This experiment was carried out at the Xifeng Soil and Water Conservation Experimental Station that is located in the Nanxiaohegou watershed, city of Qingyang, Gansu province, China. The study area belongs to a semi-arid continental climate with a mean annual temperature of 9.3 ∘C. The mean annual precipitation is 546.8 mm (1954–2014), of which precipitation from May to September accounts for 76.9 % of the total precipitation (Xia et al., 2017; Guo et al., 2019). The elevation ranges from 1050 to 1423 m. The main landforms include gentle loess tableland, steep hillslopes, and gully channels, and their areas account for 57.0 %, 15.7 %, and 27.3 %, respectively. The loess tableland is characterized by low slope (1–5∘), gentle and flat terrain and fertile soil. The main soil type is loessic soil with silt loam texture. Most of the hillslopes have been constructed as slope terraces. The main gully channel is usually U-shaped, and the branch gully is more actively developed and easily eroded as it is V-shaped by runoff from loess tableland (X. M. Xu et al., 2019). The flat loess tableland can accumulate 67.4 % of total runoff and cause serious gully erosion that can contribute 86.3 % of the total soil erosion (Guo et al., 2019). The original plant species have been seriously destroyed. Since the 1970s, the “Three Protection Belts” system, the “Four Eco-Economical Belts” system, and the “Grain for Green” project (Zhao, 1994; Fu et al., 2011) have been implemented to control soil erosion. The main land use on loess tableland has always been farmland and orchards, while the land use on hillslope was sloping farmland and orchards before 1999, which have since been changed into forested and grassy land due to the Grain for Green project. The current mean annual soil erosion rate has been reduced to 4350 Mg km-2 yr-1 in the study watershed (Guo et al., 2019). The plants are primarily artificially planted arbors and herbaceous vegetation and shrubs (Guo et al., 2021b).

The scouring experiment was conducted in August 2018. Before the formal experiment, the upstream area length was firstly adjusted to designed length of 5.0 m (Fig. 2a). Then, a self-made tent (length × width × height is 6.0 m × 3.0 m × 3.5 m) with waterproof canvas enclosed the plot to resist the effects of natural rainfall and sunshine on experimental the progress and photo shoot for 3D reconstruction (Fig. 1b). In addition, the experimental process was recorded by two Logitech C930e video cameras with a resolution of 2.0 MP (megapixels). Camera 1 was installed 2.5 m in front of plot headwall (Fig. 1a), and camera 2 was installed 3.0 m above the plot center (Fig. 1a).

Before the experiment, a watering can was used to spray each experimental plot until surface runoff was generated, and then the plot was covered for 24 h to ensure adequate water infiltration, which can assure us that the soil moisture of the five plots was approximately the same. The inlet pipeline was placed in a steady flow tank when the inflow discharge was adjusted to the designed value. A water thermometer was placed into the steady flow tank to monitor the change in water temperature during experiments. The runoff and sediment samples at the plot outlet were collected at 2 min intervals to represent the temporal change in the runoff and sediment of UA-GH-GB system, and the sampling time was recorded using a stopwatch (Fig. 2b). The runoff and sediment samples were oven-dried at 105 ∘C for 24 h and weighed to calculate the soil loss rate of the UA-GH-GB system. Besides, the timing of the collapse event was recorded during a headcut erosion. The upstream area was divided into four runoff observation sections, and the runoff width (w), depth (d), and velocity (V) of each section were measured by a calibrated scale of 1 mm accuracy and the color tracer method (Fig. 2b, c). The runoff velocity (VJ) before runoff arrived at the brink of headcut was measured 5–8 times by the flow velocity measuring instrument (LS300-A). The instrument was firstly placed perpendicular to the flow section but does not touch the underlying surface. When the flow passes through the turbine, the flow velocity can be measured by the rotating velocity of the turbine with the accuracy of 0.01 m s-1 and a measuring error of <1.5 %, and the runoff width at the headcut brinkpoint was measured (Fig. 2d). The runoff width and velocity of gully bed were also measured using the same method with the upstream area (Fig. 2e). The abovementioned measurements of runoff characteristics and sediment samples were taken at 2 min intervals. The whole experimental process was recorded by two video cameras and imported onto computers (Fig. 2f). In addition to the above runoff parameters, the runoff depth (db) at the brink of headcut, the plunge pool depth (DH), and the vertical distance (h) from the brinkpoint of the headcut to the water surface of the plunge pool were also measured 3–5 times by a steel ruler with 1 mm accuracy during each of the 2 min intervals (Fig. 3).

To obtain the dynamic change in the morphology of the erosional landform during the gully headcut erosion, the experimental duration (180 min) was divided into six stages (30–60–90–120–150–180 min). A photo-based three-dimensional (3D) reconstruction method was employed to obtain the digital elevation model (DEM) data of each plot prior to the experiment and after each 30 min test. A total of 14 target points were placed around the plot for identifying the 3D coordinates before the photos were taken. The eroded photography was recorded by a Nikon D5300 camera with a focal length of 50 mm. The following aspects were required during the photo shoot: (1) obvious water on the soil surface and direct sunshine should be avoided, (2) a minimum overlap of 60 % between subsequent photographs was required, and (3) some complex eroded photographic should be taken in detail. In this study, the upper left corner of the plot was set as the original coordinates (0, 0, 0), and the direction of the three-dimensional coordinate was determined as shown in Fig. 3d. These collected photos were imported into Agisoft PhotoScan software (Agisoft LLC, Russia, professional version 1.1.6), and then these control points and their coordinates were identified and entered into the software. The root mean square errors for the altitudes (z axis) of the target points are 0.0037, 0.0045, 0.0024, 0.0052, and 0.0030 m on average, respectively, for the experiments of the five inflow discharges, which can satisfy the study requirement (millimeter level). The DEM could be exported and was used to extract the morphological parameters and soil loss volume of three landform units at six stages (Frankl et al., 2015).

Figure 8 illustrates the temporal change in the accumulated energy consumption of the upstream area (UA), gully head (GH), and gully bed (GB). The accumulated energy consumption of the three landform units continued to increase linearly with time (R2= 0.990–0.999; P < 0.01), and the accumulated energy consumption in GH was always the highest at any time, followed by UA and GB under five inflow discharges. Moreover, the energy consumption rate (the slope value of the fitted equation) in the three landform units is basically constant, indicating that the spatiotemporal change in energy consumption maintained a relatively steady state during gully headcut erosion. Moreover, the energy consumption rate of GH was the highest, followed by UA and GB, and the energy consumption rate in the three landform units also increased with the increase of inflow discharge.

The variations in the total energy consumption of UA, GH, and GB and their proportions with inflow discharge are shown in Fig. 9. As illustrated in Fig. 9a, the total energy consumption of both the UA-GH-GB system and the three landform units increased with the increase in inflow discharge. When inflow discharge increased from 3.0 to 7.2 m3 h-1, the total energy consumption of the system, and UA, GH, and GB, increased by 3.6 %–105.3 %, 3.4 %–62.0 %, 3.5 %–108.2 %, and 9.0 %–327.5 %, respectively. Regression analysis revealed that the energy consumption of the system and the three landform units increased with inflow discharge as an exponential function (y=a⋅exp⁡(b⋅x); a= 1.14–55.41; b= 0.13–0.36; R2= 0.954–0.992; P<0.05). Furthermore, in view of the proportion of energy consumption, the energy consumption of UA accounted for 15.6 %–19.8 % of the total energy consumption and linearly decreased as inflow discharge increased (R2= 0.933; P<0.05), whereas the proportion in GB (2.8 %–5.8 %) linearly increased as inflow discharge increased (R2= 0.983; P<0.05). However, the proportion of energy consumption (77.3 %–78.6 %) in GH showed a weak change with inflow discharge (Fig. 9b), signifying that most of the runoff energy (77.5 %, on average) was consumed in the gully head position during headcut migration. Furthermore, we found that the total energy consumption (129.89–266.60 kJ) under different flow discharge conditions accounted for the 91.12 %–99.90 % of total flow energy (Fig. 9c, d), which also indicated that only 0.10 %–8.88 % of total flow energy remained at the outlet of the UA-GH-GB system. These results fully implied that most of the flow energy (> 91.12 %) upstream from the gully heads would be consumed during gully erosion, of which the gully headcut erosion (including plunge pool erosion) is the main process consuming flow energy.

This study showed that the runoff velocity at the headcut brinkpoint (Vb) firstly raised and then gradually stabilized with the experimental duration (Fig. 4a), which closely corresponded to the gradually decreasing runoff width on the upstream area over time (Shi et al., 2020a). However, this result was inconsistent with Zhang et al. (2016, 2018) and Shi et al. (2020b), who reported that the Vb decreased over time, which was mainly due to the gradually increasing roughness and resistance of the underlying surface over time, reducing the runoff velocity in their studies (Battany and Grismer, 2015; Su et al., 2015). The further analysis of the power function between Vb and time (Vb= a⋅tb; Table 1) showed that the a value increased but the b value showed a weak variation with the inflow as discharge increased, indicating that upstream flow discharge can improve initial Vb but not affect its change trend over time. Therefore, we can extrapolate the erosion process and the rule of upstream area from this simulation test to the actual ground situation. By contrast, the jet velocity entry to the plunge pool (Ve) and jet shear stress (τj) experienced a gradually decreasing process (Fig. 4c, e), which was mainly attributed to the fact that the development of several additional headcut steps caused more energy consumption in plunge pools, and the lower potential energy at the headcut brinkpoint was due to the shortened jet flow height (Guo et al., 2019; Jiang et al., 2020). This result, however, differed from the finding of Zhang et al. (2016), who stated that the Ve and τj remained stable as the experiments progressed, which was mainly attributed to the weak change of jet flow height induced by slow headcut retreat. This comparison means that the jet flow properties were strongly determined by the headcut retreat process.

For the runoff hydraulic of upstream area (UA) and gully bed (GB), the Reynolds number Re of UA and GB initially increased and gradually stabilized, but the Froude number Fr showed an opposite trend. This phenomenon was in agreement with previous studies (e.g., Su et al., 2015; Zhang et al., 2016). Besides, the Re and Fr of UA were larger than that of GB by 50.5 %–65.9 % and 1.39–2.04 times, respectively, under same inflow discharge upstream gully head, indicating that the runoff turbulence became weaker after the runoff of UA passed the gully head and experienced plunge pool erosion (Shi et al., 2020a). More evidently, the runoff on UA was in the supercritical transition and supercritical turbulent flow regime (Re > 500, Fr > 1), whereas the runoff on GB belonged to subcritical transition and subcritical turbulent flow regime (Re > 500, Fr < 1). However, Su et al. (2015) found that the steady state Re of gully bed was higher than that of the upstream area, which was mainly attributed to the difference in slope gradient. In their study, the larger gully bed slope gradient compared to the upstream area would accelerate the runoff velocity and, thus, enhance flow turbulence (Bennett, 1999; Pan et al., 2016). Furthermore, compared to UA, the τ and ω of GB increased and decreased by 2.8 %–15.7 % and 49.2 %–65.9 %, respectively. The increased shear stress was caused by the decrease in flow velocity on the gully bed, and the drastically decreased stream power can reflect the energy consumption of flow for transporting sediment on the gully bed. This result was different from some previous experimental studies on the gully and bank gully under different conditions. Previous studies have proven that lots of factors including plunge pool size, slope gradient, initial step height, and soil texture influenced the hydraulic properties from the upstream area to gully bed (Bennett and Casalí, 2001; Wells et al., 2009a, b).

Our study revealed that the accumulated runoff energy consumption of the upstream area (UA), gully headcut (GH), and gully bed (GB) linearly increased over time (Fig. 8), indicating that the spatiotemporal change in energy consumption maintained a relatively steady state during gully headcut erosion. However, the flow energy consumption of the bank gully in three landform units logarithmically increased over time (Su et al., 2015). This difference further manifested that the runoff energy consumption of different landform units depends on gully type, to some extent, and soil texture, slope, and headwall height (Wells et al., 2009a). Besides, under these flow discharge conditions, the proportion of energy consumption to the total flow energy ranged from 91.12 % to 99.90 %, indicating that almost all of the flow energy was consumed during headcut erosion. Furthermore, the proportion of energy consumption in UA, GH, and GB was 15.6 %–19.8 %, 77.3 %–78.6 %, and 2.8 %–5.8 %, respectively (Fig. 9), which was also indirectly supported by the study of Su et al. (2015), who suggested that the runoff energy consumption per unit soil loss from upstream area, headcut, and gully bed is 17.4 %, 70.5 %, and 12.0 %, respectively. This further signified that the gully head consumed most of the runoff energy (77.5 %, on average) during headcut migration. The flow energy must be consumed to surmount the soil resistance as headcut migrates, and the consumed energy was mainly focused on the headwall and plunge pool development (Alonso et al., 2002).

In terms of soil loss, our study indicated that the soil loss rate of the UA-GH-GB system initially increased to the peak value and then gradually declined and stabilized (Fig. 10), which was consistent with the results of many studies on rill and gully headcut erosion under different conditions (slope, initial step height, flow discharge, soil type, and soil stratification; Bennett, 1999; Bennett and Casalí, 2001; Gordon et al., 2007; Wells et al., 2009a; Shi et al., 2020a). Both the scour depth and sediment production increased in the initial period of underlying surface adjustment, while, once the plunge pool development was maintained, the sediment yield decreased and gradually stabilized (Bennett et al., 2000). In addition, the significant difference in the soil loss process was found among the three landform units. The soil loss of UA and GH decreased logarithmically over time, which was similar with several studies (e.g., Su et al., 2015; Shi et al., 2020b). Nevertheless, the GB was always characterized by the sediment deposition for the inflow discharge of < 4.8 m3 h-1, whereas the sediment was deposited firstly and then gradually transported as the inflow discharge increased to 6.0 and 7.2 m3 h-1. Similar phenomena were also found in some previous studies on rill headcut erosion (Bennett, 1999; Bennett and Casalí, 2001; Gordon et al., 2007; Wells et al., 2009a). This further indicated that the soil loss/deposition process of the gully system was significantly influenced by three landform units and especially by most of the flow energy (77.5 %) being consumed at gully heads, due to jet flow erosion strongly weakening the sediment transport capacity of flow on the gully bed and, thus, changing the soil loss/deposition process of gully system. However, Su et al. (2014, 2015) revealed a larger soil loss volume or soil loss rate in the gully bed than in the upstream area and headwall during the bank gully headcut erosion. This difference between our study and Su et al. (2014, 2015) is primarily caused by the difference in slope gradient. The gully bed slope (20∘) of the bank gully was larger than that (3∘) of our study, indicating that the runoff on the gully bed of the bank gully had stronger sediment transport capacity (Zhang et al., 2009; Ali et al., 2013; Wu et al., 2016, 2018). Besides, some previous research also proved that the soil type, surface roughness, slope length, and groundwater/surface runoff were the main factors influencing soil loss by gully erosion (Amare et al., 2021; Li et al., 2021). In view of the proportion of soil loss, the proportion of UA and GH was 9.5 %–11.4 % and 88.6 %–90.5 %, respectively, of which the proportion of deposited sediment on GB to the sediment yield from UA and GH can reach up to 0.4 %–10.3 %. This result fully demonstrated that the gully head is the main source of the sediment production during gully headcut erosion (Oostwoud-Wijdenes and Bryan, 1994; Zhao, 1994; Su et al., 2014) and also manifested the necessity and importance of gully headcut erosion controlling in gully-dominated regions (Amare et al., 2019).

The significantly different relationships between soil loss and jet or hydraulic characteristics were found among UA, GH, and GB. The soil loss rate of UA exponentially increased with five hydraulic parameters (runoff velocity, Reynolds number, Froude number, runoff shear stress, and stream power), indicating that soil loss of UA showed a stronger sensitive response to increasing hydraulic properties. This could be attributed to the frequent bank collapse on UA accelerating soil loss (Wells et al., 2013; Qin et al., 2018). However, the sediment deposition rate of GB linearly decreased with the five hydraulic parameters, signifying that sediment deposition on GB decreased at a stable state with the increase in hydraulic parameters. Therefore, the sediment deposition rate would reach zero when the five hydraulic parameters increased to the critical values, implying that the transformation of sediment deposition to sediment transport on GB would be triggered. Furthermore, the shear stress is the optimal parameter describing the soil loss process of UA and GB, which differed from some studies on hillslope/gully erosion hydrodynamic characteristics (Zhang et al., 2009; Shen et al., 2018; Ma et al., 2020; Sidorchuk, 2020). Most studies have verified that stream power is the superior hydrodynamic parameter describing the soil detachment process. This comparison also fully illustrated the great difference in hydrodynamic characteristics between hillslope erosion and headcut erosion. In this study, the soil loss of gully head (including plunge pool erosion) was significantly affected by jet properties. It is confirmed that the plunge pool erosion by jet flow becomes a crucial process controlling gully head migration and sediment production (Oostwoud-Wijdenes et al., 2000). Consequently, the plunge pool erosion theory is usually employed to build several headcut retreat models (Alonso et al., 2002; Campo-Bescós et al., 2013). Although the correlation between the soil loss of gully head and flow velocity at headcut breakpoint is weak, the larger flow velocity resulting from increased inflow discharge would improve the shear stress of jet flow impinging the gully bed, and thus, the gully headcut would suffered stronger incisional erosion of the plunge pool. However, in fact, the soil loss of the gully head was also affected by on-wall flow erosion (Chen et al., 2013; Guo et al., 2021a); thus, more studies should be conducted to clarify the effect of on-wall flow properties on headwall erosion.

From the energy consumption perspective, the soil loss rate of the three landform units significantly and logarithmically increased with the energy consumption, and the similar change trend was also found in the study of Su et al. (2015). This finding suggests that energy consumption could be considered as being the available parameter for estimating the soil loss of the gully headcut erosion (Shi et al., 2020b). Furthermore, we found the critical energy consumption initiating the soil erosion of UA, GH, and GB are 1.62, 5.79 and 1.64 J s-1, respectively, indicating the soil loss of the gully head (including the plunge pool) needs more flow energy consumption (Zhang et al., 2018; Shi et al., 2020a, b). This phenomenon can be attributed to the fact that the more runoff energy was consumed at the gully headwall and plunge pool erosion than UA and GB and, thus, resulted in more severe soil loss during headcut erosion. In addition, we found that the critical energy consumption activating the soil loss of UA-GH-GB system was lower than the sum of the critical energy consumption initiating soil loss and sediment transport of three landform units (9.05 J s-1). This result was closely related to mass failures, such as gully head and gully bank collapse, and can contribute the additional energy into the flow. So, the role of gravitational erosion in controlling the gully erosion process should be clarified in future studies.