Agriculture increases the bioavailability of silicon, a beneficial element for crop, in temperate soils

Silicon is considered a beneficial element for many crops, particularly under biotic and abiotic stresses, and is especially relevant for Si-accumulating staple species such as rice, sugarcane, and wheat. In soils, dissolved silicic acid (Si(OH)4) derives from weathering of silicate minerals and from recycling of plant phytoliths, while its concentration is reduced by plant and microbial uptake and by adsorption to mineral surfaces. Although various extractions (e.g., CaCl2) are used as proxies for plant-available Si (PAS), the controls on PAS—especially in temperate soils—are not fully resolved. Prior studies, largely in tropical or paddy systems, relate PAS to soil properties (pH, clay, Fe oxides, organic matter, phytoliths) and indicate potential depletion under intensive cultivation when residues are not returned. This study aims to quantify and map PAS (SiCaCl2) in France, identify the soil properties controlling its variability in non-cultivated soils, determine how cultivation alters these controls under different pedological contexts, and assess potential SiCaCl2 deficiency risk for wheat.

Existing work shows PAS proxies correlate positively with phytolith content, pH, < 2 µm fraction, organic matter, and iron oxides, and negatively with total Si due to the prevalence of low-solubility quartz. Critical PAS levels have been proposed for rice and sugarcane in tropical systems, leading to Si fertilization practices. Agricultural export of phytoliths and residue removal can reduce soil Si pools over time, with documented PAS declines under long-term intensive rice cultivation and potential contributions to yield stagnation. However, temperate regions lack comprehensive assessments of PAS spatial variability, drivers, and agricultural impacts, particularly in relation to clay mineralogy and pH-dependent adsorption.

Data source: The study used the French Soil Quality Monitoring Network (RMQS), a 16 × 16 km grid with 2111 sites sampled (0–30 cm) during 2000–2009 across metropolitan France. At each site, 25 cores (20 m × 20 m plot) were composited, air-dried, and sieved (<2 mm). Measurements included soil organic carbon (dry combustion), particle-size distribution (wet sieving/pipette), cation exchange capacity (CEC) and exchangeable cations (cobaltihexamine), pH in water, CaCO3 (volumetric), and total P, K, Ca, Mg, Fe, Al (ICP-MS after HF-HClO4 dissolution). The < 2 µm CEC was estimated as (<2 µm CEC) = (CEC − 0.15 × OC)/(<2 µm fraction) × 1000, with CEC in cmol+ kg−1 and OC and <2 µm fraction in g kg−1. Total Si was directly measured on 673 samples (ICP after alkaline fusion) and predicted for the remaining sites using a Cubist model f(Al, Fe, K, Na, Caanc, Mganc, P, SOC, CaCO3, residual water) with R2 > 0.98. Bioavailable Si (SiCaCl2) was determined on 2091 sites using 0.01 M CaCl2 extraction (solid:liquid 1:10, 16 h equilibrium, filtration 0.45 µm) and ICP-AES, with LOQ 0.5 mg kg−1 and uncertainty U = 0.0271 × SiCaCl2 + 0.25 mg kg−1. Stratification: Sites were grouped by parent material (EUSIS) and by land use (cultivated: permanent/annual crops; non-cultivated: forests, pastures, parks, natural vegetation, wetlands). Sedimentary soils were split into carbonated (>1% CaCO3) and non-carbonated (<1%). Due to limited cultivated sites, igneous extrusive and podzols were not split by land use. After removing sites with missing geology and peat soils, 1986 sites (subset 1) were retained for statistics. Digital soil mapping (DSM): A regression kriging framework combined Random Forest (RF) regression with ordinary kriging of residuals (GSIF package). Covariates (90 m resolution) followed the scorpan paradigm: soil (soil type, available water capacity), climate (climate type, precipitation, evapotranspiration), organisms/vegetation (land cover, forest type, NDVI components NDVI_1, NDVI_2, NDVI_3 from PCA of time-series), parent material (type, gravimetry), and relief (SRTM elevation, compound topographic index, slope, slope cosine, slope position, erosion). Feature selection used Boruta; RF used default mtry = 6, ntree = 500. Model validation used 30-fold cross-validation (2/3 calibration, 1/3 validation per fold) reporting RMSE, R2, concordance, and bias. Statistical analyses included correlations (SiCaCl2 with pH, < 2 µm fraction, < 2 µm CEC, SOC, Fe oxides), and evaluation of pH–SiCaCl2 relationships across < 2 µm fraction classes; Wilcoxon tests with Tukey adjustment compared medians across groups. Mapping and deficiency assessment: Regression-kriged SiCaCl2 maps were produced at 90 m; potential deficiency for wheat was assessed using two thresholds from literature (20 and 40 mg kg−1, from sugarcane and rice, respectively) applied to arable land pixels intersected with wheat-dominant municipalities (OTEX), excluding regions dominated by maize.

• SiCaCl2 in French topsoils (0–30 cm) ranged from 2.3 to 134 mg kg−1; Q1 = 9.5, median = 17, Q3 = 28 mg kg−1 (n = 1986). • DSM performance: 30-fold cross-validation yielded R2 = 0.43, concordance = 0.58, RMSE = 13 mg kg−1, bias = 0.24 mg kg−1. Most important covariates in RF: parent material (top), NDVI_1, soil type, precipitation, then land use. • In non-cultivated soils, SiCaCl2 correlated positively with: < 2 µm fraction (r = 0.59), CEC (r = 0.59), pH (r = 0.46 overall; relationship increases up to pH ~6–7 then slightly declines >8), exchangeable Ca (r = 0.54), < 2 µm CEC (r = 0.44), Fe oxides (r = 0.35), SOC (r = 0.27). • The pH effect on SiCaCl2 depended on < 2 µm fraction: a clear positive pH–SiCaCl2 relationship existed only for < 2 µm contents between 50 and 325 g kg−1; negligible effect for < 50 g kg−1; no correlation for > 325 g kg−1 (e.g., many carbonated sedimentary soils). • Clay mineralogy proxy (< 2 µm CEC) related to SiCaCl2 for most non-carbonated parent materials; acidic soils tended to be kaolinitic (lower < 2 µm CEC) with lower SiCaCl2, while near-neutral/basic soils were richer in smectite/vermiculite (higher < 2 µm CEC) with higher SiCaCl2. • Land use impact: Cultivation significantly increased SiCaCl2 for soils on sedimentary parent materials (both carbonated and non-carbonated), explaining ~12% of total variance between land uses; no significant difference for igneous intrusive and metamorphic parent materials (where SiCaCl2 levels were low overall). • Cultivated soils had higher pH than non-cultivated due to liming (except carbonated soils where pH differences were small and liming uncommon). In non-carbonated sediments, most < 2 µm fractions were 50–325 g kg−1, a range where higher pH led to higher SiCaCl2; cultivation also corresponded with higher < 2 µm CEC, indicating shifts toward smectite/vermiculite. • Potential deficiency for wheat: Using 20 mg kg−1 (lower bound) and 40 mg kg−1 (upper bound) thresholds from sugarcane/rice literature, 4% of wheat-cultivated soils were below 20 mg kg−1 (potentially deficient), while about 85% fell between 20 and 40 mg kg−1 (uncertain adequacy for wheat pending a wheat-specific critical level).

The study demonstrates that in temperate soils, the availability of Si to plants (as proxied by SiCaCl2) is primarily governed by parent material, soil texture/clay content, pH, and to a lesser extent Fe oxides and SOC. The dependence on parent material and soil type explains the strong spatial pattern captured by DSM. Crucially, agriculture—via liming—elevates soil pH in many cultivated soils developed on sediment, particularly non-carbonated ones with moderate clay contents (50–325 g kg−1 < 2 µm fraction), thereby increasing SiCaCl2 through enhanced adsorption/desorption equilibria and faster phytolith dissolution at higher pH. Concurrent shifts in clay mineralogy toward higher-CEC minerals (smectite/vermiculite) under cultivated conditions likely further enhance PAS. Conversely, in parent materials where SiCaCl2 is low (igneous intrusive and metamorphic) or where very high clay contents and carbonate buffering minimize pH effects (>325 g kg−1 < 2 µm), cultivation did not significantly alter SiCaCl2. The mapping and correlation analyses thus directly address the research question by identifying when and where cultivation increases PAS in temperate soils and by clarifying the pedological conditions underpinning this effect. The deficiency assessment suggests only a small fraction of wheat-growing areas might be Si-limited at the lower threshold, but the large share between 20–40 mg kg−1 underlines the need for crop-specific critical values for wheat.

Cultivation increases CaCl2-extractable Si in French temperate soils developed on sediments, largely through liming-induced pH rises and associated changes in clay mineralogy, while effects are negligible in soils from igneous intrusive and metamorphic parent materials. SiCaCl2 is strongly controlled by parent material, < 2 µm fraction, and pH, with pH effects most evident where the < 2 µm fraction is 50–325 g kg−1. Nationwide mapping yielded moderate predictive skill and highlighted land use and vegetation signals. Applying literature-based PAS thresholds indicates that about 4% of wheat soils could be Si-deficient, but firm conclusions for wheat require a crop-specific critical SiCaCl2 level. Future research should: (i) establish wheat-specific critical PAS thresholds; (ii) perform paired-site and chronosequence studies, including carbonated systems, to disentangle liming, mineralogical evolution, and kinetics; and (iii) investigate Si pool dynamics and adsorption/desorption mechanisms across clay mineral assemblages.

• Critical PAS thresholds were borrowed from rice and sugarcane; no wheat-specific SiCaCl2 critical level is available, limiting deficiency inference. • Limited numbers of cultivated sites on igneous extrusive rocks and podzols precluded robust assessment for those soil types. • DSM provided moderate accuracy (R2 = 0.43), so local predictions may carry uncertainty. • The study emphasizes equilibrium proxies (extractions) and correlations; kinetic aspects of Si cycling and transformations were not addressed. • Potential confounders such as residue management and Si inputs from amendments were not explicitly quantified.