Soil and management data
Soil and management data for this study were obtained from the first Agricultural Soil Inventory in Germany that was conducted between 2009 and 2018 (Jacobs et al. 2018). Soil profiles were sampled and soil samples were analysed in the laboratory. Management data for each sampling site were obtained from the farmers by means of a questionnaire. Details on the methodology and key results of the inventory are described in Poeplau et al. (2020). A total of 2,171 cropland sites were analysed. Some of these sites were excluded for the simulations due to limitations of the models (see Section 'Simulation of the effect of cover crops on soil organic carbon stocks'). In all, 24,917 years of management data were recorded for these sites, including information on crop rotation, main crop yield and fertilisation. The period for which information on the management is available ranged from one to 19 years, with an average of 12 years for each site (between 2001 and 2019).
The sites were classified in equally-sized regional groups according to the federal states. Based on the pedoclimatic conditions we defined three regions: the warm, wet and sandy North (Lower Saxony, North Rhine-Westphalia, Schleswig-Holstein, Hamburg, Bremen), the wet and clayey South (Baden-Wuerttemberg, Bavaria, Hesse, Saarland, Rhineland-Palatinate), and the dry and sandy East (Berlin, Brandenburg, Mecklenburg-Western Pomerania, Saxony, Saxony-Anhalt, Thuringia). Climate and soil characteristics of the sites in these regions are summarised in Table 1 and shown in the supporting information S1.
Table 1 Mean climate and topsoil (0–30 cm) characteristics of the sites in each region with standard deviations. Only sites that were considered in the simulations are included (n = 1267)
Scenarios for additional cover crops
We defined the time during winter fallow where cover crops could potentially be grown as “cultivation windows”. We identified cultivation windows for cover crops based on the previous and subsequent main crops and associated average harvest and sowing months. This was done in accordance with information from regional agricultural advisory services. Long cultivation windows were defined as occurring after the main crops that are typically harvested in July or August (e.g. winter wheat (Triticum aestivum L.) and barley (Hordeum vulgare L.)). Short cultivation windows were defined as occurring after the main crops that are harvested in September or October (e.g. maize (Zea mays L.)). Growing cover crops after mid-October was not considered possible since on average the vegetation period is not long enough. Thus, there are no cultivation windows after root crops, which are typically harvested late in the year (e.g. sugar beet (Beta vulgaris L.)). The main crops following both the long and the short cultivation window are summer crops sown after March (e.g. maize or oats (Avena sativa L.)) to allow potential additional cover crops to cover the soil during winter. Details on the definition of the cultivation windows according to the main crops and a schematic crop rotation are given in the supporting information (S2 and S3). We determined the average annual cover crop area [% of total cropland] and the cultivation windows based on the management years of all sites (n = 24,917).
In order to estimate the SOC sequestration potential of cover crops, four scenarios were developed with varying percentages of cover crops in the crop rotations. The aim was to estimate an easy to realise SOC sequestration potential that maintains agricultural productivity thus main crops were not changed in order to increase acceptance by farmers. In the first scenario, no cover crops, the cover crops in recent crop rotations are taken out of the rotation and only the main crops and organic fertiliser provide C input to the soil (0% cover crops on annual cropland area). In the business-as-usual scenario, the effects of the recent management are simulated. The C input in this scenario is provided by residues (roots and shoots) of recent crop rotations, including recent cover crops (10% cover crops on annual cropland area) and by organic fertiliser. In the simple scenario, it was assumed that cover crops would also be grown in long cultivation windows and therefore it includes recent cover crops and additional cover crops in long cultivation windows (10% + 13% = 23% cover crops on annual cropland area). In the ambitious scenario, both long and short cultivation windows are used for growing cover crops while the management, i.e. the main crops and fertiliser application, remains the same. Thus, it includes recent cover crops and cover crops in both long and short cultivation windows (10% + 13% + 7% = 30% cover crops on annual cropland area).
For the simple and ambitious scenarios, where additional cover crops are implemented in the crop rotations, it was assumed that the cultivation windows are filled with a set of the most common cover crops species in proportions according to their distribution in recent crop rotations. To do this, we first calculated the C input of each cover crop species for each site using the method described in subsection 'Estimating the carbon input from cover crops'. Based on this, we calculated a weighted mean C input for each site, with weights according to the prevalence of the ten most common cover crops (listed in the supporting information (S4)).
In the business-as-usual scenario, 84% of all cover crops are incorporated into the soil and 16% are harvested, e.g. for livestock feed, based on information given by the farmers. In the simple and ambitious scenarios, we assumed that all additional cover crops were incorporated into the soil. In those cases where the cover crops are harvested and removed from the field, the aboveground C input from cover crops was reduced at this site by 75% on the assumption that roughly three quarters of cover crops are harvested and the remaining aboveground biomass is left as stubble in the field.
Simulation of the effect of cover crops on soil organic carbon stocks
To simulate the four scenarios, a multi-model ensemble was used as multi-model ensembles were shown to decrease model uncertainty in SOC simulations for German croplands (Riggers et al. 2019). Our model ensemble consisted of two process-based SOC models combined with three different C-input estimation methods for the main crops, resulting in a total of three model combinations. This combination was selected based on the analysis by Riggers et al. (2019) and further checks of the ability of the models to simulate management related SOC changes on 15 long-term field experiments with 245 treatments in Europe and 139 permanent soil monitoring sites in Germany. The five-pool-model RothC (Coleman and Jenkinson 1996) was combined with the allometric functions described in Taghizadeh-Toosi et al. (2014) and initialised by an analytic solution from Dechow et al. (2019). The three-pool-model C-TOOL (Taghizadeh-Toosi et al. 2014) initialised with fixed fractions (Taghizadeh-Toosi and Olesen 2016) was combined once with allometric functions introduced by Jacobs et al. (2020) and once with allometric functions described in Rösemann et al. (2017). The allometric functions were used to calculate the C input provided by the main crops and were based on harvest information given by the farmers. To calculate the C input of the cover crops, we developed a new estimation method, which is described below. The C input from straw, manure and roots was partitioned to the SOC pools in RothC according to the partition coefficients introduced by Dechow et al. (2019).
Simulations were performed for the topsoil (0–30 cm) at 1,267 sites. Model runs were only performed for sites that had at least five years’ recorded management, thus it could be assumed that the management was representative for the site. Out of the total 2,171 cropland sites, additional sites were excluded owing to the limitations of the SOC models. These were: (1) hydromorphic soils with a groundwater level of less than 80 cm from the surface (n = 79), (2) sites that had been under cropland use for less than 60 years (n = 185), and (3) organic soils with SOC contents above 8% (n = 8). In addition, there are sandy soils in Germany with a high SOC content (black sands) and a high proportion of recalcitrant organic matter, which is characterised by slower decomposition rates than would be expected (Springob and Kirchmann 2002; Vos et al. 2018). As the current parameterisation of SOC models is not suitable for describing these conditions, these black sand sites were also excluded (n = 41).
Current climate conditions for each recorded management year at each site were used for the simulations. Weather data were sampled from gridded datasets of monthly precipitation, temperature and sunshine duration and of daily precipitation and temperature (DWD Climate Data Center, 2020a, b, c, d). No climate change scenarios were considered since climate change not only alters temperature and precipitation, and thus the mineralisation and degradation of SOC stocks (Bruni et al. 2021; Riggers et al. 2021), but can also influence the produced biomass of main crops and cover crops, e.g. due to longer growing seasons and higher CO2 concentrations (Olesen et al. 2007). An altered biomass production would lead to altered C input rates to the soil. However, based on current knowledge, there is great uncertainty about the effect size of climate change influencing C input and C decomposition, thus the impact of climate change was not included in this study. Therefore only data for a maximum of 50 years are shown.
Estimating the carbon input from cover crops
The development of cover crop aboveground biomass production mainly depends on: (1) the temperature and precipitation in the early growing phase (Koch et al. 2017; Komainda et al. 2016), (2) the species and (3) the sowing date, which limits the remaining vegetation days (McClelland et al. 2020). We developed a new method to estimate the C input from the aboveground and belowground biomass of cover crops taking these three factors into account. An illustration of this is given in the supporting information (S5).
A linear regression was used to account for the impact of temperature and precipitation on the aboveground biomass \({a}_{mustard}\) [Mg dry mass (DM) ha− 1) (Koch et al. 2017):
$${a}_{mustard}=-2.937 + 1.16 {P}_{s} + 0.021 {T}_{s}$$
(1)
where \({P}_{s}\) is the mean daily precipitation [mm d− 1] from the assumed early sowing date (18 August) up to 30 September, and \({T}_{s}\) is the sum of the air temperature [°C d] from day 19 to day 31 after sowing. These variables were identified by Koch et al. (2017). For each of the soil inventory sites, this equation was used to estimate the site-specific aboveground biomass of white mustard (Sinapis alba L.) \({a}_{mustard}\) [Mg DM ha− 1]. The maximum production of \({a}_{mustard}\) was set at 7.5 Mg DM ha− 1, in accordance with regional data. This Eq. 1 was only developed for mustard biomass so it cannot be used for all cover crop species.
The influence of species on aboveground biomass \({a}_{cc}\) [Mg DM ha− 1] was accounted for by rescaling the average aboveground biomass of the cover crop \({a}_{cc*}\) [Mg DM ha− 1] according to the ratio between the calculated site-specific aboveground biomass of mustard \({a}_{mustard}\) [Mg DM ha− 1] and the average aboveground biomass of mustard \({a}_{mustard*}\) [Mg DM ha− 1] (Eq. 2).
The site-specific root biomass was calculated for each cover crop species \({b}_{r,cc}\) [Mg DM ha− 1] based on the same ratio \(\frac{{a}_{cc*}}{{a}_{mustard*}}\) and the average root biomass \({b}_{r,cc*}\) [Mg DM ha− 1] (Eq. 3). Both the average aboveground and root biomass values were obtained from Renius et al. (1992) and Lütke Entrup (2001) who provided a large set of data on average aboveground and belowground biomasses of many cover crop species grown in Germany. The average root biomass data also contain stubble, which was considered here to be aboveground biomass. In order to estimate the aboveground and root biomasses appropriately, stubble was subtracted from the given root biomass values and added to the aboveground biomass, assuming that stubble is 10% of the aboveground bimass \({a}_{cc}\) [Mg DM ha− 1]. The aboveground biomass \({a}_{cc}\) and root biomass \({b}_{r,cc}\) were then calculated as described in Eqs. 2 and 3 below:
$${a}_{cc}=\frac{{a}_{mustard}}{{a}_{mustard*}}{a}_{cc*}\frac{1}{1-0.1}$$
(2)
$${b}_{r,cc}=\frac{{a}_{mustard}}{{a}_{mustard*}}{b}_{r,cc*} - \frac{0.1}{1-0.1}\frac{{a}_{mustard}}{{a}_{mustard*}}{a}_{cc*}$$
(3)
Only early sowing in long cultivation windows allows optimally developed cover crop biomass. The negative effect of late sowing on the biomass was accounted for by reducing the aboveground and root biomass by 35% in short cultivation windows based on Renius et al. (1992).
The calculated cover crop biomasses were considered plausible as the biomass ranges fitted well with reports by regional agricultural advisory services (Kanders and Berendonk 2013; LfL 2011; Schmidt and Gläser 2013).
The average root biomass from the literature has been evaluated for a depth of 0–60 cm (Renius et al. 1992), but since the aim here was to calculate the C input for a depth of just 0–30 cm, the root biomass was rescaled according to the root distribution introduced by Gale and Grigal (1987) (Eq. 4):
$${b}_{r,30}= \frac{1-\left({0.961}^{30}\right)}{1-\left({0.961}^{60}\right)}{b}_{r,cc}$$
(4)
The total belowground biomass \(b\) [Mg DM ha− 1] providing C input to the soil was calculated by totalling the root biomass \({b}_{r}\) and rhizodeposition. Rhizodeposition was assumed to be 31% f the root biomass \({b}_{r,30}\) [Mg DM ha− 1] (Pausch and Kuzyakov 2018). A C content of 47% was asumed for all biomasses to calculate the C input [Mg C ha− 1] (Jacobs et al. 2020).
Definition of soil organic carbon sequestration and accumulation potential
We define SOC sequestration potential or SOC accumulation potential as the simulated increase in SOC stocks [Mg C ha− 1] in relation to different scenarios (S6). Recent SOC sequestration by cover crops was accounted for by subtracting the SOC stocks of the no cover crops scenario from the SOC stocks of the business-as-usual scenario. The additional SOC sequestration by cover crops was defined in relation to the business-as-usual scenario and was calculated by subtracting the SOC stocks of the business-as-usual scenario from the SOC stocks simulated with the two scenarios of increased cover crop frequency (simple scenario and ambitious scenario). The total SOC sequestration was calculated by adding together the recent SOC sequestration and the maximum additional SOC sequestration. However, in cases where both the business-as-usual scenario and the ambitious scenario predicted decreasing SOC stocks compared with today, this was no SOC sequestration in a strict sense as C is released to the atmosphere rather than captured in the soil. We referred to these cases as SOC accumulation and reduction of SOC losses instead.
In order to provide annual SOC accumulation rates [Mg C ha− 1 a− 1], the SOC stock increase [Mg C ha− 1] was divided by the corresponding simulated time span [a], assuming a linear increase in SOC stocks over the simulated period of two to 50 years. However, in reality, accumulation rates are nonlinear and decrease over time as SOC stocks reach a new equilibrium.
Software
The simulations were run in R version 4.0.4 (R Core Team 2021). The RothC implementation from the SoilR package (Sierra et al. 2012) and the C-TOOL implementation from Riggers et al. (2019) were used. Visualisation of the results was undertaken with the tidyverse package (Wickham et al. 2019). Errors are given as standard deviation or 95% confidence interval, unless stated otherwise.