Modelling Soil Detachment of Different Management Practices in the Red Soil Region of China

Study Site

The experimental site is located at the Ecological Experimental Station of Red Soil, Chinese Academy of Sciences, Yujiang County, Jiangxi Province, China (28°15′N, 116°55′E). The average annual rainfall is 1795 mm, with approximately 50% of the total precipitation occurring from March to June. The mean annual temperature is 17·6 °C, and the elevation ranges from 35 to 60 m. The primary soils of the area include ultisols from Quaternary red clay and inceptisols from red sand shale (Shi et al., 2004). The study area is a typical agricultural region in which traditional downslope tillage practices have been adopted along sloping cropland. Severe soil loss has been reported from these lands, particularly under heavy rainfalls (rainfall intensity of about 80 mm h⁻¹ with the 50a return period).

Field plots were established in 2011 by planting peanuts with the five treatments described by Wang et al. (2016). These plots included peanuts that were fertilized with manure (PM) or covered by rice straw mulch (PC), peanut and orange intercropping (PO), peanut and radish rotation (PR) and traditional farrow peanut (PF). All plots were ploughed to a depth of 30 cm by a tractor in early April. Then, peanuts were planted at the density of approximately 125,000 plants ha⁻¹ with a row spacing of 40 cm and a plant spacing of 20 cm. Descriptions of the different treatments are shown in Table I. Each treatment was replicated in three plots for a total of 15 plots. The plots were isolated from one another by concrete banks, which extended from 80 cm below the surface to 20 cm above the surface. At sowing, all fertilizer was banded at 5 cm from the seed. The fertilizer rates used for the peanuts were 61 kg ha⁻¹ nitrogen (N), 90 kg ha⁻¹ phosphorus (P₂O₅) and 52 kg ha⁻¹ potassium (K₂O). More details on fertilizer rates of the plots can be obtained in Wang et al. (2016). According to the USDA soil texture classification, the soils under the different treatments were clay soils, and the characteristics of the soils are listed in Table II.

Soil Sampling

Experiments were conducted in July 2015 three months after the peanuts were planted, which was during the rainy season, March to July, and corresponded with the annual period of the highest risk of soil loss. Undisturbed soil samples were collected from the top layers of field plots with steel rings (10 cm in diameter and 5 cm in height) according to Cao et al. (2009). First, the ring was placed on a flat, bare patch between crops, avoiding roots of vegetation. Then, the iron ring was carefully pressed down into the soil until it was completely filled. The ring was removed, and the soil sample was carefully cut across the bottom end of the ring. A cushion of cotton cloth was placed onto the bottom of the soil ring, and the sample ends were tightly capped to avoid disturbance during sample transport. The surrounding soil was simultaneously sampled to measure soil moisture. The weight of the soil samples in the rings was measured as soon as possible to decrease the weight loss from evaporation and ensure the identical soil moisture as that in the field. The samples were wet for 8 h in a container with a water-level 1 cm below the topsoil surface and then drained for 12 h in preparation for experimentation (Wang et al., 2013; Li et al., 2015). In this study, 15 soil samples were collected from each plot for 225 total soil samples.

Measurement of Hydraulic Variables

Experiments were conducted with a 4·0-m-long and 0·30-m-wide hydraulic flume (Figure 1), which was similar to that described by Zhang et al. (2003) and Cao et al. (2009). The flume consisted of 20-mm-thick, smooth Plexiglas panels and was supported by a steel frame. Soil was collected from the cultivated land and air-dried and sieved to 2 mm, and then this soil was glued evenly onto the surface of the flume bed to simulate a similar natural hydraulic roughness to that of the test samples (Zhang et al., 2003). So as to vary the slope of the flume, the upper end of the flume was lifted manually by a pulley gear. Clean water was fed into the upper end of the flume by a pump, and the flow discharges were set at 1, 1·5, 2, 2·5 and 3 L s⁻¹ based on typical erosive storms in the study area (Wang et al., 2016). The flume bed gradient was fixed at 8·8%, 17·6% and 26·8% according to the local terrain. The flow velocity was measured using fluorescent dye in which the velocity of the leading edge of the dye was multiplied by a reduction factor of 0·8 to obtain the mean flow velocity (Luk & Merz, 1992). Flow velocity measurements were replicated 12 times for a given flow. Hydraulic variables such as the flow shear stress, stream power and unit stream power were calculated based on the flow rate and measured flow velocity as follow (Nearing et al., 1991; Hairsine & Rose, 1992; Morgan et al., 1998):

(1)

Where: τ (Pa) is the shear stress, ρ (kg m⁻³) is the water mass density, g (m s⁻²) is the gravity constant, h (m) is the flow depth and S is the tangent value of the slope gradient;

(2)

Where: ω (kg m⁻¹) is the stream power and V (m s⁻¹) is the mean flow velocity; and

(3)

Where: P (m s⁻¹) is the unit stream power.

Measurement of Detachment

Before each test, the flume was set at the desired slope gradient, and the designed flow discharge was applied to the flume from the upper edge. A pre-wet soil sample was inserted into the soil chamber, which was located at a distance of 0·3 m from the lower end of the flume, and the surface of the sample was adjusted to the identical level as that of the flume bed. During this stage, a panel was used to prevent the scouring of the soil sample by the upslope water. Once the set-up was completed, the panel was removed and the detachment experiment was initiated. The test lasted no more than 600 s for each sample. When the scouring depth reached 2 cm, the test was stopped to prevent boundary effects from sampling the ring. After each test, the soil sample was oven-dried at 105 °C for 24 h and weighed to determine the final oven-dry mass. The soil detachment capacity was calculated as the total mass of soil loss (original weight of wet soil sample minus the weight of water within the sample and minus the final oven-dry mass) divided by the time interval of the test and the cross-sectional area of the soil sample (Nearing et al., 1991). Three tests were conducted for each treatment by using soil samples from the three replicate plots. Therefore, 75 treatments were analysed (5 tillage systems × 5 flow discharges × 3 slope gradients) for a total of 225 tests.

Data Analyses

One-way analysis of variance (ANOVA) was used to compare the soil detachment rates of different treatments. Regression analyses were used to establish the relationships between detachment rates and hydraulic variables. The coefficient of determination (R²) and model efficiency (ME) were used to evaluate the proposed models. The model efficiency was calculated using the following equation (Nash & Sutcliffe, 1970):

(4)

Where: ME is the model efficiency, which equals to 1 for a perfect agreement between the measured and calculated values and is less than 1 for a weaker correlation; Q_i is the measured value; Q_c is the calculated value; and Q_m is the mean value of the measured values.

Soil Detachment Rates of Different Treatments

The soil detachment rates for the PO, PM, PC, PR and PF treatments are shown in Table III. The PR treatment showed the highest average detachment rate (0·137 kg s⁻¹ m⁻²), followed by the PF treatment, which had an average value of 0·061 kg s⁻¹ m⁻². The PC treatment showed the lowest average soil detachment rate among all treatments, which measured 0·016 kg s⁻¹ m⁻². The detachment rates of the PO and PM treatments were 0·048 and 0·034 kg s⁻¹ m⁻², respectively.

Soil detachment in rills is affected by both the flow hydraulics (Nearing et al., 1997; Zhang et al., 2003) and soil properties. Because the hydraulic conditions in this study were well controlled and were similar for each treatment, the disparities in the detachment rates for different management practices were caused by differences in the soil properties (Table II). As shown in Table II, the organic matter content, aggregate stability and bulk density of surface soil had statistically significant differences within treatments. Generally, the samples were more resistant to detachment in the soils with higher organic matter content. Therefore, the low detachment rates for PC and PM treatments might be explained by the relatively high organic matter contents because of mulch and manure decomposition. By contrast, organic matter contents were relatively low in both PR (17·17 g kg⁻¹) and PF (15·64 g kg⁻¹) treatments, which might explain the high rates of soil detachment. Additionally, the low mean-weight diameter (MWD) (1·14 and 0·73 mm for PR and PF, respectively) and the values for wet-sieved aggregates larger than 0·25 mm (WSA_0·25) (0·51 and 0·56 for PR and PF, respectively) indicated low soil structural stability in the two treatments. The findings of this study were supported by Li et al. (2015) who showed that soil detachment rate was negatively correlated with stable aggregate and aggregate median diameter. It also verified the high soil-loss potential for PR and PF treatments because of the intensive disturbances with the radish rotation and the conventional downslope furrows (Wang et al., 2016).

However, the characteristics of the aggregate stability may not be the only factor that determines the susceptibility of soil to runoff and erosion. Although the PO treatment showed the highest MWD (1·58 mm) and WSA_0·25 (0·65) values, the soil detachment rate remained relatively high. The soil bulk density of the PO treatment likely explained this result, which was the lowest among all treatments. A power relationship described the decrease in soil detachment rates in all the tests as soil bulk density increased, as shown in Figure 2. Therefore, the low soil detachment rate of the PC treatment, which had the highest soil bulk density (1·23 g cm⁻³) among all treatments, was likely explained by this relationship. Of note, although straw mulch treatments generally reduce soil bulk density (Keesstra et al., 2016), some researchers also claim that mulching effects on bulk density may vary due to soil type, antecedent soil properties, type of mulch, climate and land use (Mulumba & Lal, 2008). In our future studies, the bulk density in the PC treatment will be explored further with other properties such as aggregate stability.

Detachment Rate under Different Flow and Slopes

With a slope of 8·7%, the soil detachment rates were approximately 0·01 kg s⁻¹ m⁻², and no significant differences were detected among the five treatments (ANOVA, p = 0·8). With an increase in the slope gradient, the soil detachment rates increased, and differences among the treatments became more prominent. The soil detachment rates in the PR and PF treatments were significantly higher than those in the other treatments under slopes of 16·7% and 26·8% (ANOVA, p = 0·026 and 0·031, respectively). Especially for the PR treatment, soil detachment was much higher than that of the other treatments under a slope of 26·8% (ANOVA, p = 0·002). The soil detachment rates for the different treatments are plotted with flow discharge in Figure 3, and the detachment rates increased linearly with flow discharge for the identical slope gradient. The coefficients of linear functions increased with the slope gradient, and when the slope gradient was greater than 16·7%, the function coefficients increased sharply, particularly for the PR and PF treatments. This relation was consistent with previous results (Zhang et al., 2003; Cao et al., 2009) and demonstrated the importance of slope gradient in determining the soil detachment rate.

Additional multiple nonlinear regression analyses were conducted to predict the detachment rate based on simple hydraulic variables such as the flow discharge and slope gradient. The soil detachment rates in all treatments were described by power functions of the flow discharge and slope gradient (Table IV). For the determination coefficients (R²), the agreement between the predicted and observed values was high, particularly for the PO and PM treatments (R² = 0·943 and 0·965, respectively). The standardized regression coefficients of flow discharge and slope gradient indicated the ability of the two independent variables in explaining the variation in soil detachment rates (Stumpe et al., 2011). As shown in Table IV, when evaluating the soil detachment rates for all treatments, high standardized coefficients indicate that slope gradient contributed more than flow discharge to soil detachment.

Soil Erosion Resistance of Different Management Practices

Linear regression was conducted between the soil detachment rate and flow shear stress, with the results shown in Table V. The linear functions were essentially the excess shear stress models that are adopted in the Water Erosion Prediction Project (WEPP) model to describe rill soil detachment (Foster et al., 1995). The function coefficient is the rill erodibility (s m⁻¹), which is a measure of soil susceptibility to detachment by concentrated flow (Alberts et al., 1995).

As shown in Table V, the variation in the rill erodibility was high among the five management practice treatments. The lowest value of the rice straw mulch (PC) treatment (0·001 s m⁻¹) was consistent with that of Prosdocimi et al. (2016a), who reported that straw mulch was very effective in reducing soil erodibility. Soils from the radish crop rotation (PR) showed the highest rill erodibility (0·027 s m⁻¹), which was 27-fold higher than that of the PC treatment. The rill erodibility of the PF treatment was also higher than that of the other three treatments. As discussed earlier, the low organic matter content, MWD and WSA_0·25 might explain the high rill erodibility of PR and PF. The low soil bulk density of the PR treatment was also an important cause of the high rill erodibility, which was consistent with results of previous research in which bulk density is negatively correlated with soil erodibility during concentrated flow (Bennett et al., 2000; Sheridan et al., 2000; Li et al., 2015). The rill erodibility for the PF treatment, a conventional management practice, was lower than that reported by Zhang et al. (2008) for a silt loam loess soil (0·0164 s m⁻¹) cropland. The soils in this study contained more clay than loess soil, which possibly resulted in the higher susceptibility to detachment (Alberts et al., 1995). The soil texture was not the only determining factor; the PR treatment in this study showed higher rill erodibility than that in the loess cropland because of more tillage disturbances during the radish rotation. In this study, generally, the rill erodibility parameters were within the range of the WEPP rill erosion study (Laflen et al., 1991). Nevertheless, the values in PR and PF treatments were nearly double those in the clay or clay loam soils of the WEPP data, which implied that soils in this study were easily detached and that the soil erosion risk might be underestimated by using mode default values. However, the rill erodibility for the three treatments (PO, PM and PC) after the conservation measures were adopted was close to that reported by Laflen et al. (1991) for cropland soils of the identical texture.

Prediction of Detachment Rate Based on Hydraulic Variables

Flow shear stress, stream power and unit stream power are hydraulic variables that are widely adopted in erosion models to describe soil detachment processes. The soil detachment rates were plotted with these three parameters (Figure 4) to develop prediction models, and the nonlinear regressions are presented in Table VI. Based on the results, the detachment rates were better fitted to the shear stress with power functions than with linear functions (Table V) for the PR and PF treatments, which had extremely high detachment rates under steep slopes and high flow discharge. Compared with linear functions, the determination coefficients (R²) improved by 15% for power functions. These results are consistent with those of previous research reported under different conditions (Wilson 1993ab; Zhang et al., 2002, 2003; Cao et al., 2009; Al-Madhhachi et al., 2014); therefore, power functions of the flow shear stress are suggested to predict detachment rates in these treatments. For the PO, PM and PC treatments, the power function of the shear stress was equivalent to that of the linear function in predicting the soil detachment rate.

By contrast, as shown in Table VI, power functions of the stream power provided a good evaluation of the soil detachment rates in all the different treatments. In some experiments conducted in laboratory and field conditions, the flow stream power performed better than the shear stress in predicting soil detachment (Nearing et al., 1999; Zhang et al., 2002, 2003). Based on our data, the stream power was generally equivalent to the flow shear stress as a predictor of the detachment rate. The unit stream power is a relatively poor parameter in evaluating the soil detachment rate compared with the flow shear stress and stream power (Zhang et al., 2002, 2003; Cao et al., 2009), which was verified for the PC, PR and PF treatments by the low determination coefficients, as shown in Table VI. However, the precision of this parameter was higher (R² > 0·90) than that of stream power in predicting the detachment rates for the PO and PM treatments. Notably, the PO and PM treatments had the highest standardized coefficients of the slope gradient (Table IV), which demonstrated the effect of slope on soil detachment and might explain the performance of the unit stream power in this study. Similarly, Cao et al. (2011) report that the precision of the unit stream power is higher than that of the shear stress and stream power under steep slope conditions.

Although the hydraulic variables showed close relationships with the soil detachment rate, disparity occurred among the prediction functions when adopted in different management practices. To determine effective models for evaluating the soil detachment rate in the study area, the model efficiencies for the functions mentioned earlier were calculated and are listed in Table VII. The high ME values for the slope and flow discharge indicated that this power function was the best among all models for predicting the soil detachment rates in low-risk treatments (i.e. PC, PO and PM). However, this function performed relatively poorly compared with the shear stress and stream power for the PR and PF treatments, which had high detachment capability (ME = 0·761 and 0·775 for PR and PF, respectively). The power function of the unit stream power was similar, which reached a model efficiency of greater than 0·93 for the PO and PM treatments but was low (average ME of 0·65) for the other three treatments. These results demonstrated that management practices should be considered when selecting the best hydraulic variable to predict soil loss. Among the equations, the power functions of both shear stress and stream power provided good estimations of the soil detachment rate for all treatments (ME higher than 0·80). Therefore, these functions are recommended to evaluate soil detachment rates, particularly for high-risk treatments, such as PR and PF in this study.

Implications of This Study and Future Prospects

In this study, hydraulic flume experiments were conducted to investigate the effects of management practices on the soil detachment capacity, which was measured by using undisturbed topsoil samples from red soil cropland. As a general method adopted by many researchers (Nearing et al., 1991; Zhang et al., 2003; Li et al., 2015), the detachment rates determined for the different treatments of this study can be compared with those of other similar research. As demonstrated by our results, soils from conventional tillage cropland were highly erodible, but conservation measurements were also verified as efficient in preventing soil from detaching by overland flow. Nevertheless, because the laboratory flume is different from the actual rill conditions, the results are not suitable for a practical application in predicting soil loss. The primary objective of the experimental design was to examine the basic physics associated with detachment by flow and provide some theoretical support in erosion modelling. Therefore, the equations proposed based on our data should be helpful in selecting hydraulic variables when establishing a model to predict soil loss occurring on red soil cropland. To realistically reproduce the rill erosion process and estimate the rill erosion rate more accurately, field runoff simulation experiments would be conducted in the future. Especially for the straw mulch treatment, which showed the lowest rill erodibility in this study, different application rates, cover and types of mulches should be compared (Prosdocimi et al., 2016b) and the scale effect also should be considered when conducting experiments (Sadeghi et al., 2015). Additionally, further studies are required to clarify the relationships between soil properties of red soil cropland and detachment rates, and more soil structure indexes should be adopted and compared with one another to provide a better estimation of the physical condition of the soil.