C in cropland soil of a long-term field trial – experimental variability and implications for estimating carbon turnover

Because of their controlled nature, the presence of independent replicates, and their known management history, long-term field experiments are key to the understanding of factors controlling soil carbon. Together with isotope measurements, they provide profound insight into soil carbon dynamics. For soil radiocarbon, an important tracer for understanding these dynamics, experimental variability across replicates is usually not accounted for; hence, a relevant source of uncertainty for quantifying turnover rates is missing. Here, for the first time, radiocarbon measurements of five independent field replicates, and for different layers, of soil from the 66-year-old controlled field experiment ZOFE in Zurich, Switzerland, are used to address this issue. C variability was the same across three different treatments and for three different soil layers between the surface and 90 cm depths. On average, experimental variability in C content was 12 times the analytical error but still, on a relative basis, smaller than variability in soil carbon concentration. Despite a relative homogeneous variability across the field and along the soil profile, the curved nature of the relationship between radiocarbon content and modelled carbon mean residence time implies that the absolute error of calculated soil carbon turnover time increases with soil depth. In our field experiment findings on topsoil carbon turnover variability would, if applied to subsoil, tend to underweight turnover variability even if experimental variability in the subsoil isotope concentration is the same. Together, experimental variability in radiocarbon is an important component in an overall uncertainty assessment of soil carbon turnover.


Introduction
Long-term agricultural field trials have long been recognized as important sources for understanding long-term management effects on soil parameters such as soil organic carbon content and turnover (Jenkinson, 1991).Their special value lies in their controlled nature, the long-term record of management activities, reliable soil and crop parameter records as well as site climate data.Many experiments have indicated that soil carbon responds sensitively to agricultural management and have allowed for sustainable management prac-tices to be identified.Hence, these data sets are also valuable sources of information for developing or testing soil and ecosystem carbon models (Smith et al., 1997;Franko et al., 2011).
Soil carbon feedback to management is controlled by organic matter input as well as turnover and hence loss.Isotopes play an important role for unravelling soil carbon turnover rates.In complementation to records on carbon storage change over time, they deliver information on how fast new carbon replaces old carbon (e.g.Trumbore, 1993).When Published by Copernicus Publications on behalf of the European Geosciences Union.Hsieh (1992) the isotopic signature, i.e. the 13 C / 12 C or 14 C / 12 C ratio, of the input material to soil is not constant over time, it induces a directed shift in the soil's isotopic signature.For a known shift in input signature, the subsequent change in the soil's isotopic signature allows for the replacement rate of old by new carbon to be estimated.For example, changes from C3 to C4 vegetation or vice versa alter the 13 C / 12 C ratio of the input and allow for estimation of turnover rates (Balesdent et al., 1988).Besides stable 13 C / 12 C, the radioactive isotope 14 C also has a long history of application in soil carbon studies (Harkness et al., 1986;Jenkinson et al., 1992;Trumbore, 1993).The introduction of extra 14 C to the atmosphere via nuclear bomb testing in the 1950s and 1960s of the last century and the subsequent diffusion of that label into terrestrial ecosystems has triggered a vast amount of research that makes use of the 14 C signature of soil carbon.The beauty of 14 C is given by its ubiquity and its potential to cover the whole relevant time frame of soil carbon turnover, ranging from years to millennia.Information from both controlled long-term field experiments and the soil's 14 C signature has been combined previously with the aim to get better insight into soil carbon dynamics (Table 1).These data are particularly useful for model development as the isotope reduces the degrees of freedom in the modelling approach -i.e. it constrains the carbon turnover dynamics and reduces the risk of giving right answers for the wrong reasons.For example, Jenkinson and Coleman (2008) used 14 C from the famous Rothamsted field trials to extend the existing Rothamsted Carbon Model (RothC) by means of a subsoil module.Hsieh (1993) took advantage of the oldest cropland experiments from the USA, Morrow plot and Sanborn field, to get insight into labile carbon turnover.However, these and other applications as listed in Table 1 have not yet considered the variability of radiocarbon in the field.Experimental variability, both across the field and within the soil profile, adds an important component of uncertainty to any modelling of terrestrial carbon.This variability needs to be known for reliable estimates of management-carbon storage feedbacks.For radiocarbon, the relatively high costs of the nowadays mostly used measurement by accelerator mass spectrometry (AMS) are a major obstacle to addressing experimental variability questions.
To our knowledge, experimental variability in soil 14 C using independent treatment replicates has not yet been addressed in any of the long-term cropland experiments listed in Table 1.Here, we aim to fill that gap by using recent 14 C measurements of mineral soil from a 66-year temperate, long-term cropland trial in Zurich, Switzerland.Two questions are studied: (i) what is the variability in soil radiocarbon content in independent replicates of a long-term field trial both in top-and subsoil, and (ii) what are possible implications of experimental variability for the interpretation of soil carbon turnover estimates?

Material and methods
The Zurich Organic Fertilization Experiment, ZOFE, was commenced in 1949 at the Swiss federal research institute for agriculture, Agroscope, in Zurich.It is located at 420 m a.s.l., receives an annual precipitation of 1040 mm, and has a mean annual temperature of 9 • C (1949-2009).The soil is a welldrained, carbonate-and stone-free, homogeneous Haplic Luvisol (IUSS Working Group WRB, 2006) (texture: clay 14 %, silt 27 %, sand 57 %).ZOFE comprises 12 different fertilization treatments with five replicates each (Fig. 1), applied to an 8-year crop rotation.The experiment is arranged in a systematic block design.A detailed experiment overview is provided by Oberholzer et al. (2014).
Soil samples were taken in April 2012 from the centre of each plot using a powered rotating soil auger (Humax, Burch AG, Rothenburg, Switzerland) down to a depth of 90 cm.The auger is equipped with an outer shaft hosting a PVC inlet that gets filled with a volumetric soil sample of diameter 5.0 cm during drilling.Samples were pooled into segments of 0-20, 20-30, 30-60, and 60-90 cm.For the present study, samples from the plough pan at 20-30 cm were not analysed.After extraction, samples were sieved < 2 mm and dried at 105 • C, roots were removed by hand, and an aliquot was finely ground.Prior to radiocarbon analysis, samples were pretreated using acid fumigation with 0.5 M HCl to remove possible remnants from liming or traces of pedogenic carbonate.Soil radiocarbon content was measured by AMS at two different facilities: the radiocarbon laboratory of ETH Zurich and the radiocarbon laboratory of the University of Bern, Switzerland.Both systems operate following the protocol of Synal et al. (2007).Radiocarbon concentrations are given as percent modern carbon (pMC) as defined by Stuiver and Polach (1977).
To study effects of experimental 14 C variability on soil carbon dynamics, we applied a common, time-dependent steady-state soil carbon turnover model.This was first described by Harkness et al. (1986) and has since been used as single-or multiple-pool versions in various studies (e.g.Baisden et al., 2013;Gaudinski et al., 2000;Harrison, 1996;Trumbore et al., 1996).The model gives mean residence times (MRTs) of soil carbon.Because we have no 14 C time series available, the most simple version of that model is applied representing a single-pool assumption as described in Leifeld et al. (2013).Although soil carbon time series are better described by multiple-pool approaches (Baisden et al., 2013), the assumption followed here is sufficient to discuss possible consequences of experimental variability for the interpretation of soil carbon dynamics.Because our data represent a single point in time, they do not allow adequate parameterization of a more complex model.Hence, we do not claim that the presented turnover estimates represent the in situ situation most realistically but rather that they allow discussion of variability effects.
The effect of depth on pMC and carbon mean residence time was tested by univariate ANOVA separately for each treatment and for the aggregated sample across treatments.
Table 2 also indicates that the coefficient of variation (CV) of pMC for five independent plots, representing mostly experimental variability, lies between 1 and 7 % (mean over nine treatment-layer combinations: 3.15 %).This is, for the present data set, 3-23 times the CV of 0.3 % pMC of the analytical precision of the AMS measurement.Notably, the CV for soil organic carbon concentration is, per treatment and layer, on average 9.5 % (data not shown) and thus 3 times that of the radiocarbon content.There was no significant depth effect on the coefficient of variation (p = 0.16); hence, 14 C variability neither increases nor decreases with depth.At the same time, the CV grouped by treatment was not statistically different between "Null", "FYM+PK", and "N2P2K2Mg" (p = 0.64).The latter implies that experimental 14 C variability as measured in ZOFE is related to site-or soil-inherent properties rather than to agricultural management.
Although 14 C variability did not significantly change with depth, it influences the variability of the derived soil carbon MRTs differently in the three layers.This can be studied by calculating turnover for the range of pMC data expressed by their average confidence interval (CI).Here, we calculate the CI of MRTs from the average coefficient of variation of (i) 0.30 % of pMC (analytical error) and (ii) 3.15 % of pMC (experimental variability, average of nine individual sample sets with each n = 5) over the measured data range.We use these average CVs for calculating CIs over the whole data range because neither treatment nor depth significantly influenced pMC variability in the field.Owing to the combination of (i) a non-constant atmospheric radiocarbon concentration as a result from long-term and short-term 14 CO 2 fluctuations and (ii) exponential radioactive decay in the soil, the relationship between pMC and MRT is non-linear.This is illustrated for a series of homogeneous soil pools of different age.Radiocarbon signatures of such a pool series with range from 70 to 105 pMC (resembling the span found in the soil data, Table 2) correspond to MRTs of between 3891 and 156 years (Fig. 2).The pMC-age curve becomes steeper at smaller radiocarbon concentrations.Whereas the central curve in Fig. 2 gives results for the mean pMC, the inner and outer bands represent the 95 % CI of (i) the average variability owing to analytical error only (inner band) and (ii) the average experimental variability in the soil.These bands give upper and lower probability limits for the calculated MRT, and the older the carbon is, the further they deviate from their mean.For example, a MRT of 3891 (CI 3392-4453) years is assigned to a soil carbon pool with signature pMC = 70 (CI 67.50-72.50pMC), whereas the same relative uncertainty for a mean pMC of 105 (Cl 101.20-108.80 pMC) corresponds to a MRT of 156 (CI 92-268) years.Figure 3 further illustrates the principle.Soil carbon from 60 to 90 cm, carrying a signature of, for example, 75 pMC, has a calculated MRT of 2947 years with deviations of +477 and −428 years, referring to the variability among five independent field replicates.The uncertainty range is reduced to +193 and −129 years (mean MRT 321 years) for a pMC of 100, roughly representing the current topsoil.While the absolute uncertainty declines the younger the soil becomes, the relative uncertainty increases in the opposite direction (Fig. 3,right).Figure 3 also exemplifies the wider uncertainty band, over the calculated pMC range, when experimental variability and not only measurement error is accounted for.At pMC 70, the uncertainty range of MRTs considering experimental variability is 4.2 times that of measurement error only.This factor increases to 6.5 at pMC 105, indicating that for younger soil carbon the omission of experimental vari-ability introduces a larger relative uncertainty than for older soil carbon.

Conclusions
Soil radiocarbon dating from a long-term agricultural experiment indicates that experimental variability of this parameter is many times the analytical error.Experimental variability seems not to be controlled by either management or soil depth.Conversion of relative uncertainty in radiocarbon content to relative uncertainty in carbon turnover reveals a higher sensitivity of carbon turnover to 14 C variability in deeper soil layers that contain older carbon.Consequently, when soil samples from a long-term field trial are pooled per depth and treatment for 14 C analysis, the underestimation of the actual experimental variability of soil carbon turnover is larger for subsoil samples where long-lived C pools are more abundant.

Figure 1 .
Figure 1.Spatial arrangement of the ZOFE field trial in Zurich indicating 12 treatments with five replicates each arranged in five blocks (I-V).Plot numbers in the lower left corner are listed together with measurements in Table1.Treatments in bold (Null, FYM+PK, N2P2K2Mg) were used for the present study.For a detailed description of all treatments, please seeOberholzer et al. (2014).

Figure 2 .
Figure2.Relationship between percent modern carbon (pMC) and calculated carbon mean residence time (MRT) using a timedependent steady-state single-pool turnover model.The inner line refers to mean values, the dashed band to the 95 % uncertainty range related to the average 14 C analytical error, and the outer, solid band to the 95 % uncertainty range related to the average experimental 14 C variability in the field.

Figure 3 .
Figure 3.Comparison of absolute (left) and relative (right) deviation of calculated MRTs from the mean, expressed as 95 % confidence interval of (i) average analytical errors (inner, dashed line) and (ii) average experimental variability in the field (outer, solid lines).In the ZOFE trial, MRTs of below 200 years resemble topsoil 0-20 cm, those of ca.1200 years resemble 30-60 cm, and those of ca.2600 years resemble 60-90 cm.

Table 1 .
Long-term agricultural field experiments where radiocarbon was used to derive soil carbon turnover estimates.

Table 2 .
Percent modern carbon (%) (±1σ uncertainty * ) of organic carbon in soil samples from the ZOFE trial taken in 2012 for three treatments and three soil layers."CV"linesindicate the coefficient of variation for each treatment-depth combination.Plot number according to Fig.1."x" refers to lab ETH and "o" refers to lab Bern.