Vulnerability of soil organic matter of anthropogenically disturbed 1 organic soils 2

1 Thünen Institute of Climate-Smart Agriculture, Bundesallee 50, 38116 Braunschweig, Germany 6 2 now at: KU Leuven, Department of Earth and Environmental Sciences, Division Soil and Water Management, 7 Celestijnenlaan 200 E, B-3001 Heverlee, Belgium 8 3 University of Bonn, Institute of Crop Science and Resource Conservation, Soil Science and Soil Ecology, Nussallee 13, 9 53115 Bonn, Germany 10 4 now at: Bavarian State Research Center for Agriculture, Institute of Organic Farming, Agricultural Sciences and Natural 11 Resources, Lange Point 12, 85354 Freising, Germany 12

As previous investigations mainly have focused either on mineral (< 150 g SOM kg -1 according to the German definition, 69 Ad-Hoc-AG Boden, 2005) or "true" peat soils (> 300 g SOM kg -1 ), there are very few studies on soil properties or SOM 70 dynamics of "low C organic soils" (between 150 and 300 g SOM kg -1 ). However, measurements of GHG emissions in the 71 field have shown that organic soils with a SOC content of around 100 g kg -1 still emit large amounts of CO 2 similar to the 72 levels emitted by "true" peat soils (Leiber-Sauheitl et al., 2014;Tiemeyer et al., 2016). This is rather surprising as the 73 remaining organic matter should not be readily available for mineralisation, given that the SOC content at this stage of 74 decomposition is fairly low and CO 2 emissions and SOC content are closely related in mineral soils (Don et  The aims of this study, were i) to assess the sensitivity of SOM from anthropogenically disturbed organic soils under 81 agricultural use to mineralisation under aerobic conditions and ii) to determine the indicators and drivers of the vulnerability 82 of SOM. In this context, disturbance was defined as the effect of soil-forming processes induced by drainage and/or by the 83 mixing of peat with mineral soil. For this purpose, 91 samples of soils were examined under cropland and grassland from 84 across Germany, ranging from carbon-rich mineral soil (70 g SOC kg -1 ) to "true" peatlands (up to 560 g SOC kg -1 ). As a 85 simulation of the potential effects of drainage, all the samples were aerobically incubated in the laboratory at standardised 86 water content. Hoddesdon, UK) was used to measure CO 2 concentrations. Each replicate sample was measured hourly over an incubation 138 time of at least 40 h or until a relatively constant BR was reached (up to 90 h). 139 Afterwards soil samples were amended with a mixture of 100 mg glucose and 100 mg talcum using an electronic stir for 30 s 140 to determine the active microbial biomass using the SIR method. The mixture was then incubated again for 6 h to obtain the 141 maximal initial respiratory response of the microbial biomass (Anderson et al., 1995). 142

Data analysis 143
Statistical analysis was performed using the R software environment (version R-3.1.3, R Core Team, 2015). 144

Determination of basal soil respiration 145
The measured BR is expressed as µg CO 2 -C g soil -1 h -1 and the specific basal respiration (SBR) is normalised by the 146 sample's SOC content into µg CO 2 -C g SOC -1 h -1 . An exponential model was fitted simultaneously to all three incubation 147 replicates to determine the equilibrium values of the SBR (Figure): 148 (1) 149 where CO 2 -C (t) [µg CO 2 -C g SOC -1 h -1 ] is the specific CO 2 production per hour, a [µg CO 2 -C g SOC -1 h -1 ] is the initial 150 respiration and k [h -1 ] is the change rate of SBR. 151 To achieve an objective quantification of the (specific) basal respiration and its uncertainty, the R package "dream" was used 152 (Guillaume and Andrews, 2012), which is based on the iterative Markov Chain Monte Carlo (MCMC) approach. This 153 method is basically a Markov chain that generates a random walk through the high-probability-density region in the 154 parameter space, separating behavioural from non-behavioural solutions following the probability distribution (Vrugt et al.,155 2009b). The differential evolution adaptive metropolis (DREAM) algorithm is an efficient MCMC sampler that runs 156 multiple Markov chains simultaneously for global exploration of the parameter space. In doing so, DREAM uses a 157 differential algorithm for population evolution and a metropolis selection rule to decide whether a population of candidate 158 points is accepted or not. After the burn-in period, the convergence of individual chains is checked using the Gelman and 159 Rubin (1992) convergence criterion, which examines the variance between and within chains (Vrugt et al., 2008(Vrugt et al., , 2009a. 160 Once the convergence criterion of Gelman and Rubin was < 1.01, another 500,000 simulations were run to determine the 161 posterior probability density functions of the model parameters, which were used to calculate the median and the 2.5 and 162 97.5 % quantiles. 163 For the evaluation of the SIR experiment, the value of the maximum initial respiratory response was identified manually and 164 then transcribed to microbial biomass (SIR-C mic ) [µg g -1 soil] as follows (Kaiser et al., 1992): 165 SIR-C mic = µl CO 2 g −1 soil * 30.
(2) 166 To quantify the efficiency of microbial respiration per unit biomass, the metabolic or respiratory quotient q(CO 2 ) [mg CO 2 -167 C h -1 g -1 biomass SIR-C mic ] was calculated by dividing the BR by the SIR-C mic (Anderson and Domsch, 1985): 168 (3) 169 The higher the value of q(CO 2 ), the higher the CO 2 emissions per unit microbial biomass, indicating a lack of available C for 170 metabolism in the soil. 171

Degree of disturbance 172
The present classification of anthropogenic disturbance is based on the mapped soil horizons from which the samples 173 originated. The soil horizons and the degree of decomposition after von Post were mapped according to the German manual 174 of soil mapping (Ad-Hoc-AG Boden, 2005 has not yet been a secondary transformation of the peat structure. Earthified topsoils are defined as "moderately disturbed". 181 Strong disturbance is characterised by blocky to prismatic aggregates and/or the formation of shrinkage cracks, and is only 182 found in subsoils. In the present sample set, this level of disturbance only occurred in fen peat. Finally, both highly 183 decomposed dusty "moorsh" and mixtures of peat and mineral soil have been defined as "heavily disturbed". This class also 184 only occurred in fen peat. Overall there are five fen classes, three bog classes, and one class each for gyttja (organic or 185 calcareous sediments) and other samples. Samples from the class other were organic marsh soils or could not be assigned to 186 any disturbance class (e.g. buried organic soils). For further information see Table A1 in the appendix. Given that this 187 classification was developed after the sample selection, the distribution among the groups is not uniform. The von Post scale 188 from H1 to H10 was altered by adding H11 for low C organic soils deriving from peat. Gyttja and other remaining samples

Statistical and multivariate analysis 195
In a first step, Spearman's rank correlation coefficient r s was evaluated for the specific basal respiration and all measured 196 explanatory variables using the R package "Hmisc" (Harrell, 2016). The p-values were adjusted using the method after 197 Bonferroni. Differences between the results of disturbance classes for BR, SBR, SIR-C mic and q(CO 2 ) were determined using 198 an analysis of variance. P-values were computed with the Tukey 'honest significant differences' test (α = 0.05) and adjusted 199 with the Bonferroni correction using the R package "multcomp" ( Afterwards, the best model was cross-validated using the "leave-one-site-out" approach. 212 All the results given below are medians with standard errors, unless otherwise stated.

Vulnerability of SOM as determined by respiration rates 216
For all classes, BR was highly variable, ranging from 0.3 to 7.0 µg CO 2 -C g soil -1 h -1 (Fig. 1a). The BR rates of fen samples 217 decreased with an increasing degree of disturbance due to concomitantly decreasing SOC content (Table 2), while bog 218 samples behaved inversely. Overall, bog samples (2.0 ± 0.3 µg CO 2 -C g soil -1 h -1 ) had similar BR rates to fen samples 219 (2.5 ± 0.2 µg CO 2 -C g soil -1 h -1 ). Gyttja (1.3 ± 0.3 µg CO 2 -C g soil -1 h -1 ) and other samples (1.1 ± 0.2 µg CO 2 -C g soil -1 h -1 ) 220 showed significantly lower BR rates than undisturbed, slightly and strongly disturbed fen samples (D0F, D1F, D3F). 221 Overall, fen samples had significantly higher (p < 0.01) average SBR rates of 8.3 ± 0.7 µg CO 2 -C g SOC -1 h -1 than bog 222 samples (5.1 ± 0.9 µg CO 2 -C g SOC -1 h -1 ). This difference was especially clear for undisturbed and slightly disturbed 223 samples. SBR rates were also highly variable between and within classes, and ranging from 1.5 to 25.1 µg CO 2 -C g SOC -1 h -224 von Post decomposition degree of H3 and H7 had significantly lower (p < 0.05) SBR rates of 6.3 ± 1.0 µg CO 2 -C g SOC -1 h -1 250 and 7.8 ± 0.8 µg CO 2 -C g SOC -1 h -1 , respectively, than samples mapped as H10 (13.4 ± 1.2 µg CO 2 -C g SOC -1 h -1 ). 251 With decreasing C:N-ratios, SBR rates increased in an exponential manner (Fig. 3c). However, when splitting the samples 252 into two groups at C:N = 25, there was no longer any correlation for any of the groups. Again, the highest and most variable 253 rates of 10.4 ± 0.6 µg CO 2 -C g SOC -1 h -1 were measured for highly disturbed samples with low C:N ratios < 25, which 254 mainly belong to all fen classes, gyttja, D2B and other (see Table 2). In contrast, samples with a C:N-ratio > 25 were bog 255 samples with low or minimal disturbance, which had significantly lower (p < 0.001) and less variable SBR rates of 256 4.1 ± 0.6 µg CO 2 -C g SOC -1 h -1 . In detail, there was a strong negative correlation between SBR rates and the C:N-ratio for 257 D0F (r s = -0.73) and gyttja samples (r s = -0.85, p < 0.05). Moderate correlations were found for other (r s = -0.56), D1F (r s = -258 0.47), D2F (r s = -0.32) and D0B (r s = -0.47) but these correlations were not significant (Fig. 4).  Even though there was no general relationship between N t and SBR rates (Fig. 3d), N t concentrations were positively 265 correlated with the respiration rates of the disturbance classes D0F (r s = 0.57), D4F (r s = 0.37), D0B (r s = 0.47) and gyttja 266 samples (r s = 0.55) (Fig. 4). In contrast, the samples of D3F (r s = -0.6) were negatively correlated. 267 There was a significant (p < 0.001) difference in δ 15 N values between samples from undisturbed and disturbed horizons 268 (Fig. 3e). The mean values for undisturbed horizons were 0.0 ± 0.6 ‰ (D0B) and -0.3 ± 0.4 ‰ (D0F) respectively. All the 269 other disturbance classes showed higher δ 15 N values up to 11.2 ‰, and a slight overall increase in SBR rates with increasing 270 δ 15 N. However, mainly within (relatively) undisturbed classes, there were negative correlations between δ 15 N and SBR rates 271 for D0B (r s = -0.5), D1B (r s = -0.7) and D3F (r s = -0.3) (Fig. 4). In contrast, there were slightly positive correlations in the 272 case of heavily disturbed D4F samples (r s = 0.33). 273 Overall, P CAL showed significant positive correlations with the SBR rates (Fig. 2)  linear increase of SBR with P CAL (Fig. 3f). In the case of bogs, the correlation increased with increasing disturbance (D0B: 276 r s = 0.42, D1B: r s = 0.5, D2B: r s = 0.77, Fig. 4). Overall the bog samples (r s = 0.71, p < 0.05) had a significantly strong 277 dependence and the fen samples (r s = 0.49, p < 0.01) a significantly moderate dependence on P CAL . The effect of the 278 disturbance class was less consistent in the case of fens compared to bogs, with the strongest correlation in D3F (r s = 1, 279 Considering all the samples, SBR rates increased linearly with increasing pH (Fig. 3g), reflecting the general differences 281 between bogs and fens. This increase in SBR was most distinctive for D0F (r s = 0.85, p < 0.05), D1B (r s = 0.5) and gyttja 282 samples (r s = 0.78). The correlation with pH values was moderate for D0B (r s = 0.47), other (r s = 0.37) and D2F samples 283 (r s = -0.46), which were negatively correlated (Fig. 4). Overall, the SBR rates of bog samples correlated strongly with pH 284 (r s = 0.74, p < 0.05), even though bog samples covered only a small range (3.5 ± 0.1) of the overall pH values, whereas other 285 samples had the widest range (5.2 ± 0.3) followed by gyttja (4.8 ± 0.4) and fen samples (5.6 ± 0.2). 286 The concentration of iron oxides had a linear positive relationship with SBR rates (Fig. 2), which was especially noticeable 287 in the strong correlation with D2F samples (r s = 0.96, p < 0.05). Furthermore, rates of D1F (r s = 0.37), D1B (r s = 0.4) and 288 other samples (r s = 0.36) showed moderate positive dependence on Fe O (Fig. 3h; Fig. 4). However, this effect was not 289 systematic as the SBR rates of samples of disturbance class D3F (r s = -0.5) were negatively correlated with iron   D2=moderate disturbance, D3=strong disturbance, D4=heavy disturbance) and soil properties: SOC: soil organic carbon content, C:N-301 ratio: carbon to nitrogen ratio, δ 15 N, N t : total nitrogen content, P CAL : calcium acetate lactate (CAL) extractable phosphorus content, pH-302 value, Fe O : oxalate extractable iron content, ρ: bulk density, as well as the determined specific microbial biomass (SIR-C mic ) 303

Relation between microbial biomass and mineralisation rates 304
Overall, the specific microbial biomass (Fig. 5a) was positively correlated with SBR (r s = 0.75) and thus followed a similar 305 pattern to SBR (Fig. 1b) across disturbance classes: values were higher for fen samples than for bog samples and tended to 306 increase with increasing disturbance, especially in the case of bog samples. There were strong positive correlations between 307 SIR-C mic and SBR for all disturbance classes (Fig. 4), except D2F (r s = 0.45), which was moderately correlated with SIR-C mic 308 and D3F. Samples of the classes D2B (r s = 1, p < 0.001) and gyttja (r s = 0.88, p < 0.05) showed the highest dependencies. 309 Overall, specific SIR-C mic was highest for D4F samples (3249 ± 411 µg C g -1 SOC) followed by D3F (2293 ± 612 µg C g -310 1 SOC) and D2B samples (2265 ± 400 µg C g -1 SOC). Gyttja (1907 ± 339 µg C g -1 SOC) and other samples 311 (1528 ± 533 µg C g -1 SOC) had relatively high values of microbial specific SIR-C mic , with the latter being the most variable 312 of all the groups. 313 Specific SIR-C mic showed similar relationships with explanatory variables as SBR (Fig. 2). 314 There were no significant differences in the metabolic quotient between disturbance classes (Fig. 5b). The values of the 315 classes D0F, D1F, D2F, D4F and D2B were slightly lower than those of the other classes, and there was a slight tendency for 316 a decreasing metabolic quotient with increasing disturbance for bog samples.

Enhanced SOM vulnerability to decomposition with increasing disturbance of peat 325
The most striking result of the present experiment was that specific basal respiration rates increased both in magnitude and 326 variability with increasing degradation and disturbance of the peat soils, irrespective of whether this degradation was 327 expressed as a von Post value, a disturbance class or SOC content (Figs. 3 and 4). This finding was best illustrated in the bog 328 samples and manifested itself in the more variable specific basal respiration rates with a smaller SOC content. Second, in 329 contrast to less disturbed peat samples, it proved to be practically impossible to describe the specific basal respiration or 330 specific microbial biomass for heavily disturbed fen samples with the set of explanatory variables used here (Fig. 4). In 331 contrast to specific basal respiration, it is noteworthy that the basal respiration rates tended to increase with increasing 332 disturbance for bog samples, while there was a significant decrease for fen samples, i.e. the effects of disturbance on total 333 basal respiration were soil specific (Fig. 1a). 334 (Wells and Williams, 1996) and effects of physical disturbance (Ross and Malcolm, 1988;Rovdan et al., 2002) might 356 contribute to this effect. Furthermore disturbance might reactivate enzymes that were previously inactive, thus enhancing 357 mineralisation rates under favourable conditions (Freeman et al., 1996(Freeman et al., , 2001. Freeman et al. (2001) found that the enzyme 358 phenol oxidase has a tremendous effect on increasing the peat decomposition under aerobic conditions. These effects might 359 cause the strong increase in specific basal respiration rates and specific microbial biomass values from undisturbed/slightly 360 disturbed to moderately disturbed bog samples. changes during decomposition differ between peat-forming plants (Bohlin et al., 1989), that can no longer be identified and 388 are more diverse in bogs than in fens. Furthermore, the class of heavily disturbed fen samples combines samples which have 389 been amended by mineral soil by different processes (e.g. ploughing, application from external sources, or natural 390 sedimentation in riverine fens) and those which have become "moorshy", but were not amended. To disentangle different 391 processes, a larger number of samples and more detailed information on the sites' history will be required in future studies. 392 Finally, the DREAM-fits showed the largest uncertainty for some of the samples of the class D4F (Fig. 3a), which might 393 have contributed to the difficulty in finding appropriate explanatory variables. 394

Nutrients and acidity as indicators of SOM vulnerability 395
Agriculturally used peats drained for a long time are often enriched in N and (labile) P concentrations (Laiho et al., 1998;396 negative relationship by a restricted efficiency of C metabolism of the microbial biomass in bogs due to the acidic 420 environment, which was however not the case in the present study's samples set (see section 4.4; Fig. 5b), and contradict the 421 common observation that bogs show higher respiration rates than fens (e.g. Bridgham and Richardson, 1992; Urbanova and 422 Barta, 2015). Although all three studies used undisturbed peatland sites, their samples had a smaller pH range and they only 423 sampled to a depth of 30 cm, instead of up to 200 cm as in the present study. This and the fact that a broader sampling site 424 basis was used here could explain the contrasting correlation. However, with increasing disturbance the influence of pH 425 diminishes in the present samples, possibly due to better nutrient availability and increased pH-values overall in bogs. 426 As already mentioned, it was impossible to identify any strong correlations between soil properties and specific basal 427 respiration or SIR-C mic of heavily disturbed fens samples (D4F Fig. 4). Since these soils have a comparably low SOC content 428 (142 ± 12 g kg -1 , Table 2), they have become increasingly similar to mineral soils. It could therefore be expected that 429 stabilisation mechanisms for SOM become more similar to mineral soils. However, Fe O , which has been shown to be 430 important for SOM stabilisation (Wagai and Mayer, 2007 and references therein), is of minor importance for specific basal 431

4.5
Implications for peatland management 452 The high specific basal respiration rates of heavily disturbed samples confirm the vulnerability of "low C organic soils" that 453 has already been identified in field studies (Leiber-Sauheitl et al., 2014;Tiemeyer et al., 2016). Potential emissions do not 454 reach a constant level, and do not always decrease or stop with increasing disturbance. The SOC content below which such 455 soils behave like mineral soils does not seem to be within the studied SOC range. However, there were heavily disturbed 456 samples in this present study that showed low potential emissions which agrees with the finding that the variability of CO 2 457 emissions from "low C organic soils" field studies is high. Therefore mixing organic soil with mineral soil does not seem to 458 mitigate respiration rates by the potential stabilisation effect of clay, but on average increases the vulnerability of SOM. 459 However, for specific samples the respiration rates are still rather unpredictable. By ploughing mineral soil into the peat 460 layer, a whole new soil horizon develops, that may include modified microbial communities and potentially fresh SOM after 461 disaggregation of the peat takes place (Ross and Malcolm, 1988). Applying N and especially P fertilisers on peatlands might 462 increase the specific basal respiration rates (Amador and Jones, 1993)

5
Conclusions 471 This study examined the vulnerability of SOM of organic soils to decomposition by determining the specific basal 472 respiration rates under aerobic conditions in the laboratory. It was shown that the specific basal respiration increased in 473 magnitude and variability with increasing disturbance, and that it was at its highest and most variable at the boundary 474 between mineral and organic soils. At this boundary heavily degraded organic soil or peat soils mixed with mineral soil 475 prevailed, and therefore it was surprising that a decreasing trend of specific basal respiration with higher SOC was identified. 476 Furthermore, bog samples seemed to be more sensitive to anthropogenic disturbance than fen samples as indicated by a 477 stronger increase of specific respiration rates with increasing disturbance. Overall, the most important indicators for the 478 vulnerability of SOM identified in the present study were narrow C:N-ratios, higher pH-values, lower SOC content, and 479 higher concentrations of available phosphorus. There seems to be a positive feedback loop of disturbance and increased 480 mineralisation. However no explanation could be found for the very variable specific basal respiration of heavily disturbed 481 ("moorshy") fen peat and mineral soil-peat mixtures. For these types of soils, more sophisticated indicators of vulnerability 482 still need to be identified. Given the continued drainage and disturbance of peatlands and the considerable potential of CO 2 483 emissions from heavily disturbed organic soil presented here, future research needs to be concentrated on identifying 484 hotspots within these very heterogeneous soils for correctly targeting mitigation measures. Furthermore mixing peat with 485 mineral soils does not seem to be a promising option to mitigate emissions.   Topsoil horizon of intensively drained sites, "moorshy", dusty or small-grained structure when dry, intensive aerobic decomposition, plant residuals not visible anymore ("Mulm") OR horizons with a high content of mineral soil due to ploughing, mineralization, anthropogenic addition from external sources, or addition from natural sources (sedimentation in riverine fens or translocation by wind). All earthified ploughed horizons (Hvp) in our dataset had a low SOC content pointing to mixing with mineral soil and were thus classified as "heavily disturbed". Due to the strongly disturbed conditions of these topsoils, it was impossible to distinguish between the underlying different processes.

Gyttja fF
Organic or calcareous lacustrine sediments mainly in terrestrialization peatlands. Due to the lack of an English translation of the German term "Mudde", the term "gyttja" was used here for all these sediments, although it describes calcareous sediments only in the