 yr −1 ) and at least 25 times higher than observed for AR and GL (< 1.2 ± 0.6 g CH 4 -C m −2 yr −1 ; Table 2). However, as the high annual CH 4 emissions at HS in 2007/08 result from a peak described by three measurement campaigns during the flooded period (Fig. S3), they are also associated with a higher uncertainty (± 3.7 in 2007/08 vs. ±0.5 and ±0.2 g CH 4 -C m −2 yr −1 in 2008/09 and 2009/10; Table 2).
The modelled CO 2 exchange rates (for model evaluation statistics, see Table S3) reflected the daily and seasonal dynamics of air temperature and PAR, with generally higher fluxes in the growing season compared to fall and winter (Fig. 2a, b). In summer, peak GPP fluxes considerably exceeded the amplitude of R eco fluxes. At all sites, the CO 2 ex-change was also influenced by management events, with particularly pronounced peaks of R eco following tillage. In addition, GPP was immediately reduced to 0 after maize harvest due to the removal of the photosynthetically active aboveground plant biomass. In general, the organic GL and HS sites showed the highest CO 2 exchange intensity, with maximum R eco and GPP fluxes of 23 and −46 g CO 2 -C m −2 d −1 , respectively, observed at the HS site (Fig. 2a, b). However, during the wet summer of 2007, the mineral AR site featured the highest intensity of CO 2 exchange, resulting in cumulated annual R eco and GPP fluxes that were 25-44 and 52-61 % higher, respectively, than in the following years (2008-2011, Table 2). In contrast, at HS, the 2007 flooding resulted in strongly reduced CO 2 flux intensities and large net annual CO 2 -C losses (NEE of 493 ± 83 g CO 2 -C m −2 ) compared to the following years. Although the CO 2 fluxes measured during the flooded period are associated with higher error values compared to periods without flooding (Table 2), the modelled results are plausible, clearly reflecting the negative effects of flooding on plant growth and thus plant C exchange. Hence, in 2007/08, cumulated annual R eco and GPP fluxes at AR were 76 and 49 % higher than at the HS site (Table 2). Excluding 2007/08, the average NEE during the study period at the mineral AR site was close to 0, with 50 ± 32 g CO 2 -C m −2 yr −1 (Table 2), whereas the organic sites were net CO 2 -C sinks, with −385 ± 133 (GL) and −334 ± 61 g CO 2 -C m −2 yr −1 (HS). Including the flooddominated year of 2007/08 resulted in a 62 and 21 % reduction in the overall NEE at the HS and GL sites, respectively. In contrast, when 2007/08 is included in the overall 2007-2011 average for the AR site, cumulated R eco and GPP increase by 63 and 67 %, respectively, while NEE remains unaffected.

Impact of environmental controls on carbon gas exchange
Despite the wide range of control parameters included in the complex analysis, site (i.e. soil) had a significant (p value ≤ 0.05) effect on all gas fluxes (Table 3). The generally highly significant (p value ≤ 0.001) interactions between site and controls such as biomass, GWL and soil parameters show that the selected study sites represented a wide range of the respective control parameters. Especially annual CH 4 -C emissions were dominated by site, suggesting the presence of additional important control factors not considered in this analysis. However, little residual variability indicates that most of the variability in annual R eco and GPP was explained by the factors included in the GLM analyses, with more residual variability remaining for NEE and the GPP : R eco ratio. While climate played a minor role in determining annual CH 4 -C emissions via the effect of precipitation on GWL, climate controls were more relevant for CO 2 exchange (Table 3). There, the importance of climate was higher for cumulated GPP and R eco than for NEE and the GPP : R eco ratio. The impact of climate variability on CO 2 exchange was even more pronounced on the monthly scale, as indicated by highly significant interactions between climate controls and month of year (data not shown). Biomass was as imwww.biogeosciences.net/12/2737/2015/ Biogeosciences, 12, 2737-2752, 2015   portant as climate in determining annual GPP, whereas for R eco , biomass and its interactions were less relevant than climate (Table 3). In contrast, the derived variables NEE and GPP : R eco were less influenced by biomass than the individual fluxes R eco and GPP.
Direct groundwater influence was particularly pronounced for R eco , GWL being by far the most important GLM parameter (Table 3). Groundwater influence on CH 4 -C emissions and the GPP: R eco ratio was expressed mainly through the interaction between GWL and site. Table 4. Summary statistics of multiple nonlinear regression analysis of the form NEE = poly(GWL) + liny (1;2;3 or 4) describing the influence of GWL and one environmental parameter, either (1) SOC dyn , (2) N dyn , (3) SOC dyn : N dyn or (4) biomass, on cumulative annual NEE: mean absolute error (MAE), RMSE observations standard deviation ratio (RSR), adjusted coefficient of determination (Adj. R 2 ), modified index of agreement (md), percent bias (PBIAS), Nash-Sutcliffs model efficiency (NSE), Akaike Information Criterion (AIC) and Bayesian information criterion (BIC). Groundwater-dependent soil parameters and their interactions with site and GWL dominated annual CH 4 -C emissions (Table 3). Soil parameters were also the main controls on NEE, particularly the SOC dyn : N dyn ratio and its interactions with site. Dynamic soil parameters and their associated interactions thus were of higher relevance for the derived variables NEE and GPP : R eco than for the NEE flux components R eco and GPP. This indicates differences between R eco and GPP with respect to their reaction to changing GWL and soil parameters, i.e. a shift in the ratio between R eco , and GPP throughout the range of GWL, SOC dyn and N dyn stocks. In contrast, static SOC stocks and N stocks showed no significant (p value ≥ 0.05) effect on cumulated annual or monthly fluxes of either R eco , GPP or NEE (data not shown).
Nonlinear regression analysis of annual NEE versus GWL and either SOC dyn , N dyn , SOC dyn : N dyn or biomass across all sites resulted in highly significant two-parameter models (Table 4; Fig. 3). While all models explained > 86 % of the overall variability of annual NEE, model fit was best for GWL and SOC dyn , likely because the study sites represent a wide range of SOC dyn . For all sites, the model shows a negative NEE optimum for a GWL of 0.8-1.0 m below the soil surface, with NEE increasing at higher or lower GWL (Fig. 3). In contrast, the model reflects a linear effect of SOC dyn on NEE with more negative NEE for higher SOC dyn . Depending on SOC dyn , NEE changes to positive values at a GWL above −0.43 m (for SOC dyn = 60 kg C m −2 ) or −0.61 m (SOC dyn = 30 kg C m −2 ). However, the relations shown cannot be assumed as valid outside the measured ranges of SOC dyn and GWL.  Table 4). Displayed grid represents the derived model surface with (i) estimated model area covered by direct measurements (solid black) and (ii) non-empirically approved model area computed by extrapolation (grey). Modelled NEE is separated according to positive (solid lines) and negative (dashed lines) values.

Soil influence on C gas exchange
As indicated in the introduction, data about the CO 2 exchange of groundwater-influenced arable soils are generally scarce, particularly for maize, although some data are available for organic soils. Although the maximum CO 2 fluxes observed during a 1-year study of maize cultivated on a Haplic Gleysol in the Netherlands  are ∼ 25 % lower compared to the Gleysol studied (Fig. 2), the flux dynamics and the cumulative net CO 2 exchange of the organic soil types are relatively similar in both studies, with a mean annual NEE of −385 (Gleysol) and −334 g CO 2 -C m −2 yr −1 (Histosol) in this study (Table 2) vs. −332 g CO 2 -C m −2 yr −1 . Moreover, the dynamics and the intensity of the CO 2 exchange observed for the groundwater-influenced soils in this study are of the same order of magnitude as reported for maize cultivated on soils without groundwater influence (Gilmanov et al., 2013;Kalfas et al., 2011;Zeri et al., 2011;Ceschia et al., 2010). The observed biomass yield of maize (257-3117 g DM m −2 yr −1 : DM -dry matter) is also in line with previous studies (500-2800 g DM m −2 yr −1 ; Zeri et al., 2011;Verma et al., 2005). According to Gilmanov et al. (2013) and Ceschia et al. (2010), maize cultivation generally resulted in a net annual CO 2 sink across a wide range of sites in America and Europe but -like in this studywith considerable variability between sites and years (+89 to −573 g CO 2 -C m −2 yr −1 ).
The results of this study demonstrate for the first time a considerable influence of groundwater-influenced soils on crop CO 2 exchange, particularly on cumulative NEE (Tables 3, 4, Fig. 3), thus clearly providing a positive answer to research question (1) regarding the soil effect. Surprisingly, the C-rich drained organic soils showed a strong net CO 2 uptake (Table 2), while the C-poor Arenosol was a small net CO 2 source. This observation cannot be entirely explained by the interaction between GWL and the potentially mineralisable soil C stocks. Hence, an integrated consideration of all relevant C gas fluxes and their regulation within the plantsoil system is required, which is discussed in detail below. We are unaware of any previous study ever reporting such an effect, likely because any systematic effects may only be observed in longer-term studies due to the high interannual variability of C gas fluxes. This strongly supports the high relevance of such investigations for the accurate evaluation of the C dynamics of groundwater-influenced arable soils.

Relevance of interactions between GWL and maize ecophysiology
Apart from soil type and SOC content, the study sites are mainly differentiated by different average GWL, which our study results show to be a crucial factor determining the high short-and long-term variability of maize C gas exchange across the entire range of groundwater-influenced soils. Previous studies have mainly shown an influence of GWL on CH 4 fluxes from peat soils, mainly reporting an exponential increase in CH 4 fluxes for rising GWL with particularly high CH 4 losses for GWL ≥ -0.2 m (Couwenberg et al., 2011;Jungkunst and Fiedler, 2007;Drösler, 2005;Fiedler and Sommer, 2000). Annual CH 4 emissions (−0.2 to 1.2 g CH 4 -C m −2 yr −1 ) for GWL between −1.6 and −0.6 m and peak fluxes during flooding (≤ 28 g CH 4 -C m −2 yr −1 ; GWL of −0.3 m) observed at the HS site are similar to values of Couwenberg et al. (2011) andDrösler (2005). However, for crops cultivated on groundwater-influenced mineral soils, little data are available on the impact of GWL on CH 4 fluxes (e.g. Pennock et al., 2010). CO 2 exchange has also been intensively studied for organic soils, but mostly for pristine peatlands and grasslands on peat soils (e.g. Leiber-Sauheitl et al., 2014;Berglund and Berglund, 2011;Couwenberg et al., 2011), while data on maize are lacking. For peatland NEE, one study reports a linear decrease with rising GWL over a range of −0.4 to −0.1 m, with maximum NEE observed at −0.4 m (Leiber-Sauheitl et al., 2014). Couwenberg et al. (2011) also observed decreasing NEE when GWL rose above −0.5 m, but net CO 2 -C uptake was only reported for very high GWL above −0.1 m. In contrast, in this study, maize NEE was largely negative across the entire range of GWL recorded in the groundwater-influenced soils studied (−2.1 m to +0.2 m), changing to positive values when GWL rose above −0.4 m to −0.6 m. Moreover, the GWL-NEE relationship for maize shows a clearly nonlinear relationship to GWL, with a distinct optimum at considerably lower GWL (between −0.8 and −1.0 m; Fig. 3) than observed for grasslands. Further studies are required to determine whether this is a general pattern applicable to other groundwater-influenced soil types and crops.
Our study results further indicate that R eco and GPP also feature specific GWL optima (data not shown). For example, maximum R eco fluxes were observed for GWL of −0.8 to −1.0 m, similar to data from grassland on four GWLinfluenced soil types (Fiedler et al., 1998). Similar to the R eco of maize at the organic HS and GL sites, R eco fluxes in grasslands on organic soils typically decrease with rising GWL (Leiber-Sauheitl et al., 2014;Berglund and Berglund, 2011;Laine et al., 1996;, particularly if GWL rises above the soil surface (Koebsch et al., 2013). The impact of GWL on GPP was relatively small in this study (Table 3), except for the effect of the 2007 flooding, which resulted in a drastic reduction in GPP (Table 2) as also observed by Koebsch et al. (2013) after rewetting.
Most of the study results concerning the individual CO 2 fluxes can be explained by the interactions between GWL and maize plant activity because the magnitude and the variability of GPP and R eco are most pronounced during the short period from May to September, which corresponds to the growing period of maize (Fig. 2). For example, the drastic reduction in the CO 2 fluxes during the flooding in 2007 at HS and GL (Figs. 1, 2) was very likely caused by the previously mentioned negative effect of anoxic soil conditions on maize metabolism. On the other hand, the lower CO 2 fluxes during the summer of 2009, especially at the AR site, probably result from an inhibition of maize gas exchange due to drought stress (Vitale et al., 2008;Jones et al., 1986), i.e. long periods of very low GWL (Figs. 1, 2). Apart from these extreme situations, GWLs were mostly at soil depths which were favourable to the metabolism and the productivity of a C 4 plant such as maize (Tollenaar and Dwyer, 1999).
For example, maize features considerably higher gas exchange activity under maximum PAR and temperature conditions than all C 3 grasses and crops (Zeri et al., 2011;Kutsch et al., 2010). As a consequence, although the main growing period of maize (∼ 2 months) is much shorter than that of most C 3 plants (3-4 months), the CO 2 flux intensity of maize throughout this short active period is large enough to result in higher annual cumulative R eco and GPP values compared to C 3 crops (Beetz et al., 2013;Klumpp et al., 2011;Zeri et al., 2011;Flanagan et al., 2002). It is very likely that the GWL optima of GPP and R eco can be traced back to this fact, e.g. as indicated by the enhanced amplitudes of the GPP as well as the R eco fluxes at the AR site during the wet summer 2007 compared to years with lower GWL (Fig. 2). However, the interactions between GWL and maize growth do not offer explanations for the observed differences in cumulative NEE among sites and the functional relationship between NEE and GWL.

Relevance of interactions between GWL and dynamic soil C and N stocks
The strong effect of GWL on the C gas exchange is likely also the reason for the lack of any effect of total, i.e. static, soil C and N stocks on daily, monthly or annual C gas exchange. In the few existing studies on this subject, an impact of soil C and N stocks on C gas fluxes was only found if GWL was either constant (Mundel, 1976) or irrelevant to the soil water regime (Lohila et al., 2003). Moreover, in agreement with the results of this study, Leiber-Sauheitl et al. (2014) found no relationships between static soil C and N stocks and the C gas exchange of Gleysols with highly variable GWL during a 1-year study. In contrast, our study revealed a very strong effect of mainly GWL-determined dynamic soil C and N stocks on C gas dynamics (Table 3), thus indicating a higher relevance of SOC and N stocks located in the aerobic zone above the GWL for plant and soil gas exchange than of total soil SOC stocks and N stocks in the soil profile. However, the functional GWL-related mechanisms mentioned in the introduction cannot fully explain the results of this study. Several observations indicate that the influence of the dynamic soil C and N stocks on the C gas exchange extends beyond the mere GWL effect: 1. All C gas fluxes are differently and specifically influenced by the dynamic soil C and N stocks (Tables 3, 4).
2. Compared to the GWL, the effects of dynamic soil C and N stocks on NEE are considerably stronger than on the individual R eco and GPP fluxes, also reflected by the associated shift in the GPP : R eco ratio (Table 3).
It must be pointed out that these two parameters differ in their informational value: while NEE is the absolute difference between the opposing CO 2 fluxes R eco and GPP, the GPP : R eco ratio reflects the relative proportion of these fluxes, thus giving indications for the reasons for changing NEE values. Interestingly, the dynamic C : N ratio shows a similarly strong effect on these two parameters. The potential relevance of these observations for explaining the study results is also discussed in Sect. 4.2.
3. The effects of the GWL and the dynamic soil C or N stocks on the cumulative CO 2 fluxes clearly differ with respect to their type and direction (Fig. 3, Table 4).
Despite a limited number of sites, clustering of sites with respect to GWL range and a single crop, the results of this study are considered consistent and plausible for the range of measured GWL and soil C stocks, as the results from several very different statistical methods point to the same conclusions. Still, subsequent studies which consider other sites and plants are required to determine whether the discussed conclusions regarding the type and intensity of the effect of dynamic soil C and N stocks on cumulative NEE, their differentiated effects on GPP and R eco as well as their interactions with GWL are generally valid. A reassessment of data from previous studies using continuous GWL data (if available) for the calculation of dynamic soil C and N stocks could be helpful to determine whether similarly strong effects of dynamic soil C and N stocks on C gas dynamics exist for other sites and plants. System-oriented investigations, which aim to understand the underlying processes and mechanisms, might reveal whether and how the observed phenomena are related and from which underlying processes they originate.

4.4
The nature and relevance of mechanisms causing the effect of the dynamic soil C and N stocks

Potential mechanisms
A common observation may be used as a starting point for a comprehensive explanation: crop growth on groundwaterinfluenced soils is mainly influenced by rooting depth, which in turn is mostly influenced by GWL (e.g. for maize, see Kondo et al., 2000). In this context, stress due to O 2 deprivation only plays a minor role, i.e. via the GWL-defined lower limit of the rootable soil volume (Glaz et al., 2008;Livesley et al., 1999). More importantly, larger root systems enable the improved supply of plants with nutrients and water (especially at the AR site), likely resulting in increased photosynthetic capacity and thus higher primary productivity. The link between increasing N content and increased GPP was previously documented in studies by Flanagan et al. (2002) and Ashraf et al. (1999). Interestingly, several long-term field trials with crops grown on mineral soils also show that changing SOC stocks not only depend on crop rotation and organic fertiliser amount but also on the nutrient supply to the crops per se. In these trials, the mere application of mineral fertiliser results in a significant increase in soil organic matter compared to non-fertilised treatments (Jung and Lal, 2011;Banger et al., 2010;Thomas et al., 2010;Christopher and Lal, 2007;Sainju et al., 2006). In addition to other crops, this also applies to maize (Kaur et al., 2007). In particular, the N supply plays a key role: up to a threshold, the gradual increase in mineral N fertiliser amount generally results in higher SOC and SON stocks (e.g. for maize, see Kaur et al., 2007;Blair et al., 2006a, b). Pot experiments with maize indicate that N fertilisation increases the input of newly assimilated C more than CO 2 emissions from root respiration and the mineralisation of soil organic matter (Gong et al., 2012;Conde et al., 2005), thus resulting in the accumulation of SOC. Moreover, in field trials, mineral N fertilisation reduced the decomposition rate of maize residues in the soil (Grandy et al., 2013). Therefore -apart from the impact of C export (removal during harvesting) and import (input through organic fertilisation) on the soil C budgetit seems highly likely that the N fertilisation of arable crops contributes to an increase in SOC stocks by promoting C input through gross and net primary productivity more than C loss via ecosystem respiration. Although this has not yet been experimentally confirmed in its entirety, scientific evidence on the individual effects of N fertilisation on the SOC stocks of arable soils without groundwater influence makes this hypothesis plausible.

Indications of similar mechanisms on groundwater-influenced soils
Several results of this study suggest a strong N influence on C gas fluxes. All sites received a total of 122 kg N ha −1 yr −1 throughout the entire study period, providing sufficient N for plant growth. The dynamic soil N stocks and the SOC dyn : N dyn ratio had strong effects on cumulative NEE and the GPP : R eco ratio (Table 3). Formally, this also holds true for the dynamic SOC stocks, but -unlike for N -this effect results from the tight correlation of soil C and N contents rather than from direct effects of organic-matter production or decomposition. The large influence of GWL on dynamic soil N stocks, reflected by strong interaction, indicates that both parameters control N mineralisation. It has been repeatedly observed both for organic and mineral soils that the lowering of GWL, i.e. an increase in the dynamic N stocks due to improved soil aeration, increases N mineralisation, while a rising GWL, i.e. decreasing dynamic N stocks, results in the opposite (Eickenscheid et al., 2014;McIntyre et al., 2009;Venterink et al., 2002;Hacin et al., 2001;Goettlich, 1990;Reddy and Patrick, 1975). Increased dynamic soil N stocks are equivalent to an improved N supply to plants and microorganisms, which should be similar in effect to the N fertilisation in the above-mentioned long-term field trials. In this study, the tight correlation between the dynamic soil N stocks and the maize biomass development during the vegetation period (r 2 = 0.817; data not shown) indicates that most of the N that was mineralised when GWLs were low and root systems deep likely played a significant role in plant N supply and thus plant development -regardless of the fertilisationinduced N impulse and the fact that the monthly biomass values were not measured but calculated using a simple linear approach. Similarly strong biomass and dynamic C and N stock effects on cumulative NEE (Table 3) further support this line of thought, as an increased biomass production stimulated by higher N availability is always associated with increased CO 2 input into the plant-soil system via gross primary production.
In other words, the N supply in the plant-soil system and its effects on C formation and transformation processes likely also play a key role in the C gas exchange of groundwaterinfluenced soils by promoting CO 2 input via gross primary production more than CO 2 emission via ecosystem respiration. The observed effects of the dynamic soil C and N stocks on cumulative NEE can thus be plausibly explained. However, the relatively low optimum GWL for minimising NEE (Fig. 3) likely requires additional explanatory mechanisms. For example, an improved plant water and nutrient supply, e.g. with macronutrients such as P and K, could increase root and shoot growth and thus CO 2 input, as observed for soils without groundwater influence (Ladha et al., 2011;Poirier et al., 2009;Al-Kaisi et al., 2008;Reay et al., 2008;Kaur et al., 2007).

Future improvements of the dynamic stocks concept
Most of the functional mechanisms discussed above are somewhat speculative and require subsequent validation by means of experiments which consider all mentioned processes of the plant-soil system and their respective regulating factors. Special attention should be paid to the determination of the scope of all relevant processes, as several studies state that the input of N and other nutrients does not always have only positive effects on net CO 2 exchange and the C sink function of arable soils (Thangarajan et al., 2013;Hofmann et al., 2009;Mulvaney et al., 2009;Al-Kaisi et al., 2008;Khan et al., 2007). Moreover, the concept of dynamic soil C and N stocks is only an indicator of real dynamic stocks because in this study dynamic stocks were modelled exclusively based on GWL dynamics. Further developments might include precipitationrelated topsoil water dynamics or soil hydraulic properties (e.g. capillary fringes), which might considerably reduce dynamic soil C and N stocks. The concept of dynamic stocks could also be expanded to other plant nutrients such as plantavailable P or K. However, these suggested refinements require very detailed high-resolution data on soil and plant properties and processes, including their vertical variability in the soil profile, and were thus beyond the scope of this study.

Conclusions
Results clearly showed that the soils studied differ considerably with respect to the intensity and dynamics of C gas exchange. In order to accurately assess the climate impact of arable sites on drained peatlands, it is therefore necessary to consider the entire range of groundwater-influenced mineral and organic soil types and their respective areal extent within a heterogenous soil landscape.
While climatic controls such as PAR, temperature and precipitation mainly have short-term effects on C gas fluxes, the effects of dynamic soil C and N stocks are clearly observable on all temporal scales. It is to be determined by future studies in how far this also applies to (i) crops other than maize, (ii) other land use forms such as grasslands and iii) other groundwater-influenced sites. Dynamic soil C and N stocks may be major controlling factors of C gas fluxes and of the CO 2 source or sink function of the entire range of wetlands, and they may potentially be of higher and more global relevance than GWL and vegetation, which are the main factors favoured to date (Couwenberg et al., 2011;Byrne et al., 2004). The insight that the effect of the dynamic soil C and N stocks very likely results from the regulation of C formation and transformation processes by N and -potentially -the nutrient and water supply as such may be of particular importance. This mechanism would be a favourable prerequisite for the development of generalisable process-based models, which would be very useful in providing more precise estimates of the impact of important factors such as climate, site conditions and land use on the C gas fluxes in wetlands.
Overall, the results presented and subsequent analyses show the enormous potential of combining long-term measurements of C gas fluxes with process-oriented analyses of the functional mechanisms and their regulation within the soil-plant system when aiming for an improved understanding of the biogeochemistry of wetlands.
The Supplement related to this article is available online at doi:10.5194/bg-12-2737-2015-supplement.