Sub-soil irrigation does not lower greenhouse gas emission from drained peat meadows

Current water management in drained peatlands to facilitate agricultural use, leads to soil subsidence and strongly increases greenhouse gas (GHG) emission. High-density, sub-soil irrigation/drainage systems have been proposed as a potential climate mitigation measure, while maintaining high biomass production. In summer, sub-soil irrigation can potentially reduce peat 15 decomposition by preventing groundwater tables to drop below -60 cm. In 2017-2018, we evaluated the effects of sub-soil irrigation on GHG emissions (CO2, CH4, N2O) for four dairy farms on drained peat meadows in the Netherlands. Each farm had a treatment site with perforated pipes at 70 cm below soil level spacing 5-6 m to improve both drainage (winterspring) and irrigation (summer) of the subsoil, and a control site drained only by ditches (ditch water level -60/-90 cm, 100 m distance between ditches). GHG emissions were measured using closed 20 chambers (0.8 x 0.8 m) every 2-4 weeks. C inputs by manure and C export by grass yields were accounted for. Unexpectedly, sub-soil irrigation hardly affected ecosystem respiration (Reco) despite raising summer groundwater tables (GWT) by 6-18 cm, and even up to 50 cm during drought. Only when the groundwater table of sub-soil irrigation sites was substantially higher than the control value (> 20 cm), Reco was significantly lower (p<0.01), indicating a small effect of https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c © Author(s) 2020. CC BY 4.0 License.

). In the Netherlands, 26% of the surface area is currently below sea level, an area currently inhabited by 4 million people (Kabat et al., 2009). This area is expected to increase due to further land subsidence, while sea level is rising at the 50 same time, which is a general issue of coastal peatlands (Erkens et al., 2016). Additionally, peatland subsidence alters hydrology, leading to drainage problems, salt water intrusion and loss of productive land (Dawson et al., 2010;Herbert et al., 2015). This will result in strongly increased societal costs and difficulties in maintaining productive land use ( Van den Born et al., 2016;Tiggeloven et al., 2020).

55
The peatland area used for agriculture is estimated at 10% for the USA and 15% Canada, and varies from less than 5 to more than 80% or Europe . In the Netherlands, 85% of the peatland areas are in agricultural use (Tanneberger et al., 2017), leading to CO2 emissions of 7 Mt CO2-eq per year, amounting to 4% of total national greenhouse gas (GHG) emissions (Couwenberg, 2009). Fundamental changes in the management of peatlands are required if land use, biodiversity and socio-economic values including GHG emission reduction are to be maintained. A higher groundwater table (GWT) 60 creates anaerobic conditions (Berglund and Berglund, 2011b), which could lower peat oxidation rates and therefore CO2 emissions and soil subsidence (Van den Bos and van de Plassche, 2003;Lloyd, 2006b;Wilson et al., 2016b;Van Huissteden et al., 2006).
To reduce peat oxidation, drastic rewetting (raising the water table to -20 cm below soil surface or higher) would be the ideal 65 option (Hendriks et al., 2007a;Jurasinski et al., 2016). However, current agricultural use would then no longer be feasible.
Therefore, there is a incentive to explore options where the effects of peat oxidation are mitigated but land use is not changed.
A solution suggested to reduce C loss and land subsidence, which is already in use in the Netherlands, is sub-soil irrigation (SSI). The aim of this management option is to raise the GWT during summer when CO2 emissions are highest due to high temperatures in concert with low GWT. Raising the GWT in the summer could prove effective to limit aerobic peat oxidation 70 (Hoving et al., 2015;Kechavarzi et al., 2007). Irrigation pipes are placed in the soil at a depth of 70 cm below the soil surface, and 10 cm below ditchwater level. This will have two effects: drainage when there is excess water (mostly in autumn, winter and spring), and irrigation in dry periods (summer). This will force the GWT towards the ditch water level at around -60 cm https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.
below the soil surface. The drainage effect results in more of the peat being exposed to oxygen, but since this happens in a colder period, it is expected that the effect of irrigation on CO2 emissions during summer will be much larger. There are, 75 however, few comprehensive studies that report on the effect of sub-soil irrigation on total GHG emissions and C balances for peat soils (Van den Akker et al., 2010;Hendriks et al., 2007b). The hypothesis for the effectiveness of SSI is based on the assumption that peat layers below -70 cm contribute most to GHG emissions. However, this is only based on soil subsidence data, and until now there have not been any studies that directly measured GHG fluxes to test the expected GHG reduction.

80
The aim of our study was therefore to quantify the effect of sub-soil irrigation as an alternative drainage technique on the GWT and the GHG balance. The main research questions were whether, compared to traditional drainage, sub-soil irrigation of peat meadows can 1) achieve the intended regulation of GWT within each year and between years (i.e. irrigation during summer and drainage during winter), and 2) lead to a significant reduction of peat oxidation and GHG emission? 2 Material and methods 85

Study area
The study areas are located in a peat meadow area in the province of Friesland, the Netherlands. The climate is humid Atlantic with an average annual precipitation of 840 mm and an average annual temperature of 10.1°C (KNMI, reference period 1999(KNMI, reference period -2018. About 62% of the Frisian peatland region is now used as grassland for dairy farming (Hartman et al., 2012). Agricultural land 90 in Friesland is farmed intensively, with high yields, and intensive fertilization (>230 kg N ha -1 yr -1 ). It is characterized by large fields with deep drainage, as one third of the fields are drained to -90 --120 cm below soil surface. Large parts of these grasslands are covered with a carbon rich clay layer, ranging from 20-40 cm thick. The peat layer below has a thickness of 80-200 cm, which consists of sphagnum peat on top of sedge, reed and alder peat. The top 30 cm of the peat layer is strongly humified (van Post H8-H10) and the peat below 60 -70 cm deep is only moderately decomposed (van Post H5-H7). On two 95 locations (C and D, see below), there is a 'schalter' peat layer present, highly laminated peat (compacted/ hydrophobic layers of Sphagnum cuspidatum remnants) with poor degradability and poor water permeability. The grasslands are dominated by https://doi.org/10.5194/bg-2020-230 Preprint.  https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Experiment setup
Four sites were set up at dairy farms with land management and soil types representative for Friesland (see Table 1 and Fig.   1). Each location consisted of a treatment site with sub-soil irrigation pipes and a control site. The irrigation pipes were installed 110 at a depth of 70 cm below the surface and 6 m (2,000 m drains ha -1 ) apart from each other, except for the D location where pipes were 5 m apart. The pipes were either directly connected to the ditch (A and C) or connected to a collection tube before connected into the ditch (B and D). The connections with ditches were placed 10 cm below the maintained ditchwater level.
The control sites are fields that have traditional drainage, through a system with deep drainage ditches with convex fields and small shallow ditches. 115 On the treatment sites, three gas measurement frames in 80x80 cm squares were placed on 0.5 m, 1.5 m and 3 m distance from the chosen irrigation pipe (Fig. 2), representing best the variation in the environmental conditions and vegetation.
Dip well tubes were installed to monitor water levels 0.5, 1.5 and 3 m from the pipe, pairing with the locations of gas measurement frames (Fig. 2). The nylon coated tubes were 5 cm wide and perforated filters placed in the peat layer. The tube 120 1.5 m from the irrigation pipe was equipped with a pressure sensor and a data logger (ElliTrack-D, Leiderdorp instruments, Leiderdorp, Netherlands) that measures and records the GWT every hour. Ten more dip well tubes were further placed at intervals 0.5 and 3 m from the pipes in the field, which were manually sampled every 2 weeks during gas sampling campaigns, to obtain the variation on field scale. from the weather station Leeuwarden (18 to 30 km distance from research sites) were used. (KNMI, data). The location specific precipitation was estimated using radar images.
The sites were managed with 4-5 cuts per year. Due to grazing disturbance in 2018, an estimation instead of measurements 140 was made for the C-export of location A in consultation with the farmer, but excluded from statistical analysis. Four times per year slurry manure from location C was applied to all plots. The slurry was diluted with ditchwater (2:1 ratio) and applied above ground in the gas measurement frames and the surrounding area. (0.61 kg m -2 yr -1 dw for 2017 and 0.62 kg m -2 yr -1 dw for 2018 with a C/N ratio of 16.3±1.3 Figure 2 Overview field site SSI. Blue dashed line = irrigation pipe, blue circle = dipwell, Adipwell with data logger, Bgas measurement frame, Cdata logger, -5 -10 -20 soil temperature and soil moisture https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Flux measurements 145
CO2 exchange was measured from January 2017 to December 2018, at a frequency of two measurement campaigns a month during growing season (April -October) and once a month during winter. This resulted in 34 (A), 35 (C and D) and 38 (B) campaigns over the two years. A measurement campaign consisted of flux measurements with opaque (dark) and transparent (light) closed chambers (0.8x0.8x0.5 m) to be able to distinguish ecosystem respiration (Reco) and gross primary production (GPP) from net ecosystem exchange (NEE). During winter an average of 9 light and 10 dark measurements, and during summer 150 18 light and 20 dark measurements were carried out over the course of the day, to achieve data over a gradient in soil temperature and PAR.
The chamber was placed on a frame installed into the soil and connected to a fast greenhouse gas analyzer (GGA) with cavity ring-down spectroscopy (GGA-24EP, Los Gatos Research, Santa Clara, CA, USA) to measure CO2 and CH4 or to a G2508 155 gas concentration analyzer with cavity ring-down spectroscopy (G2508 CRDS Analyzer, Picarro, Santa Clara, CA, USA) to measure N2O. To prevent heating and to ensure thorough mixing of the air inside the chamber, the chambers where equipped with two fans running continuously during the measurements. For CO2 and CH4, each flux measurement lasted on average 180s. N2O fluxes were measured on all frames at least once during a measurement campaign, with an opaque chamber for 480s per flux. 160 PAR was manually measured (Skye SKP 215 PAR Quantum Sensor, Skye instruments Ltd, Llandrindod Wells, United Kingdom) during the transparent measurements, on top of the chamber. Within each measurement, a variation in PAR higher than 75 µmol m -2 s -1 would lead to a restart of the measurement. Soil temperature was measured manually in the frame after the dark measurements at -5 and -10 cm depth (Greisinger GTH 175/PT Thermometer, GMH Messtechnik GmbH, Regenstauf, 165 Germany). Crop height was measured before starting the measurement campaign. The biomass was harvested five times per year. These samples where weighed and dried at 70 °C until constant weight. Total nitrogen (TN) and total carbon (TC) was determined in dry plant material (3 mg) using an elemental CNS analyzer (NA 1500, Carlo Erba; Thermo Fisher Scientific, Franklin, USA) https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Flux calculations
Gas fluxes were calculated using the slope of gas concentration over time (Almeida et al., 2016) (eq.1).
Where F is gas flux (mg m 2 d -1 ), V is chamber volume (0.32 m 3 ), A is the chamber surface area (0.64 m 2 ), slope is the gas 175 concentration change over time(ppm second -1 ); P is atmospheric pressure (kPa); F1 is the molecular weight, 44 g mol -1 for CO2 and N2O and 16 g mol -1 for CH4; F2 is the conversion factor of seconds to days; R is gas constant (8.3144 J K -1 mol -1 ); and T is temperature in Kelvin (K) in the chamber.

Reco modeling
To gap-fill for the days that were not measured for an annual balance for CO2 exchange, Reco and GPP models needed to be 180 fitted with the measured data for each measurement campaign. Reco was fitted with the Lloyd-Taylor function (Lloyd and Taylor, 1994) based on soil temperature (Eq. 2): where Reco is ecosystems respiration, Reco,Tref is ecosystem respiration at the reference temperature (Tref) of 281.15 K and was 185 fitted for each measurement campaign, E0 is long term ecosystem sensitivity coefficient (308.56, (Lloyd and Taylor, 1994)), T0 Temperature between 0 and T (227.13, Lloyd and Taylor, 1994), T is the observed soil temperature (K) at 5 cm depth and Tref is the reference temperature (283.15 K). If it was not possible to get a significant relationship between the T and the Reco with data from a single campaign, data were pooled for two measuring days to achieve significant fitting (Beetz et al., 2013;Poyda et al., 2016;Karki et al., 2019) 190

GPP modeling
GPP was obtained by subtracting the measured Reco (CO2 flux measured with the dark chambers) from the measured NEE (CO2 flux measured with the light chambers). For the days in between the measurement campaigns, data were modeled with https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License. the relationship between the GPP and PAR using a Michaelis-Menten light optimizing response curve Beetz et al., 2013). For each measurement location per measurement campaign, the GPP was modeled by the parameters 195 and GPPmax (maximum photosynthetic rate with infinite PAR) of (eq.3): where NEE is the measured CO2 flux with light chamber, α is ecosystem quantum yield (mg CO2 m -2 s -1 ) which is the linear 200 change of GPP per change in PAR at low light intensities (<400 µmol m -2 s -1 as in (Falge et al., 2001), PAR is measured photosynthetic active radiation (µmol quantum m -2 s -1 ), GPPmax is gross primary productivity at its optimum, Reco is ecosystem respiration measured for light response curve and for the year budget calculated with the Lloyd-Tayler function where used.
The fitted parameters were linearly interpolated between the measurement campaigns. Due to low coverage of the PAR range in a single measurement campaign in data from year 2017, the complete data set of 2017 were divided into summer and winter 205 periods, and the two datasets (instead of every field campaign) were fitted for the corresponding period per location.

Yearly budget calculations
The calculated parameters were used to interpolate the data for a yearly budget. For the GPP, an important factor of grass growth was added by assuming a linear development of the model parameters α and GPP max, since the plant biomass continued growing between the measurement dates. The NEE year budgets were calculated using the interpolated hourly Reco 210 and GPP values. Extrapolated values at times between two modeled measurements are weighted averages, where the weights are temporal distances of the extrapolated time spots to both of the measurements.
Besides the campaign-wise gap-filling strategy introduced above, other approaches exist to calculate NEE year budget that may result in different values (Karki et al. 2019), which is considered an important source of uncertainty in our study. To 215 was accepted and calculated into gap-filled NEE. Not all sites and years have acceptable models due to large variations of measured fluxes within a year. The remaining NEE values were averaged per site per year and compared with the campaignwise NEE year budgets as a range of uncertainty. 220 CH4 and N2O fluxes per site and measurement campaign were averaged per day. The annual emissions sums where estimated by linear interpolation between the single measurement dates. Global Warming Potential (GWP) of 34 t CO2-eq and 298 t CO2-eq per ton for CH4 and N2O was used according to IPCC standards (Myhre et al., 2013) to calculate the yearly GHG balance. 225

Statistics
The effect of the treatment on gap-filled annual Reco and GPP, the resulting NEE, the C-export data, CH4, N2O exchanges and the combined GHG balance were tested by fitting linear mixed-effects models, with farm location as a random effect.
Effectiveness of the random term was tested using the likelihood ratio test method. Significance of the fixed terms was tested via Satterthwaite's degrees of freedom method. The treatment effect was further tested using campaign-wise Reco data. 230 Measured Reco fluxes from SSI and Control were calculated into daily averages and paired per date. The data pairs were grouped based on the GWT differences between SSI and control of the dates. Differences between treatments were then analyzed by linear regression of the Reco flux pairs without interception and testing the null hypothesis 'slope of the regression equals to 1'. All statistical analyses were computed using R version 3.5.3 (Team, 2019) using packages lme4 (Bates et al., 2014), lmerTest (Kuznetsova et al., 2017), sjstats (Lüdecke, 2019), and car (Fox and Weisberg, 2018). 235 https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Weather conditions
Mean annual air temperature was 10.3 °C for 2017 and 10.7 °C for 2018, which were higher than the 30-year average of 10.1 240 °C. The growing season (April-September) in 2017 was slightly cooler with 14.3 °C than the average 14.6 °C, while the temperature during the growing season in 2018 was 1.1 °C warmer than average. Precipitation was slightly higher for 2017 840-951 mm compared to the 30-year average of 840 mm (KMNI data). There was a small period of drought in May and June (see Fig.3). In contrast, 2018 was a dry year with average of 546-611 mm. The year is characterized by a period of extreme drought in the summer, from June to the beginning of August, and precipitation lower than average in the fall and winter. 245    https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Groundwater table (GWT)
Deploying subsurface irrigation (SSI) affected the GWT during the two years for all farms (Fig. 4). However, there was a large 255 variation in effect-size between years and locations. The effect of SSI can be divided into two types of periods. Periods with drainage, in the wet periods, coincided with the autumn (in 2017) and winter period (2017 and 2018). Irrigation periods, where the SSI leads to a higher water table than control, occurred during spring and summer when the GWT dipped below the ditch water level. In 2017, the effectiveness differed per farm. For locations A and B, GWT was more stable in summer around the -60 and -70 for SSI compared to the control, while locations C and D the GWT fluctuated more like in the control fields. 260 During the dry summer of 2018, in contrast, all locations showed a strong effect of irrigation, especially after the dry period in the beginning of august. In this period the water table recovered quickly while the control lagged behind.

Measured Reco 275
Despite these observed differences in GWT, there was no overall effect of SSI on the total C budget. Comparing days and locations where the Reco was measured with the measured the GWT, provide insight of the effect of SSI. There is variation between emission rates depending on temperature and grass height, but these differences were small during the measurement day due to the regular harvests. However, the Reco values for the measurement days can give an indication for the effectivity of the differences in GWT (Fig. 6). The division between the groups was based on the function of the irrigation pipes, the 280 difference of the GWT between SSI and control on the measurement days (similar to the groups used in Fig. 2). There was a slightly higher Reco for SSI during drainage periods when GWT was lower, which compensates for the lower Reco during summer. For moments where there was no GWT difference and those showing moderate irrigation, there was no effect of SSI on Reco. However, when the GWT of the SSI was more than 20cm higher than the control, the emissions of the control where significantly higher than SSI (p < 0.01), indicating an effect of the irrigation. However, this effect of the raised GWT was 285 small, even though in some cases the GWT was raised more than 60 cm. Fig. 5 shows how often the different groups of GWT effects occurred. For 2017 the majority of the days were dominated by drainage (increasing Reco), or by no difference or small irrigation resulting in no effect on the Reco. However, the moments with increased irrigation, when there was a reduced Reco effect of SSI were sparse compared to the other dominating periods. 290 https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Ecosystem respiration (Reco)
Reco was generally high for all the farms measured during the two years, with the average Reco of 131±1 t CO2 ha -1 yr -1 for 2017 315 being significantly higher than 101±4 t CO2 ha -1 yr -1 for 2018 (p < 0.001) ( Table 2). However, no effect of SSI on Reco was found (p = 0.350), with no difference among farm locations (random effect p = 0.627). Reco showed a strong seasonal pattern; in 2017 Reco peaked in June and July, while in 2018 the highest Reco was found in May ( Fig. 7 Appendix B).

Net ecosystem exchange (NEE) 325
All locations functioned as large C sources during the measurement period. The annual NEE of all sites and years amounted, on average, to 47.1 t CO2 ha -1 yr -1 , with an uncertainty of 3-16 t CO2 ha -1 yr -1 . The overall explanatory power of year, treatment and location was low (conditional r 2 = 0.531 for fixed and random effects combined) after combining Reco and GPP into NEE.
There was, again, no treatment effect of SSI (p = 0.329), but there were small differences between both years (p = 0.040). NEE values were 67.9±1.6 t CO2 ha -1 yr -1 in 2017 and 56.4±5.1 t CO2 ha -1 yr -1 in 2018 for the treatment plots. No differences between 330 locations were observed (random effect p = 0.076). On average, for all sites and both years, the emission was 62 t CO2 eq. ha -1 yr -1 with an uncertainty of 3-16 t CO2 ha -1 yr -1

Methane exchange
The total exchange of CH4 was very low during both years. During most periods, the locations functioned as a sink of CH4.

Nitrous oxide exchange
The fluxes for N2O showed a high spatial variability between (random effect p = 0.010) and within all locations, and showed 340 an erratic pattern with mostly low emissions with some high peaks. The highest emissions were measured on the frame closest to the irrigation pipe in the treatment plot of location D, with 4.4 t CO2 eq. ha -1 yr -1 for 2017 and 4.9 t CO2 eq. ha -1 yr -1 for 2018.
The highest peak was measured in August for SSI of location D, showing 55±15 mg N2O m -2 d -1 . The peaks observed were erratic, and cannot be explained by year or treatment effect (p = 0.060 and p = 1.000 respectively, marginal r 2 = 0.107 for the fixed effects). Emissions did not correspond to fertilization management with slurry before measurement campaigns. 345

Total GHG balance
All sites showed high emissions, without an effect of SSI (p = 0.332) which was consistent for all farms, without location effect (random effect p = 0.099) (table 3). However, there was a large difference between both years, with higher emission rates in 2017 amounting to 63±2 t CO2 eq. ha -1 yr -1 , compared 52±3 t CO2 eq. ha -1 yr -1 for 2018 (p<0.001). Table 2 Overview of all processes contributing to the carbon balance calculated for both years. Ecosystems respiration (Reco), gross primary production (GPP), net ecosystems exchange (NEE, sum of GPP and Reco), C-exports from harvest and C-addition from manure for subsoil irrigation (SSI) and control plots at farm locations A-D.

Carbon exchange
Year Location treatment Reco GPP NEE C-export C-manure t CO2 ha -1 yr -1 t CO2 ha -1 yr -1 t CO2 ha -1 yr -1 t CO2 ha -1 yr -1 t CO2 https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License. Table 3 All GHG emissions contributing to the total GHG balance for subsoil irrigation (SSI) and controls for the four locations (A-D) for both years. The sum of NEE, C-export and C-manure form the total CO2 flux. The total GHG balance per year, location and treatment is the sum of CO2, CH4 and N2O fluxes in CO2 equivalents, using radiative forcing factors of 34 for CH4 and N2O 298 according to IPCC standards (Myhre et al., 2013).

GHG fluxes
Year

Figure 7 Reco and GPP for location B in g CO2 m -2 d -1 on the primary y-axis, for control and SSI. Accumulative NEE in t CO2 ha -1 yr -1 , for control and subsoil irrigation (SSI)
, every year starting at 0.

Discussion
For both years, SSI had a clear irrigation effect during summer at the four farms, increasing the GWT on average by 6-18 cm.
During winter, there was a moderate but consistent drainage effect, reducing the average GWT in the wet/winter period by 1-20 cm. Despite the irrigation effects and higher water levels in summer, there was no effect of SSI and total GHG balances remained high (62 t CO2 eq. ha -1 yr -1 on average of all sites and years with an uncertainty of 3-16 t CO2 ha -1 yr -1 ). We found 370 no evidence for a reduction of CO2 emissions, nor for higher yields, on an annual base by implementing SSI.

SSI does not reduce annual Reco
Despite the higher summer GWT, there was no effect of SSI on the annual Reco at all sites. We found a modest 5-10% reduction in Reco only when GWT differences were larger than 20 cm, based on the direct comparison using raw Reco fluxes (Fig. 6).
When the irrigation effect was smaller, no effect on the Reco was found. An earlier study in the Netherlands on the role of GWT 375 also showed small effects of higher summer GWT on Reco and NEE (Net Ecosystem Exchange) despite substantial differences in soil volume changes/soil subsidence (Dirks et al., 2000). Similarly, the 4-year study (Schrier-Uijl et al., 2014) found little differences in NEE estimates despite substantial large variations in summer GWT and soil moisture contents.
Our findings contradict the general assumption that a higher GWT leads to lower CO2 emissions, which is often found in near-380 natural peatlands with the presence of peat-forming vegetation (Wilson et al., 2016a;Lloyd, 2006a;Moore and Dalva, 1993).
However, most studies discuss the effect of lower annual average GWT. In addition, there are also studies that did not find an effect of GWT on CO2 emissions during the season (Parmentier et al., 2009;Lafleur et al., 2005;Nieveen et al., 2005)). This lack of effect is explained by the fact that there is only a small variation in soil moisture values above the GWT. A large number of studies report lower CO2 emissions when water levels were structurally elevated, concomitant with substantial 385 differences in vegetation/land use following higher water levels (Beetz et al., 2013;Schrier-Uijl et al., 2014;Wilson et al., 2016a). In our study, SSI seems to have an effect of a similar magnitude trending towards higher emissions during periods with lower GWT at the SSI sites.
The small effect size in our study can most probably be explained by differences in peat oxidation rates along the soil profile. 390 Some other studies suggest that the top 30-40 cm layer of the peat profile plays an important role in C turnover rates in drained peatlands, due to more readily decomposable C sources and higher temperatures (Saeurich et al., 2019;Karki et al., 2016;Lafleur et al., 2005;Moore and Dalva, 1993). This soil layer was, however, not affected by higher summer GWTs in our study. Moreover, the top soil layer was even exposed to oxygen for longer periods due to extra drainage during wet seasons.
As the infiltrating water will affect the soil moisture content of these layers, it is even expected that this content will approach 395 the optimum for C mineralization more often at the locations where SSI is applied. (Saeurich et al., 2019) speculated that the highest CO2 production in the top 10 cm is reached when GWTs are approximately 40 cm below the surface (Silvola et al., 1996).
In contrast to surface irrigation where the topsoil is replenished with moisture, the SSI effect is limited to deeper parts of the 400 peat soils, at -60--100 cm depth. However, the role of this layer as a C source is only limited. Its potency to act as a C source https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.
is reduced by lower temperatures, limited O2 intrusion, and the fact that water content of this layer is already close to saturation (Taggart et al., 2012;Berglund and Berglund, 2011a;Saeurich et al., 2019). This layer shows low levels of stronger electron acceptors such as O2 and nitrate used for the microbial oxidation of organic compounds, and of labile organic matter (Fontaine et al., 2007;Leifeld et al., 2012). Visually, the layers deeper than 60 cm are less decomposed (plant macrofossils still visible) 405 compared to the highly degraded uppermost 40 cm.
In addition, lower CO2 production in the deeper peat layers that are saturated due to the higher water level may be compensated for by the increased CO2 production in the top 20-40 cm due to the higher moisture levels resulting from elevated water levels.
The dry year of 2018 with very low GWT in the control sites (and thus an expected maximized effect of SSI) provides 410 additional evidence that SSI contributes little if any to the mitigation of CO2 emission from drained peatlands.

SSI effects on CH4 and N2O emissions
Findings of this experiment agree with the generally accepted idea that intensively drained peatlands have low levels of CH4 emissions, and often these systems even function as a small CH4 sink (Couwenberg et al., 2011;Couwenberg and Fritz, 415 2012;Tiemeyer et al., 2016;Maljanen et al., 2010). The SSI site in farm C showed the highest N2O emissions with 23 kg N2O ha −1 yr −1 for 2017. In the current study the average N2O emission from the drained peatland grasslands was 9 kg N2O ha −1 yr −1 falling with the range of annual N2O emissions from drained peatlands in Northern Europe (4-18 kg N2O ha −1 ) (Kandel Tanka et al., 2018;Leahy et al., 2004;Maljanen et al., 2010). Fertilization, temperature and water table fluctuations play major roles in the total N2O emission (Regina et al., 1999;Van Beek et al., 2011). No distinct peaks were measured after 420 application of fertilizer, and fertilizer was applied on all locations on the same day, so missing peak fluxes would not influence the comparison. The mechanisms of N2O production and consumption in organic soils are, however, complex and there is high temporal and spatial variability as influenced by site conditions and management (Leppelt et al., 2014;Taghizadeh-Toosi et al., 2019).

High CO2 emissions, but lack of effect of SSI on GHG emission 425
The GPP of the sites (-45.2 --92.6-80.7 and -56.5 t CO2 ha -1 yr -1 in 2017 and 2018, respectively) was in line with values found by (Tiemeyer et al., 2016) for productive and drained peatlands (-70 ± 18 t CO2 ha -1 yr -1 ) and within the range of grasslands from Europe (45-78 t CO2 ha -1 yr -1 ) (Eze et al., 2018;Ma et al., 2015;Byrne et al., 2005). The Reco values of the sites (131.5 and 100.6 t CO2 ha -1 yr -1 in 2017 and 2018, respectively) are, however, at the higher end of the range (97 ± 33 t CO2 ha -1 yr -1 in Tiemeyer et al., 2016). This leads to a relatively high NEE contributing to the generally large annual GHG budgets found 430 in our study. There was, however, a large difference between 2017 and 2018 (-80.7 and -56.5 t CO2 ha -1 yr -1 , respectively), which was due to the strong drought effect in 2018. In contrast to our expectations, no effect of SSI was found on GPP. The net GHG budgets from the current study (42.4 -70.4 t CO2 eq. ha -1 yr -1 ) fall in the upper range of reported emissions from drained? temperate peatlands (Hiraishi et al. 2014, Wilson et al. 2016a. Intensively drained peatlands with productive grassland vegetation tend to emit more CO2 (40-70 t CO2 ha -1 yr -1 ) (Hoffmann et al., 2015;Tiemeyer et al., 2016;Wilson et 435 al., 2016a;Tiemeyer et al., 2020) than IPCC Tier default values (Hiraishi et al. 2014). Emissions found in the current study were substantially higher than those reported earlier for drained peatlands in the Netherlands (20-25 t CO2 ha -1 yr -1 in (Jacobs et al., 2007;Schrier-Uijl et al., 2014). There are a number of reasons for the high emissions found here. Abiotic conditions that favor high CO2 emissions were present, with high temperatures for both years and optimal moisture conditions for 2017.
Research from (Pohl et al., 2015) found a high impact of dynamic soil organic carbon (SOC) and N stocks in the aerobic zone 440 on CO2 fluxes. In our case, the peat soils contained a high amount of C, especially in the upper 20 cm layer. This layer was also aerobic for long periods during the experiment, thus promoting C formation and transformation processes in the plantsoil system.

Uncertainties
GHG emissions on peat grasslands are highly variable (Tiemeyer et al., 2016) given the uncertainties from the wide ranges of 445 land use and management activities (Renou- Wilson et al., 2016) and gap filling techniques (Huth et al., 2017). In this study, only uncertainties from gap-filling techniques in terms of data-pooling strategies and model selections were considered. https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.
Campaign-wise fitting of Reco and GPP models can best represent the original data sets, while pooling data for a longer period can provide better model fitness and less bias toward single measurements (Huth et al., 2017;Poyda et al., 2017). However, in this study, different responses of vegetation and soil processes to drought, especially to the extreme drought in 2018, caused 450 abnormal data points that do not fit the classic models, resulting in the generally poor performances of annual models. For this reason, we reported the annual budgets with campaign-wise gap-filled NEE values. The uncertainties of NEE estimates from model differences were on average 14 tons and up to 25 tons of CO2. Nevertheless, no SSI effect was found considering NEE estimates from annual models. The model differences quantified here were in good agreements with other model tests (Karki et al., 2019;Görres et al., 2014) and match the magnitude of NEE uncertainties calculated with other methods (e.g. the 23-30 455 tons CO2 variances reported by (Schrier-Uijl et al., 2014) using eddy co-variance techniques).

Costs and benefits SSI
The intensity of land use (intensity and timing of drainage and fertilization, plant species composition, mowing and grazing regimes) influence the grassland's ability to accumulate or lose C (Renou- Wilson et al., 2016;Smith, 2014;Ward et al., 2016).
SSI can increase the load-bearing capacity of the field surface for fertilizing equipment, facilitating earlier fertilization 460 compared to management under current drainage systems. This can also cause increased leaching of water due to earlier drainage in a wet spring. However, the general land-use intensity will not change with the use of SSI. It was expected that Cexport via crop yields due to extra drainage could increase in a wet autumn. However, we did not find any indication for an increase in land-use intensity or yield as a result of SSI.

465
The use of SSI is considered impractical for use in most regions outside of the Netherlands due to the high investment costs for irrigation pipes and the intensive water infrastructure needed for controlling the water level. In addition, irrigation pipes will increase the water demand in summer for these agricultural fields. Both land-use intensity and an increase in yield are related to an increase in CO2 emissions on drained peat (Beetz et al., 2013;Couwenberg, 2011). The land-use history of our sites favors high CO2 emission: tillage (cultivators, sod-renewal, and some plowing), cumulative fertilization and well-470 maintained drainage (Provincie Fryslân ,2015). https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.

Main conclusion
Unfortunately, the implementation of SSI does not lead to a reduction of GHG emissions from drained peat meadows, even though there was a clear increase in GWT during summer (especially in the dry year of 2018). We therefore conclude that the use of SSI is ineffective as a mitigation measure to sufficiently lower peat oxidation rates and, therefore, also soil subsidence. 475 Most likely, the largest part of the peat oxidation takes place in the top 70 cm of the soil, which stays above the GWT with the use of SSI. This layer is still exposed to higher temperatures, sufficient moisture, oxygen and alternative electron acceptors such as nitrate, and nutrient input. We expect that SSI may only be effective when the GWT can be raised permanently to levels close to the soil surface (-20-35 cm below the surface). https://doi.org/10.5194/bg-2020-230 Preprint. Discussion started: 2 July 2020 c Author(s) 2020. CC BY 4.0 License.