What is a good level of soil organic matter? An index based on organic carbon to clay ratio

Simple measures of appropriate levels of soil organic matter are needed for soil evaluation, management and monitoring, based on readily measurable soil properties. We test an index of soil organic matter based on the soil organic carbon (SOC) to clay ratio, defined by thresholds of SOC/clay ratio for specified levels of soil structural quality. The thresholds were originally delineated for a small number of Swiss soils. We assess the index using data from the initial sampling (1978–83) of the National Soil Inventory of England and Wales, covering 3,809 sites under arable land, grassland and woodland. Land use, soil type, annual precipitation and soil pH together explained 21% of the variance in SOC/clay ratio in the dataset, with land use the most important variable. Thresholds of SOC/clay ratio of 1/8, 1/10 and 1/13 indicated the boundaries between “very good”, “good”, “moderate” and “degraded” levels of structural condition. On this scale, 38.2, 6.6 and 5.6% of arable, grassland and woodland sites, respectively, were degraded. The index gives a method to assess and monitor soil organic matter at national, regional or sub‐regional scales based on two routinely measured soil properties. Given the wide range of soils and land uses across England and Wales in the dataset used to test the index, we suggest it should apply to other European soils in similar climate zones.

K E Y W O R D S land use, soil clay content, soil organic carbon, soil structure

| INTRODUCTION
What is a good level of soil organic matter? Maintaining and, if possible, increasing the level of soil organic matter is generally a good thing for most functions expected of soils, including carbon sequestration, and increased levels improve soil structure. Farmers, food producers and governments need to know their soil status in relation to a critical value of soil organic matter. However, as soil organic matter varies with land use, soil type, location and other variables, an index for gauging the level of soil organic matter under given conditions needs to account for these variables. Verheijen, Bellamy, Kibblewhite, and Gaunt (2005) derived indicative ranges of soil organic carbon (SOC) content for arable soils of England and Wales that are potentially attained under different types of management and environmental conditions, and they found that clay content, precipitation and depth of topsoil could explain 25% of the variation in SOC content. Clay soils under wetter conditions had higher values than more-sandy soils, and grassland soils had higher values than arable soils with similar clay content. Clay content is a key factor because of its effects on SOC protection, including adsorption on mineral surfaces and within soil aggregates (Dungait, Hopkins, Gregory, & Whitmore, 2012;Six, Conant, Paul, & Paustian, 2002). Under constant land management and organic matter inputs, soils tend towards a steady-state SOC content, with a capacity for stabilizing SOC modelled as a function of clay content (Hassink, 1997;Hassink & Whitmore, 1997;Six et al., 2002;Stewart, Paustian, Conant, Plante, & Six, 2007). Dexter et al. (2008) found that soil physical properties (bulk density, water retention characteristics and clay dispersibility) could be better explained by the relative amounts of SOC and clay content to each other than by their total contents. In their analysis of data on French and Polish arable and grassland soils, maxima of correlations between the mass of clay per unit mass of SOC and soil physical properties corresponded to SOC/clay = 1/10, the SOC/clay content ratio was a good indicator of soil physical conditions, and this ratio gave a general separation between the different land uses. These findings were subsequently supported by others in Denmark (de Jonge, Moldrup, & Schjønning, 2009;Schjønning et al., 2012) and a study in England (Jensen et al., 2019). Johannes et al. (2017) developed the approach further and, in an analysis of Swiss soils, defined SOC/clay thresholds of 1/ 8, 1/10 and 1/13 as indicating the boundaries between "very good", "good", "suggest improvement" and "poor" levels of structural condition.
In this paper, we assess the three SOC/clay thresholds of Johannes et al. (2017) for soils across different land uses and climates in England and Wales. We use data from the original sampling of the National Soil Inventory (NSI), which contains information on soils at 5,662 sites under agricultural and non-agricultural land uses across the two countries (Bellamy, Loveland, Bradley, Lark, & Kirk, 2005). This is a far larger dataset with greater variation in soils, environments and land use than the datasets used by Dexter et al. (2008) and Johannes et al. (2017), and so provides a more comprehensive test of the SOC/ clay ratio. We have three objectives. First, to assess the variation in SOC/clay ratio and its drivers across the NSI dataset. Second, to test its ability to delineate soils of different structural quality. Third, to illustrate the use of the SOC/clay index for mapping soil carbon across England and Wales, and for gauging changes in a long-term experiment with contrasting organic and inorganic fertiliser treatments.

| National-scale data
The NSI was first conducted between 1978 and1983. Topsoil (0-15-cm depth) samples were collected at the intersections of an orthogonal 5-km grid over the entire area. A full description of the survey methods, analytical methods and available data is given in the LandIS database (www.landis.org.uk; Proctor, Siddons, Jones, Bellamy, & Keay, 1998). We considered only arable, ley grassland, permanent grassland and woodland sites, and excluded sites without measurements of soil clay content, pH or depth of topsoil, or that were classified as "peat". To reduce the impact of sites with very high SOC content relative to clay content, we excluded 290 outliers with SOC/clay > third quartile +1.5 × interquartile range. This gave 3,809 sites. Figure 1 shows the distribution of the sites across the two countries and Table 1 gives summary statistics for SOC and clay contents.
Soils at each site were classified by major soil group (Avery, 1980). Data on soil carbonate content were obtained from field observations of fizzing on addition of HCl to samples on a 5-point scale from non-calcareous to very calcareous.
Soil structural quality was characterized using the Agricultural Land Classification of England and Wales (MAFF, 1988), which gave scores of good, moderate or poor structural quality according to the texture and shape, size and development of aggregates, and friability of subsoil. The NSI contains values for each of these except friability; therefore, we made estimates based on the shape and size criteria (and where possible development of aggregates was taken into account) (Table S1).
Monthly average precipitation was obtained from the UKCP09 dataset (Met Office, 2017). Mean accumulated annual precipitation was calculated for the years 1910-1983 and values at each NSI site were intersected using ArcGIS version 10.4. (ESRI, 2015). Ranges for precipitation classes were taken from Verheijen et al. (2005): < 650, 650-800 and 800-1,100 mm year −1 , with the addition of "very wet" for annual precipitation >1,100 mm year −1 .

| Data analysis
Statistical analyses were performed in R version 3.5.0 (R Core Team, 2017). Random forest analysis (package: randomForest; Liaw & Wiener, 2002) was used to analyse the variance of SOC/clay with land use, average annual precipitation, major soil group, pH, lower depth of topsoil, calcareous score and risk of flooding. A square-root transformation was applied to SOC/clay to reduce the skewness of the data. Three-quarters of the data (n = 2,857) were used as a training set, and the RMSE and R 2 values of predictions of the remaining set (n = 952) were calculated. Training and sample sets were randomly selected. Spatial or other correlations across training and validation sets were unlikely because only topsoil samples were used and the minimum distance between sites was 5 km.
Chi-squared tests were used to compare numbers of sites within SOC/clay ranges under different land uses and precipitation classes and to test the relationship between the SOC/clay thresholds and soil structure. We used the results of statistically significant chisquared tests to interpret interactions between variables, with contributions from specific combinations of variables to the chi-squared statistic inferred from the differences between observed frequency and that expected if there was no interaction between the variables.

| Field-scale data
We assessed the effects of field-scale soil management on SOC/clay ratios relative to the threshold values using data from a long-term organic manuring experiment at Woburn, Bedfordshire, UK (Mattingly, 1974). The experiment had eight treatments with four replicates: (a) peat for 6 years then ley, (b) farmyard manure (FYM), (c) grass ley plus nitrogen, (d) grassclover ley, (e) green manure (GM) for 6 years then ley, (f) straw, and (g) and (h) two inorganic fertiliser treatments (details in Mattingly, 1974). Treatments were applied in two cycles (1965 to 1972 and 1979 to 1987); the second cycle of treatment is denoted by "then" above if different from the first. We calculated SOC/clay ratios for each plot and then averaged the values for each treatment. The plot-level soil clay content ranged from 78 to 131 g kg −1 and the initial SOC content ranged from 5.7 to 8.6 g kg −1 (Table S2).
F I G U R E 2 Soil organic carbon (SOC) content as a function of clay content for different land uses. Lines are SOC/clay thresholds:

| Variation in SOC and clay contents and SOC/clay ratio
Mean SOC contents increased in the order arable << ley grass < permanent grass ≈ woodland soils (Table 1). Mean clay contents and their ranges were similar across land uses, except those of woodland soils were smaller. The dominant soil types in all land uses were brown soils and surface-water gleys; the proportions of other soil groups varied (Table S3).
Arable sites tended to have smaller average annual precipitation than the other land uses (Tables S4  and S5). The proportions of sites above and below the three SOC/clay thresholds differed between land uses, particularly for the SOC/clay = 1/13 threshold ( Figure 2 and Table 2). A greater proportion of arable sites had SOC/clay <1/13 (i.e., depleted in SOC for their clay content) and a greater proportion of permanent grassland and woodland sites had SOC/clay >1/8 (i.e., enriched in SOC for their clay content; X 2 (9) = 681.3, p < .001).
Analysis of the influence of land use, soil and other variables on SOC/clay ratio by random forest analysis showed that 21.0% of the variance was explained by the variables examined (Table 3). Land use, average annual precipitation, major soil group and pH were more important than carbonate score, flood risk and depth of topsoil. When the model was run with just the top four variables, the variance explained did not change; however, the importance of land use increased relative to the other variables.

| Effects of land use and precipitation
The effect of land use was clear, with lower SOC/clay ratios observed for arable land and predominantly higher SOC/clay ratios observed for grassland and woodland ( Table 2). As there was some geographical relationship between the distributions of land use and precipitation, the effects of each on numbers of sites relative to the SOC/clay thresholds were considered. Verheijen et al. (2005) suggested that dry sandy soils were more at risk of lower SOC content than wetter clayey soils and that grassland soils would have higher SOC content than (ley-) arable soils. Comparing SOC/clay threshold ranges, land uses and precipitation classes (< 650, 650 to 800, 800 to 1,100 and > 1,100 mm year −1 ; Table S6), two questions were asked. (i) Were arable soils under dry climate conditions more likely to have SOC/clay <13 than arable soils under wetter climate conditions? (ii) For soils under dry climate conditions (< 650 mm year −1 ), were arable soils more likely to have SOC/clay <13 than other land uses?
In answer to the first question, chi-squared analysis showed that precipitation class was not independent of SOC/clay ratio for arable soils (X 2 (9) = 78.9, p < .001). Comparing the contributions of each combination to the chi-squared statistic showed that a larger number of soils receiving less than 650 mm year −1 and smaller numbers of soils receiving more than 650 mm year −1 than expected had SOC/clay <1/13. Also, a smaller number of dry soils and larger number of soils with greater than 800 mm year −1 had SOC/clay >1/8. This suggests that T A B L E 2 Percentages of sites above, below and between soil organic carbon (SOC)/clay thresholds of 1/8, 1/ 10 and 1/13 for each land use and landuse precipitation class combination Chi-squared analysis to answer the second question showed that land use was not independent of SOC/clay ratio for soils under dry climate conditions (X 2 (9) = 94.0, p < .001). A larger number of arable soils and smaller number of grassland and woodland soils than expected had SOC/clay <1/13 than if the land use was independent of SOC/clay ratio range for soils receiving <650 mm year −1 annual precipitation. For soils with SOC/ clay >1/8, the reverse was true (i.e., arable < grassland or woodland). This suggests that land use was affecting the number of dry climate soils with SOC/clay <1/13.
The relative effects of land use, precipitation and soil type were evident from the distribution of the 820 sites with SOC/clay ratio < 13 across England and Wales (Figure S1). These sites were predominantly arable, and their distribution across eastern and central England confirmed the lesser statistical effect of precipitation and major soil group observed. Northwest England and Wales had notably few degraded sites, although soils sampled there were mostly under non-arable land uses. F I G U R E 3 Box plots of soil organic carbon (SOC)/clay ratio for each major soil group. Horizontal lines are SOC/clay thresholds: Solid = 1/8, dashed = 1/10, dot-dash = 1/13. Abbreviated major soil groups: Terr. Raw, terrestrial raw; Lith., lithomorphic; SW gley, surface-water gley; GW gley, ground-water gley F I G U R E 4 Box plots of soil organic carbon (SOC)/clay ratio for each structural quality score. Horizontal lines are SOC/clay thresholds: Solid = 1/8, dashed = 1/10, dotdash = 1/13. Numbers of samples in each group were n = 2,250, 1,111, 229 and 208 for good, moderate, moderatedegraded and degraded, respectively

| Effects of soil type and soil pH
The statistical effect of major soil group appeared to be driven by two of the soil groups and some of this might already have been accounted for by land use (Figure 3). Podzolic soils tended to have SOC/clay >1/ 8 and were mostly not arable, whereas clay-rich pelosols were more likely to have SOC/clay <1/13 and a higher proportion were arable. The lower importance of soil group might be linked to the smaller sample sizes of the podzolic and pelosol soils compared with brown and gley soils, for which SOC/clay ratios showed similar variation.
As pH decreased below pH = 5, the SOC/clay ratio tended to increase ( Figure S2). Above pH = 5 there was less of a trend when considering permanent grass and woodland soils; however, arable and ley grass soils showed decreasing minimum SOC/clay ratio particularly above pH = 7, although sites with SOC/clay >1/8 were still observed.

F I G U R E 5 Maps of soil organic carbon (SOC)/clay ratio across England and Wales under (a) arable land, (b) ley grass, (c) permanent grass and (d) woodland coloured by SOC/clay index
3.4 | Relation between structural quality and SOC/clay ratio Structural quality, classified as good, moderate, moderate-degraded and degraded, tended to improve with increasing SOC/clay ratio, as shown by the box plots in Figure 4 and the chi-squared test result for the relation between SOC/clay range between the thresholds and structural quality (Χ 2 (9) = 129.3, p < .001). Most (82%) of the relationship between SOC/clay and structural quality was explained by (a) a larger than expected frequency of sites with SOC/clay <1/13 and moderate-degraded or degraded structure, (b) a smaller than expected frequency of sites with SOC/clay >1/8 and degraded structure, and (c) smaller than expected frequency of sites with SOC/clay <1/13 and good structure.

| Variation in SOC/clay ratio across England and Wales
Mapping the index across the two countries ( Figure 5) showed the effect of land use and geography at the time of survey. For any land use, degraded sites were not limited to a particular region. But, as previously mentioned, there were fewer degraded sites towards the northwest and in Wales. Calculating summary values of SOC/clay by land use (Table 4) showed that the minimum value increased slightly in the order: arable = ley grass < permanent grass < woodland. The median results showed a stronger difference between arable sites and the other land uses, with arable in the moderate category and the other land uses equal to or above the very good threshold. The different land uses had similar upper SOC/clay values as a result of excluding outliers.
3.6 | Changes in SOC/clay ratio with field management Figure 6 shows changes in SOC/clay ratios over 30 years of the Woburn organic manuring experiment. Leys and treatments with organic matter application (straw and manures) showed similar trends of increasing SOC/clay ratio during the application period and a decreasing ratio after the treatment was stopped, but with differing magnitudes. Peat and farmyard manure gave the largest increases, followed by the ley treatments and then straw. Horizontal lines are thresholds separating degraded (SOC/clay <1/13), moderate (SOC/ clay = 1/13-1/10), good (SOC/ clay = 1/10-1/8) and very good (SOC/clay >1/8) soil conditions. Treatments were applied in two cycles (1965 to 1972 and 1979 to 1987); peat and green manure (GM) treatments were replaced by grass ley for the second cycle. Our finding that few grassland and woodland sites had SOC/clay <1/13 supports the use of SOC/clay = 1/13 as an indicative threshold for degradation, as grassland and woodland soils are not generally subject to major disturbance and are close to semi-natural systems. Our analysis shows that many arable soils were depleted in SOC compared with the more natural systems. Ley grassland soils were intermediate between arable and permanent grassland soils. The NSI survey did not include information on the length of leys nor the time under ley at sampling, but typically this is between 3 and 8 years. Some proportion of arable sites will have been part of ley rotations at the time of sampling. The large variation of SOC/clay ratio within each land use and soil group demonstrates that clay content is not the only determinant of SOC dynamics, especially considering that land-use history before the sampling will have big effects too. As discussed above, despite the scatter, the thresholds show differences between soils under different land managements.
The variance of the SOC/clay ratio explained by random forest analysis was similar to the variance of SOC content explained by Verheijen et al. (2005) with stepwise general regression modelling, using similarly derived precipitation data, and the same soil dataset (although a different subset). We would expect the variance explained to increase with more specific measures of land management within land-use classes (crop type, residue treatment, land-use history and, for grassland systems, grazing management). Interpolated precipitation data is another estimation which could be improved; however, this is what is generally available at this scale.
The effect of the major soil group on the SOC/clay ratio suggests some consideration should be given to soil type, as highlighted by Johannes et al. (2017). Comparing the variation in the SOC/clay ratio between major soil groups, similar variation and medians were found for lithomorphic, brown, gley and man-made soils. The tendency for a higher SOC/clay ratio of podzolic soils might be attributed to concentrated organic horizons in the topsoil. The tendency for lower SOC/clay ratios of Pelosols might be attributed to higher clay contents combined with a higher proportion (62%) of them being arable compared with most major soil groups. Although acidic soils had a tendency for higher SOC/clay ratios, there appeared to be little relationship between pH and SOC/clay ratio in agriculturally productive pH ranges (circa. pH = 5.5 to 7).

| Significance of the threshold values
The fact that the empirical threshold values found by Johannes et al. (2017) for Swiss soils also hold for the wide range of soils and land uses across England and Wales in our study suggests that they have some fundamental basis, and that they may apply in soils in similar climate zones across Europe. An association of soil structural quality with the SOC/clay = 1/10 ratio was expected from physicochemical considerations (de Jonge et al., 2009;Jensen et al., 2019). Intuitively, there will be some minimum range of SOC/clay ratio below which soil structure is impaired, and some maximum range above which the capacity of soil clays of given mineralogy to bind SOC is exceeded. However, there are no obvious reasons why the precise threshold values indicated by our and the Swiss study should be absolute.
The observed decrease in soil structural quality with decreasing SOC/clay ratio was statistically significant, although there was overlap of the boxplots of SOC/clay ratio between structure classes. Our analysis was limited by the quality of the available data on structure. This was based on the scheme defined for the Agricultural Land Classification of England and Wales, which includes a measure of friability. Because friability was not recorded in the NSI, we had to estimate structural quality without it, introducing error.
The mechanistic link between structural quality and SOC/clay ratio should reduce errors due to cross-correlation with spatial and temporal variations in the data. We found, as did Verheijen et al. (2005), that SOC content tended to decrease with decreasing precipitation across England and Wales, partly in interaction with land use. However, low SOC/clay ratios were not limited to particular combinations of land use and precipitation; therefore, we would not consider precipitation to limit the SOC/clay ratio in this dataset and geographical range. Land management was shown to affect proportions of very good and degraded soils under dry (< 650 mm year −1 ) climate conditions. So, SOC/clay ratios of at least 1/10 should be attainable in such soils.

| Practical usefulness of the index
The SOC/clay index is a simple measure to evaluate the SOC status of any given soil in England and Wales, independent of the land use. It will therefore be meaningful for experts and non-experts and has consequences for many soil functions beyond agricultural uses. It could allow farmers to identify degraded soils on their farms and adjust their management accordingly. It could also be used to monitor and understand the state of soils at a national scale to inform decision making and policy.
Application of the index to data from the long-term Woburn experiment showed it was consistent with expectations, with an improving index in treatments favouring organic matter accumulation, and a deteriorating index in soil-degrading treatments. This illustrates the time taken for the various contrasting managements to change SOC and the index. The soil in the Woburn experiment is a sandy loam; the results show that the index can be used for soils with low clay content, despite the narrowing of the SOC/clay thresholds with decreasing clay content, and the relatively small changes in SOC content between the treatments. It should be noted that, to be useful for monitoring purposes, measurements of SOC and clay over time and between sites need to be consistent.
It would be interesting to look at other longer-term studies to explore a wider range of clay contents, treatments and time periods. Saturation concepts suggest that a soil closer to steady state or saturation limit should accumulate carbon more slowly than one further from saturation (Six et al., 2002;Stewart et al., 2007). Hence, whether sites with lower index values (higher degradation) improve more quickly could be tested.

| CONCLUSIONS
An index of soil organic matter with threshold SOC/clay ratios of 1/8, 1/10 and 1/13 satisfactorily separates the soils of England and Wales into very good, good, moderate and degraded classes of SOC content and physical structure condition. In agreement with previous publications, grassland and woodland soils mostly had an SOC/ clay ratio > 1/10, indicating that their SOC contents are close to or above the capacity for protection of SOC by interaction with clay particles. That these more natural systems tend to have an SOC/clay ratio > 1/10 supports this as a suitable threshold for good condition.
Arable soils and soils receiving less annual rainfall were most likely to be physically degraded, although rainfall was a less important factor determining the SOC/ clay ratio. Very-good status soils (SOC/clay >1/8) occurred in low rainfall areas, even under arable management, suggesting that rainfall does not fundamentally limit SOC content in this climate.
The SOC/clay ratio index allows the evaluation of soils on a scale from degraded to good soil conditions. It therefore gives a ready metric for communication to experts and non-experts, enabling users to adjust their practices and decision makers to develop adequate policies. An SOC/clay ratio greater than 1/10 should be achievable for all managed soils of different textures. Many arable soils in England and Wales evidently have a substantial SOC deficit, suggesting a significant opportunity to increase SOC storage to both improve soil conditions and sequester carbon.
Being based on two routinely measured soil properties, the index provides a suitable means of monitoring SOC at national, regional or sub-regional scales. Given the wide range of soils and land uses across England and Wales in the dataset used to test the index and agreement with literature using French, Polish and Swiss soils, it should apply to other European soils in similar climate zones.