Coupled modelling of soil erosion, carbon redistribution, and turnover has received great attention over the last decades due to large uncertainties regarding erosion-induced carbon fluxes. For a process-oriented representation of event dynamics, coupled soil–carbon erosion models have been developed. However, there are currently few models that represent tillage erosion, preferential water erosion, and transport of different carbon fractions (e.g. mineral bound carbon, carbon encapsulated by soil aggregates). We couple a process-oriented multi-class sediment transport model with a carbon turnover model (MCST-C) to identify relevant redistribution processes for carbon dynamics. The model is applied for two arable catchments (3.7 and 7.8 ha) located in the Tertiary Hills about 40 km north of Munich, Germany. Our findings indicate the following: (i) redistribution by tillage has a large effect on erosion-induced vertical carbon fluxes and has a large carbon sequestration potential; (ii) water erosion has a minor effect on vertical fluxes, but episodic soil organic carbon (SOC) delivery controls the long-term erosion-induced carbon balance; (iii) delivered sediments are highly enriched in SOC compared to the parent soil, and sediment delivery is driven by event size and catchment connectivity; and (iv) soil aggregation enhances SOC deposition due to the transformation of highly mobile carbon-rich fine primary particles into rather immobile soil aggregates.
Soil organic carbon (SOC) is the largest terrestrial carbon (C) pool and has
been identified as a cornerstone for the global C cycle. Globally, approx.
1400 Pg C is stored in the upper meter of soil, with approx. 700 Pg C in
the upper 0.3 m (Hiederer and Köchy, 2011). As a result, exchange rates
between soil and the atmosphere are a major concern with regards to climate
change (Polyakov and Lal, 2004a). Earth system model-based estimates for
terrestrial C storage in the year 2100 vary widely, ranging from a sink of
approx. 8 Pg C yr
Most challenging in developing and especially testing models that couple process-oriented SOC redistribution with SOC dynamics are the different spatial and temporal scales of the processes at play (Doetterl et al., 2016). Process-oriented erosion models need event-based data to be validated, while SOC dynamics can hardly be observed on timescales smaller than several decades. Consequently, most existing models that couple soil erosion and SOC turnover processes are based on long-term, USLE-type erosion models that ignore event dynamics. The most widespread of these is SPEROS-C, which was applied on scales ranging from micro- to mesoscale catchments (Fiener et al., 2015; Nadeu et al., 2015; Van Oost et al., 2005b).
The conventional approach to modelling coupled soil erosion and SOC turnover is to treat SOC as a stable part of the bulk parent soil and statistically model (long-term) erosion. However, this approach is likely to lead to biased estimates of both water-erosion-induced SOC redistribution and its effect on vertical C fluxes. Numerous studies have shown that the transport of SOC is selective (Schiettecatte et al., 2008), controlled by event characteristics (Sharpley, 1985; Van Hemelryck et al., 2010) and soil aggregation (Hu and Kuhn, 2014, 2016). The enrichment of SOC during transport has been explicitly addressed by a few modelling studies, using different approaches (Fiener et al., 2015; Lacoste et al., 2015). The effects of tillage erosion on vertical C fluxes have not yet been evaluated in detail, although a representation has been accounted for in some modelling studies (Lacoste et al., 2015; Van Oost et al., 2005a).
The aim of this study is to couple a spatially distributed, process-oriented and event-based water erosion model with a tillage erosion model and a SOC turnover model in order to analyse the importance of individual erosion processes in the erosion-induced C balance of agricultural catchments. The study intends to identify relevant processes that should be implemented in less data-demanding, more parsimonious models.
The test site is located about 40 km north of Munich in the Tertiary Hills,
an intensively used agricultural area in southern Germany. The site consists
of two small arable catchments (48
Land use, topography, and tillage direction for modelled catchments C1 and C2. In catchment C2, a grassed waterway (GWW) is located along the thalweg, while vegetated filter strips (VFS) are located along the upslope and downslope field borders.
Modelling scheme of the Multi-Class Sediment Transport and Carbon dynamics model (MCST-C).
For our study, we coupled three different models: (i) the process-oriented Multi-Class Sediment Transport Model (Fiener et al., 2008; Van Oost et al., 2004; Wilken et al., 2017), a spatially distributed and event-based water erosion model with a specific emphasis on grain size selectivity using the Hairsine and Rose equations (Hairsine et al., 1992; Hairsine and Rose, 1991); (ii) a tillage erosion model following a diffusion-type equation adopted from Govers et al. (1994), which derives tillage erosion from topography and tool-specific tillage erosion coefficients; and (iii) the Introductory Carbon Balance Model (ICBM; Andrén and Kätterer, 1997; Kätterer and Andrén, 2001), which models SOC turnover. The ICBM calculates yearly SOC dynamics using two SOC pools (“young” and “old”) and four C fluxes (C input from plants, mineralization from the young and the old pool, and humification). Both the tillage erosion and ICBM model were adapted from SPEROS-C, which couples annual water erosion (based on the RUSLE; Renard et al., 1996), tillage erosion and SOC turnover (Fiener et al., 2015; Nadeu et al., 2015; Van Oost et al., 2005b). In the following, we describe only those features of the coupled MCST-C model (Multi-Class Sediment Transport and Carbon dynamics model) that had to be adapted in order to couple the models or for the introduction of SOC-specific transport mechanisms. An overview of the main model concepts of MCST-C is given in Fig. 2. For more details regarding the three coupled models and processes modelled therein, we refer the reader to the original publications (see above).
The representation of soil texture and SOC in the model is three-dimensional.
The horizontal distribution of grain-size-specific soil and SOC is grid-based, while the vertical distribution is represented by ten 10 cm layers.
The two uppermost layers are assumed to be homogeneously mixed due to
tillage. The grain size distribution is represented in 14 primary particle
classes, described by class median particle diameter, particle density, and
the class proportion relative to the bulk soil (kg kg
SOC transport is associated with various primary particle and aggregate
classes. Based on the literature (Doetterl et al., 2012; Von Lützow et al.,
2007), it is assumed that mineral bound SOC is primarily attached to fine
particles (
In its original version, the MCST model treats events individually without considering changes caused by previous events. For a continuous application, the water erosion module of MCST-C simulates single events and keeps track of the following redistribution related changes in the catchment: spatial and vertical changes in (i) the grain size distribution and (ii) SOC content and (iii) the development of a rill network, which remains until the next tillage operation. A layer-specific mixture module continuously updates for spatial changes in the vertical grain size distribution and its associated SOC content, changes which are caused by selective redistribution of water and non-selective tillage erosion. In the case of net deposition, new material with a different grain size distribution is added to the top of the plough horizon (layer 1 and 2). Subsequently, the grain size distribution of the plough layer is mixed and assumed to be homogeneous. Furthermore, deposition leads to an upward movement of the layer borders such that soil material from the plough layer becomes incorporated into the subsoil layers. Any C content moving below 1 m depth is summarized and assumed to be stable in time. In contrast, erosion lifts new material from the subsoil horizons upwards. Assuming that the deepest horizon represents the original loess, the properties of uplifted subsoil remain constant, delivering infinite material of the same grain size distribution and C content.
For a truly rigorous validation of MCST-C, there are numerous long-term data requirements: event-based data for surface runoff, sediment delivery and SOC delivery, long-term data regarding changes in spatially distributed SOC stocks, spatially distributed C loss and gain due to crop harvesting, and C input via plants and manure application. In addition to these validation data requirements, model input data would also be required over decades for a long-term validation. The research project (Auerswald et al., 2000) which is the basis of this study provided a very comprehensive database. However, continuous monitoring was “only” carried out for 8 years (1994 to 2001), and SOC inventories span roughly a decade (first inventory in 1990/91, second in 2001). Therefore, measured changes in SOC stocks are too small to be used for a long-term model validation (requires approx. 50 years; see implementation).
In consequence, we only use the measured continuous event-based surface
runoff and sediment delivery from catchment C1 to validate the modelled
erosion. The runoff was collected at the lowest point of the catchment
(Fig. 1), which was bordered by a small earthen dam. From the dam, the runoff
was transmitted via an underground tile outlet (diameter 0.29 m) to a
measuring system consisting of a Coshocton-type wheel runoff sampler (for
details regarding the procedure and the precision of the measurements see
Fiener and Auerswald, 2003). Corresponding precipitation was measured using a
tipping bucket rain gauge of 0.2 mm volume resolution. To determine single
erosion events, the precipitation data are filtered in two steps: first, all
events with cumulative precipitation
Main input data and parameters used in the Multi-Class Sediment Transport and Carbon dynamics model (MCST-C).
As the original MCST model was previously tested in catchment C1 (Fiener et al., 2008), we did not explicitly calibrate the surface runoff and erosion model. Instead, observed runoff and sediment delivery data was used to test whether our changes to the model still result in a reasonable model performance.
To run and test MCST-C, a variety of measured input data and parameters are
required. This input data are partly calculated from measured data at the
research farm and partly taken from literature (Table 1; Fig. 2). To model
surface runoff and erosion, the most important input data requirements are
(i) precipitation, measured at two meteorological stations about 100 to
300 m from the catchments using 0.2 mm tipping buckets, (ii) a lidar
5 m
Model parametrization to analyse the effects of different erosion
processes upon C fluxes. Model runs are abbreviated as follows: reference
run (Ref), without tillage erosion (Til
As indicated above, it is difficult, if not impossible, to identify erosion-induced changes in SOC and vertical C fluxes if measurements or modelling efforts do not cover decadal time spans. Therefore, a 50-year synthetic input data set and parameter set was created for MCST-C in order to analyse C dynamics. This data set is based on the 8 years of measured data used to validate the erosion component of the model. First, a time series of precipitation was established by randomly choosing the data of one of the eight measured years (see Sect. 2.5: Model validation) and applying it for the first 42 years of the time series. This was followed by the original 8 measured years to reach the total of 50 years. Next, this precipitation time series was combined with synthetic land use and soil management data representing two full crop rotations (1994 to 2001), which were repeatedly used for all 50 years. This combination leads to a wide variety of precipitation events (time step 1 min) occurring for different daily soil covers by vegetation as a major driver of soil erosion. In contrast to the erosion dynamics, C inputs via plants and manure are repeated every 8 years, which ignores any potential change in management and yields within the modelling period. The synthetic input data were applied for both catchments for the purpose of comparability.
Various model setups were chosen (Table 2) to analyse the effects of
different erosion processes upon lateral SOC redistribution and the resulting
vertical C fluxes. All of these model runs were compared to the 50-year
reference run that was validated for the 8-year monitoring phase at the
research farm (1994–2001). In general, we tested the effect of a number of
water erosion processes and compared the relevance of water vs. tillage
erosion. Firstly, the critical shear stress of rill initiation (
Median class diameter distribution (14 primary particle and 2 aggregate classes) in the plough layer assuming different aggregation levels, as described in Table 2.
Spatial patterns of tillage and water erosion for the 50-year simulation period of the reference run.
To compare vertical C fluxes from erosional and depositional sites, the
corresponding total and mean C flux was calculated on an annual basis. To
isolate the C fluxes that result solely from erosion processes, we first
calculate all vertical C fluxes excluding erosion processes and then subtract
these from the vertical C fluxes including erosion processes. In the
following results section, positive C fluxes indicate an erosion-induced C
gain for the catchment (input to the soil), while negative fluxes indicate an
erosion-induced loss (from soil to the atmosphere or SOC delivery from the
catchment by runoff). Subsequently, erosional and depositional sites were
spatially subdivided and an average vertical C flux in kg C m
A number of goodness-of-fit parameters (Table 3) indicate a sufficient model
performance to simulate event runoff and sediment delivery for the 8-year
observation period. The Nash–Sutcliffe efficiency and coefficient of
determination for runoff (NSE
Model performance, as described by the Nash–Sutcliffe efficiency
(NSE; Nash and Sutcliffe, 1970), root mean square error (RMSE),
coefficient of determination (
The simulated tillage and water erosion shows distinct spatial patterns (Fig. 4). The highest rates of tillage erosion are found along the upslope boundaries of the arable field and on hilltops. The main areas for tillage-induced deposition are at the downslope arable field boundaries and in concavities (Fig. 4). Due to the well-established soil conservation system, water erosion takes place over a much smaller spatial extent and is limited to the main hydrological flow path, while deposition is dominantly found in the vegetated filter strips and grassed waterway (Fig. 4).
Simulated cumulative vertical C fluxes for erosional (Ero1, Ero2) and depositional (Dpo1, Dpo2) sites, lateral C delivery (Del1, Del2), and catchment C balance (Bal1, Bal2) for catchment C1 and C2. For details regarding the model runs and corresponding abbreviations see Table 2.
The reference run (validated against sediment delivery in catchment C1,
1994–2001) shows positive vertical C fluxes at erosional sites over the
50-year simulation period, with a cumulative flux of 40 g m
The event-based SOC enrichment in delivered sediments, compared to parent
soil, ranges from 1.1 to 2.7 (2.4 mean) for C1 and from 2.5 to 2.7 (2.7 mean)
for C2 over the 50-year time span (Fig. 6). Subdividing the events into
tertiles (33 % parts) according to sediment delivery, the mean enrichment
in C1 is 2.5 (
Vertical C fluxes show a large response to changes in the
Lateral SOC delivery is solely caused by water erosion. The model shows its
smallest levels of lateral SOC delivery when grain size selectivity is
ignored (GS
Event-size-specific simulated mean SOC enrichment in delivered
sediments of catchment C1 and C2. Error bars indicate one standard
deviation. Panels
Variations in SOC enrichment of delivered sediments is generally rather small
for all model runs (Fig. 6). The most pronounced effect on SOC enrichment
results from different aggregation levels (Agg
Tillage erosion dominates the erosion-induced vertical C fluxes in both
catchments. Without water erosion (Wa
Overall, to achieve accurate estimates of vertical erosion-induced C fluxes, it seems to be more important to improve the representation of tillage erosion in the model, rather than focusing on detailed process-oriented water erosion modelling, which is less important for vertical C fluxes.
In contrast to vertical C fluxes, lateral erosion-induced C fluxes are substantially affected by a number of event-specific processes. To assess these processes, a spatially distributed process-oriented modelling approach is needed.
Our synthetic 50-year data set (based on the 1994–2001 observations) produces three large SOC delivery events, representing nearly 60 % of the total SOC delivery in both catchments (Fig. 5: Del1–Del2). This underlines the importance of accounting for individual events, particularly for the enrichment of SOC in delivered sediment (Fig. 6). However, it should be noted that SOC enrichment is mostly affected by catchment characteristics (Fig. 6b–c). While catchment C1 follows the expected behaviour, i.e. decreasing SOC enrichment with increasing event size (Auerswald and Weigand, 1999; Menzel, 1980; Polyakov and Lal, 2004b; Sharpley, 1985), and is in good agreement with the results of Wang et al. (2010) for similar soils in the Belgian loam belt, event size had hardly any effect on the SOC enrichment in catchment C2, where any larger particles, including aggregates, are deposited in the grassed waterway due to consistently high hydraulic roughness throughout the year. Hence, a parsimonious approach solely relating annual erosion magnitude to SOC enrichment (e.g. Fiener et al., 2015, using the model SPEROS-C) might fail on the landscape scale due to varying inter-field connectivity characteristics of catchments. Underlining the results of recent studies (e.g. Hu and Kuhn, 2016), it seems to be essential to take detailed processes into account during erosion, transport, and deposition in order to accurately capture the SOC enrichment of delivered sediments. In our modelling example, neglecting enrichment would lead to a 36 % underestimation of the total SOC delivery in catchment C1 and an even more extreme 63 % underestimation in catchment C2. This large difference between catchment C1 and C2 suggests that the relevance of SOC enrichment in delivered sediments is controlled not only by event size but also by the catchment connectivity to the outlet.
SOC enrichment in delivered sediments is mainly controlled by the physical
properties (e.g. soil texture) of the parent soil (Foster et al., 1985).
Soil aggregation transforms unconsolidated fine primary particles, a highly
mobile SOC fraction, into soil aggregates, a fraction in which SOC is far
less mobile. Hu and Kuhn (2016) showed that soil aggregation reduces the
transport distance and potentially enhances terrestrial SOC deposition up to
64 %. We found a similar trend: upon increasing the aggregation level of
the model from non-aggregated (Agg
Under the same precipitation and field conditions, the simulated erosion-induced C balance of catchment C1 and C2 show opposing results (Fig. 5: Bal1–Bal2). While catchment C1 acts as a C source for the majority of simulated processes (controlled primarily by SOC delivery), the presence of the grassed waterway for catchment C2 substantially reduces lateral SOC delivery and leads the catchment to function as a C sink for most simulated processes. For both catchments, the majority of simulation years show a positive erosion-induced C balance (sink). However, three heavy erosion events in catchment C1 exceeded the positive cumulative vertical flux. Therefore, we underline that any analysis of landscape-scale erosion-induced C balances must consider inter-field connectivity.
In this study, the effect of individual SOC redistribution processes on SOC dynamics is assessed by utilizing a coupled process-oriented erosion and C turnover model. The erosion component of the model was successfully validated against a continuous 8-year data set of surface runoff and sediment delivery. The model was able to estimate the relevance of different processes in terms of their impact on vertical and lateral C fluxes for two catchments with distinct characteristics over an artificial time series of 50 years. We found that tillage erosion dominates on-field soil redistribution and vertical erosion-induced C fluxes on arable land, while water erosion processes have a much more limited effect. However, episodic lateral SOC delivery is critically important for the carbon balance. Ignoring SOC enrichment in delivered sediments leads to a pronounced underestimation of delivered SOC. Soil aggregates substantially reduce SOC delivery by turning highly mobile fine primary particles into less mobile soil aggregates. In general, the erosion-induced C balance is largely affected by inter-field deposition related to catchment connectivity.
Our results underline the importance of having an accurate and spatially distributed representation of tillage erosion. The episodic nature of water erosion calls for a sufficiently long simulation period and the inclusion of grain-size-selective transport in order to address the enrichment of delivered SOC. Furthermore, we stress the need for future investigations on seasonal and spatial variations in soil aggregation for a conceptual model implementation.
The MCST-C model is still under development and utilized in current studies. If someone is interested in cooperation, the authors would be happy to share the recent model version and the data used in this study.
The authors declare that they have no conflict of interest.
The study was supported by the Terrestrial Environmental Observatory TERENO-Northeast of the Helmholtz Association. We would like to acknowledge the large number of scientists and technicians who collected the data used in this study, which was funded by the Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (BMBF No. 0339370) and the Bayerische Staatsministerium für Unterricht und Kultus, Wissenschaft und Kunst. Edited by: N. J. Kuhn Reviewed by: two anonymous referees