Soil organic carbon (SOC) sequestration across
agroecosystems worldwide can contribute to mitigate the effects of climate
change by reducing levels of atmospheric CO
Carbon (C) sequestration in soils offers a significant opportunity to remove
CO
Hassink (1997) compared pairs of Dutch arable and grassland soils and found that while soil bulk SOC contents significantly differed among soils, fine-fraction OC did not. These findings led to the idea that the saturation point of the fine soil fraction could be estimated by linear regression using the mass proportion of the fine fraction in a soil sample (%) and the current fine-fraction OC (g C per kg soil). Several iterations of the concept have been proposed to overcome the limitations of linear regression. For example, boundary line analysis uses a defined upper or lower subset of a data set to estimate the boundary line, when a limiting response to one or more independent variables along a boundary is supported (Lark and Milne, 2016; Schmidt et al., 2000). Using the upper 90th percentile of a data set, boundary line analysis overcomes the limitation of linear regression, depicting the mean response to the independent variable (Feng et al., 2013; Shatar and Mcbratney, 2004), which is thought to cause an underestimation of sequestration potential. Quantile regression estimates the response of a specific quartile using the entire data set. It also makes no assumptions regarding homogeneity of variance, thus increasing the robustness of the estimated maximum fine-fraction OC. In quantile regression sample size is not reduced as in boundary line analysis (Beare et al., 2014; Cade and Noon, 2003). Using a forced zero intercept overcomes the contradiction of a positive intercept, indicating the presence of fine-fraction OC without any fine soil fraction (Beare et al., 2014; Feng et al., 2013; Liang et al., 2009). To our knowledge no comparisons between the equation developed by Hassink (1997) and one generated in the same way with a different data set have been done within the UK. This type of assessment would help to determine the suitability of the Hassink (1997) linear regression equation to predict maximum fine-fraction OC in UK soils. Without this, carbon sequestration potentials may be both over- and underestimated.
In the UK, human-managed grasslands are the dominant land use, covering 36 % of the land area (Ward et al., 2016). Managed grasslands are planted and maintained to increase agricultural productivity through fertiliser and liming applications and the reseeding of swards. The high levels of disturbance associated with reseeding events by mould board ploughing and harrowing in particular, resulting in changes in soil structure, nutrient cycling and SOC mineralisation (Carolan and Fornara, 2016; Drewer et al., 2017; Soussana et al., 2004). Organo-mineral associations form the basis of microaggregates (Baldock and Skjemstad, 2000). The destruction of aggregates makes the organic carbon protected within the aggregates more accessible for mineralisation by the soil microbial community. This may result in the increased mineralisation of existing SOC, known as the priming effect (Kuzyakov et al., 2000) The long-term effect of such a reseeding event on SOC dynamics is understudied; it is therefore important to understand how disturbance might affect OC in the fine fraction and the SOC sequestration ability of managed grasslands.
The objectives of this study were (i) to assess the suitability of the
Hassink (1997) equation to estimate maximum fine-fraction OC in UK
grassland soils; (ii) to evaluate the linear regression, boundary line and
quantile regression methods to estimate maximum fine-fraction OC; and (iii) to explore the relationship between sward age (time since the last reseeding event) and current and predicted maximum fine-fraction OC. We hypothesised
that (i) the linear regression equation developed using UK grassland soils
would be significantly different to that of Hassink (1997) and that (ii) grasslands with an older sward age would have a greater proportion of total
SOC stabilised in the fine fraction (
Summary of UK grassland site characteristics.
Ten grassland chronosequences covering a wide range of soil types, land use
and climatic conditions were identified across the UK in 2016. The sites
included the range of agricultural activity associated with UK grasslands
(upland grazing, dairy and mixed grazing), variations in soil type
(organo-mineral, mineral and chalk) and the majority of UK climatic zones
(Table 1). At each location, five to eight individual fields of different
sward age (represented by years since a ploughing and reseeding event),
ranging from 1 to 179 years, were identified for sampling. In each field,
areas were avoided which had different applications of manure, soil types or
topography, headlands, areas near gates; where lime or manure had previously
been dumped; or where livestock congregate. Two replicate soil cores were
collected to a depth of 30 cm using a soil auger with a 2.5 cm diameter
steel core and bulked to give a single composite sample. This was repeated
10 times in each field at regular intervals in a “W” shape across the field
totalling 10 replicate samples per field per site. Intact soil cores for
determining bulk densities were collected at three locations in each field
at two depths (10 to 15 cm and 20 to 25 cm) using intact rings (7.5 cm diameter and 5 cm height). Replicate samples were sieved to 2 mm, and fresh
subsamples were used to determine soil pH in water. Remaining sieved soils
were dried at 40
The fine fraction (
Hydrochloric acid (HCl) fumigation was used to remove carbonates from the
Plumpton samples. Ball-milled samples, in silver capsules, were moistened
with deionised water (
All statistical analyses were carried out using R software version 3.5.3
(Team, 2019). Significant differences were determined by ANOVA's
and post hoc Tukey's tests (
Linear regression was used to predict maximum fine-fraction OC, with the
mass proportion of the fine fraction (
Boundary line analyses were performed as an alternative to linear
regression, both with and without a forced zero intercept to predict maximum
fine-fraction OC for all UK sites. The data were organised by mass proportion
of the fine fraction (%) and divided into subgroups at 5 %, 10 % and 15 %
intervals. The 10 % interval reflects the method of Feng et al. (2013),
whilst the 5 % and 15 % intervals were used to assess the effect of the interval
on estimation of maximum fine-fraction OC. The groups were then ordered by
measured fine-fraction OC (g C per kg soil), and the values in the 90th
percentile were used to plot the boundary line. Boundary line analysis was
not used for individual sites, as it resulted in too few data points.
Quantile regression analysis was performed in RStudio using the “quantreg”
package (Koenker, 2019), for the 90th and median percentiles
(
The carbon saturation ratio was determined to identify the degree of
saturation across the sites. The carbon saturation ratio was calculated by
dividing the current fine-fraction OC by the estimated maximum fine-fraction
OC content. Values
Correlation matrix of Kendall's tau (
Measured total SOC (g C per kg soil)
The measured total SOC and fine-fraction OC concentrations varied within the
grassland sites (Fig. 1). Total SOC ranged from 8.2 to 85.8 g C per kg soil, with a median of 32.7 g C per kg soil. Hillsborough, Overton and
Plumpton had significantly higher total SOC, whilst Harpenden and Llangorse
had the lowest total SOC (
Relationships between mass proportion of the fine fraction (%) and fine-fraction organic carbon (g C per kg soil) in the soil types used in this study.
The significance of correlations between the measured soil properties, time
since reseeding and known environmental factors were analysed. The matrix of
Kendall's tau (
Analysis coefficients for the estimation of maximum fine-fraction
organic carbon by linear regression (LR), linear regression with a forced zero
intercept (LR_0), boundary line (BL) and quantile regression
(QR). Lettering indicates slopes that were significantly different within a
method (
SEM: standard error of the mean. RMSE: root mean square error. Level of significance:
The slope generated from the UK data used to estimate maximum fine-fraction
OC (Table 3) was significantly different (
Relative proportion of measured fine-fraction organic carbon of
the total SOC content of the bulk soil for the different soil types used in
this study. Lettering indicates significant differences at
Coefficients from boundary line analysis are presented in Table 3. There was no significant difference in slopes between the 5 %, 10 % and 15 % fine-fraction intervals used. The median-percentile quantile regression analysis had a similar slope to the boundary line and linear regression with a forced zero intercept. Quantile regression using the 90th percentile resulted in the steepest slope of all estimation methods (Table 3). The C saturation ratios revealed the difference in the number of samples with the potential to sequester more C (Table 4). The Hassink (1997) linear regression equation, without a forced zero intercept, predicted the greatest number of unsaturated sites, followed by the 90th percentile quantile regression, with a forced zero intercept. There was no clear relationship between oversaturated sites and the proportion of silt and clay contents, as oversaturation occurred across all proportions, indicated by points above the lines in Fig. 4.
Measured fine-fraction organic carbon (g C per kg soil) in
relation to mass proportion of the fine fraction of a soil sample (%). Lines
of best fit represent the
Carbon saturation ratios calculated from the estimated maximum fine-fraction organic carbon by linear regression (LR), linear regression with a forced zero intercept (LR_0), boundary line (BL) and quantile
regression (QR). Values
Sward age (years since the last reseeding event) had a weak positive correlation
with the mass proportion of the fine fraction (%) (Table 2). When grouped
in 5-year intervals, significant differences were found between the age group
and the mass proportion of the fine fraction (%), measured fine-fraction
OC (
Effect of sward age grouped at 5-year intervals on selected soil
properties. Values are means
Age: years since the last reseeding event. The
Determining the potential C sequestration capacity of soils is essential to predict the influence of land management for climate change mitigation. The determination of the saturation deficit using the mass proportion of the fine fraction and current fine-fraction OC content is an established method with a strong grounding in correlation between the variables. As mentioned earlier, previous studies have examined methods to improve estimates of maximum fine-fraction OC. However so far no comparison has been made between the Hassink (1997) linear regression equation and one developed using grassland soils in the UK.
The significantly different slopes for the linear regression equations (Table 3) shows that the Hassink (1997) regression equation is not suitable for estimating maximum fine-fraction OC in UK grasslands. Previous concerns have focused on the potential for the equation developed by Hassink (1997) to underestimate maximum fine-fraction OC, as least-squares linear regression represents the mean response of the independent variable, rather than the maximum. For the UK grasslands in this study, estimating maximum fine-fraction OC using the Hassink (1997) regression approach resulted in a significant overestimation of fine-fraction OC sequestration potential. Future work using maximum fine-fraction OC prediction equations reported in the literature (e.g. Beare et al., 2014; Feng et al., 2014; Hassink, 1997; Six et al., 2002) should first conduct a validity test and determine if the regression equations match the soils in question or a subset of the data to ensure results are not significantly over- or underestimated.
To overcome the contradiction of an intercept greater than zero, indicating
that C is stabilised in the fine fraction without any fine fraction, a
forced zero intercept was used. The linear regression slopes with a forced
zero intercept were not significantly different and were similar to that of
Feng et al. (2013) at 0.42
Boundary line analysis and quantile regression have been suggested as
alternatives to overcome the limitations of linear regression. The
estimation of maximum fine-fraction OC was greatest when using quantile
regression (
The strength of using quantile regression analysis is that it makes no assumptions of homogeneity of variance and uses the entire data set to estimate the upper limit of a response. The measured fine-fraction OC in the UK sites lacks homogeneity of variance (Fig. 4) where the variation in the measured fine-fraction OC increases with the proportion of the fine fraction. Standard deviation of the proportion of the fine fraction in the 10th percentile is 0.4 compared to 6.9 in the 90th percentile. Of the methods explored in this study for our grassland soils, we consider the quantile regression at the 90th percentile estimate of maximum fine-fraction OC to be the most robust. This method results in the greatest number of unsaturated samples (Table 4), suggesting great potential for additional sequestration.
When examining the estimated OC input versus existing fine-fraction OC using
estimates generated by quantile regression at the 90th percentile, a
positive correlation between current fine-fraction OC and estimated C input
(Kendall's tau (
Estimating maximum fine-fraction OC based on the mass proportion of the fine fraction is likely to be an oversimplification of the dynamics of fine-fraction OC accrual. Other parameters such as mineralogy, soil microbial community, environmental conditions (e.g. precipitation, Table 2) and land management can significantly influence fine-fraction OC stabilisation (Cotrufo et al., 2015; Kallenbach et al., 2016). This work has identified some soil and environmental properties that may play a role in fine-fraction OC stabilisation such as median annual temperature, %N and %C in the bulk soil, mean annual rainfall, and %N in the fine fraction (Table 2). Warmer median annual temperatures may enhance plant productivity and microbial processing, the byproducts of which are important precursors to fine-fraction OC (Cotrufo et al., 2013). It would be interesting to know at which point higher temperatures have a deleterious effect on fine-fraction OC accumulation. Mean annual rainfall and %N in the fine fraction were negatively correlated to fine-fraction OC. It was anticipated that fine-fraction OC would be positively correlated with fine-fraction N, as N-rich microbial byproducts have been found to form new organo-mineral associations onto which OC preferentially binds (Kopittke et al., 2018). These bonds may have been disturbed during the fractionation process, resulting in an N-rich fine fraction with less OC content.
The influence of soil type on fine-fraction OC was also evident in our
results, as all soil types had statistically significant positive
correlations between the mass proportion of the fine fraction and measured fine-fraction OC, except for Leptosols and Luvisols (Fig. 2). However, these soil
types exhibited the greatest proportion of total SOC stabilised in the fine
fraction (Fig. 3). Luvisols have a high base saturation facilitating more
fine-fraction OC stabilisation via complexation of organic ligands by free
Ca
Whilst we consider the quantile regression at the 90th percentile method to provide the most robust estimate of maximum fine-fraction OC in the sites studied, further experimental work to test the saturation level of these soils would help to validate this. Incubation studies that force an unsaturated soil to its “saturation” level and the effect of influencing variables mentioned above will help to elucidate the factors controlling fine-fraction OC saturation. Further empirical evidence of practical methods to manipulate fine-fraction OC stabilisation processes to promote the formation of new organo-mineral associations and understand their stability is necessary to guide grassland management to enhance SOC sequestration.
It was anticipated that for fields of an older sward age, a greater
proportion of total SOC would be stabilised as fine-fraction OC, as tillage
breaks up macroaggregates, making OC in the fine fraction available for
mineralisation. Alternatively, fine-fraction OC is less sensitive to
disturbance than particulate organic matter (POM), resulting in the
accumulation of POM as the fine-fraction OC pool remains stable, if
sufficiently saturated. The results seem to support neither hypothesis. The
proportion of total SOC stabilised in the fine fraction was not consistently
higher in the oldest field and, in some instances, was significantly less,
such as at Aberystwyth (Table A2). When grouped in 5-year intervals,
significant differences in the
Fine-fraction OC only accounted for 4.5 % to 50.12 %, indicating high OC
storage in other soil pools such as POM or different aggregate fractions.
The fine roots of grassland flora species promote aggregate formation
(O'Brien
and Jastrow, 2013; Rasse et al., 2005), which may be a dominant
stabilisation process in grasslands. However previous work has found no
effect of sward age or the frequency of grassland reseeding on the %C in
differing aggregate fractions (
Estimating the long-term sequestration of soil C in the fine fraction is difficult due to the lack of reliable methodologies that can be widely applied to all soils. Our study has demonstrated that the Hassink (1997) linear regression equation is not suitable to estimate maximum fine-fraction OC in a range of UK grassland soils. Therefore, caution should be applied to estimates of maximum fine-fraction OC obtained using the Hassink (1997) equation, in instances where it may not accurately reflect fine-fraction OC of the soil in situ. After exploring various univariate estimation methods, we recommend the use of quantile regression at the 90th percentile to overcome the shortfalls of least-squares linear regression. However, such a simple estimate is unlikely to accurately reflect the dynamics of fine-fraction OC stabilisation. This work has helped to identify some key parameters that play a role in fine-fraction OC stabilisation, such as median annual temperature, mean annual precipitation, bulk-soil %C and %N and fine-fraction %N. Further work to understand how these parameters influence fine-fraction OC dynamics will help to accurately assess the feasibility of achieving soil carbon sequestration targets. Our results showed little evidence of the impact of time since the last reseeding event on the OC in the fine soil fraction. However, improving our understanding of SOC stabilisation processes and their resilience to grassland management is essential to ensure that current SOC is not only enhanced but also protected.
Bulk-soil properties for each UK site. Values are means of 10 replicated in each field
Fine-fraction (
% Fine fraction: mass proportion of the fine fraction in a sample (%).
Linear regression coefficients for the estimation of maximum fine-fraction organic carbon (g C per kg soil). Lettering indicates
slopes that are significantly different (
RMSE: root mean square error. Level of significance:
Linear regression coefficients for the estimation of maximum fine-fraction organic carbon (g C per kg soil) with a forced zero
intercept. Lettering indicates slopes that are significantly different (
RMSE: root mean square error. Level of significance:
Estimated fine-fraction OC input (g C per kg soil) compared to
measured fine-fraction OC (g C per kg soil) in each of the sites
studied. The estimated fine-fraction OC input (g C per kg soil) was
calculated by subtracting the maximum fine-fraction OC (g C per kg soil) from the current fine-fraction OC (g C per kg soil). The
maximum fine-fraction OC (g C per kg soil) was estimated using the
quantile regression equation (
All data resulting from this study are available from the authors upon request to Sarah Buckingham (sarah.buckingham@sruc.ac.uk).
KCP, SB, JMC, RMR and EMB formulated the research question and study design. KCP conducted the experimental work, data analysis and prepared the draft of the paper. All authors contributed to editing and reviewing of the paper.
The authors declare that they have no conflict of interest.
We are grateful to John Parker and Lydia Guo for their assistance in both the field and laboratory. We also thank Steve Freeman for technical assistance and Margaret Glendining and Sarah Perryman for access to information and data from the electronic Rothamsted Archive (e-RA). The Rothamsted Long Term Experiments National Capability (LTE-NCG) is supported by the UK Biotechnology and Biological Sciences Research Council (grant no. BBS/E/C/000J0300) and the Lawes Agricultural Trust. The authors are also grateful to collaborators at the local field sites for supplying soil samples and management information.
This research has been supported by SRUC's postgraduate studentship programme and the Global Academy of Agriculture and Food Security, University of Edinburgh. Funding has also been provided by Business Environment, Industry and Strategy (grant no. TRN1133); Ricardo-AEA; and the Rural & Environment Science & Analytical Services Division of the Scottish government.
This paper was edited by Sara Vicca and reviewed by Emanuele Lugato and Steffen A. Schweizer.