The organic matter trial (OM trial; 50.560∘ N, 4.726∘ E) was set up in 1959, with the initial goal of addressing the issue of decreasing organic matter inputs (farmyard manure, crop by-products) on cropland soils of the silt loam region and related consequences for soil properties, crop yields and farm profitability . The trial includes six contrasting treatments of SOM restitution in plots of 70 m × 10 m, repeated six times, following a Latin square design. From 1959 to 1974, the field was cropped according to a 4-year rotation with sugar beet (Beta vulgaris) as the starter crop, followed by 3 years of winter cereals – wheat, oat (Avena sativa), barley (Hordeum vulgare) – or two winter cereals – wheat, barley/oat – and one legume – horse bean (Vicia faba). Cultivation cycle shifted from 1975 onwards to a 3-year sugar beet–winter wheat–winter barley rotation. Among the six treatments described by , three were selected for soil sampling in 18 plots. The “residue exportation” (RE) treatment consists in the maximal exportation of by-products (straws and sugar beet heads and leaves) and no farmyard manure application nor green manure during the intercropping period. Since 2009, however, sugar beet heads and leaves have been left on the field. The “farmyard manure” (FYM) treatment consists in one application of 30 to 60 t ha-1 of composted cattle manure once per rotation, after the harvest of the winter barley, in order to enrich the soil for the sugar beet. The last application before soil sampling occurred on the 26 July 2017. In the “residue restitution” (RR) treatment, all crop by-products (cereal straws and sugar beet heads and leaves) are left on the fields, and one cover crop acting as a green manure is sown once per rotation during the intercropping period between the winter barley and the sugar beet. Cover crops were vetches (Vicia sp.) until 2009, except (i) mustard (Sinapis alba) in 1980; (ii) phacelia (Phacelia sp.) in 2011 and 2014; and (iii) a mix of oat, vetch and clover (Trifolium sp.) in 2017. The estimated mean and standard deviation of annual total carbon (C) input amount to 315 ± 76, 472 ± 82 and 487 ± 93 gC m-2 for the RE, FYM and RR treatments, respectively . Since the start of the trial, yearly measurements of topsoil properties (0–25 cm) have shown a drop in SOC content for the RE treatment, an increase for the FYM treatment and a steady state for the RR treatment . For all treatments, the soil has been ploughed annually with a moldboard plough.

The soil tillage trial (50.560∘ N, 4.727∘ E) was set up in 2004 and follows a 2-year rotation of winter wheat and a spring crop, generally sugar beet or flax (Linum usitatissimum), with an exception in 2018 where corn (Zea mays L.) was cultivated as spring crop. A green manure is sown after tillage following the harvest of the cereal and destroyed during winter time before the spring crop. The trial includes four tillage treatments in plots of 24 m × 21.5 m repeated four times, following a Latin square design. Among the four treatments, the two most contrasting ones were sampled in eight plots: (i) annual ploughing (P) to a depth of 25–30 cm with a moldboard plough and (ii) annual reduced tillage (RT) with a spring tine cultivator tilling to a depth of about 10 cm.

The P–K mineral fertiliser trial (50.582∘ N, 4.687∘ E) was set up in 1967, with the initial goal of assessing the effect of the rate of P and K mineral fertiliser application on crop quality and yield, nutrient exportation with harvest, soil properties and farm profitability . The trial comprises three levels of phosphorus (P) fertiliser (applied as superphosphate 18 % or triple superphosphate 45 %) crossed with three levels of potassium (K) fertiliser (applied as KCl 40 % or 60 %), namely nine different treatments repeated six times in two randomised complete blocks (i.e. three repetitions in each block), for a total of 54 plots of 7.5 m × 50 m. The lower level of P and K fertilisation has received no P and K mineral fertiliser since 1975 (P0 and K0). The intermediate level of fertilisation consists in balancing P and K outputs and inputs, according to the nutrient balance method (P1 and K1). The higher level of fertilisation is over-fertilised, multiplying by 2 (until 2000) or 1.5 (onwards) the amount of P and K applied to the P1 and K1 treatments (P2 and K2). The last application of P–K fertilisers before soil sampling occurred on the 15 July 2016. The whole field is cropped according to a 3-year rotation cycle similar to that of the organic matter trial, with sugar beet as the starter crop followed by two winter cereals (winter wheat and winter barley). Since the start of the trial, the soil has not received any exogenous organic matter, but all by-products (cereal straws and sugar beet heads and leaves) have been left on the field to maintain sufficient SOC contents. For all treatments, soil is ploughed annually with a moldboard plough, except before the seeding of sugar beet in 2017 (soil was prepared by deep decompaction with a heavy tine cultivator at about 30 cm depth in August 2016). In this study, we focussed on the potential effect of contrasting levels of KCl application on soil structural stability . Therefore, we selected three repetitions of each level of K within the trial for soil sampling in nine plots.

After homogenisation, about 500 g of each disturbed soil sample was gently crushed with a pie roll and sieved to 2 mm, and the fraction < 2 mm was sent to the Centre interprovincial de l'agriculture et de la ruralité in La Hulpe (Belgium) to be analysed. Soil pH was measured in water (pHH2O) with a 1 : 5 soil : solution mass ratio, according to the norm NF-ISO-10390:2005 . Total C content was determined by dry combustion according to the norm NF-ISO-10694:1995 . Inorganic C content was measured by infrared quantification of CO2 emitted from soil after addition of orthophosphoric acid, according to the norm NF-EN-15936:2012 . SOC content was calculated as the difference between total and inorganic C content. Soil texture (sand, silt and clay contents) was determined by sedimentation and sieving, according to Stokes law, by a method derived from the norm NF-X31-107:2003 . Bulk density was measured from one structured soil sample per plot, dried at 105 ∘C until constant weight, by dividing the mass of dry soil by the core volume (100 cm3). The main properties of the soils of the experimental fields of the three trials are shown in the Table  (open data available: https://doi.org/10.5281/zenodo.7405113).

Soil aggregate stability was measured according to the method of , following the norm ISO-FDIS-10930:2011 . For each soil gently crumbled by hand, 5 to 10 g of soil aggregates from 3 to 5 mm in size was subjected to three contrasting disaggregating treatments. The first test consists in the fast wetting of soil aggregates in water, exacerbating the effect of slaking. The second test is a slow wetting of soil aggregates by capillarity, to test their resistance to clay dispersion and differential swelling during wetting independently from the slaking effect. The third test consists in a standardised shaking of the aggregates in water after rewetting them in 95 % v/v ethanol for 30 min, to test their mechanical strength while minimising slaking, differential swelling and dispersion. After each disaggregation treatment, the resulting aggregates were immersed in ethanol and dried at 40 ∘C for 2 h. The size distribution of the remaining aggregates was measured by way of dry sieving, with sieves of 2, 1, 0.5, 0.2, 0.1 and 0.05 mm.

Two main indicators were calculated from the fractions. The first is the mean weight diameter (MWD) of the aggregate fraction that survived each individual test, according to the following equation . 1MWD=∑(mean diameter between two sieves×[weighed percentage of particles retainedon the sieve])100

The second indicator is the percentage of macro-aggregates (MA) remaining after each individual test, calculated as the mass fraction of soil aggregates > 200 µm.

The Le Bissonnais (1996) method has two main advantages: (i) the three tests target the three main mechanisms of soil disaggregation in field conditions, namely slaking, mechanical breakdown and clay differential swelling and dispersion, and (ii) it measures the size distribution of particles remaining after the disaggregation treatment, which provides further insight into soil susceptibility to water erosion .

The QST method consists in plunging an undisturbed soil sample supported by an 8 mm metallic mesh basket into demineralised water and measuring soil mass continuously using the underfloor weighing hook of the balance. The balance is connected to a computer for data logging (Fig. ). For each plot, the five air-dried structured soil samples were left to slake for approximately 1000 s (around 17 min), with recording time step decreasing from less than 1 s at the start of the experiment to approximately 30 s at the end. Due to some electronic or computer issues during the experiment, some samples were lost. In total, the data from 157 QST curves could be processed and constituted the main database of our study. Most of the 35 plots had five or four usable QST curves (20 and 13 plots respectively). One plot from the OM trial and one from the P–K mineral fertiliser trial had only three and two usable curves, respectively.

The evolution of the relative soil mass from before to after immersion in water is presented in Fig. . Immediately after soil immersion in water, the soil mass drops due to Archimedes' upward buoyant force (Fig. ). The first record of soil mass under water is defined as the time 0 (t0) of the QST (Fig. ). The general shape of one QST curve from t0 is presented in Fig. , as well as the main indicators that have been calculated from the curves. In the initial phase, soil mass generally increases due to water filling the porosity. After a few seconds or minutes, the soil mass reaches a maximum (Wmax⁡ at tmax⁡) before decreasing, once mass loss due to disaggregation becomes dominant compared to mass gain by wetting. Soil mass was normalised relative to Wmax⁡ (Wmax⁡=1 [–]), so that mass values are relative soil masses, varying between 0 and 1.

QST indicators calculated from the QST curves were split into four categories (Fig. ): i.

Indicators related to the early increase in soil mass soon after soil immersion in water. They include the time to reach the maximum mass value (tmax⁡), the increase in soil mass between t0 and tmax⁡ (Wmax⁡-Wt0), and the slope between t0 and tmax⁡ (slope0–max⁡).

A draft of the “slaker” R package (Vanwindekens, 2023) containing the code used to run the SlakingLab and analyse QST data is available online.

For samples from the soil tillage trial, root biomass retained in the 8 mm mesh metallic basket was weighed after running the QST by cleaning remaining soil with a water jet. The roots were dried carefully with Tork paper, air-dried and weighed.

Between continuous variables, correlation coefficients were determined. For QST indicators, average values were calculated at the plot level for comparison with data that were measured only at the plot level: aggregate stability indicators from and physico-chemical soil properties. Since many QST indicators are calculated from the same curve, a correlation matrix was drawn for QST indicators, to evaluate the level of redundancy between them and propose a selection to be used for the statistical analysis of soil management practices in the long-term experiments.

In order to test whether soil management practices affect QST indicators, linear mixed-effects models were fitted. For each model, the QST indicator was used as the outcome variable and the treatments of the trials were used as a fixed explanatory variable, whereas the blocks were defined as a random effect. As several samples were related to one single plot (157 QST in total from 35 plots), the plot identifier was added as a random effect of the model to take into account the dependence between field replicates from one plot.

The normality and the homoscedasticity of the residuals of the models were verified using Shapiro–Wilk and Bartlett tests respectively. For all models, the significance of differences in QST indicators between soil management practices was tested using Type II Wald F tests with the Kenward–Roger estimation of degrees of freedom method . When the F test was significant (p<0.05), post hoc comparisons were performed: treatments of the trial were compared pairwise at 0.05 probability level of significance using estimated marginal means (EMMs; also named least-squares means; ).

All statistical analyses were performed using R version 4.3.0 (21 April 2023) software . The linear mixed-effect models were performed with the lme4 package , the Wald F tests with the car package and contrast analyses with the emmeans package .

The correlation matrix between Le Bissonnais' indicators and soil properties is shown in Appendix A, in Table . A positive correlation exists between total SOC content and both MWD1 (r=0.75) and MWD2 (r=0.70), whereas MWD3 and MA3 correlate poorly with SOC content (r=0.11 and -0.07, respectively). In contrast, clay content correlates positively with MWD3 and MA3 (r=0.52 and 0.66, respectively) but poorly with MWD1 and MWD2 (r=-0.35 and -0.12). A linear relationship with the SOC : clay ratio, evidenced as a proxy for predicting field soil structural quality by visual assessment methods , was also tested. The SOC : clay ratio correlated positively with both MWD1 (r=0.67) and MWD2 (r=0.48) and negatively with MWD3 (r=-0.33) and MA3 (r=-0.55). No clear linear relationship was found between Le Bissonnais' indicators and pH or bulk density.

Generally, indicators derived from QST curves correlate positively with SOC content, except for slope0–max⁡ and slope60–300 and slope300–600. Coefficients remain low to moderate though, with the stronger coefficient obtained for Wmax⁡-Wt0 (r=0.56) and t95 (r=0.55) (Table ). In contrast, most QST indicators correlate negatively with clay content. The stronger coefficients were found for Wmax⁡-Wt0 (r=-0.83), tmax⁡ (r=-0.67), slopemax⁡–30 (r=-0.64) and AUC (r=-0.58) (Table ). This antagonist effect of SOC and clay contents on soil resistance to disaggregation under water is well captured by the SOC : clay ratio, which correlates strongly with indicators from the start of QST curves, particularly Wmax⁡-Wt0 (r=0.92; Table  and detailed in Fig. ) but also tmax⁡ (r=0.82), slopemax⁡–30 (r=0.67) and dtmax⁡–25 (r=0.68). While we observe a clear relationship between Wmax⁡-Wt0 and SOC : clay ratio (Fig. ), we observe a residual variability in the repeated test of a same plot that could be explained by local soil conditions of the sampling sites (microsite heterogeneity related, for example, to micro-relief, the presence of crop residues, roots, earthworm galleries, …).

Similarly to indicators of soil aggregate stability from Le Bissonnais, all indicators from the QST curves correlated poorly with pH. Except for slope0–max⁡, a moderate to poor negative correlation is observed between QST indicators and bulk density, with the lower values obtained for Wmax⁡-Wt0 (r=-0.61) and tmax⁡ (r=-0.49).

Soils of the three treatments of OM inputs in the OM trial have different contents of total SOC, with the FYM treatment having the highest SOC content (11.32 g kg-1), the RE treatment having the lowest SOC content (8.41 g kg-1) and the RR treatment having an intermediate value (9.95 g kg-1). Accordingly, tmax⁡ tends to respect this gradient of total SOC, with the FYM showing the best scores on average (p = 0.047; Fig. a–d). Counter-intuitively though, this order is not respected anymore for QST indicators related to intermediate or late stages of the curves (slope60–300, Wend). The response of treatments follows the order RR > RE > FYM for slope60–300 (p = 0.005, Fig. c) and for Wend (p = 0.098, Fig. d). Antagonistic results were also obtained between the three tests of Le Bissonnais, with the MWD scores from the fast-wetting test (MWD1) slightly in favour of the FYM treatment (FYM ∼ RR > RE; not significant – n.s., results in Appendix B, Fig. ) and the scores from the slow-wetting test (MWD2, n.s.) and from the mechanical breakdown (MWD3, p < 0.05) in favour of the RR treatment (RR > FYM ∼ RE; results in Appendix B, Fig. ).

Remarkably, the QST responds very well to contrasting tillage treatments, with all QST indicators having a better score for reduced tillage (RT) than for ploughing (P) (Figs.  and ). This result is in agreement with total SOC content, because RT has an average SOC content of 12.16 g kg-1, whereas P treatments have an average SOC content of 10.07 g kg-1. Similarly, a higher root biomass content was measured in the topsoil under RT, with 0.42 ± 0.19 mg cm-3 of root biomass for the RT treatment against 0.31 ± 0.16 mg cm-3 for the P treatment (p=0.168). However, slope30–60 (p < 0.007) and Wend (p < 0.008) are more sensitive to tillage than tmax⁡ and dt50–75 (Fig. ). Indicators from the three tests of Le Bissonnais provide similar results, with the most contrasting response between RT and P tillage treatments obtained for the fast-wetting test (Appendix B, Fig. ). However, slope30–60 and Wend discriminate better between tillage treatments than MWD1 (p < 0.05).

In the P–K mineral fertiliser trial, soil structural stability follows the order K2 > K0 > K1 regardless of the QST indicator (data not shown) but without any significant differences (n.s.). Similar results were obtained with the three tests of Le Bissonnais but with a smaller standard deviation on average than that of QST indicators.

Right after immersion in water, soil mass increases due to the replacement of air by water in soil porosity. Sooner or later, soil mass then reaches a maximum before decreasing when mass loss by disaggregation exceeds mass gain due to filling of soil porosity with water. QST indicators from the start of the curves (e.g. Wmax⁡-Wt0, tmax⁡, slopemax⁡–30) are the most correlated to the fast-wetting test of Le Bissonnais (MWD1 and MA1; Table ), which suggests that slaking plays a major role in the initial stage of the QST. In contrast, QST indicators from the intermediate to late stages of the curves (e.g. t75 to t95) are more correlated to the slow-wetting test of Le Bissonnais (Table ), specifically targeting clay dispersion and differential swelling. This indicates that after a longer time period under water, when soil is saturated, the effect of slaking decreases, and clay dispersion and differential swelling become the dominant mechanisms of soil disaggregation. Nevertheless, both mechanisms overlap, with air release from soil further interfering with the measurement of soil mass loss when running QST. This may explain the relatively low correlation coefficients obtained between QST slopes and indicators from the fast- and slow-wetting tests of Le Bissonnais. It is also worth mentioning that the time of wetting of the soils of our study was relatively short (less than 2 min, as indicated by the release of air bubbles from soil). We therefore argue that indicators from the initial stage of the curve, like slope30–60, may provide information much more specific to slaking than indicators from the fast-wetting test of Le Bissonnais, lasting 10 min, which largely exceeds the time during which slaking is the dominant driver of disaggregation.

Except for Wmax⁡-Wt0, QST indicators correlate very poorly to the third test of Le Bissonnais, targeting soil mechanical resistance . This indicates that little information on soil resistance to raindrop impact or shear strength from agricultural machinery can be inferred from QST curves, which is not surprising. For the soils of this study, soil mechanical resistance (as estimated by the third test of Le Bissonnais) seems to be somehow controlled by the absolute clay content of soil, since clay content correlates positively to MWD3 (r=0.52) and MA3 (r=0.66).

Soil mass evolution under water as captured by QST indicators responds in an antagonist way to SOC and clay contents. Indeed, most QST indicators are positively correlated to SOC content and negatively correlated to clay content, with the absolute value of correlation coefficients decreasing for indicators of the later part of the curves (slope60–300, slope300–600 and Wend). Similar trends were observed in other contexts, with the resistance to slaking increasing with SOC content and decreasing with clay content . In light of the comparison between QST curves and Le Bissonnais's indicators, the amplitude of the early mass loss under water is mainly controlled by soil resistance to slaking. Accordingly, the absolute SOC content increases soil resistance to slaking, as highlighted by the positive correlation between SOC content and MWD1 (r=0.75). The role of SOM in promoting soil aggregate stability is well-known , as SOM has long been recognised as one of the main binding agents in micro-aggregates . The increase in SOC along a concentration gradient has been shown to decrease the wettability of individual aggregates 3 to 5 mm in diameter and of SOM-associated clay in the < 2 µm fraction of soil . This decrease in clay wettability might explain, in part at least, the higher aggregate stability under water of soils rich in SOC, with the slower wettability of macro-aggregates explaining their improved resistance to slaking and the slower wettability of clay decreasing its dispersive character .

In contrast, while the absolute clay content increases soil mechanical resistance (supported by the positive correlation with MWD3 and MA3), it also tends to decrease soil structural stability under water (as indicated by the negative correlation with most QST indicators). This supports the view that, for cropland soils of this study with low SOC content on average, clay dispersivity and differential swelling are strong drivers of soil disaggregation in wet conditions. This is in agreement with the findings of , who found that, for a variety of soils from France and Poland, about 1 g of SOM was necessary to decrease the dispersive power of 10 g of clay by organo-mineral association. This threshold value of 0.1 for the mass SOC : clay ratio was reported as pivotal between good and medium structural quality as estimated by field visual soil assessment by the CoreVESS method for 161 agricultural soils of Switzerland and for a large number of forest, grassland and cropland soils from England and Wales . Additionally, both and found a linear increase in soil structural quality scores with increasing SOC : clay ratios in the range 1 : 13 to 1 : 8, suggesting that SOM has beneficial effects on soil structure beyond the threshold value of 1 : 10 determined empirically by . We assume that these results can be extrapolated to other temperate European soils under similar pedoclimatic conditions and clay mineralogy, as supported by the linear increase of QST indicator Wmax⁡-Wt0 with the SOC : clay ratio in the range 0.04–0.12 (Fig. ). This supports the idea that Wmax⁡-Wt0, as a predictor of the SOC : clay ratio, has the potential to estimate the overall soil structural quality as measured in field conditions with visual soil assessment methods.

It is important to underline that the close linear relationship found between Wmax⁡-Wt0 and the SOC : clay ratio (see Fig. ) has probably no general character and was obtained here because cropland soils of the current study were sampled under standardised conditions of seeding and cover (winter wheat). It is very unlikely to find an identical relationship for the same soils under contrasting conditions of soil preparation, sampling dates or crop type, since soil structure does also relate to more or less dynamic external factors such as tillage, root and hyphae development, and biological activity. To sum up, we suggest that the SOC : clay ratio, a proxy for soil intrinsic “potential” structural stability, is a valuable indicator of the organic and structural status of agricultural soils . On the other hand, QST indicators such as Wmax⁡-Wt0 provide a quantitative, direct measurement of the overall structural stability of a soil under a given set of conditions. Both parameters are therefore relevant in terms of appreciation of soil resistance to water erosion and structural damage by farm machinery.