Journal of Geophysical Research: Earth Surface

Soil organic carbon mobilization by interrill erosion: Insights from size fractions

Authors


Abstract

[1] Sediments mobilized by interrill erosion are often highly enriched in soil organic carbon (SOC) in comparison to source soils. This selectivity may lead to the preferential mobilization of SOC with specific properties, e.g., SOC that is especially susceptible to decomposition. This may then have important implications with respect to the role of soil erosion in the global carbon cycle. We addressed this issue by investigating the behavior of different SOC components in field rainfall simulation experiments on arable fields in loess-derived soils. We characterized the mobilization of mineral-bound organic carbon (MOC) and particulate organic carbon (POC) by interrill erosion using size fractionation and we used the C:N ratio as a tracer variable to determine the composition of the SOC in eroded sediments. MOC was found to be preferentially mobilized by interrill erosion in comparison to POC. The enrichment ratio (i.e., the ratio of the concentration of a soil constituent in the eroded sediment to its concentration in the original soil) of MOC decreased with increasing sediment concentration. The enrichment ratio of POC displayed a similar pattern to that of MOC but enrichment was less pronounced. Furthermore, sediments were found to be enriched in fine POC while they were impoverished with respect to coarse POC. The selective MOC mobilization together with the dominance of MOC in the total SOC pool in the soil explained the dominance of MOC in interrill eroded sediment. The fact that it is mainly MOC that is mobilized by interrill erosion implies that the SOC in the interrill eroded sediments is on average at least as recalcitrant than that in the source soils which may have important implications for the fate of the mobilized SOC. In order to understand the role of soil erosion in C cycling, MOC and POC need to be considered separately not only because they are chemically different but also because of their different behaviors with respect to geomorphic processes.

1 Introduction

[2] Interrill erosion is an important component of the water erosion process on hillslopes. It occurs when soil particles are detached by raindrop impact while subsequent transport of these particles occurs in overland flow. Kinnell [2005] distinguished two sediment transport modes, i.e., raindrop-induced flow transport and overland flow transport. The first mode occurs when flow competence is very low; after detachment or reentrainment from the flow bed or protruding roughness elements by raindrop impact, grains (or aggregates) fall back to the flow bed. Therefore, the distance that particles can travel between the moment when detachment occurs and the moment the particle is (re-)deposited will, for a given water depth, be proportional to the water velocity and inversely proportional to the particle's fall velocity. The second mode occurs when the flow competence exceeds the threshold for a given grain size flow transport, where sediments are transported by the flow without any additional energy input by raindrops. Evidently combinations of raindrop-induced flow transport and flow transport may also occur. The combination of both transport patterns explains why interrill erosion can be highly selective at low flow competences, (i.e., the sediment removed from a surface by interrill erosion is often significantly finer than the parent material) and why size selectivity decreases with increasing flow energy [Gabriels and Moldenhauer, 1978; Alberts et al., 1980; Heilig et al., 2001]. When raindrop-induced flow transport dominates, the equilibrium sediment concentration in the flow for a given size fraction will depend on the ratio between the sediment detachment rate by drop impact and the redeposition rate. Assuming no limiting water depth effects, the latter may, as a first approximation, be assumed to be proportional to the particle's fall velocity [Hairsine and Rose, 1992]. Assuming that detachment of grains or aggregates of a given size class is proportional to the relative frequency of the size class within the soil, the equilibrium sediment concentration and the resulting transport rates will then be much larger for the finer-size classes. With increasing flow competence, flow transport of the coarser grain size fractions will become more important, thereby decreasing size selectivity. When the flow transporting capacity is sufficient to transport all grain size at the rates at which they are supplied by drop impact, size selectivity will no longer be observed [Schiettecatte et al., 2008]. The above is a significant simplification of how interrill detachment and sediment transport interact to produce size selectivity as other factors such as water depth, slope length and soil roughness are also important in controlling the size selectivity of the interrill erosion process [Kinnell, 2009; Heng et al., 2011]. Nevertheless, these key mechanisms explain why interrill erosion is size selective and why size selectivity decreases with increasing flow competence.

[3] The preferential mobilization of fine-grained materials by interrill erosion may have important implications for the mobilization and transport of organic carbon by soil erosion. The redistribution of soil organic carbon (SOC) by soil erosion has provoked particular interest, not only because of its importance for on-site soil degradation, but also due to its potentially significant impact on the global C cycle [Stallard, 1998; Lal, 2003]. SOC is primarily associated with the fine soil fraction [Six et al., 2002] and it may therefore be expected that interrill erosion does not only preferentially remove the fine soil fraction but also SOC. Enrichment of eroded sediments in SOC has been observed in several experimental studies of interrill erosion [Jacinthe et al., 2002; Polyakov and Lal, 2004; Kuhn, 2007; Schiettecatte et al., 2008; Jin et al., 2009; Berhe, 2012] as well as during field monitoring at the plot and catchment scales [Wang et al., 2010; Nadeu et al., 2011; Pelletier, 2012].

[4] Wang et al. [2010] noticed that, in comparison to the source soil, sediments eroded from interrill surfaces were, in general, much more strongly enriched in SOC than in clay and hypothesized that this enhanced selectivity may be due to the fact that particulate organic carbon (POC) is preferentially removed due to its very low density in comparison to mineral soil. This would be in line with the findings of Jacinthe et al. [2004] who found that sediment eroded from small watersheds (0.55–17 ha) was strongly enriched in light and labile SOC. However, the data available in the study by Wang et al. [2010] did not allow us to confirm or to reject this hypothesis.

[5] A better understanding of the composition of interrill eroded SOC is important not only because it may help to explain selective SOC erosion but it may also determine the fate of the eroded carbon. If interrill erosion preferentially mobilizes labile SOC, as suggested by Jacinthe et al. [2004] and Wang et al. [2010], this SOC may be rapidly mineralized in the runoff or river water, resulting in an extra release of carbon to the atmosphere. If, on the other hand, the eroded carbon is mineral-bound and less prone to mineralization, it may be largely preserved during transport and become stored at a sediment deposition locus, thereby indirectly contributing to carbon sequestration [Stallard, 1998; Van Oost et al., 2007]. There are indications that the quality of eroded SOC (i.e., its composition in terms of organic substrates) may be different from that of the parent soil and may also vary with the scale as well as the intensity of the erosion event considered [Jacinthe et al., 2004; Chaplot et al., 2005; Nadeu et al., 2011].

[6] While existing studies suggest that carbon characteristics may differ between parent soils and interrill eroded sediments, they do not allow us to assess to what extent variations in eroded SOC composition may control the selective nature of interrill SOC erosion, nor are they sufficient to assess the resistance of the eroded SOC against mineralization as it is not known whether the characteristics of the eroded SOC change with a varying degree of enrichment or selectivity. The main objective of this study is therefore to test the hypothesis that the high enrichment of SOC in interrill eroded sediments is explained by the preferential mobilization of light POC.

[7] We carried out field rainfall simulations designed to mimic interrill erosion under natural high-intensity rainfall on arable land as realistically as possible. We measured soil losses as well as the grain size distribution of the interrill eroded sediment and we assessed the quantity and the quality of the organic carbon mobilized by erosion to investigate carbon enrichment in the eroded sediments and how the quality of the eroded carbon was related to that of the parent soil. Furthermore, we investigated whether or not variations in eroded SOC quality could explain the selective erosion of SOC by interrill erosion as suggested by Wang et al. [2010]. Finally, we discussed the implications of our experimental findings for carbon dynamics at larger scales.

2 Materials and Methods

2.1 Study Area

[8] Our study was performed in the Loam Belt area in Central Belgium (Figure 1). The area has a temperate climate with a mean annual temperature of approximately 9.5 °C and an average annual precipitation of 750–800 mm, < 5% of which is, on average, snow (Data are available at http://www.meteo.be/meteo/view/nl/6042865-Klimaat+in+de+wereld.html). Soils in this area are mainly loess-derived Luvisols with a high silt content (> 70%) and relatively low clay (< 15%) and sand (< 20%) content [e.g., Beuselinck et al., 2000]. The land is mainly used for growing maize, sugar beets, cereals and potatoes. Due to the rolling topography of the area, with slopes up to 25%, soil erosion is a significant problem. On sloping fields, average yearly soil erosion rates by water may well exceed 10 t ha-1 yr-1 [Govers, 1991b]. Soil erosion mainly occurs during spring and early summer on summer crop fields (maize, sugar beets, potatoes) which have a low vegetation cover during this period when high-intensity rainfall events (>30 mm h-1) lasting 30 min or more occur relatively frequently [Takken et al., 1999]. Recently, conservation tillage (which is a general term describing tillage techniques reducing soil disturbance and maintaining a good soil cover) has been introduced in the area as one possible strategy to reduce erosion rates [Leys et al., 2007].

Figure 1.

Map of locations (filled circles) where field rainfall simulations were performed.

2.2 Field Experiments

[9] Field rainfall simulations were conducted with a 3 m high nozzle rainfall simulator (Lechler 460 788 nozzle, Figure 2) on a wide range of arable land surfaces, covering different crops as well as two different tillage techniques (conventional vs. conservation tillage). Rainfall was applied to small runoff plots of approximately 0.75 by 0.75 m on slopes of 5–25 %. The runoff plots were constructed by vertically inserting 0.1 m high aluminum plates approximately 0.05 m into the soil surface to ensure that no loss of overland flow could occur. At the downslope end, a gutter was installed with a sufficient slope to prevent significant sediment deposition within the gutter. The gutter drained into a 0.5 m pipe at the end of which a plastic container was placed wherein runoff was collected: by measuring the water height over time, runoff rates could be determined. For most experiments, all runoff water (and the sediment within it) was ultimately transferred to a large plastic container. Alternatively, this set up also allowed collecting several smaller runoff samples throughout the duration of the experiment (Figure 2). The dimensions of the plots used result in an average travel distance of sediment and water of approximately 0.4 m. Given that rill spacing on these fields is often approximately 1 m [Govers, 1991a], this realistically reflects interrill travel distances on fields where rill erosion does occur: once interrill sediment and water reach a rill, they will be efficiently transported to the foot of the hillslope where a large fraction of the rill and interrill eroded sediment may be deposited [Steegen et al., 2001]. If no rills are present, interrill travel distances may be much longer, which may result in lower average interrill erosion rates in combination with a more pronounced size selectivity [Parsons et al., 2006; Kinnell, 2009].

Figure 2.

Pictures of (a) the nozzle rainfall simulator, and (b) a runoff plot installed in the field.

[10] Rainfall simulations were carried out with a design rainfall intensity of 45 mm h-1 which is similar to the intensity of the rainstorms causing most of the erosion observed in the area. However, significant variation occurred due to variations in weather conditions so that real rainfall intensity averaged 54 ± 14 mm h-1. At the design intensity of 45 mm h-1, the simulated rainfall had a median drop size of approximately 1.5 mm and a kinetic energy of 18–20 J mm-1 m-2, which is approximately 75–80 % of the kinetic energy of natural rainfall occurring at the same intensity (approximately 25 J mm-1 m-2 ) [Van Dijk et al., 2002]. It may be assumed that the rain drop size distribution and hence the rainfall kinetic energy were not strongly affected by the actual rainfall intensity as the same nozzle and water pressure were used throughout all experiments, although some variation will inevitably have occurred as smaller drops are more likely to be affected by wind than larger ones. Experiments were continued until a steady-state runoff rate was reached and the durations of the experiments were between 0.5 and 1 h. All experiments were carried out with deionized water.

[11] At each field site that was tested, we conducted 2–3 rainfall simulation experiments on different plots that were under the same tillage treatment and having a similar slope and microtopography. Thus, in total 240 rainfall experiments were carried out in the spring between 2007 and 2010 (Table 1). After each rainfall simulation experiment, the plastic container containing the runoff-sediment mixture was thoroughly stirred, after which a well-mixed subsample of approximately 1 L was taken. Repeated subsampling after selected experiments showed that variations in sediment concentrations calculated from subsamples were < 10%. These samples will be referred to as bulk samples. For six selected experiments in 2010, a time series of runoff-sediment samples was collected in order to investigate the temporal evolution of the characteristics of the mobilized SOC during the rainfall simulation: these samples were taken by collecting approximately 100 ml of runoff and sediment below the outlet pipe of the plot at 5–10 min intervals. These samples were used for SOC fractionation and will be referred to as time series samples: in total, 66 of such samples were taken. Before each of the rainfall simulations, a sample of topsoil was taken near to each plot to a depth of 0.05 m (experiments where bulk samples were taken) or 0.01 m (experiments where time series samples were taken) in order to determine topsoil properties (SOC content and grain size distribution). The average topsoil SOC contents measured over 0.05 m and 0.01 m can be assumed to be similar, given that all the fields were at least minimally tilled, leading to the thorough mixing and a homogeneous distribution of SOC within the tilled soil layer [e.g., Oorts et al., 2007]. Throughout the paper we refer to the samples taken from the parent soil as topsoil samples and samples taken from the runoff-sediment mixtures as sediment samples.

Table 1. Number of Rainfall Simulations Conducted in Fields With Various Crop Types and Tillage Techniques
YearConventional TillageConservation TillageTotal
WheatSugar BeetsMaizePotatoWheatSugar BeetsMaizePotato
2007963 1863 45
20089985181882297
200912 9 17 12 50
2010639 9129 48

2.3 Laboratory Analysis

[12] The sediment concentration of the bulk samples was determined by weighing the samples before and after oven drying at 105°C for 24 h. The time series samples, which were far smaller in volume, were first filtered at 2.5 μm (ashless filter paper Machery-Nagel MN640d) after which the residue was oven dried at 60°C.

[13] Before grain size analysis was carried out, a small amount of deionized water was added to the dried samples. The mixture was then boiled for at least 15 min which allowed us to obtain full dispersion [Beuselinck et al., 1998]. The sample was then introduced into a Coulter LS 13 320 laser diffraction apparatus (Beckman Coulter, USA) to measure grain size distributions where sample suspension was maintained by ultrasonic stirring. During the measurement, a narrow beam of laser light is passed through a sample cell containing an upward moving suspension. The diffracted light is focused onto 116 detectors, which allows for a very high resolution measurement of grain size (116 measurement channels between 0.04 and 2000 μm). The methodology we used thus assesses the primary grain size distribution of the sample: this was our objective, as we were investigating the relationship between carbon enrichment and the primary particle size distribution of the sediment as the latter is the primary control on SOC storage within the soil [Saggar et al., 1996; Torn et al., 1997]. However, it should be kept in mind that this size distribution is likely to be different from the size distribution of the eroded sediment which is not completely dispersed under natural condition [e.g., Beuselinck et al., 2000].

[14] For the samples collected in 2007 and 2008, the SOC content was determined with the Walkley and Black method [Walkley and Black, 1934], while for those taken in 2009 and 2010 SOC and nitrogen contents were measured with a varioMAX CN Macro Elemental Analyzer (Elementar Analysensysteme GmbH, Germany).

[15] For the time series sediment samples and the corresponding topsoil samples, the SOC was separated into different fractions using the method proposed by Cheng et al. [2010]. Approximately 3 ml solution of sodium hexametaphosphate (3.3%) and sodium carbonate (0.7%) were added to 1 g of dry soil or sediment sample. After shaking for 18 h, the suspension was passed through a 53 μm sieve. The material left on the sieve was recovered by careful back washing. The organic carbon found in the > 53 μm fraction was considered as POC not bound to mineral surfaces while the fraction passing the 53 μm sieve was considered as mineral-bound organic carbon (MOC). The amount of MOC in the fraction > 53 μm was very small as (1) most soils contained only a small mineral fraction > 53 μm and (2) the specific surface area (SSA) of this fraction (and hence the possibility for SOC to be bound to it) was negligible in comparison to the clay fraction to which most MOC is associated. We also assumed that all the carbon present in the fraction < 53 μm is MOC. While this is most probably not entirely correct, various studies (see Six et al. [2002] for an overview) have shown that the SOC associated with the silt and clay fraction is chemically well protected against mineralization due to its incorporation in organomineral complexes. For 10 selected sediment and corresponding parent topsoil samples, the POC fraction was further subdivided into two fractions: 53–710 μm and > 710 μm, making thereby a distinction between fine and coarse POC. After oven drying at 60°C, the separated fractions were manually ground and C and nitrogen contents were measured with an elemental analyzer (GSL-ANCA2020, Sercon Ltd., UK) after removal of inorganic carbon using the HCl-fumigation method proposed by Harris et al. [2001]. The mass lost during fraction separation was always < 3% of the total sample mass: this loss was proportionally distributed over all size classes.

[16] Incomplete oxidation of SOC in the Walkley and Black method tends to lead to an underestimation of C content. Therefore we used a correction factor of 1.33 as proposed by Sleutel et al. [2007] to make the Walkley-Black data comparable to those obtained using the Variomax analyzer. While it cannot be excluded that some residual bias may still be present, this does not have a significant effect on our analysis as we always compared the SOC content of the eroded sediment with that of the corresponding parent topsoil and the SOC content of both was always measured using the same methodology. We also checked whether the differences in the drying temperature between the bulk samples and the time series samples had an effect on the measured SOC content and found that there was no effect (Figure S1S3).

[17] The drying and filtration of soil and sediment samples may have induced some loss of SOC as dissolved organic carbon. It should be kept in mind though that the dissolved organic carbon pool in the soils we studied is several orders of magnitude smaller than the SOC pool [Mertens et al., 2007].

2.4 Data Analysis

[18] MOC is generally assumed to be stored in so-called organomineral complexes [Six et al., 2002]. The capacity of a particle to bind SOC may be assumed to be proportional to its SSA. Assuming the shape of the particles to be spherical, a first approximation of the SSA of each sample was calculated by summing the surface area of each grain size fraction as follows:

display math(1)

where Di is the average grain diameter for measurement channel i, Ni is the number of grains of diameter Di needed to fill a unit volume V0, and Pi is the volumetric fraction of grains of diameter Di present in the sample.

[19] Ni can be calculated as,

display math(2)

where Vi is the volume occupied by a single grain of diameter Di.

[20] Thus, SSA can also be calculated as

display math(3)

[21] We also calculated enrichment ratios (ER) as the ratio between the amount of a given fraction present in the eroded sediment to that present in the topsoil. Throughout the paper a distinction between different enrichment ratios is made by using appropriate subscripts (G = grain size, SSA = specific surface area, SOC = soil organic carbon, MOC = mineral-bound organic carbon, POC = particulate organic carbon).

[22] An enrichment ratio was calculated to assess average enrichment over a set of sample pairs whereby each pair consists of a sample of the source soil and a sample of the sediment mobilized by erosion originating from this soil as follows:

display math(4)

where SediPij and SoilPij are the volumetric fractions of particles in a given grain size range i (e.g., 0–2 μm) for the paired sediment and topsoil Sample j, respectively, and n is the number of the measured paired samples.

[23] Various statistical tests have been performed to analyze the obtained data. Most simple tests were done in Excel while more advanced statistical analysis were done in SAS [SAS Institute, 2002].

3 Results

3.1 Runoff and Erosion Rates

[24] A full analysis of the variation of runoff and erosion rates and controlling factors is beyond the scope of this paper and is presented elsewhere [e.g., Leys et al., 2007; Van den Putte et al., 2013]. Average soil loss rates varied between 0 and 5.8 kg m-2 h-1, with an average of 0.42 kg m-2 h-1 (Table S1). For only 15 % of the experiments, soil loss rates exceeded 1 kg m-2 h-1. Corresponding carbon loss rates varied between 0 and 54 g C m-2 h-1 with an average of 6 g C m-2 h-1. Runoff coefficients varied between 0 and nearly 100%, the latter value indicating that almost no infiltration occurred due to the high degree of crusting of the soil.

[25] Soil loss rates and sediment concentrations were significantly related to rainfall intensity (R = 0.35, n = 234, p < 0.0001, Pearson correlation) and even more strongly so to runoff rates (R = 0.63, n = 234, p < 0.0001, Pearson correlation). No direct statistical association with other independent variables was found. When relevant independent variables were combined through (stepwise) multiple regression analysis, the following model could be established for describing the variation in soil loss rates (R = 0.80, n = 136, p < 0.001):

display math(5)

where SLR is the interrill soil loss rate (kg m-2 h-1), RR is the runoff rate (mm h-1, p < 0.0001), I is the rainfall intensity (mm h-1, p < 0.02), Clay is the clay content of the soil (%, p < 0.0001), and Cov is the vegetation and residue cover on the soil's surface (%, p < 0.05).

[26] Even when accounting for confounding effects, neither slope gradient nor SOC content were identified as significant controls on soil loss rates.

3.2 Particle Selectivity

[27] The clay enrichment ratio was not significantly related to rainfall intensity or runoff rates, while it was significantly related to the soil loss rate (R = -0.25, n = 211, p < 0.0003, Pearson correlation), and even more strongly coupled to sediment concentration (R = -0.33, n = 211, p < 0.0001, Pearson correlation). Figure 3 shows the variation of particle selectivity with sediment concentration for different size groups. Generally, ERG of fine particles (<20 μm) decreases with increasing sediment concentration while that of coarse particles (>20 μm) increases with increasing sediment concentration. However, the extent of this enrichment does not vary monotonically with particle size but shows fluctuations with a maximum at approximately 0.2 μm for low sediment concentrations and approximately 6–8 μm for all sediment concentrations (Figure 4).

Figure 3.

Examples of enrichment ratio (ERG) variation with sediment concentration (SC) for grains of different size classes ((a) 0.2–0.7 μm; (b) 2.7–10.8 μm; and (c) 43.7–176.9 μm). Since the Coulter grain size analyzer has 116 measurement channels between 0.04 and 2000 μm, here we give examples of three size classes. The regression lines were derived by combining the samples of both conventional and conservation tillage techniques given the fact that there was no significant difference in particle selectivity between different tillage techniques.

Figure 4.

Variation of the average grain enrichment ratio of interrill eroded sediments with particle diameter grouped by sediment concentration (SC) class.

3.3 SOC Selectivity

[28] As is the case for fine grains, the ERSOC of the bulk samples decreased exponentially with increasing sediment concentration (Figure 5). Carbon enrichment was very significant: when interrill erosion intensity was low, ERSOC could be as high as 8 and the average value for experiments with an average sediment concentration < 10 g l-1 was 2.8. These values were clearly higher than the enrichment ratio of SSA (p < 0.001, Mann-Whitney U-test, Figure 5).

Figure 5.

Relationships between sediment concentration (SC) and enrichment ratio of SOC (ERSOC, full line) and SSA (ERSSA, dotted line) for the bulk sediment samples.

[29] For the time series samples, results were very similar to bulk samples when total SOC is considered (Figure 6a). However, there were clear differences in the behavior between POC and MOC. Both ERMOC and ERPOC decreased nonlinearly with increasing sediment concentration, but enrichment was significantly higher for MOC than for POC (p < 0.001, Mann-Whitney U-test, Figures 6b and 6c). Furthermore, the relationship between ERMOC and sediment concentration was much stronger (R = 0.70, p < 0.0001) than that between ERPOC and sediment concentration (R = 0.30, p = 0.024). Neither ERMOC nor ERPOC did show a clear trend with time (Figure 7, Table 2, and S2).

Figure 6.

Relationships between sediment concentration (SC) and enrichment ratio of (a) SOC (ERSOC), (b) POC (ERPOC), and (c) MOC (ERMOC) in the time series sediment samples.

Figure 7.

An example of the temporal evolution of sediment concentration (SC) and enrichment ratio of MOC (ERMOC) and POC (ERPOC) during a rainfall simulation experiment.

Table 2. Statistical Results of the Linear Regression of Sediment Concentration, Enrichment Ratio of POC (ERPOC) and MOC (ERMOC) With Time in the Rainfall Simulation Experimenta
ExperimentSlopeRp
  1. aSlope indicates the slope of the linear regression, R indicates the multiple correlation coefficient, and p indicates the probability of being wrong that the slope coefficient is not zero.
SC
1Positive0.340.29
2Negative0.610.15
3Negative0.820.023
4Negative0.0400.94
5Positive0.150.69
6Positive0.160.66
ERPOC
1Positive0.930.0002
2Positive0.730.16
3Positive0.650.11
4Negative0.050.92
5Positive0.980.09
6Negative0.0460.91
ERMOC
1Negative0.780.013
2Positive0.110.81
3Positive0.0580.90
4Negative0.130.81
5Positive0.480.19
6Negative0.780.0082

3.4 C:N Ratios and C Fractions

[30] The average C:N ratio of the particulate organic matter in the sediments was 18.83 ± 3.17, whereas for the mineral-bound organic matter it averaged 9.30 ± 0.40 (Figure 8). For the topsoil samples, the particulate organic matter C:N ratio averaged 20.72 with a standard deviation of 5.45, whereas it was 9.28 ± 0.64 for the mineral-bound organic matter. When the particulate organic matter of topsoil samples was further separated, the C:N ratio of the fine particulate organic matter (53–710 μm) was 17.53 ± 2.07, whereas for coarse particulate organic matter (> 710 μm) it was 31.08 ± 7.20. A Mann-Whitney U-test showed that the C:N ratio of the particulate organic matter was significantly higher than that in mineral-bound organic matter (p < 0.001) both for topsoil and sediment samples, which confirms the validity of the size separation. Also, the C:N ratio of coarse particulate organic matter was significantly higher than that of fine particulate organic matter (p < 0.001). On the other hand there was no statistically significant difference in C:N ratio when particulate organic matter or mineral-bound organic matter fractions in topsoils and sediments were compared (p = 0.35 for particulate organic matter and p = 0.89 for mineral-bound organic matter).

Figure 8.

C:N ratios in mineral-bound organic matter and particulate organic matter fractions of eroded sediments and source soils and in fine (53–710 μm) and coarse (>710 μm) particulate organic matter fractions of source soils. The error bar denotes the standard deviation of different samples.

[31] The POC content in the topsoils has an average of 0.35% ± 0.19%, while the POC content in the sediments averaged 0.48% ± 0.27% (Figure 9). The MOC content in the topsoil samples averaged 0.65g% ± 0.07% whereas it was 1.83% ± 0.86% in the sediments. In the topsoils 33.0% ± 10.1% of the total C consisted of POC while in the sediments this value was 22.0 ± 7.8% (Figure 10). Thus, the eroded sediments contained, on average, relatively more MOC (as compared to POC) than the source topsoils (p < 0.001, t test). The difference between sediments and topsoils was even bigger when only POC was considered: in the topsoils, 33.7% ± 14.4% of the total POC was coarse POC while the presence of coarse POC in the sediments was negligible.

Figure 9.

Average POC and MOC content of interrill eroded sediments and source soils. The error bar denotes the standard deviation of different samples.

Figure 10.

Fractions of (a) MOC and POC as a percentage of total SOC and (b) fine (53–710 μm ) and coarse (>710 μm) POC as a percentage of total POC in interrill eroded sediments and source soils. The error bar denotes the standard deviation of different samples.

4 Discussion

[32] The soil losses we observed under simulated rainfall were significant, with losses exceeding 1 kg m-2 h-1 for approximately 15 % of the experiments. However, it should be kept in mind that applied rainfall intensities were relatively high (although not unrealistic): a rainfall event of 45 mm h-1 has a return period of approximately 5–10 years for a duration of 30 min and approximately 100 years for a duration of 1 h [Willems, 2000]. Chaplot et al. [2007] measured an interrill soil loss of 0.6–2.4 kg m-2 over a whole year on hilly arable land in Laos subject to tropical rainfall. Interrill erosion rates are known to be strongly dependent on rainfall intensity. Under low-intensity rainfall (< 10 mm h-1), Chaplot and Le Bissonnais [2000] observed interrill soil loss rates < 0.1 kg m-2 h-1 on 1 m2 plots in Normandy, France. They measured rates similar to those observed in this study under simulated rainfall at 50 mm h-1 (0.1–0.3 kg m-2 h-1). It is also important to consider that these relatively high rates are also explained by the fact that sediment delivery on the small plots used in this study is very efficient. The interrrill soil loss rates observed here will only be representative for larger fields or hillslopes if a dense network of rills or concentrated flowpaths is present so that interrill eroded sediment can be efficiently transported over greater lengths. However, under such conditions, rill erosion will often become rapidly more important than interrill erosion [Govers and Poesen, 1988]. If such a flow network is not present, erosion rates measured at the field scale may be much lower than the rates observed in our experiments [Chaplot and Poesen, 2012]. Extrapolation of observed rates over larger areas is therefore only possible by explicitly accounting for these factors [e.g., Cerdan et al., 2002].

[33] The statistical analysis shows that rainfall intensity and runoff rates are, as expected, major controls on interrill erosion rates. The effect of rainfall intensity can simply be related to the direct relationship between rainfall intensity and the amount of soil that is detached from the soil surface. Sediment concentration generally increased with increasing runoff rate, which demonstrates that flow competence is also important: in a system with only raindrop-induced flow transport, sediment concentrations would be expected to decrease with runoff rates, as was observed by Cerdan et al. [2002].

[34] The negative relationship between interrill soil loss rates and vegetation cover is also as expected, given the vast literature demonstrating the protective effect of soil cover by vegetation and residue against erosion at scales ranging from interrill plots [Hussein and Laflen, 1982] to catchments [Molina et al., 2008]. What may seem surprising is the absence of a clear slope effect on the interrill soil and carbon losses we observed. The dominant detachment mechanism in interrill erosion is raindrop impact and this is not strongly dependent on slope gradient for the range of slopes tested in this study [Torri and Poesen, 1992]. Thus, the amount of material available for transport is slope-independent. Nevertheless, one might expect soil loss rates to be positively correlated to slope gradients from the moment that flow transport is present, as flow competence and transporting capacity can be expected to be controlled by both discharge (runoff rate) and slope. The most likely explanation for the absence of such an effect is that it is confounded by other factors such as crop type, and soil properties and soil roughness. Such covariation is indeed present: there is a positive correlation between plot slope gradient and the sand content of the topsoil (R = 0.41, n = 234, p < 0.0001, Pearson correlation), a similar negative correlation with silt content (R = -0.41) and a somewhat weaker negative relationship with clay content (R = -0.23). Sandy soils may be expected to deliver relatively less sediment for a given rainfall intensity, runoff rate and soil cover, as sandy particles are more difficult to transport.

[35] We measured the grain size distribution of the soils and the eroded sediments after dispersion and they therefore were expected to be different from the size distribution of the particles as they were actually eroded in the field. However, general tendencies may be expected to be similar [Govers, 1985]. As reported in other studies [Jin et al., 2009; Warrington et al., 2009], we found that interrill eroded sediments were highly enriched in particles < 20 μm (Figure 3 and 4) indicating preferential removal of fine particles and that the size selectivity decreased with erosion intensity. We found that clay enrichment was more strongly negatively related to sediment concentration in comparison to runoff rate or soil loss rate. Low sediment concentrations are associated with a relatively high contribution of rainfall induced flow transport to the total sediment export. Theoretical analysis shows that, when only raindrop-induced flow transport is present and a plane land surface is considered, the sediment concentration for a given size fraction may be expected to be inversely related to the fall velocity of the particles. At the same time the effect of absolute soil loss rates on size selectivity will be limited [Hairsine and Rose, 1992; Kinnell, 2005]. Thus, a similar size distribution may result from different erosion rates. As flow competence rises, sediment concentrations increase and selective raindrop-induced flow transport becomes relatively less important so that overall size selectivity is reduced. It may therefore be expected that size selectivity is more closely related to sediment concentration than to absolute soil loss rates. We found that size selectivity did decrease with increasing sediment concentration but remained present over the whole range of sediment concentrations that was measured (Figure 3 and 4).

[36] One might expect that the grain enrichment ratio should decrease monotonically with increasing particle diameter, and that is indeed what is observed if the soil material is completely dispersed during the erosion and transport process (see experiments of Beuselinck et al. [1999]). However, this was not the case for our experiments: enrichment ratios were lower for (primary) particles < 2 μm as compared to those for fine silt particles (Figure 4). We hypothesize that the strong enrichment in fine silt particles may be due to the fact that these particles were relatively easily dislodged as individual particles by drop impact, while most of the clay fraction was locked up in slightly larger aggregates which are more difficult to transport. We do not have a clear explanation as to why the grain enrichment ratio shows two distinctive peaks (at particle diameter of 0.2 and 8 μm) for relatively low sediment concentrations.

[37] The C:N ratio of soil organic matter decreases when fresh organic matter is transformed into more stable residues [Natelhoffer and Fry, 1988]. Therefore, the C:N ratio can be used as an indicator of the degree of soil organic matter decomposition, provided that all soil organic matter is derived from the same plants. Our observation that the C:N ratio increases significantly from mineral-bound organic matter to coarse particulate organic matter is consistent with this general tendency as well as with findings of previous studies [Christensen, 2001; Baisden et al., 2002; von Lutzow et al., 2007] and suggests that there is a difference in the degree of decomposition and hence also in recalcitrance between these three fractions.

[38] Enrichment in SOC is, at low sediment concentrations, much higher than enrichment in fine grains. We calculated the SSA of the sediment as we hypothesized that the SOC might be primarily associated with the very fine sediment fraction that is preferentially exported at low sediment concentrations. However, the enrichment in SOC was much higher than the enrichment of SSA. Thus, the presence of large amounts of SOC in the sediments at low sediment concentrations cannot be simply explained by assuming that the preferential erosion of SOC is proportional to the preferential erosion of the very fine soil fraction. Alternatively, the high enrichment in SOC may be explained by the preferential mobilization of the POC fraction [Wang et al., 2010]: this hypothesis is primarily based on the fact that POC has a low density and might therefore be preferentially transported by interrill flow. However, we found that interrill eroded sediment was strongly enriched in MOC and that the ERMOC was strongly negatively related to sediment concentration (Figure 6). There was also enrichment of POC at low sediment concentrations, but the POC enrichment ratio is, on average, significantly lower than the MOC enrichment ratio for a given sediment concentration. As a result, interrill eroded sediments contain, on average, more MOC in comparison to the source topsoil (p < 0.001, Mann-Whitney U-test), while their average POC content is only slightly higher than that of the source topsoils (p = 0.163, Mann-Whitney U-test, Figure 9). This lower enrichment in POC in comparison to MOC is mainly due to the fact that interrill erosion does not mobilize the coarse POC fraction: while coarse POC constitutes approximately 35% of the total POC in the soils, its presence in interrill eroded sediments is almost negligible (Figure 10).

[39] The preferential removal of MOC as compared to POC, in combination with the fact that the MOC reservoir in the soil is much larger than the POC reservoir explains why MOC is largely dominant in interrill eroded sediments (Figure 10). Due to the dependency of enrichment on erosion intensity, this dominance will be more important during low-intensity erosion events as compared to high-intensity events. Our findings can be summarized in a conceptual model as illustrated in Figure 11. Enrichment of various SOC fractions follow different patterns with increasing sediment concentration (erosion intensity) and therefore the composition of bulk eroded sediments varies with increasing sediment concentration. The conceptual model shows that the contribution of coarse POC to carbon export is expected to become important only for very high-intensity erosion events.

Figure 11.

Conceptual model describing the variation of (a) enrichment ratios and (b) C pools in interrill eroded sediments with sediment concentration based on observations of this study.

[40] Our findings invalidate the hypothesis put forward by Wang et al. [2010], i.e., that the highly selective mobilization of SOC by interrill erosion would be explained by the high erodibility of POC. Our data show that, on the contrary, the high erodibility of MOC is the most important explanation for the observed selectivity. While our observations do allow rejection our earlier hypothesis, they do not allow pinpointing the precise mechanism for this preferential mobilization: the enrichment in SOC of interrill eroded sediment appears to be much more important than the enrichment in very fine mineral soil particles. This is surprising, given the fact that several studies have shown that there is a strong association between MOC and clay particles in so-called organo-mineral complexes [Six et al., 2002]. A possible mechanism that may explain this observation is that, on interrill surfaces, drop impact leads to aggregate breakdown and crust formation [Govers, 1991b; Kuhn and Bryan, 2004]. This process may lead to the production of some very fine particles containing a disproportionate amount of SOC, which may, due to the lower density of SOC, be preferentially removed by the flow. Experiments whereby the size distribution of the eroded sediments and the distribution of SOC within these classes are directly measured and compared to the primary particle size distribution and the presence of SOC within these primary classes may help to further elucidate these processes.

[41] The preferential removal of the fine mineral fraction (and associated MOC) and fine POC should ultimately lead to the depletion of these fractions at the soil surface. We did not find any evidence that such depletion would occur within a single rainfall event as enrichment ratios for POC and MOC did not show a systematic variation over time (see Figure 7 for an example). This is not unexpected: it has long been known that raindrop impact dislodges much more material than is transported off by interrill flow, thereby supplying ample fine material to the flow. Over a longer timescale, the material at the soil surface is being continuously renewed by tillage operations.

[42] Our findings have implications with respect to the long-term fate of eroded SOC. As interrill erosion mainly mobilizes MOC, it may be expected that the SOC mobilized by interrill erosion is, overall, quite recalcitrant with respect to mineralization. The mobilized SOC may therefore be transported over considerable distances and most of it may become buried in a depositional environment, thereby contributing to the stabilization of soil C in a low-mineralization context [Stallard, 1998]. This result may appear to be contradictory to earlier findings suggesting that eroded C is more prone to mineralization. Some of these studies relied on the incubation of soil samples which were collected with a depositional layer on top after rainfall events [Polyakov and Lal, 2004; Mora et al., 2007]. These studies showed that the fraction of mineralizable C in eroded sediments was higher than in the source soils and that a larger fraction of this mineralizable C was decomposed soon (< 30 days) after the erosion event. However, the fraction of mineralizable C in these studies was always < 5 % of the total amount of C present in the deposited soil, suggesting that the largest part of the SOC was relatively stable and not affected by erosion and deposition. Van Hemelryck et al. [2010] and Van Hemelryck et al. [2011] measured respiration rates both in field and laboratory conditions and obtained similar results, i.e., only a small fraction (< 10%) of the SOC was mineralized after erosion and deposition. Van Hemelryck et al. [2011] also showed that differences between sites under field conditions were clearly lower than differences observed during incubation experiments. The stability of eroded SOC is not an universal finding, however, Juarez et al. [2011] studied the stability of eroded SOC in a grassland area in South Africa and obtained different results: they observed that eroded SOC was much more sensitive to mineralization than in situ SOC. One of the reasons explaining this was the fact that the eroded SOC consisted dominantly of very labile C, related to fresh litter. Jacinthe et al. [2002] studied the mineralization of C in sediment samples collected from laboratory rainfall simulations and found that a large fraction of the eroded C (29–35%) was mineralized when the samples were incubated over a period of 100 days. Jacinthe et al. [2004] found similarly high fractions (13–40%) of mineralizable C to be present in runoff samples, whereby most variation of the fraction of mineralizable C was explained by the rainfall intensity and not by the soil management type. They attributed the high mineralization rates of eroded C to the preferential removal of labile, rapidly mineralizable C by erosion, with enrichment ratios for this fraction > 20. This suggests a very important contribution of a labile C pool, derived from residue and litter to the C present in the runoff, similar to what was reported by Juarez et al. [2011]. It may then be questioned whether the high mineralization rates measured by these authors do really imply that erosion leads to significant additional C mineralization. Rather it seems that erosion may, under some circumstances, lead to the concentration of a highly labile C pool in the runoff water, where this C is subsequently decomposed. It appears reasonable to assume, however, that this material would also have been mineralized had it remained in place so that erosion merely caused a displacement of the mineralization locus within the landscape from the terrestrial to the aquatic environment. Clearly, the export of such a highly labile C to the river network may have important implications for the aquatic ecology [Battin et al., 2008].

5 Conclusions

[43] Interrill erosion has long been known to be a selective process and it is therefore not surprising that the process mobilizes a disproportionate amount of soil organic carbon (SOC). Our study shows that interrill erosion can be far more selective with respect to carbon than with respect to the mineral soil fraction: while the enrichment ratios for the fine mineral fractions are generally not higher than 3, they can go up to 8 for SOC. The degree of enrichment is, not unexpectedly, directly coupled to the erosion intensity: the sediment concentration in interrill flow is the best predictor of carbon enrichment.

[44] Our data show that this high SOC enrichment is not primarily due to the mobilization of particulate organic carbon (POC): the SOC fractions in mobilized sediments show that most of this carbon, contrary to what was expected, is mineral-bound organic carbon (MOC). Fine POC is also preferentially exported, but the coarse POC fraction is relatively immobile.

[45] The fact that interrill erosion preferentially mobilizes stable MOC has important implications with respect to the effect of soil erosion on carbon cycling. As the mobilized carbon is relatively resistant to mineralization, one may expect it to be relatively stable during transport and after deposition. Clearly, this SOC stability is not absolute and more research is needed to determine the fate of the eroded POC in aquatic and depositional environments, also over longer time scales. Also, SOC selectivity and mineralization show different dynamics in other agroecological conditions. The presence of an important labile SOC pool at the soil surface (litter) appears to be a dominant control on how interrill erosion affects SOC dynamics. Only when these processes and their interactions are fully understood can the impact of soil erosion on the carbon cycle be completely assessed.

Acknowledgments

[46] This research is funded by the China Scholarship Council and Katholieke Universiteitg Leuven. Their support is gratefully acknowledged. We are also grateful to M. Ayyad, L. Fondu, and S. Vandevelde for their help in rainfall simulation and laboratory analysis.

Ancillary