12 plant species were grown in mesocosms for two months in three closed climate simulators in an ecolab simulation system . Ecolab systems allow a realistic simulation of climate and ecosystem conditions within a large range of parameters. Here, we simulated a continental temperate summer climate by averaging six years of meteorological data from 15 August to 10 October, recorded on the CEREEP Ecotron IdF site (48°16^′56.5^′′ N, 2°40^′15.9^′′ E) (Fig. S1). Temperature and relative humidity were programmed to vary within a 10 min resolution. Photosynthetic Photon Flux Density (PPFD) was also simulated on a 10 min resolution step and discretised into 10 steps over a day, from 0 % to 100 % of the measured intensity. The maximum (100 %) average PPFD was equal to 1181 µmol of photons s⁻¹ m⁻² at the top of the mesocosms (soil surface). When light intensity meteorological data were higher than this maximum value, light intensity was set to 100 %. CO₂ was set at 415 ppm during daytime and at 487 ppm during nighttime, corresponding to on-site measurements. Mesocosms were irrigated with deionized water by drippers and received in total 83 mm of water, which corresponds to the cumulative precipitation data over 56 d. Soil moisture was measured every 15 min in 12 separate mesocosms that were not used for C analysis and each contained one of the 12 plant species. We used two sensors (METER Teros 12) per mesocosm, one at the surface and one at 30 cm depth. As mesocosms soils generally dried faster than soils in the field, the irrigation frequency was smoothed over time to avoid the soil surface water content decreasing below 0.05 cm³ cm⁻³ to ensure plant survival. If the soil moisture level dropped below this value for any mesocosm, regardless of the species, irrigation was launched for all mesocosms. At day 16, the plants were fertilised with an ammonium nitrate solution, at a dose equivalent to 40 kg of Nitrogen (N) per hectare to ensure growth and to simulate mineral residual soil N following a crop harvest.

To label plant material, we performed multiple injections of 13C-CO₂ (99 %). Whereas to assess the dynamic transfer of recent photoassimilates, a single pulse of enriched CO₂ is well suited, continuous labelling is preferred to ensure that all root-derived products are labelled and the rhizodeposition can be quantified . Nonetheless, multi-pulse is a robust alternative if pulses are injected very regularly within a short period of time . To do so, the δ13C isotopic signature of the air in the chambers was measured on line with a cavity ring-down spectroscopy CO₂ analyser (PICARRO G2201-i) coupled to a multiplexer switching between the different chambers during the whole experiment, in order to control the duration and number of injections (Fig. S2). 13C-CO₂ signature and atmospheric CO₂ concentration were controlled separately.

The 12 plant species, listed in Table , are crops from three families (Fabaceae; Poaceae and Brassicaceae). They were selected to cover a wide range of plant traits and for their ability to be grown in the targeted period. Except for Medicago sativa, which is perennial, each species could potentially be selected for an intermediate cropping in temperate regions. The growth period of 56 days was chosen, as it is the minimum period permitted before the crop can be destroyed, according to French regulations in areas vulnerable to nitrate pollution. They were sown in the mesocosms at a different time to target a simultaneous plant emergence. We sowed twice the amount of seeds recommended by French technical institutes for a pure intermediate crop (Table S2) . In the first two weeks, some seedlings have been removed if the seed germination rate exceeded 50 %. The number of seeds ranged from two for faba bean to 64 for alfalfa. The number of plants that have been maintained is available in Table .

Polyvinyl chloride mesocosms of 20 cm diameter were filled with 13 L of sandy soil (6.9 % clay, 19.0 % silt, 74.1 % sand for the upper horizon) . This soil was excavated from a semi-open habitat (grassland, shrubs and individual trees) of the research station by distinguishing two operational horizons and was sieved at 1 cm. Soil was then packed in the mesocosms to recreate the two horizons after homogenisation. The upper horizon was 20 cm deep, and the lower one 25 cm deep. Both were packed at a density of 1.2, corresponding to realistic values found on site. C concentrations and δ13C isotopic signatures are 7.2 gC kg⁻¹ and -27.4 ‰ for the upper horizon and 4.3 gC kg⁻¹ and -26.2 ‰ for the lower horizon, respectively. This design of 12 mesocosms was replicated three times across three climate chambers simulating the same atmospheric conditions.

After 56 d, mesocosms were unpacked with a minimal destruction of the structure, by extracting the entire soil core from the pot. Standing aboveground biomass and dead leaves, if any, were collected after cutting the stems at the soil surface. The phenological stage was recorded according to the BBCH scale . Roots were at first extracted manually. The soil was then sieved at 2 mm with a minimal shaking, so that remaining fine roots could be sampled with tweezers. All roots were then washed with tap water. We distinguished roots from the upper and lower horizon. Soil was collected in both horizons in several places to obtain a composite sample of 500 g. Out of each planted mesocosm, the root system of one plant was carefully isolated, and the rhizosheath, which is the soil adhering to the roots after root extraction and shaking , was collected in both horizons when possible with a paintbrush for this specific plant, after a gentle shaking of the root system. The rhizosheath and its corresponding roots were weighted. It allowed us to obtain a Rhizosheath:Root ratio and therefore to estimate the whole rhizosheath soil mass for each mesocosm with Eq. (). 1mrhizosheath=mroots×isolated.mrhizosheathisolated.mroots mroots is the whole mesocosm root mass and isolated.m_rhizosheath and isolated.m_roots are the masses of rhizosheath and roots subsamples respectively. More rhizosheath soil was collected on the roots from the other plants of the mesocosm, but without associating it with a root mass. Soil samples of both horizons and of both locations (bulk vs. rhizosheath) were dried at 40 °C, sieved at 2 mm, milled and analyzed for total C and isotopic signature δ13C with cavity ring-down spectroscopy (PICARRO G2201-i/COSTECH). For every soil compartment, a humidity correction was applied after heating a soil subsample at 105 °C, to obtain a soil mass. Rhizosheath soil was analysed only when more than 13 g of sample was collected in one horizon. Plant material was dried at 40°C, weighted, milled and was analyzed for C and δ13C, as well as for total N (Thermo Fisher Scientific FlashHT).

We grouped net rhizodeposition and non-collected fine root debris under the term SOC_new, which is in our case the labelled SOC remaining after roots extraction and sieving at 2 mm. We calculated this amount (kg) of root-derived C with Eq. () in four compartments (2 horizons × 2 localisations: bulk soil vs. rhizosheath soil). We obtained the total SOC_new of one mesocosm by summing all the compartments. 2SOCnew=δ13Csoil-δ13Creferenceδ13Croot-δ13Creference×OCsample With δ13Csoil and δ13Croot being the δ13C isotopic signature (‰) of the soil sample and of the corresponding labelled roots, respectively. δ13Creference is the mean δ13C of the soil before the plants were sown, at the corresponding horizon (n=20). OC_sample is the mass of OC contained in the soil compartment (kg), obtained with the OC concentration of the sample. This equation assumes that SOC_new has an isotopic signature similar to that of roots. As this is a strong assumption, we also provide a quantification of SOC_new calculated with the δ13C of the shoots instead of δ13Croot for comparison (Fig. S4). We calculated specific SOC_new (kg kg⁻¹) by dividing SOC_new by root C. We expressed C quantities per hectare (kg ha⁻¹) in Fig.  and S4 by doing a cross product with the surface of the mesocosms.

Following the first experiment, we used the labelled material to perform a litterbag incubation in the field. Out of the 12 species, six species (2 per family) were chosen to be incubated: Vicia sativa; Vicia faba; Secale cereale; Avena sativa; Brassica napus and Raphanus sativus var. longipinnatus. Plant and soil material was placed in 100 mm × 100 mm nylon bags with a mesh size of 50 µm. This mesh size only enables microfauna to access the fresh material, but limits the diffusion of SOC_new particles out of the bags.

The design comprises two general treatments, applied to each of the six plants and their corresponding soil: (1) To assess SOC_new decomposition, 50 g of labelled soil was directly placed in the bags. (2) To assess root decomposition, 50 mg of dry labelled roots were mixed with around 50 g of sieved unlabelled soil, corresponding to the soil used in the first labelling experiment. This represented from 0.9 % to 10.3 % of the C in the bags Broots at the start of the incubation. The roots were fine roots (≤ 2 mm), except for Raphanus sativus, where some of the taproot was mixed with fine roots due to a lack of material. In the following, we will use BSOCnew and Broots to refer to the bags of these two treatments. Each combination of treatment and plants was replicated three times, using plants and soils coming from different mesocosms. Besides, three bags were filled with control soil. This whole design was replicated four times, to allow four sampling dates.

Incubation started on 20 December 2023. The bags were buried at a 10 cm depth in the soil of a Poaceae-dominated grassland of the research station CEREEP-Ecotron IDF and recovered by means of a wire emerging from the soil. For year 2024, the mean annual temperature was 13.1 °C and cumulative rainfall was 844.7 mm. The incubation site was in the direct vicinity from the excavation site of the soil inside the litterbags. We consider that they are similar. Four sets of bags were recovered after 124, 195, 330 and 524 d. The whole content of each bag was dried at 40 °C, milled and analyzed for total C and δ13C. When retrieving the bags, attention was paid to preserve clods of soil and thus vegetation above the bags. After 124 d, plant cover (Vicia sativa; Sinapis alba; Secale cereale) was sown to maximise vegetation recovery.

We first calculated the proportion of C originating from the plant for each bag, Fplant, be it SOC_new or roots, with Eq. (). This was done for all collected bags and for the initial labelled soil of BSOCnew, before the incubation (t0). 3Fplant=δ13Cbag-δ13Ccontrolδ13Croot-δ13Ccontrol

δ13Cbag, δ13Ccontrol, and δ13Croot are the C isotopic signatures of the labelled and control soil in the bag and the corresponding roots respectively. For bags Broots, δ13C_root was measured independently of the first experiment to account more precisely for the roots subsample specifically used for the incubation. We used the mean of all controls for δ13Ccontrol, irrespective of the collection time, for bags Broots, whereas for bags BSOCnew, we used -27.4 ‰, in accordance with the previous labelling experiment.

We then calculated the concentration of plant-derived products, [SOC_plant] (gC kg⁻¹), with Eq. (). 4SOCplant=Fplant×OCsample [OC_sample] is the OC concentration of the sample (gC kg⁻¹).

To assess SOC_new loss, we calculated its remaining proportion, SOCnew_remaining (%), by dividing [SOC_plant] of the bags BSOCnew at day ti by the initial [SOC_plant] at harvest (t0) with Eq. (). 5SOCnewremainingti=[SOCplant]ti[SOCplant]t0×100

If SOCnew_remaining was above 100 % or below 0 %, we chose to assign it 100 % or 0 % respectively, instead of removing the data point.

To estimate the C loss of roots Rootsremainingti (%), we devided the root C remaining at the sampling date ti, estimated with Fplant, by the original amount of root C added in the bag, according to Eq. (). 6Rootsremainingti=masssoilt0×[OCsample]×Fplantmassrootst0×[OCroots]×100 masssoilt0 and massrootst0 are the initial masses (kg) of the control soil and the roots mixed in the bags and [OC_root] the respective root C concentrations (gC kg⁻¹).

We used mesocosms as the statistical unit for our analyses. Consequently, we had 35 data points, corresponding to 12 species replicated three times each, except for mustard, where plant growth failed in one mesocosm. One mesocosm may contain one single plant or several plants, according to the sowing density (Table ). For the latter case, quantitative values, C amounts or Root : Shoot (R : S) ratios for instance, stand for the entire mesocosm. In the text, the error associated with means is ± the standard deviation. For the figures, it is specified in the caption. All statistical analysis were performed with R language . For p-values of models or models parameters, we set the significance threshold at 0.05. However, in certain cases that are specified in the text, we report a trend when p-values <0.1.

Comparisons between species (n=3) were performed using one-way analysis of variance. Posthoc comparisons were performed with Tukey HSD tests. To compare families (n=12), we adopted a linear mixed-effects model with the species as a random effect to account for the hierarchical structure, with the package lme4 . Only the intercept was allowed to vary. We then performed an anova on the model with the package car and a multiple comparison with the package emmeans that uses the marginal means .

To test for linear correlations between quantitative variables, we also selected linear mixed-effects model with the species as a random effect to let the intercept vary. For these models, we report the marginal R2, that accounts for the variability explained by the fixed effects, calculated according to the recommendations of with the package Performance .

For the incubation experiment, we did not fit any decomposition models given the low number of sampling dates. To compare decomposition status of SOC_new and roots, we performed paired Student's t-tests for each sampling date. To compare species, we performed one-way analysis of variance followed by Tukey HSD tests. The list and results of statistical analysis are available in .

The root incubation revealed a fast C loss, with 32.7 ± 23.2 % of the root C remaining after 124 d, all plants combined (Fig. , Table ). This tended to reach a plateau as we observed similar values for the other dates of harvest, with 23.7 ± 13.8 % remaining after 524 d. We observed significant differences in mineralisation status between plant species at days 124 and 524 only (p=4×10-3 and 3×10-3 respectively). These differences were mainly driven by faba bean, which exhibited slower decomposition than non-legume species, and, to a lesser extent, by vetch at day 524. However, no significant pattern was detected for other species and sampling dates.

A first important result for SOC_new decomposition is that we did not see any significant difference between species. However, a high uncertainty is associated with the results, and the decomposition curves do not show a clear decreasing exponential pattern, as often seen in such experiments (Fig. ). Indeed, its calculation yielded nine incoherent values (≤ 0 % or ≥ 100 %) out of 65 bags recovered, due to a δ13C lower than the control or to an increase of the δ13C compared to t0. We attributed this to the fact that fresh SOC_new is a very small fraction of SOC (between 0.34 % and 1.71 % at the start of the incubation), resulting in a small δ13C difference with the control, exacerbating errors due to analysis and handling. Indeed, initial δ13C of the bags containing SOC_new varied from -21.929 ‰ to -25.9 ‰, whereas the mean value of native SOC is -27.4 ‰.

Nevertheless, we were able to observe that SOC_new tended to have a significantly slower decomposition than roots, at least for three sampling dates (Fig. , Table ). After 524 d, 48.6 ± 35.2 % of SOC_new was remaining.

We compared our results on root decomposition with data from literature reviews from which we recalculated a percentage of dry mass presumed to be remaining at day 524 (Table S1). Our decomposition rates for graminoids are in line with literature values reporting 12 % to 17 % of the C remaining at day 524 against 17 % in our case (Rye and Oat). However, for forbs, we observed a slower decomposition rate compared to who reported a mean of 1 %, against 27 % in our case (faba bean, vetch, rapeseed and radish). The only differences between species that we observed were for legumes whose roots exhibited lower decomposition at two sampling dates. This was unexpected as legumes have the lowest C : N ratios (Table ). Even though C : N is explaining less than 5 % of the variance of decomposition for fine roots at a global scale, the expected trend is in the opposite direction . However, no pattern was detected for other species and sampling dates, which assigns a limited role to litter quality as a driver of SOC storage, compared to the quantity of the inputs, in accordance with findings from .

The fact that SOC_new tends to have a significantly slower decomposition than roots may seem counterintuitive as rhizodeposition is often considered a labile pool, notably because it is composed of small molecules. Two hypotheses could explain this result. First, the fraction of SOC_new that remains at harvest may have already been depleted of labile compounds. As a result, we compared the decomposition of roots, that have not undergone any transformation and therefore contain easily decomposable products, and of SOC_new which has undergone a first decomposition during the growth of the plants. Secondly, another hypothesis is proposed by the Microbial Efficiency-Matrix Stabilisation framework : SOC_new would have decomposed more slowly precisely because gross rhizodeposition may contain in proportion more labile compounds than roots. These compounds would have been preferentially incorporated in the microbial loop and would thus have had a greater likelihood of forming associations with minerals. At the start of the incubation, SOC_new would then already contain a portion of relatively stable OM. Results of studies tracing labelled C in stable OM fractions support this assumption . These two hypotheses rely on two different concepts: recalcitrance versus C use efficiency associated to stabilisation mechanisms. Nevertheless, they are not contradictory but complementary. There is little incubation data available in the literature for comparison. found that after 69 d of incubation, 80 % to 57 % of net rhizodeposition from perennial grasses was remaining (against 74 % to 89 % for the roots). For rice, found that 54 % was remaining after 240 d (against 58.1 % for the roots). These two studies, together with our results, show that net rhizodeposition is a sub-compartment of gross rhizodeposition which stands out for its longer decomposition time. Whereas more than the half of gross rhizodeposition is respired within 15 d , about half of the fraction remaining at harvest, here embedded in SOC_new, is not mineralised after 524 d (Table ). Nevertheless, the high variability in our SOC_new results tempers the strength of our conclusions and calls for further research to confirm them.

We propose that SOC_new, as defined in our study and in other continuous labelling studies , is not a C pool that derives from a physiological process. Indeed, it encompasses a broad variety of compounds and depends on root sampling, which is itself soil- and operator-dependent . Rather, it is an operational parameter that helps refine estimates of C inputs in annual crops by accounting for all inputs that are not mineralised at crop harvest. Our results showed that it is a significant C input with greater persistence in soils than roots during the first 524 d of decomposition, although previous incubation studies temper this conclusion . This makes it a relevant and a necessary pool to consider when reasoning SOC sequestration. For instance, our observation of inter-species differences in specific SOC_new reveals that a fixed allocation coefficient is not suitable for SOC_new quantification from root C. In our case, the belowground C inputs of Brassicaceae, that exhibited high specific SOC_new, would be relatively underestimated with such an approach.

However, including SOC_new in C assessments requires estimating it from crop traits or other C pools, as its direct quantification is costly. We highlighted here that the task remains challenging for annual crops, and it is likely to be exacerbated if we try to extrapolate our results to a complex cultivation system. Nevertheless, the positive correlation with root C provides initial guidance. Crop selection favouring high root inputs to the soil seems to be a promising practice to sequester additional C in certain pedoclimatic contexts without being at the expense of the yield . Considering that there is more knowledge on root selection than on SOC_new and considering that there is a coupling between the two pools in terms of quantity and vertical distribution, adjusting the reasoning of additional SOC sequestration on root inputs is likely not to be contradictory with a reasoning based on SOC_new. Moreover, we highlighted that root inputs are the main contributors of the fresh root-derived C pool which persists in the soil in the year following the harvest (Fig. ): even though SOC_new tends to remain longer in the soil based on our results, its low net production compared to that of roots (27.8 % of net belowground C production) restrains its relevance as a lever to sequester additional SOC.

A missing link of our analysis is the priming effect. This process can greatly restrict C sequestration and should therefore be taken into account. We can nonetheless note that priming effect is mostly driven by aboveground biomass or by rhizodeposition and less by roots . This further reinforces the previous statement that belowground inputs, roots like SOC_new, are decisive to increase SOC stocks. Moreover, the rhizosphere priming effect is closely linked to nutrient acquisition and may, therefore, be more of a return on investment associated with biomass production rather than a net C loss.