Slope–velocity equilibrium is hypothesized as a state that evolves naturally over time due to the interaction between overland flow and surface morphology, wherein steeper areas develop a relative increase in physical and hydraulic roughness such that flow velocity is a unique function of overland flow rate independent of slope gradient. This study tests this hypothesis under controlled conditions. Artificial rainfall was applied to 2 m by 6 m plots at 5, 12, and 20 % slope gradients. A series of simulations were made with two replications for each treatment with measurements of runoff rate, velocity, rock cover, and surface roughness. Velocities measured at the end of each experiment were a unique function of discharge rates, independent of slope gradient or rainfall intensity. Physical surface roughness was greater at steeper slopes. The data clearly showed that there was no unique hydraulic coefficient for a given slope, surface condition, or rainfall rate, with hydraulic roughness greater at steeper slopes and lower intensities. This study supports the hypothesis of slope–velocity equilibrium, implying that use of hydraulic equations, such as Chezy and Manning, in hillslope-scale runoff models is problematic because the coefficients vary with both slope and rainfall intensity.
Hillslopes in semi-arid landscapes evolve in various ways, one of which is the formation of surface roughness through soil erosion. As surface erosion occurs, surficial material is preferentially detached and transported according to particle size, detachability, and transportability. This process results in a hillslope with an erosion pavement that is characterized by greater surface rock cover than that of the original soil. An erosion pavement is “a surface covering of stone, gravel, or coarse soil particles accumulated as the residue left after sheet or rill erosion has removed the finer soil” (Shaw, 1929). Once formed, the erosion pavement acts as a protective cover against erosive forces, which reduces subsequent rates of soil erosion. Erosion pavements are analogous to desert pavements formed in arid regions by wind erosion.
Because erosion potential is greater with steeper slope, the rock cover
resultant from the process of erosion pavement formation along the hillslope
profile is positively correlated with slope steepness in many semi-arid
environments, which has been documented in previous studies at the
USDA-Agricultural Research Service's Walnut Gulch Experimental Watershed
(Walnut Gulch hereafter) in south-eastern Arizona. Simanton et al. (1994)
measured rock cover at 61 points along 12 different catenas at Walnut Gulch,
with slopes at points ranging from 2 to 61 %. They found that rock
fragment cover (> 0.5 cm),
They also measured the rock fragments in the upper 50 mm of soil and found that total rock fragments (> 0.5 cm) in the soil were also correlated with slope steepness. Nearing et al. (1999) measured surficial rock content at the Lucky Hills site in Walnut Gulch on a range of slope gradients and also found a positive logarithmic relationship between rock cover and slope. Poesen et al. (1998) also reported a positive correlation between rock cover and slope gradient for a semi-arid site in Spain. Van Wesemael et al. (1996) also looked at hillslopes in the field in south-eastern Spain and found that both rock cover and surface roughness increased as a function of slope gradient.
Increased rock cover is associated with increased hydraulic friction because the rocks act as form roughness elements in shallow flow and thus create drag, and reduced soil erosion, both because the rocks protect the surface and because they dissipate flow energy. Rieke-Zapp et al. (2007) conducted a laboratory study with a soil mixed with rock fragments to investigate the evolution of the surface as the material was eroded by shallow flow. They measured flow rates and velocities, surface morphology (with a laser scanner), erosion (by measuring sediment concentrations from runoff samples), and rock cover change. The soil was a relatively highly erodible silt loam, which formed rills in the flume. They reported that rock fragment cover increased with time during the experiments, resulting in an armouring effect that greatly reduced erosion rates as flow energy was dissipated on the rocks. They also found that for treatments with no initial rock content in the soil, flows were narrower and formed headcuts that also acted to reduce flow velocity when rocks were not present. They reported that the addition of only small amounts of rocks in the initial soil matrix greatly reduced erosion rates compared to the soil with no rocks. Rill widths were greater for treatments with more rock.
The most common methods for mathematically describing the velocity of runoff on a hillslope are with either the Chezy (or the related Darcy–Weisbach) or Manning equations (e.g., Graf 1984; Govers et al., 2000; Hussein et al., 2016). Both of these equations relate flow velocities to slope gradient and flow rate using a hydraulic roughness factor, which is generally considered to be related to the morphological roughness of the surface in some way. The equations work well for fixed channel or rill beds (flow surfaces), but this approach when applied to eroding rills is problematic. In an eroding rill the flow interacts with the bed to change the surface morphology, which also changes the hydraulic roughness and hence flow velocity. In other words, there exists a dynamic feedback between the rill flow and the bed morphology on an eroding surface (see, e.g., Lei et al., 1998).
The dynamic feedback between flow, bed morphology, and erosion was discussed in a hypothesis testing study conducted by Grant (1997), in a broad way, from the perspective of mobile-bed river channels. Grant's hypothesis stated that “in mobile-bed river channels, interactions between the channel hydraulics and bed configuration prevent the Froude number from exceeding one for more than short distances or periods of time.” In other words, when the kinetic energy of flow exceeds the gravitational energy, flow instability is created that results in “rapid energy dissipation and morphologic change” that counteracts flow acceleration and “applies the `brake' to flow acceleration.” Grant argues that this system is consistent with the concept of energy minimization, because flow rate relative to flow energy is maximized at critical flow. Grant suggests that this general mechanism may be applicable in channels ranging from boulders to sand bed streams, with structures including step pools and antidunes. It has also been suggested that supercritical flow is necessary for the development of headcuts in upland concentrated flows (Bennett et al., 2000), which then act to retard flow velocities.
Govers (1992) used a compilation of data from laboratory experiments on
eroding rills and determined that the measured flow velocities were
independent of slope and related well to flow discharge alone:
Nearing et al. (2005) hypothesized that stony hillslopes in the semi-arid
environment evolve to a state of “slope-velocity equilibrium”. We define
slope–velocity equilibrium as a state that evolves naturally over time due
to the interactions between overland flow, erosion, and bed surface
morphology, wherein steeper areas develop a relative increase in physical and
hydraulic roughness such that flow velocity is a unique function of overland
flow rate independent of slope gradient. Under these conditions runoff flow
rates and velocities would increase approximately linearly down the slope
under normal runoff conditions, independent of slope gradient. This
hypothesis was based in part on a study of erosion rate estimations on two
small, gaged watersheds at Walnut Gulch using
Given that the slope characteristics had been largely determined prior to the
time period spanned by the field experiment with the
In this study we will investigate the development and hydrologic nature of slope–velocity equilibrium. We hypothesize that as the slopes evolve to a state of slope–velocity equilibrium through the process of preferential erosion of fines and resultant increase in surface roughness, flow velocity becomes independent of slope gradient. Rainfall simulation experiments were designed and conducted to test this hypothesis.
The soil used in the experiment was Luckyhills–McNeal gravely sandy loam
formed on a deep Cenozoic alluvial fan. It contains approximately 52 %
sand, 26 % silt, 22 % clay and less than 1 % organic carbon. The
soil was collected from the top 15 cm layer on level ground in the Lucky
Hills area of the Walnut Gulch Experimental Watershed
(310
Rainfall was delivered using a portable, computer-controlled, variable
intensity rainfall simulator (Walnut Gulch Rainfall Simulator). The WGRS is
equipped with a single oscillating boom with four V-jet nozzles that can
produce rainfall rates ranging between 13 and 190 mm h
Runoff rate from the plot was measured using a V-shaped flume equipped with an electronic depth gage and positioned at a 4 % slope. The flume was calibrated prior to the experiment to determine the depth to discharge relationship. Throughout the simulation timed volumetric samples of runoff were taken as a control of runoff rates.
Overland flow velocity was determined using a salt solution and electrical sensors at the end of the plot. Two liters of the solution were uniformly applied on the surface using a perforated PVC pipe placed across the plot. The application distances were 1.65, 3.5, and 5.8 m from the outlet. Salt transport was monitored through electrical resistivity of the runoff water measured with sensors imbedded in the outlet flume. The data were collected at 0.37 s intervals with real-time graphical output using LoggerNet3 software and a CR10X3 data logger by Campbell Scientific. Peak values from the salt curve were used because they were consistently more reliable, and hence comparable, relative to computation of the centroid of the salt curve, which is sometimes used.
Surface rock cover was measured at 300 points on a
Surface elevations were measured along three 2 m long transects oriented across the plot at 0.9, 2.9 and 4.9 m from the lower edge of the plot. Elevation points along these transects were measured at 5 mm intervals with at 0.2 mm vertical resolution using a Leica3 E7500i laser distance meter mounted on an automated linear motion system. The data were logged by a Bluetooth3 enabled mobile device using Leica1 software. A photo of the experiment in progress is shown in the Supplement.
The entire experiment included three treatments with soil slope gradients of 5, 12, and 20 % with two replications for each treatment. A replication consisted of three (for 20 % slope) or four (for 5 % and 12 % slopes) individual rainfall simulations (a simulation is considered to be one continuous application of rainfall on the plot) that progressively increased from 1.5 to 5 h. The reason for the increasing duration was the expectation that over time the slope would evolve at a progressively slower rate. The 20 % slopes were simulated only three times because they initially eroded more quickly, and the surface evolved more quickly, than did the other two slopes. In this paper we will use the words “treatment” to refer to the three slope gradients, “replication” to refer to one of the two sets of simulations performed on each slope, and “simulation” as the single application of rainfall, applied three to four times for each replication. A schematic example of the sequence of measurements can be found in the Supplement.
Prior to the experiment the soil was placed in the box and spread evenly in
an approximately 20 cm layer. The box was positioned horizontally, covered
with cloth to prevent splash, and low-intensity rainfall (35 mm h
The experimental procedure was as follows. Immediately before the first
rainfall simulation of each replication, soil surface measurements (rock
cover and laser elevation transects) were conducted. The first rainfall
simulation of the treatment started with 60 mm h
When all simulations of a treatment were completed, the top soil layer in the box was removed and replaced.
Statistical analyses were performed using SAS and Excel. Differences
reported in the paper are based on
Plot-average rock cover (> 0.5 cm) as a function of cumulative runoff for the six experiments.
Values of Chezy C and Manning's
Relationships involving rock cover and random roughness with each other and with cumulative runoff were developed using the measured values, which were made prior to the first simulation and at the end of each simulation for each slope and replication. Relationships involving the measured flow velocities were identified using interpolated values of rock cover and roughness based on the timing of the velocity measurements relative to the timing of the rock cover and roughness measurements. The velocity measurements were made near the end of each simulation (and toward the beginning of the first simulation), so the values of the rock cover and roughness were very nearly the same as the values measured at the end of each simulation (or prior to the first), but were adjusted slightly based on the measurements of rock cover and roughness that were made at the end of the previous simulation (or prior to the simulation, in the case of simulation 1.)
Random roughness was calculated as the standard deviation of all the trend-adjusted elevation measurements from the laser distance meter after removing 10 cm of data from the edges of the plot to remove plot edge effects.
The initial rock covers for the six experiments ranged from 16 to 40 %, and final covers ranged from 78 to 90 % (Fig. 1). There was no relationship between the final rock cover and either the slope gradient (Table 1) or the initial rock cover (Fig. 1). Increasing the threshold size for defining “rock” from 0.5 cm to 1 and 2 cm did not change this result. There were also no trends in final rock cover as a function of the distance down the plot (Table 1), and there were no consistent trends in the rate of rock cover development as a function of downslope distance (Fig. 2). An example photo of rock cover taken after simulation 3 at 20 % slope, replication 2, is shown in the Supplement.
The final random roughness was quite different as a function of slope gradient treatments, with the steeper slopes resulting in a rougher final surface (Table 1, Fig. 3). There were no consistent differences in random roughness measured in the lower, middle, and upper sections of the plots (e.g., Fig. 4).
There were significant, but relatively weak, relationships between rock cover and random roughness (Fig. 5).
Measured runoff velocities tended to decrease as the slopes evolved to
relatively consistent values of approximately 0.035 to 0.04 m s
Percentages of rock cover greater than 0.5 cm for the full, lower, middle, and upper portions of the plots, and laser-measured random roughness measured at the end of each experimental replication.
Corresponding to the changes in runoff velocities, hydraulic friction factors indicated an increase in hydraulic roughness as the surfaces evolved (Fig. 8). By the time that cumulative runoff reached 1000 mm, according to the Chezy and Manning coefficients, the surfaces were hydraulically rougher on the steeper slopes (i.e., Chezy values were less and Manning values greater on steeper slopes compared to shallower slopes). Also, Chezy and Manning coefficients were different for the two rainfall intensities (and hence runoff rates), with lesser Chezy values and greater Manning values (e.g., apparently hydraulically rougher) for the lower rainfall and runoff rates. These results were statistically significant.
There were very clear and strong relationships between the hydraulic roughness coefficients (Manning and Chezy) and measured random roughness from the laser measurements taken at the end of each of the six replications (i.e., each replication for each of the three slopes) (Fig. 9). Both the random roughness and hydraulic roughness (as quantified by Chezy and Manning) were greater on the steeper slopes (Fig. 8, Table 1). Hydraulic resistance, as indexed by the roughness coefficients, was greater on physically rougher surfaces. These data (Figs. 8 and 9) show that the roughness coefficients were different for the two different rainfall rates; hence, there was no unique hydraulic coefficient associated with a given surface roughness for these plots.
Surface rock cover (> 5 mm) on the three sections of
the plots as a function of cumulative runoff for
Averages of the three cross-section (lower, middle, and upper) laser-measured random roughness measurements (mm) as a function of cumulative runoff for the six experiments.
Laser-measured random roughness measurements (mm) at the three cross sections (lower, middle, and upper) as a function of cumulative runoff at 5 % slope, replications 1 and 2.
Random roughness as a function of rock cover on all
sections of the plots and both replications for
Flow velocities down the full plot as a function of cumulative
runoff depth for
Our results that show no dependence of final rock cover percentages, after the development of the erosion pavements, as a function of slope gradient (Fig. 1, Table 1) appear to contradict previous findings that indicate greater rock cover on steeper slopes. However, the final rock covers measured in this experiment are greater than those reported in previous work at Walnut Gulch, where the previous relationships were determined. Final rock covers in this experiment ranged from 78 % to 90 %, with an average of 85 %. Rock covers measured by Simanton and Toy (1994) and Simanton et al. (1994) ranged from 2 to 77 %. We conclude from these facts that the surfaces that evolved in this experiment were at maximum or near maximum coverage possible for this soil under natural hillslope conditions and climate of the area. A possible explanation for the somewhat lower rock cover values on the natural hillslopes may be related to factors at work that bring new material to the surface in the natural landscape, such as bioturbation and heaving associated with freeze/thaw, both of which are active in this and many other semi-arid landscapes (Emmerich, 2003).
Given the fact that rock cover did not vary with slope gradient, it is quite interesting that both random roughness of the surface (Fig. 3, Table 1) and hydraulic roughness (Fig. 9) did so. This in fact is evidence of the development of slope–velocity equilibrium. The steeper slopes evolved to both physically and hydraulically rougher surfaces compared to shallower slopes, even with similar rock covers, and in doing so maintained a constant relationship between flow velocities and unit discharge irrespective of slope (Fig. 7). We note that rainfall intensity also did not influence this unique relationship between velocity and discharge.
We are not sure yet exactly what process is taking place to effect the differences in roughness with slope independent of rock cover, but a corollary may be drawn to our understanding of flow rates in rills on non-rocky soils. Govers (1992) and Govers et al. (2000) reported on experiments with rills that showed a relationship between flow velocity and flow discharge independent of slope gradient on non-rocky soils. This was attributed largely to the development of headcuts (see also Rieke-Zapp et al., 2007, and Nearing et al., 1999). In our experiment it appears that development of greater form roughness occurred on the steeper slopes compared to the shallower slopes. One is tempted to compare the exponent determined from this experiment (Eq. 3) to exponents from previous work on rill flow velocities. However, it should be noted that Eq. (3) uses unit width discharge, while rill flow experiments typically reported relationships using total discharge (e.g., Eq. 2). Because of the complexity and variability in flow in interrill areas, it is not clear that a direct comparison of these values is entirely valid or robust. Nonetheless, under the specific conditions of this experiment, since width is a constant, the use of total discharge for these data would also result in an exponent of 0.696 (see Eq. 3), though the equation would have a different linear coefficient. This is greater than values previously reported for rill studies, including a value of 0.294 determined by Govers (1992), 0.459 determined by Nearing et al. (1999), and 0.39 reported by Torri et al. (2012).
Our data are not inconsistent with the hypothesis that has been proposed for channel beds, that the feedback between bed morphology, erosion, and flow velocities is associated with or controlled by the Froude number. Average Froude numbers calculated from the data tended to decrease as a function of the surface development and stabilize toward the end of the replications. Average values of the Froude number were less than one in all but two cases, both of which were measured during the early stage of the experiments when the surface was just beginning to evolve to its final state (Fig. 10). Of course there is no way of using the data presented here to know the spatial variations in the Froude number occurring on the plot at any given time, nor where the Froude number may have approached unity. With detailed measurements of the surface morphology at various times during such an experiment as this, possibly combined with distributed measurements or modeling of the flow velocity field, it might be possible to better investigate the role of energy minimization and the Froude number threshold in the development of these types of surfaces and control of the runoff velocities.
The results of this study do not suggest that the overland flow relationships, such as Chezy or Manning equations, are physically incorrect for flow modeling, but they do suggest that there are significant practical difficulties associated with their application to hillslope surfaces, which erode. Hydraulic friction factors, such as Chezy and Manning coefficients, are commonly used on runoff models for stony rangeland (and other) soils (Nouwakpo et al., 2016), but this usage is problematic because the coefficients depend on slope gradient and runoff rates, and hence with distance downslope on the hill and time during the runoff event. The data clearly showed that there was no unique hydraulic coefficient for a given surface condition. Generally, Chezy and Manning roughness coefficients are presented in tables based on surface conditions, suggesting that they are not related to either slope or runoff rates. In order to accurately implement the use of these equations in models, the coefficients used should be adjusted for slope and runoff rates, which means that they should be adjusted both temporally and spatially on a hillslope within each runoff event.
Final flow velocities for all six experiments at each of the three velocity transect measurement transects and two rainfall rates.
Hydraulic friction factors, Chezy and Manning coefficients, for the
full length of the plots as a function of cumulative runoff for 59 and
178 mm h
Relationships between the roughness coefficients, Chezy and Manning, to laser measured random roughness.
The fact that there is a unique relationship between velocity and discharge would suggest that routing runoff over hillslopes in models using such a relationship (e.g., Fig. 7) would provide more consistent and realistic results. It would appear to be counterintuitive and convoluted to apply an equation that relates velocity to flow depths and slope when the coefficients in the equations used are dependent on slope, discharge rate, and physical roughness. As pointed out previously with respect to flow in rills, if we assume a constant roughness coefficient, then “velocities will be over-predicted on steep slopes and under-predicted on shallower slopes” (Govers et al., 2000).
The results of this study are consistent with the hypothesis of slope–velocity equilibrium as a state that evolves naturally over time due to the interaction between overland flow and bed surface morphology, wherein steeper areas develop a relative increase in surface and hydraulic roughness such that flow rates and velocities increase approximately linearly down the slope under normal runoff conditions, independent of slope gradient. If this were not the case, then either supercritical or backwater flow would occur, which would cause erosion or deposition to bring the slope into slope–velocity equilibrium.
Velocities were dependent on discharge rates alone. The relationship between velocity and discharge was independent of slope gradient or rainfall intensity.
Use of hydraulic friction factors, such as Chezy and Manning coefficients, are problematic because they depend on both slope and runoff rate. The data clearly showed that there is no unique hydraulic coefficient for a given generalized surface condition.
All data used in this paper are available in the Supplement.
The authors declare that they have no conflict of interest. Edited by: Roger Moussa Reviewed by: Dino Torri and two anonymous referees