The study was conducted in the Tässi peat extraction area located in central Estonia (58∘32′ 16′′ N, 25∘51′43′′ E). The region has a temperate climate with long-term mean (1981–2010) annual temperature and precipitation of 5.8 ∘C and 764 mm, respectively (Estonian Weather Service). Peat extraction in the peatland started in late 1960s and today peat continues to be harvested for horticultural purposes using the milling technique on about 264 ha.

The current study was carried out on a 4.5 ha area which was set aside from peat extraction in the early 1980s. The residual Sphagnum peat layer depth is about 2.5 m. A section approximately 0.24 ha in size within the abandoned site was restored in April 2012. The restoration was done following a slightly modified protocol of the moss layer transfer technique (Quinty and Rochefort, 2003) aimed at restoring the growth of Sphagnum mosses and initiating the development of a natural bog community. The first restoration steps included stripping the uppermost oxidized peat layer (20 cm) and flattening the freshly exposed surface. In addition, the peat along the borders of the restoration area was compressed and the outflow drainage ditch was dammed with peat material to reduce the lateral water outflow from the experimental site.

To study the impact of WTL on restoration success in terms of vegetation development and GHG fluxes, the restoration site was divided into wetter and drier sections by lowering the peat surface by 10 cm for approximately one-third of the area. This resulted in restoration treatments with high and low WTLs (i.e., ResH and ResL). In addition, an unrestored BP site was included in the study as a reference. Two replicate plots (20 × 20 m) were established for each of the three treatments.

To enhance vegetation succession, living plant fragments from Sphagnum-dominated hummocks were collected from a nearby (10 km) donor site (Soosaare bog) and spread out in the ratio of 1 : 10 (i.e., 1 m2 of collected plant fragments were spread over 10 m2) in the ResH and ResL treatments. As the last step, straw mulch was applied to protect plant fragments from solar radiation and to improve moisture conditions. Further details about the restoration procedure at this study site have been given in Karofeld et al. (2015).

Three years following restoration, the bryophyte species found at the restored site were dominated primarily by Sphagnum mosses (e.g., S. fuscum, S. rubellum and S. magellanicum). The common vascular plant species observed post-restoration included shrubs and trees such as common heather (Calluna vulgaris L.), common cranberry (Oxycoccus palustris Pers.), downy birch (Betula pubescens Ehrh.), bog rosemary (Andromeda polifolia L.) and Scots pine (Pinus sylvestris L.), with a minor cover of accompanying herbaceous sedge and forb species such as tussock cottongrass (Eriophorum vaginatum L.) and round-leaved sundew (Drosera rotundifolia L.) (Karofeld et al., 2015).

A meteorological station to continuously monitor environmental variables was set up on-site in June 2014. This included measurements of air temperature (Ta; model CS 107, Campbell Scientific Inc., Logan, UT, USA), photosynthetically active radiation (PAR; model LI-190SL, LI-COR Inc., Lincoln, NE, USA) and precipitation (PPT; tipping bucket model 52202, R. M. Young Company, Traverse City, MI, USA) at 1.2 m height above the ground. Soil temperature (Ts; depths of 5 and 30 cm) was measured with temperature probes (model CS 107, Campbell Scientific Inc., Logan, UT, USA) and soil volumetric water content (VWC; depth 5 cm) with water content reflectometers (model CS615, Campbell Scientific Inc., Logan, UT, USA). All automated abiotic data were collected in 1 min intervals and stored as 10 min averages on a data logger (CR1000, Campbell Scientific Inc., Logan, UT, USA). In addition, continuous 30 min records of the WTL relative to the soil surface were obtained with submerged water level loggers (HOBO, Onset Computer Corporation, Bourne, MA, USA) placed inside perforated 1.0 m long PVC pipes (∅ 5 cm; sealed in the lower end).

The on-site meteorological measurements were complemented by Estonian Weather Service data to obtain complete time series of Ta, PPT and PAR over the entire year. Hourly means of Ta and daily sums of PPT were obtained from the closest (∼ 20 km away) Viljandi meteorological station. Global radiation (hourly sums) data from the Tartu meteorological station (∼ 40 km away) were converted to PAR based on a linear correlation relationship to on-site PAR.

In addition, manual measurements of Ts (depths 10, 20, 30 and 40 cm) were recorded by a handheld temperature logger (Comet Systems Ltd., Rožnov pod Radhoštěm, Czech Republic) and VWC (depth 0–5 cm) using a handheld soil moisture sensor (model GS3, Decagon Devices Inc., Pullman, WA, USA) during each sampling campaign. Furthermore, groundwater temperature, pH, redox potential, dissolved oxygen content, electrical conductivity as well as ammonium and nitrate concentrations were measured in observation wells (∅ 7.5 cm, 1.0 m long PVC pipes perforated and sealed in the lower end) installed at each sampling location using YSI Professional Plus handheld instruments (YSI Inc., Yellow Springs, OH, USA). In addition, soil samples (depth 0–10 cm) in three replicates were taken from each of the treatments and analyzed for pH as well as total C, total N, P, K, Ca and S contents at the Tartu Laboratory of the Estonian Environmental Research Centre. Three additional samples were taken from the same depth to determine bulk density in each treatment. Mean values for these soil properties are summarized in Table 1.

To assess the effect of vegetation development on GHG fluxes, vegetation cover (%) and species composition were recorded inside each of the flux measurement collars (see Sect. 2.4) in late spring. In each collar, the cover was estimated visually for each species and rounded to the nearest 1 %. Bryophyte, vascular plant and total vegetation cover were computed as the sum of their respective individual species coverages.

To evaluate the impact of WTL on NEE in the restored ResH and ResL treatments, CO2 flux measurements were conducted biweekly from May to December 2014 at three sampling locations within each replicate plot (i.e., six locations per treatment) using the closed dynamic chamber method. At each sampling location, a collar (∅ 50 cm) with a water-filled ring for air-tight sealing was permanently installed to a soil depth of 10 cm. NEE measurements were conducted in random plot order (to avoid diurnal effects) using a clear Plexiglas chamber (95 % transparency; h 50 cm, V 65 L) combined with a portable infrared gas analyzer (IRGA; EGM-4, PP Systems, Hitchin, UK). The chamber was equipped with a sensor to measure PAR and Ta (TRP-2, PP Systems, Hitchin, UK) inside the chamber. Ambient Ta was also recorded with an additional temperature sensor placed on the outside of the chamber. Cooling packs placed inside the chamber were used to avoid a temperature increase inside the chamber during measurements. The chamber was also equipped with a low-speed fan to ensure constant air circulation. After every NEE measurement, Re was determined from a subsequent measurement during which the transparent chamber was covered with an opaque and light reflective shroud. CO2 concentrations, PAR, Ta, pressure and relative humidity were recorded by the IRGA system every 4.8 s over a 4 or 3 min chamber deployment period for NEE and Re measurements, respectively. Since the aim of this study was to assess the atmospheric impact of restoration, all fluxes are expressed following the atmospheric sign convention in which positive and negative fluxes represent emission to and uptake from the atmosphere, respectively.

GPP was derived from the difference between NEE and Re (i.e., GPP = NEE -Re). In addition, an estimate of NPP was derived from the difference between NEE and heterotrophic respiration (Rh; see Sect. 2.5) (i.e., NPP = NEE - Rh).

Re estimates during the non-growing-season months of March to April 2014 and January to February 2015 were determined by closed static chamber measurements (described in Sect. 2.6). Air samples collected during these measurements were analyzed for their CO2 concentrations on a Shimadzu GC-2014 gas chromatograph with an electron capture detector. These Re estimates also represented non-growing-season NEE for all treatments.

In the BP treatment, Re was determined by measurements using a separate closed dynamic chamber setup as described below in Sect. 2.5. Due to the absence of vegetation, GPP as well as NPP were assumed to be 0 and NEE subsequently equaled Re in the BP treatment.

From May to December 2014, Rh was measured simultaneously with NEE from separate PVC collars (∅ 17.5 cm) inserted to a depth of 10 cm beside each NEE collar. The soil around the Rh collars was cut with a sharp knife to a depth of 30 cm in April 2014 to exclude respiration from the roots. The area inside the collars was cleared of living moss and vascular plants and kept free of vegetation during the remaining year. For Rh measurements, a second set of instrumentation was used which included an opaque chamber (h 30 cm, V 0.065 L; equipped with a low-speed fan) combined with an EGM-4 infrared gas analyzer. During each Rh measurement, CO2 concentration and Ta inside the chamber were recorded every 4.8 s over a period of 3 min. Autotrophic respiration (Ra) was derived from the difference between the measured Re and Rh fluxes (i.e., Ra = Re - Rh). Due to the absence of vegetation, Ra was not determined in BP.

To assess the impact of WTL on CH4 and N2O exchanges, flux measurements were conducted with the closed static chamber method at a biweekly to monthly interval from March 2014 to February 2015 at the same locations (i.e., same collars) as were used for the NEE measurements (described in Sect. 2.4). During each chamber deployment period, a series of air samples were drawn from the chamber headspace (h 50 cm, V 65 L; white opaque PVC chambers) into pre-evacuated (0.3 mbar) 50 mL glass bottles 0, 0.33, 0.66 and 1 h after closing the chamber. The air samples were analyzed for CH4 and N2O concentrations with a flame ionization detector and an electron capture detector, respectively, using a Shimadzu GC-2014 gas chromatograph combined with a Loftfield automatic sample injection system (Loftfield et al., 1997).

Fluxes of CO2, CH4 and N2O were calculated from the linear change in gas concentration in the chamber headspace over time, adjusted by the ground area enclosed by the collar, volume of chamber headspace, air density and molar mass of gas at measured chamber Ta. The linear slope in case of the dynamic chamber measurements was calculated for a window of 25 measurement points (i.e., 2 min) moving stepwise (with one-point increments) over the entire measurement period after discarding the first two measurement points (i.e., applying a 9.6 s “dead band”). The slope of the window with the best coefficient of determination (R2) was selected as the final slope for each measurement. In the static chamber method, the linear slope was calculated over the four available concentration values.

All dynamic chamber CO2 fluxes with a R2≥ 0.90 (p < 0.001) were accepted as good fluxes. However, since small fluxes generally result in a lower R2 (which is especially critical for NEE measurements), dynamic chamber fluxes with an absolute slope within ±0.03 ppm s-1 were always accepted. The slope threshold was determined based on a regression relationship between the slope and respective R2 values. For static chamber measurements, the R2 threshold for accepting CO2, CH4 and N2O fluxes was 0.90 (p < 0.05), 0.80 (p < 0.1) and 0.80 (p < 0.1), respectively, except when the maximum difference among the four concentration values was less than the gas-specific GC detection limit (i.e., < 20 ppm for CO2, < 20 ppb for CH4 and < 20 ppb for N2O), in which case no filtering criterion was used. Based on these quality criteria 11 % of NEE, 9 % of Re, 21 % of Rh, 33% of CH4 and 6 % of N2O fluxes were discarded from subsequent data analysis.

To obtain estimates for the annual CO2 fluxes, nonlinear regression models were developed based on the measured CO2 flux, PAR, WTL and Ta data following Tuittila et al. (2004). As a first step, measured GPP fluxes were fitted to PAR inside the chamber using a hyperbolic function adjusted by a second term which accounted for additional WTL effects (Eq. 1): GPP=α×Amax×PARα×PAR+Amax×exp⁡-0.5×WTL-WTLoptWTLtol2, where GPP is the gross primary production (mg C m-2 h-1), PAR is the photosynthetically active radiation (µmol m-2 s-1), α is the light use efficiency of photosynthesis (i.e., the initial slope of the light response curve; mg C µmol photon-1), Amax is the maximum photosynthesis at light saturation (mg C m-2 h-1), WTL is the water table level (cm), WTLopt is the WTL at which maximum photosynthetic activity occurs and WTLtol is the tolerance (i.e., the width of the Gaussian response curve of GPP to WTL).

Secondly, Re fluxes were fitted to Ta using an exponential function (Eq. 2): Re=R0×exp(b×Ta), where Re is the ecosystem respiration (mg C m-2 h-1), Ta is the air temperature (∘C), R0 is the respiration (mg C m-2 h-1) at 0 ∘C and b is the sensitivity of respiration to Ta. Both GPP and Re were modeled with hourly resolution using hourly PAR, WTL and Ta as input variables. Growing season (1 May to October 31) GPP and annual Re were then derived from the cumulative sums of these modeled fluxes. The balance between growing season GPP and annual Re estimates resulted in the annual NEE in ResH and ResL, whereas annual Re represented annual NEE in BP. The GPP and Re model parameters for the different treatments are summarized in Table 2.

Annual sums of CH4 and N2O fluxes were estimated by scaling their hourly mean and median flux values, respectively, to annual sums. The median flux was used for N2O to avoid a positive bias caused by episodic high peak fluxes measured directly after rainfall events. The annual sums were converted to CO2 equivalents (CO2 eq) using the global warming potentials (over a 100-year time frame including carbon–climate feedbacks) of 34 and 298 for CH4 and N2O, respectively (IPCC, 2013).

Collar flux data were averaged for each plot before conducting further statistical analysis to avoid pseudoreplication. The non-parametric Friedman one-way analysis of variance (ANOVA) by ranks test for dependent samples was used to account for repeated measurements in time when testing for treatment effects (i.e., ResH, ResL and BP) on the growing season or annual means of the various component fluxes. This analysis was followed by a Bonferroni post hoc comparison to determine significant differences among treatment means. The Mann–Whitney U test was used when comparing only the restoration treatments for significant effects (i.e., on GPP, NPP and Ra fluxes). Pearson's correlations were used to investigate the effects of vegetation cover on mean growing season fluxes. The significance level was P < 0.05 unless stated otherwise. All calculations and statistics were computed using the Matlab software (Matlab Student version, 2013a, Mathworks, USA).

The annual mean Ta and total PPT from March 2014 to February 2015 were 7.2 ∘C and 784 mm, respectively, which suggests warmer conditions with normal wetness when compared to the long-term climate normal (5.8 ∘C and 764 mm). PAR peaked in the first week of July while the seasonal Ta curve peaked at around 23 ∘C in late July (Fig. 1a). A prolonged warm and dry period occurred from early to late July with a mean Ta of 20.0 ∘C and total rainfall of 43.3 mm.

The WTL ranged from -2 to -52 and from -8 to -59 cm in the restored ResH and ResL treatments, respectively, while remaining between -26 and -69 cm in the unrestored BP site (Fig. 1b). The mean WTLs in ResH and ResL were -24 and -31 cm, respectively, resulting in a mean annual difference of 7 cm between the restored treatments. Throughout the year, the WTL in ResH was always higher than in ResL with the difference varying between 3 and 10 cm. The mean WTL in BP was -46 cm resulting in mean differences of -22 and -15 cm compared to ResH and ResL, respectively.

The total surface cover, i.e., the fraction of re-colonized surface area, inside the flux measurement collars was higher in the wetter ResH (63 %) than in the drier ResL (52 %) treatment. Bryophytes were more abundant in ResH (62 %) than in ResL (44 %) (Table 3). The bryophyte cover consisted primarily of Sphagnum species which contributed 98 and 96 % in ResH and ResL, respectively. Vascular plants occurred more frequently in the drier ResL (14 %) than in the wetter ResH (4 %) treatment and were dominated by woody plants (i.e., shrubs and tree seedlings) (Table 3). The cover of sedges was < 1 % in both restored treatments.

Daytime NEE was positive indicating CO2 emissions during the non-growing-season months (November to April) in all three treatments (Fig. 2a). During the early (i.e., June) and late (i.e., mid-August to September) summer, net CO2 uptake occurred in both ResH and ResL with maximum rates of -42 and -41 mg C m-2 h-1, respectively. However, during the warm and dry mid-summer period, CO2 emissions of up to 36 and 27 mg C m-2 h-1 were observed in ResH and ResL, respectively. In contrast, NEE remained positive in BP throughout the growing season and followed the seasonal pattern of Ta with maximum emission rates of 104 mg C m-2 h-1 occurring in early August. The annual mean midday NEEs in ResH and ResL were significantly lower than in BP but not significantly different between the two restored treatments (Table 4).

Midday Re was similar for all treatments during the non-growing-season months (Fig. 2b). During the growing season, however, midday Re differed among treatments with lowest and highest Re observed in ResH and BP, respectively. Re in ResH and ResL reached maximum values of 74 and 96 mg C m-2 h-1 during early July, respectively, whereas Re peaked at 104 mg C m-2 h-1 in early August in BP. The annual mean midday Re was significantly lower in ResH and ResL than in BP (Table 4).

From early June to late August, both the daytime GPP and NPP were more negative (i.e., representing greater production) in the drier ResL than in the wetter ResH treatment (Fig. 2c, d). Greatest GPP occurred in late June and mid-August reaching -90 and -98 mg C m-2 h-1 in ResH and ResL, respectively. GPP temporarily decreased (i.e., resulting in less negative values) to -14 and -41 mg C m-2 h-1 during the warm and dry mid-summer period in both ResH and ResL. The seasonal patterns of NPP followed closely those of GPP, reaching -65 and -68 mg C m-2 h-1 in ResH and ResL, respectively. The growing season mean GPP in ResH (-49.3 mg C m-2 h-1) was significantly higher than that in ResL (-65.5 mg C m-2 h-1) (Table 4). The difference in the growing season means of NPP in ResH and ResL was not statistically significant.

Midday Ra was more than 2 times greater in the drier ResL than in the wetter ResH treatment for most of the growing season sampling dates (Fig. 2e). The seasonal pattern of Ra coincided with that of GPP in both restored treatments with greatest Ra occurring in late June and mid-August, reaching maximum values of up to 27 and 36 mg C m-2 h-1 in ResH and ResL, respectively. The growing season mean Ra was significantly higher (by about 2 times) in ResL than in ResH (Table 4). The ratio of Ra to Rh was on average 0.21 and 0.42 in ResH and ResL, respectively.

Midday Rh was consistently lower in ResH and ResL than in BP throughout the growing season (Fig. 2f). Maximum Rh of up to 61, 73 and 104 mg C m-2 h-1 in ResH, ResL and BP, respectively, were observed in early July (restored treatments) and early August (unrestored BP). The growing season mean Rh was significantly lower (by about 50 %) in ResH and ResL than in BP (Table 4).

Throughout most of the year, CH4 fluxes were observed in the range of -13 to 60 µg C m-2 h-1 in all three treatments (Fig. 3a). However, occasional peak CH4 emission of up to 170 and 92 µg C m-2 h-1 occurred in ResH and ResL, respectively. During the non-growing-season months, CH4 exchange was variable, showing both small uptake as well as large emission (-6 to 138 µg C m-2 h-1). The mean annual CH4 exchange was about 2 times greater in the wetter ResH than in the drier ResL treatment although the differences among the three treatments were not statistically significant (Table 4).

N2O fluxes in ResH and ResL remained within the range of -2.8 to 25 µg N m-2 h-1 for most of the year (Fig. 3b). In contrast, high N2O emissions of 66 to 133 µg N m-2 h-1 occurred during July and August in BP. The annual mean N2O exchanges of -0.12 µg N m-2 h-1 in ResH and 2.13 µg N m-2 h-1 in ResL were not significantly different (Table 4). Meanwhile, the mean N2O exchanges in the two restored treatments were significantly lower (by 1–2 magnitudes) compared to the 27.1 µg N m-2 h-1 in BP (Table 4).

The differences in mean growing season NEE, GPP, NPP and Ra among individual collars (i.e., the spatial variability) were significantly correlated to bryophyte but not to vascular plant cover in ResH (Table 5). In contrast, spatial variations in NEE, GPP, NPP and Ra were significantly correlated to vascular plant but not to bryophyte cover in ResL. In addition, Re was significantly correlated to vascular plant cover in ResL. Meanwhile, the CH4 and N2O exchanges were not significantly correlated to vegetation cover in either ResH or ResL.

Ts measured at 10 cm depth was the abiotic variable that best explained variations in Re (R2= 0.79, 0.84 and 0.81 in ResH, ResL and BP, respectively) in the form of an exponential relationship (Fig. 4) with higher temperatures resulting in higher respiration rates. The basal respiration and temperature sensitivity parameters were lowest in the wetter ResH treatment and highest in BP.

N2O fluxes correlated best with VWC measured at 0–5 cm soil depth in ResL (R2= 0.60) and in BP (R2= 0.39) (Fig. 5). In contrast, N2O fluxes were not correlated to soil VWC or any other abiotic variable in ResH. Similarly, the CH4 exchange did not show any significant relationships with any abiotic variable for any of the three treatments.

In the restored ResH and ResL treatments, the modeled annual Re estimates were 188.6 and 213.2 g C m-2 yr-1, respectively, whereas in the unrestored BP treatment annual Re was 267.8 g C m-2 yr-1 (Table 6). The annual GPP was estimated at -78.0 and -110.5 g C m-2 yr-1 in ResH and ResL, respectively. This resulted in annual NEEs of 110.6, 102.7 and 267.8 g C m-2 yr-1 in the wetter ResH, drier ResL and BP treatments, respectively. The growing season net CO2 loss represented 45 and 37 % of the annual NEE in ResH and ResL, respectively, while it accounted for 67 % in BP. The additional C losses via CH4 emission were 0.190, 0.117 and 0.137 g C m-2 yr-1 in ResH, ResL and BP, respectively. In total, all treatments acted as C sources; however, the annual C balance was lower in the restored ResH (110.8 g C m-2 yr-1) and ResL (102.8 g C m-2 yr-1) treatments than in the unrestored BP (268.0 g C m-2 yr-1) treatment. The total GHG balance, including NEE as well as CH4 and N2O emissions expressed as CO2 eq, was 4.14, 3.83 and 10.21 t CO2 eq ha-1 yr-1 in ResH, ResL and BP, respectively (Table 6). The GHG balance was driven by NEE (96 to 98 %) in all three treatments. The contribution of CH4 emission was highest (2.1 %) in the wetter ResH treatment, while the contribution of N2O emission was highest (3.9 %) in the unrestored BP treatment.

Both restored treatments were C sources during the growing season, which indicates that the CO2 uptake by the re-established vegetation was not able to compensate for the C losses via respiration and CH4 emissions 3 years following restoration. Several studies have previously reported estimates for the growing season C sink–source strength of restored peatlands, with contrasting findings due to different restoration techniques, environmental conditions during the study year and time passed since the initiation of the restoration (Tuittila et al., 1999; Bortoluzzi et al., 2006; Yli-Petäys et al., 2007; Waddington et al., 2010; Samaritani et al., 2011; Strack et al., 2014). For instance, restored peatlands in Finland (Tuittila et al., 1999) and Canada (Waddington et al., 2010; Strack et al., 2014) were C sinks during the growing season 3 to 6 years after restoration. In contrast, other studies suggested that several decades may be required before restored peatlands resume their functioning as C sinks (Yli-Petäys et al., 2007; Samaritani et al., 2011). However, while growing season studies can provide important information on processes governing the fluxes, it is necessary to quantify and compare full annual budgets to better evaluate the climate benefits of peatland restoration relative to abandoned peatland areas (and other after-use options, e.g., afforestation or energy crop cultivation).

In our study, the annual C source strength of the restored and BP treatments was about 1.5 to 2.5 times greater than on the growing season scale. This highlights the importance of accounting for the considerable non-growing-season emissions when evaluating the C sink potential of restored peatlands. In comparison, the annual C source strength of the two restored treatments (111 and 103 g C m-2 yr-1) was lower than the annual emissions of 148 g C m-2 yr-1 reported for a restored cutaway peatland in Canada 10 years following restoration (Strack and Zuback, 2013). Similarly, the C balance of BP (268 g C m-2 yr-1) in our study was about half of the 547 g C m-2 yr-1 emitted at the Canadian unrestored site. However, high emissions in the study of Strack and Zuback (2013) were partly attributed to the dry conditions during the study year. This indicates that restored peatlands are unlikely to provide an annual C sink during the first decade following restoration of peat extraction sites. However, compared to naturally re-vegetating peatlands, which may require 20–50 years to reach a neutral or negative C balance (Bortoluzzi et al., 2006; Yli-Petäys et al., 2007; Samaritani et al., 2011), initiating the restoration by rewetting in combination with re-introduction of peatland vegetation might reduce the time required for the ecosystem to return to being a C sink similar to that of a natural peatland (Tuittila et al., 2004; Roulet et al., 2007; Nilsson et al., 2008).

The similar GHG balances in the two restored treatments ResH and ResL suggest that the differences in the mean WTL had a limited effect on the GHG balance within the few years following restoration of the peat extraction area. Moreover, the GHG balances in the restored treatments were driven primarily by the NEE, while the contribution of CH4 and N2O exchanges remained minor in our study. In contrast, 30 years after rewetting of a German bog, high CH4 emission were reported as the main component of the GHG balance (Vanselow-Algan et al., 2015). The same study also reported GHG balances ranging from 25 to 53 t CO2 eq ha-1 yr-1 which are considerably higher compared to our study. This indicates that the GHG balances of restored peatlands may vary greatly over longer time spans. Moreover, this also suggests the GHG balance of peatland restoration with differing WTL baselines is likely to further diverge over time due to contrasting trajectories in vegetation development and changes in soil biogeochemistry (e.g., pH, nutrient contents and soil moisture dynamics).

While the two restored treatments had similar GHG balances, the difference between the GHG balances in restored and BP treatments was considerable. Only 3 years following restoration, the GHG balance in the restored treatments was reduced to about half of that in BP. This reduction was mainly due to lower annual CO2 emissions (i.e., lower NEE) in the restored treatments compared to BP likely as a result of increased WTL and vegetation development. In addition, annual N2O emissions were also significantly reduced in the restored treatments, although, compared to the differences in the CO2 balance, the impact of the reduction in N2O emissions on the GHG balance was relatively small. Overall, our study suggests that peatland restoration may provide an effective method to mitigate the negative climate impacts of abandoned peat extraction areas in the short term. However, due to the lack of long-term observations and recent reports of potential high CH4 emissions occurring several decades after rewetting (Yli-Petäys et al., 2007; Vanselow-Algan et al., 2015), it remains uncertain whether restoration of abandoned peat extraction areas may also provide an after-use solution with climate mitigation potential in the long term.