The soil samples used for this study were taken from organic soils distributed across Switzerland that were identified using the map of Wüst et al. (2015). Apart from current land use (grassland, cropland, forest), they differed in the type of drainage system (ditches in forest, pipes in crop- and grassland), time since drainage onset and drainage intensity, altitude (MASL) and climate (Table 1). All sites were classified as fens, although we found bog-derived peat layers within the top 30 and 40 cm of the soil profiles at two sites (SK_FL, K_FL). Cropland management comprised rotations typical for Switzerland with maize, winter wheat, ley and rapeseed as major crops. Sites were conventionally tilled. Grasslands were used for cutting and haymaking, not grazing, and are fertilized according to the Swiss Fertilization Recommendations (Flisch et al., 2009). Forest sites were managed and their vegetation was not peat-forming.

Between October 2013 and June 2015, we sampled a total of 84 peat cores from all 21 sites (4 cores per site). All cores were taken to a maximum depth of 1 m. If the underlying mineral layer was reached before 1 m depth, coring was discontinued. We used a Belarusian peat corer (cuts a half-cylindrical undisturbed core of diameter 4 cm) for soils with low bulk densities and a motorized Humax corer (cuts a cylindrical core of diameter 5 cm) for denser soils. The samples were stored at 4 ∘C for up to 2 months until analysis. We applied the method of Rogiers et al. (2008) to account for soil compaction during sampling for any sample and divided the cores into segments corresponding to 5–10 cm depth increments. This corresponded, depending on the type of soil corer used and length of the increment, to sample volumes of between 31 and 196 cm3 per segment. In total this resulted in 1605 soil samples. Some cores had interlayers of mineral sediment identified by a different colour (grey), a high bulk density and their SOC content was lower than 150 g kg-1. These interlayers were excluded from the analysis. The soil of one site (BI_FL) had no limnic layer and therefore was classified as a murshic Histosol; all others were classified as murshic limnic Histosols (WRB, 2014).

Soil pH was measured for two to three samples of each core (307 samples in total) using a flat surface electrode (pH 100, Extech Instruments, USA) calibrated at pH 7.00 and pH 4.01. Aliquots of fresh soil (10 g dry matter) were diluted in distilled water (2.5 parts water to 1 part material by mass), shaken, left for 20 h and shaken again, before the pH measurements were carried out.

Prior to further chemical analysis, the samples were oven-dried at 105 ∘C and weighed to determine bulk density (g cm-3). The dried samples were ground for 2 min at 25 rotations s-1 in a ball mill (Retsch MM400) and subsampled to determine total carbon (Ctot), SOC, hydrogen (H), nitrogen (N) and oxygen (O) contents. Ctot, H and N were analysed after dry combustion of ground subsamples in an elemental analyser (Hekatech, Germany). To determine SOC, we hydrolysed ground aliquots with 36 % HCl (acid fumigation) in a desiccator to remove any carbonates before the samples were analysed in the elemental analyser. A third set of ground subsamples were used to determine the O contents by means of the same analyser after pyrolysis at 1000 ∘C. We corrected O contents for inorganic O, assuming that all inorganic O was present in form of CaCO3. The O / C and H / C ratios given in this paper represent mole ratios, whereas the C / N ratios represent mass ratios. For analysis O / C ratios and H / C ratios of samples having a SOC content lower than 150 g kg-1 were excluded from analysis. Soil carbon stocks (t C ha-1) refer to the organic horizons summed over each profile and thus do not include sediment layers that interspersed the profiles.

We selected at least two soil segments of each soil core from depths between 0–30, 30–60 and 60–100 cm for incubation to determine SOM decomposability. Each segment was divided in two subsegments, whereas one subsegment was incubated at 10 ∘C and the other subsegment at 20 ∘C for between 6 and 13 months. From the one location (M_CL) where we had taken cores of > 100 cm length, we selected six additional samples from the depth below 100 cm for incubation, resulting in a total of 560 incubated samples. Prior to incubation, we thoroughly mixed every segment, removed visible roots and adjusted the water potential to -10 kPa using a hanging water column. The sample weight was 53.9 ± 0.7 g (mean ± standard error) at -10 kPa. Following the method of Chapman (1971), we measured CO2 emission rates by means of a Respicond VII analyser (Nordgren Innovation, Sweden) over three to four measurement cycles of several weeks between November 2013 and March 2016. The measurement principle is based on the change in electrical conductivity of the NaOH solution with increasing uptake of CO2. In each cycle, we vented the alkali CO2 traps (NaOH 0.6 M) of the analyser regularly after 50–60 mg of CO2 had been emitted to prevent O2 deficiency. In addition, we exchanged the NaOH solution while the traps were vented. Between measurement cycles, we kept the soil samples at the same temperature and moisture level as during the cycles.

We only used CO2 data taken after the first 3 days of each measurement cycle for data analysis to avoid artefacts that might have resulted from moving the samples and adjusting their water content. Furthermore, we excluded all negative emission rate values (0.45 % out of 1700 CO2 measurements taken on average per sample). Data gaps (83 % of the timeline) between measurement cycles were filled by means of interpolation using a robust linear regression on the log-transformed data. The specific amount of SOC which was emitted from a sample as CO2 during 10 000 h of incubation at 10 or 20 ∘C [mg CO2-C g-1 SOC], L, was calculated as L=(CO2sample-CO2blank)×12.0144.01SOCsample×msample, where CO2sample is the amount of CO2 emitted from the sample over 10 000 h of incubation [mg CO2-C g-1 SOC], CO2blank is the median of ambient CO2 accumulation collected in six blank vessels over more than 6 months and extrapolated to 10 000 h (on average 27 mg), SOCsample is the SOC content of the sample [g kg-1], and msample is the mass of the soil sample [kg].

To determine Q10 values we applied the method used by others (e.g. Hogg et al., 1992; Scanlon and Moore, 2000; Wang et al., 2010; Wetterstedt et al., 2010; Hardie et al., 2011), dividing the 10 000 h length of the incubation period at 10 ∘C by the time span over which samples incubated at 20 ∘C emitted the same amount of CO2-C per mg SOC as those incubated at 10 ∘C emitted during 10 000 h. Given that the same amount of SOC is lost at both temperatures, changes in OM composition during incubation are also assumed to be the same and thus differences in the rates are assumed to reflect only the influence of temperature and not that of differences in composition. Q10 values are known to depend on incubation temperatures. In order to compare our results with those of other studies we calculated the activation energy (Ea in kJ mol-1) required for decomposition of SOC using Q10 values.

While R is the gas constant (8.314 J K-1 mol-1) and T is the temperature used for incubation (K). Ea=R×ln⁡(Q10)(1T1-1T2)1000 Mixed linear models were used to analyse the effects of the various soil parameters on SOM mineralization and their interactions with land use. The function lmer from the package lme4 (Bates et al., 2015) was implemented using the software R (R core Team, 2015) to run mixed linear models. Heteroscedasticity or departure from normality was assessed graphically. In order to avoid heteroscedasticity, we log-transformed topsoil C stocks and bulk density data. We tested whether the factor “land use” had a significant influence on the variation of each of the analysed variables (α=0.05). To do this, the following two mixed models, 2 and 3, were run for each dependent variable and compared using an ANOVA. variable∼land.use+randomeffectsvariable∼randomeffects Sampling depth, sampling location and site repetition were included as random effects to account for the dependence among segments of the same core and among cores from the same sampling location, respectively. In addition, we included bulk density, SOC, nitrogen, hydrogen and oxygen contents as well as the emitted CO2 as further random effects, given that there was no collinearity with the tested variable and that the Akaikes criterion (AIC) of the models revealed smaller scores with additional random effects. The additional random effects used for each model are given in Tables 2 and 3. Further, we determined the significance of land-use-specific differences (CL vs. FL, CL vs. GL, FL vs. GL) using a least square means test for linear models (lsmeans package).

We used the same approach to test the influence of the factor “soil depth” on the target variables with interactions between the three sampling depths (0–30, 30–60, > 60 cm) using the model variable∼depth.interval+randomeffects in addition to Eq. (3). To determine the significance of depth-specific differences, we used a least square means test as mentioned before.

Soil pH, SOC content, C / N ratio and bulk density showed significant land-use effects (Fig. 1, Table 2, Table S1). The lowest soil pH values were found in the forest topsoil samples, whereas SOC content and C / N ratio were the highest in these samples. Bulk density was highest in the cropland topsoils. Below 30 cm depth, soil pH, SOC content, C / N ratio and bulk density showed no land-use effect.

In the forest soil profiles, soil pH overall increased with depth, whereas it decreased in the grassland and cropland soils (Table 3). Also bulk density decreased with depth in the grassland and cropland soils, while SOC content and C / N ratio increased. In the forest soils, SOC content, bulk density and C / N ratio did not differ between topsoil (0–30 cm depth) and subsoil samples (30–60 cm depth); however, below 60 cm depth SOC was slightly lower than above, while bulk density and C / N ratio were higher than above 60 cm depth (Fig. 1, Table 3). The cumulated topsoil C stocks showed no land-use effects but tended to be larger in cropland and forest than in grassland soils over the entire profile (Fig. 1, Table 2).

The molar H / C and O / C ratios of the organic matter fell between the typical values of the ratios for carbohydrates and lignin, which is displayed in a Van Krevelen plot (Fig. 2). The lowest values of both ratios were found in the forest soils, the highest in the grassland and cropland topsoils. Both ratios were lower in the topsoils than in the subsoils of the cropland and grassland sites, while there was no difference between the two depths in the forest soils (Table 3). At depths below 30 cm, the O / C ratio was lower in the forest soils than in the other soils but without a land-use effect in the H / C ratio.

The samples incubated at 10 ∘C emitted 32.56 ± 1.39 mg CO2-C g-1 SOC, while samples incubated at 20 ∘C emitted 74.06 ± 2.98 mg CO2-C g-1 SOC (Fig. 1). At 10 ∘C we did not observe a land-use effect on CO2 emission (Fig. 1, Table 2), but at 20 ∘C the topsoil samples from croplands emitted less CO2 than those from forests. This effect occurred due to extraordinarily high emissions of the samples from two grassland and two forest sites (VW_GL, VW_F, SK_GL, SK_F) (Fig. 3). Those four sites experienced the least intensive drainage. Furthermore, these sampling sites were situated at high altitude in a pre-alpine environment with lower mean annual temperatures and higher precipitation than at the other sites (Table 1). In pairwise comparisons between adjacent sites of different land use (i.e. VW_GL vs. VW_FL, SK_GL vs. SK_FL, C_CL vs. C_GL and G_CL vs. G_GL and G_FL; Table 1), land-use effects were only found for the last site (Fig. 3).

At 10 ∘C, CO2 emissions of the topsoil samples from all sites together were higher than from samples taken at 30 to 60 cm depth, independent of land use (Table 3). When analysing the influence of depth separately by land-use type, this effect was found to only manifest in grassland and forest but not in cropland soils. We found no overall depth effect at 20 ∘C, but CO2 emissions of topsoil samples from forests were higher than those of samples taken at lower depths, whereas we found the opposite case for the cropland soils. Despite these depth effects, the general relationship between emissions and soil depth was weak and not consistent in its sign (Table 4).

Over the course of the incubation, CO2 emissions increased for 40 % of the samples, as revealed in Table S1 by positive slopes of the regression lines. These increases were independent of land use. In total, the CO2 emissions from these samples were almost 50 % higher than those from the other samples that instead showed a trend of decreasing emissions.

Mean Q10 values were 2.57 ± 0.05. The Q10 did not differ between the three land use types. It was lower below 60 cm depth in the cropland and grassland but not in the forest soils (Fig. 1, Table 3). Activation energies (Ea) calculated from Q10 values ranged around 48.1 and 123.5 kJ mol-1 and like Q10 values decreased with depth. There were significant relationships between CO2 emission and SOC content, bulk density and C / N ratio in general but they were weak (Table 4). The Q10 values showed similar relationships to these soil variables as CO2 emission.

The SOC contents, bulk densities and C / N ratios found in the deeper parts of soil profiles presented here were close to values that are typical for undisturbed peat (Grover and Baldock, 2012; Loisel et al., 2014). They also indicate that soils of our study sites were characteristic of European fens and resembled typical properties of managed organic soils (Berglund, 1995; Kechavarzi et al., 2010; Eickenscheidt et al., 2015; Krueger et al., 2015; Wüst-Galley et al., 2016; Brouns et al., 2016). Several studies assume that deeper layer peat of managed organic soils is less decomposed (Ewing and Vepraskas, 2006; Rogiers et al., 2008; Leifeld et al., 2011a, b; Krueger et al., 2015; Wüst-Galley et al., 2016). We therefore interpret the different SOM characteristics found in the topsoils of our samples as indicators of advanced decomposition triggered by drainage.

The land-use-specific differences manifested in different topsoil SOC contents and C / N ratios (highest under forest) and topsoil bulk densities (lowest under forest). The higher forest C / N ratios might be explained by absence of the use of N fertilizers and lower bulk densities by lower traffic with field machinery. In addition, differences in C / N between land-use types may also suggest that peat decomposition was less advanced in forests compared to croplands and grasslands. Further, depth effects are lowest in forest soils, indicating a lower impact of soil management that could also result in a lower decomposition of forest topsoils. The relatively high carbon stocks found in cropland top soils are most likely the result of subsidence after drainage and compaction from field traffic, leading to increased soil bulk density in the uppermost layers. This effect, with respect to C stocks, overrides the overall much smaller C concentration in agriculturally managed organic soils.

The H / C and O / C ratios in the deeper layers of the studied soils were similar to those found in undisturbed bogs and drained bogs used for forestry in Switzerland (Zaccone et al., 2007; Wüst-Galley et al., 2016). They indicate an enrichment of polyphenols and aromatic carbon with depth, which is in line with the current understanding of peat development in peatlands (Cocozza et al., 2003; Zaccone et al., 2007; Klavins et al., 2008; Delarue et al., 2011; Leifeld et al., 2012, 2017). The increased H / C and O / C ratios in the grass- and cropland topsoils can be attributed to inputs of fresh plant litter to the topsoil via above- and belowground residues, as such residues are rich in carbohydrates (Koegel-Knabner, 2002). In a previous study, in which we used stable and radiocarbon isotopes to label the SOC of two of the studied soils (C_CL and C_GL in Table 1), at least 20 % of topsoil organic matter was not peat but derived from recent plant litter (Bader et al., 2017). The results further indicated that the OM derived from these fresh plant residues was a source of labile C that contributed more to decomposition than the old, peat-derived SOM.

The H / C and O / C ratios reflect the mixing ratio of these two SOM sources. The H / C and O / C ratios in forest topsoils were lower than of those under cropland and grassland and did not change with depth. Interpreting these lower H / C and O / C ratios in the forest topsoils as indicators of more advanced peat decomposition (Klavins et al., 2008; Leifeld et al., 2012; Biester et al., 2014; Wüst-Galley et al., 2016) would be in contradiction to our conjecture that land management effects on peat decomposition, revealed by SOC, bulk density and C / N ratio, are less pronounced for forests. We rather argue that the reason for the low H / C and O / C ratio in the forest soils is a higher abundance of lignin rich (wood derived) plant residues. A second mechanism for comparably higher O / C and H / C ratios in cropland and grassland soils could be that peat loss in the uppermost layers was higher under agriculture than under forest, resulting in a relatively higher share of SOM from recent inputs. Considering all the available evidence of SOM characteristics, we conclude that peat decomposition is less advanced in forest soils than in agricultural soils and also in line with field flux measurements on managed organic soils that typically show faster decomposition in croplands and grasslands than in forests (IPCC, 2014).

The studied soils emitted, on average, ca. 5–10 % of their SOC (20 ∘C) as CO2, calculated for an incubation duration of 1 year. The advanced decomposition state of many of the samples might give reason to expect that CO2 rates are below that of more intact peat or mineral topsoils that contain a larger fraction of recent plant residues. To understand whether SOM in the studied organic soils is particularly stable, we compared its average daily carbon loss with data from studies that used undisturbed to extensively managed organic soils or mineral soils (Table 5, Fig. 4). Indeed, our values are on average below those from other organic soil studies. However, their range overlaps with the uncertainty of the regression line that is plotted through results from studies from other, mostly intact or little degraded organic soils. Hence, the pronounced oxidative decomposition after long periods of drainage might result in a relatively smaller labile SOC pool, but the large variability between experimental set-ups, incubation lengths and water contents among incubation studies prevents a stronger line of interpretation. Interestingly, the regression lines modelled for organic and mineral soils did not deviate significantly from each other. Therefore, the pools size of labile carbon, indicated by the decomposition rates, seem not to differ between these soil classes. This comparison suggests that accumulation of recent, labile plant materials that presumably account for most of the evolved CO2 is not systematically different between mineral and organic soils.

Samples showing an increase in CO2 emission rate over time were predominantly of subsoil origin, where SOC contents and C / N ratios indicate a lower decomposition than in the topsoil. Furthermore, based on the information we have on land use and drainage depths, it appears that most of these samples were taken from soil layers that were protected from intensive decomposition by water saturation. The long incubation period in our study might have given aerobic decomposer communities time to develop and grow, whereas time might not have been sufficient in shorter studies.

Like other studies on extensively managed or undisturbed organic soils that investigated depth interaction of decomposition rates in the top 30 to 50 cm (Hogg et al., 1992; Scanlon and Moore, 2000; Wang et al., 2010; Hardie et al., 2011), we found a decrease of specific CO2 release with depth. However, the relationship between CO2 emissions and depth was rather weak in our case and not consistent for both incubation temperatures and the different land uses. Compared to the studies on unmanaged organic soils, reporting declines of a factor 2 to 30 (Hogg et al., 1992; Scanlon and Moore, 2000; Wang et al., 2010; Hardie et al., 2011), our differences were substantially smaller. Drainage and decadal agricultural use of the studied soils led to more intense decomposition processes in the topsoil, resulting in little depth interaction or, for croplands, sometimes maybe even a reversal of decomposability. Further, the abundance and decomposability of crop residues have to be considered as a substantial CO2 source. For two topsoils (C_GL and C_CL), Bader et al. (2017) showed that at least 20 % of the SOM is crop residue derived and responsible for 40 % of the emitted CO2. Assuming that the abundance of crop and plant residues is highest in topsoils, it might be possible that decomposability of peat-derived SOM either does not depend on depth or topsoil peat decomposes at smaller rates. Therefore, we cannot confirm our first hypothesis of decreasing decomposition rates with depth.

As Table 5 shows, the Q10 values found in our study (2.74 ± 0.06) were higher than Q10 values found elsewhere for similar sampling depths but in unmanaged organic soils (Chapman and Thurlow, 1998; Hamdi et al., 2013; Hardie et al., 2011; Hilasvuori et al., 2013; Hogg et al., 1992; Scanlon and Moore, 2000; Yavitt et al., 2000). Also the temperature-independent Ea was higher in the studied samples (69.4 ± 3 kJ mol-1) than in most other studies on undisturbed organic soils (47.4 ± 7.2 kJ mol-1) (Table 5). However, three studies (Chapman, 1971; Hardie et al., 2011; Hogg et al., 1992) found similar or higher Ea values in northern organic soils. In the case of Chapman and Thurlow (1998) they were also managed as grassland or forest, whereas the other studies used peat from undisturbed organic soils. Nevertheless, the high Ea of the studied samples might reflect the change in chemical peat composition with decomposition after drainage towards higher recalcitrance. In contrast to other studies on unmanaged organic soils reporting no trend or increasing Q10 values with depth (Scanlon and Moore, 2000; Wang et al., 2010; Hardie et al., 2011; Hilasvuori et al., 2013), the cropland and grassland profiles in our study had a lower Q10 below the 60 cm depth. Various studies on SOM decomposition used Q10 values as an indicator of SOM recalcitrance (Hogg et al., 1992; Biasi et al., 2005; Davidson and Janssens, 2006; Conant et al., 2008, 2011; Hartley and Ineson, 2008; Hilasvuori et al., 2013). Considering that the presence of labile crop residues would decrease Q10 in the topsoil rather than in the subsoil, the higher topsoil Q10 may be explained by an extended accumulation of recalcitrant moieties. This proposed a high abundance of recalcitrant moieties in topsoils of degrading organic soils is in line with the reported differences in SOM composition in different layers as well with the pattern of CO2 emissions. The latter show no substantial difference with depth and indicate that a higher fraction of recent and labile plant residues in topsoils is counterbalanced by a high recalcitrance of the highly degraded peat. Comparing radiocarbon concentrations in SOC and emitted CO2 of two sites also used for this study (C_CL, C_GL), Bader et al. (2017) estimated that SOC from plant residue inputs is more labile than peat. The measured radiocarbon contents for SOC were 75 to 80 pMC and therefore indicated that peat of the topsoil must have experienced a substantial decomposition.

It is remarkable that despite the controlled conditions in our incubation experiment the variation in cumulative loss of initial SOC of between 0.6 and 42.3 % (Fig. 4) was similar to or even larger than that observed in field flux measurements (IPCC, 2014). This large variability suggests that the composition of SOM is of similar importance to drainage, climate and other site factors in controlling CO2 emissions from drained organic soils. Nevertheless, the relationships between the measured SOM parameters used to assess the biochemical decomposability, CO2 emissions and Q10 values were rather weak and thus do not support our second hypothesis. This stands in contrast to other studies which concluded that chemical composition is a major factor of SOM decomposability in organic soils (Scanlon and Moore, 2000; Koch et al., 2007; Reiche et al., 2010; Hardie et al., 2011; Leifeld et al., 2012). However, these studies focused mainly on single profiles of undisturbed or extensively used organic soils. A recent study investigated relationships between SOM parameters and decomposition rates of German organic soils under controlled conditions (Säurich et al., 2017). These authors mostly studied strongly disturbed fens with similar properties to the soils in our study. Besides SOC contents, soil pH and C / N ratios, Säurich et al. (2017) focused on other soil nutrients, stable isotopes and microbial biomass. In line with our results, they could not identify strong proxies for SOC decomposition by means of simple chemical attributes.

In order to explain the weak relationships between SOM composition and CO2 release it should be considered that, in our case, the emitted CO2 comprised, on average, only 3.2–7.4 % of the total SOC, while the analysed SOM parameters in this and other studies represent bulk SOM. Our methods allowed us to gain a broad overview of the chemical composition of SOM, while decomposition might more tightly be bound to the abundance of specific OM moieties.

Although land-use-affected SOM characteristics, such as elemental contents and their ratios, the amount of CO2 emitted from the soils did not differ among the three types of land use. We therefore have to also reject our third hypothesis of a higher SOM decomposition rate in forest topsoils. We assume that long-lasting drainage and management might have resulted in an equivalent decomposition of most of the labile OM, along with its intrinsic decomposability.