Quantification of red soil macropores affected by slope erosion and sediment using computed tomography

Soil structure is an important factor interacting with soil erosion and sediment processes. However, few studies have focused on the relationship between soil macroporosity and soil erosion across different terrains. The aim of this study was to quantify and compare soil properties and macroporosity characteristics in collapsing gully areas and to explore their impact on the formation and development of collapsing gullies. Soil cores were excavated at different positions of a typical collapsing gully and then scanned to analyze soil macropores. Soil properties and saturated hydraulic conductivity were also investigated. The results showed that the contents of sand, silt, and clay, the mean weight diameter of aggregates, and the infiltrate rates varied at different positions. The valley had the greatest macroporosity (1.09% ± 0.33%), the number (5919 ± 703), volume (1468 ± 194 mm3), and surface area (10.4 ± 2.6 m2) of macropores, as well as the mean volume (16.8 ± 7.4 mm3) of macropores >1 mm3, whereas these indices were lowest at the slope (0.15% ± 0.14%, 1189 ± 747, 266 ± 188 mm3, 1.7 ± 1.4 m2, and 10.6 ± 2.9 mm3, respectively). The macroporosity and the number of macropore decreased with increasing depth but were also influenced by the erosion and sediment processes. The processes of sediment and the roots of vegetation also influenced the orientation of the macropores. Macropore characteristics at different sites of the collapsing gullies affected the soil water infiltration and hydraulic conductivity and further affected the processes of water erosion and mass erosion.


INTRODUCTION
Collapsing gullies, also called Benggang by locals, are the most extensively eroded areas of the seven provinces of South China (Figure 1a,b).Both mass erosion and water erosion occur synergistically on the weathering crust of granite, at a depth of dozens to hundreds of meters, in tropical and subtropical regions.This area contains more than 2.4 × 10 5 collapsing gullies, covering an area of 1.1 × 10 3 km 2 (Figure 1b), 45% of which are in Guangdong Province, which contains 74% of their total area (Liang et al., 2009).More than 6.7 × 10 7 t year −1 of soil is lost because of these collapsing gullies, which damage farmland, houses, roads, bridges, reservoirs, and ponds, and causes economic losses of $3.2 × 10 9 , affecting 9.2 × 10 6 residents (Liang et al., 2009).The occurring and developing mechanisms of collapsing gullies have rose many efforts of researchers, and soil macropore and soil water movement were thought to be important factors (Tao et al., 2017).
Many factors influence both the spatial distribution and movement of soil water and other soil hydrological processes (Ghasemizade & Schirmer, 2013).The soil properties that particularly affect the soil hydrological process include soil structures, particle size distributions, porosities, soil layer characteristics, oxidation-reduction characteristics, organic matter contents, soil water repellency, and the soil water content itself (Zhu et al., 2012).Among these factors, the influence of macropores on soil water movement is of key importance (Lin et al., 2010).Macropores allow rainfall to infiltrate into deeper soil layers, whereas impervious layers without macropores, such as caliche, prevent the continuous infiltration of water, thus leading to a higher soil water content above the caliche (Hu et al., 2018;Li et al., 2013;Zhang et al., 2012).The macropore structures of soil in different slope positions differ because of the different characteristics of soil aggregates, gravel contents, and plant roots.This results in differences in soil water conductivity and preferential flow characteristics (Holden, 2009;Hu et al., 2020a;Mei et al., 2018;Zhang et al., 2016).However, soil subsurface flow can form different macropore structures along the preferential channel due to internal erosion (Nguyen et al., 2019).
The tropical and subtropical regions of South China are subject to a lot of storm, but the alternations between dry and wet are obvious.Water is the limiting factor of the ecosystem service function during the dry season and also the driving factor of collapses, landslides, debris flows, and other disasters during the rainy season (Liang et al., 2014;Tang et al., 2014).The granite weathering crust in the study area can be divided into a topsoil layer, red soil layer, sand layer, debris layer, and spherical weathering layer from the top to the bottom (Tao et al., 2017).It has been reported that the infiltration capacity of the red soil layer, sand layer, and debris layer

Core Ideas
• The architecture of soils at an erosive collapsing gully was studied using CT.• The ridge had middle macropores and was collapsing due to high infiltration.• Soil at the slope had low macroporosity and Ksf and was eroded by runoff.• Sediment at the valley had highest macroporosity and was eroded by subsurface flow.• Erosion and sediment processes interacted with soil structure properties.
decreases sequentially from surface to the deeper soil layers (Duan et al., 2018).The water holding capacity of the topsoil layer and the red soil layer is stronger than those of the sand layer and the debris layer.The soil water content gradually increases from the top to the bottom within the topsoil layer.However, the soil water content of the red soil layer to the debris layer gradually decreases.The farther away from the collapsing wall the site is, the higher its soil water content will be (Deng et al., 2015;Ni, 2016).The preferential flow varies by slope positions and soil layers (Tao et al., 2017).There are more preferential flows at the red soil layer than at the sandy soil layer and the clastic layer (Zhao, 2016), which are affected by soil properties, precipitation characteristics, and the initial water content (Duan et al., 2016).At present, studies of the soil hydrology of collapsing gullies focus mostly on the characteristics and processes of hydrology, or on the characteristics of the physical and chemical properties of the soil.However, little research focuses on the interplays among core-scale soil structures, profile-to hillslope-scale hydropedological processes, and landscape-scale soil erosion processes.Bridging these multiscale processes has been relatively less explored in collapsing gullies.
The loose and deep granite crust forms the material basis of collapsing gullies, and both rainstorms and soil water accumulation are the main drivers of the development of collapsing gullies.Soil structural properties (e.g., soil particle compositions, bulk densities, and porosities) significantly vary in different layers of the deep granite crust (Liao et al., 2018).This variation results in greatly differing soil mechanical characteristics in response to wetting (Xia et al., 2018).The increase of the soil water content after rainfall and infiltration will lead to an increase of upper soil weight and a decrease of soil shear resistance, which in turn results in the collapse of the granite soil wall (Zhang & Zhong, 1990).At the same time, the decrease of the cohesion and internal friction angle of the granite soil, as well as the increase of the pressure of the soil fissure flow when subjected to water are the main causes of the granite soil wall collapse (Li, 1992).Preferential flow along soil macropores is considered to be an important reason for the spatial-temporal variation of the soil water content and soil fissure water movement (Hencher, 2010).
Therefore, the hypothesis of this study was that different erosion and deposition processes existing at different sites of the collapsing gullies would lead to different soil structures, which in turn have an impact on hydropedological processes and soil erosion processes, and further on the development of collapsing gullies.The aims of this study were to quantify the macropore features of soil using computed tomography (CT) at different sites of a region with collapsing gullies, and to further explore the effects of soil erosion and sediments on the soil structure and clarify an aspect of their interaction.It is of great practical importance to study the soil macropores of granite red soil, which will aid investigations of the hydropedological process toward a better understanding of the occurrence and development law of collapsing gullies.This study also will help to effectively prevent, control, and restore erosion-related damage in these regions.

Study sites
The soil cores were sampled at a collapsing gully at Huacheng Town (24˚04′N, 115˚38′E), north of Wuhua County, Guangdong Province, China (Figure 1b).Collapsing gullies are often close to residential and agricultural regions, and the often devastating soil erosion and mud sand flow events (Zhang & Liu, 2014) increase the risk for houses, roads, and farmlands (Figure 1c).Wuhua County is a representative subtropical monsoon region with a mean annual air temperature of 21.3˚C, and a mean annual precipitation of 1507.2 mm (Zhang et al., 2020).The soil parent material is biotite granite and the weathering crusts are very deep, often exceeding 200 m.The soil is barren sandy loam with only 1.71 g kg −1 of soil organic carbon (Zhang et al., 2020).The vegetation is secondary with a coverage varying from 20% to 80% under different topographic and erosion conditions (Figure 1c).Pinus massoniana is the dominant constructive species, whereas Dicranopteris dichotoma, Rhodomyrtus tomentosa, Miscanthus floridulus, and Baeckea frutescens form the main undergrowth (Zhang et al., 2020).Three sites along a slope near Benggang, the ridge site, slope site, and the valley site (Figure 1c), were chosen and a total of nine undisturbed soil cores were collected in three duplicates per site.The ridge and slope sites were covered by sparse P. massoniana with rare undergrowth vegetation, whereas the valley site was mainly covered by Phyllostachys edulis with dense D. dichotoma and Nephrolepis auriculata as undergrowth.

Analysis of basic soil properties
For each site of the ridge site, slope site, and the valley site, three replicated soil samples, in the same slope location, with similar vegetation cover, and no far way from 20 m, were collected and soil properties were analyzed.Undisturbed soil cores of each layer (0-20, 20-40, 40-60 cm) were collected using a stainless steel cutting ring, 5 cm in diameter and 5 cm in depth to measure soil bulk density, capillary porosities, and filed capacity.Bulk density was calculated as the dry mass of undisturbed soil with known volume after ovendrying at 105˚C.Capillary porosity was determined based on the measured volume of water retained after placing the 5cm undisturbed soil core in a tray with a 5-mm water table until the filter paper at the top of the core became wet.Total porosity was determined based on the measured volume of water retained after placing the 5-cm undisturbed soil cores in trays filled with water at a depth of 5 cm for 24 h.The water of the tray was poured in slowly after the soil cores were placed and kept very close to the top of the soil cores to ensure essentially no air trapping and complete saturation.The noncapillary porosity was calculated as the difference between the total porosity and the capillary porosity.The field capacity was determined based on the measured the remaining volume of water after putting the saturated soil core on another cutting ring with dry soil for 8 h.Soils were air-dried at room temperature before sieving to 2 mm to analyze the soil particle size distribution with the pipette method (Gee & Bauder, 1986).The aggregate size distribution was measured by separating different aggregates via wet sieving following a previously published method (Dong, 1997).First, field-moist soil samples were carefully passed through an 8-mm sieve (without breaking the soil structure) and then air-dried.Then, about 50 g of air-dried soil for each replicate was wet-sieved through a series of four sieves, which separated the samples into five different aggregate size fractions: >2000-μm large macroaggregates, 1000-2000 μm middle macroaggregates, 500-1000 μm small macroaggregates, 250-500 μm tiny macroaggregates, and 106-250 μm microaggregates.The four sieves were stacked so that the largest was at the top and the smallest at the bottom, and the soil sample was put into the top 2-mm sieve.Before wet sieving, the soil samples were immersed in water on the top sieve for 5 min to break down aggregates via slaking pressure.Sieving was controlled by an Aggregate Analyzer (QT-WSI021, Qudaotech) with an up and down movement frequency of 30 times per min for a sieving period of 2 min.After sieving, each sieve was backwashed, and the fraction that remained on the top of the sieve was collected, oven-dried at 105˚C, and weighed.The mean weight diameter (MWD) was computed as the following equation: where X i is the diameter of each size fraction (mm), and W i is the pro-portion of the total sample mass in the corresponding size fraction; n is the number of sieves.

Field-saturated soil hydraulic conductivity measurements by DualHead Infiltrometers
The field-saturated soil hydraulic conductivity (K fs ) was measured using a DualHead Infiltrometer (Decagon Devices) following an optimized method of Zhang et al. (2019).Before data collection, grass and vegetative litters were carefully removed, and the surface soil temperature and water temperature were measured using ECH2O 5TE sensors (Meter Group).An insertion ring with a depth of 5 cm and a radius of 7.5 cm was gently driven into the soil to ensure good contact with the soil and minimize disturbance.It was ensured that the ring was level in all orthogonal directions.K fs was measured using a single cycle with 35 min holding time at each pressure head of 10 and 15 cm after 15 min of soaking.Only measure-ments without leakage and with coefficients of variation of the pressure head and water level below 15% were assumed to be reliable and used for further analyses.All K fs values were viscosity-corrected to a standard temperature of 25˚C before comparison among different sites.This correction prevented the influence of changes in water effluent viscosity at different temperatures according to a previously described method (Zhang et al., 2019).

Soil core samples
Polyvinyl chloride (PVC) cylinders with 5-mm wall thickness, 10-cm internal diameter, and 50-cm length were used to house intact soil.After a plot had been selected, the aboveground grass (if it existed) was clipped to ground level, and the litter was removed to expose bare soil before soil sampling.The PVC cylinder was pushed into the soil using the instruments shown in Figure 2. First, the ground spears (8) were revolved into the ground, and the support system (7, 6) was set up.An air-bubble level (9) was affixed to the top of the iron roof plate (6) to ensure that it was level.A steel tube (1) under the roof was used to house a PVC cylinder (4) with a window (2) on the upper side.The internal diameter of the steel tube (1) was only marginally larger than the external diameter of the PVC cylinder to limit its shifting.A steel cap was affixed on the top of the PVC cylinder.Then, a vertical hydraulic jack (3) was inserted between the roof (6) and the steel cap through the window (2) of the steel tube (1).The PVC cylinder (4) was pressed into the soil by driving the hydraulic jack (3) as the roof (6) was fixed by the struts (7) and the ground spears (8).
As the hydraulic jack (3) reached its maximum lifting limit, it was taken out and reset.A log (5) with the same diameter of the PVC cylinder (4) and a height below the maximum lifting limit of the hydraulic jack (3) was put on the steel cap.The hydraulic jack (3) was inserted into the window (2) and driven again.These operations were repeated until the whole PVC cylinder (4) had penetrated the soil.To reduce the resistance of the soil layer, the roots of plants were cut, and to avoid damage from gravel, a steel ring knife (10) was installed at the bottom of the PVC cylinder.The inner diameter of the front blade of this steel ring knife is identical to the inner diameter of the PVC cylinder.The inside of the upper end of the ring knife featured a groove, the inner diameter of which equals the outer diameter of the PVC cylinder (4).There were also two small vertical windows at the bottom of the steel tube (1) for observing whether the PVC cylinder (4) had been completely pressed into the soil.This instrument can smoothly and ver-tically push the PVC cylinder into the soil without shock and rotation and can thus obtain the soil core with minimal disturbance.After the procedure, the PVC cylinder was dragged out, covered by a PVC cap at each end, and wrapped in a sponge to protect the soil from mechanical disturbances.

X-ray computed tomography scanning and image analysis
A Philips Medical Scanner (Royal Dutch Philips Electronics Ltd., Model: iCT 256), a 256-slice spiral X-ray CT, was used to scan the obtained soil cores.The energy level was set to 140 kV, and the X-ray tube current was 153 mA, which can provide detailed and low-noise projections, 16-bit and 1024 × 1024 pixels per slice.The scanning was continuous with a scanning spacing of 0.3 mm.After scanning and reorientation, an image sequence was produced with more than 1667 slices for each soil core, in a coronal view.The images were then resampled to 0.122 mm × 0.122 mm × 0.3 mm (width × length × depth) voxel cubic to facilitate computation.Images were analyzed using ImageJ software version 1.4.3.67 to examine the pore characteristics according to the following steps (Figure 3): 1. Slice selection: To avoid voids and minimize disturbance caused by sampling, 100 slices (3 cm in depth) at the soil core head and 60 slices at the soil core end were removed using the "Slice Remover" tool in ImageJ.Finally, 1500 slices (45 cm in depth) were selected for the following analyses.2. Region of interest selection: The "Region of Interest" tool was used to exclude voids near the core walls and to minimize the effects of beam hardening.A square area in the center of the soil core, with a diameter of 80 mm, was selected as a region of interest and cropped for further analysis.3. Thresholding: The attenuation value of the obtained images ranged from −1024 to more than 3071.The intensity value of the water phantoms (mean = 0) was set as threshold value to differentiate air-filled spaces from other regions for image analysis.Values below the threshold were identified as air-filled pores, whereas values above the threshold were identified as non-pores (Feng et al., 2003;Hu et al., 2015).This threshold was also used by Udawatta et al. (2008) and performed well in the experiments for this study.Small PVC cylinders with known diameters were inserted into an undisturbed soil core and scanned, and this threshold analysis enabled the interpretation of their pore sizes.After thresholding, the images were translated into binary images, where black areas represent the soil matrix and white areas represent macropores.4. Macropore analysis: After segmentation, the connectivity of soil macropores was estimated using the plugins "Purify" and "Connectivity" of ImageJ (in turn).The numbers of macropores, the macropore area, macropore perimeter, circularity, Feret, and minimum Feret of each slice were calculated using the "Analyze Particles" tool of ImageJ in 2D.A slice porosity was calculated by the pore area of slices dividing the soil core area (5024 mm 2 ).
Previous studies classificated pores with equivalent pore diameter (φ) >0.1 mm as macropores (e.g., Harvey & Vadose Zone Journal Nuttle, 1995), and thus, the pores detected using CT (at a resolution of 0.122 mm × 0.122 mm) could all be defined as macropores.Other researchers reported that only pores with φ > 1 mm can be divided into macropores (Carey et al., 2010), and therefore, the present study classificated pores with φ > 1 mm as wide macropores to better distinguish them.The equivalent diameter was calculated separately according to the pore shape, which is related to the ratio of pore area (A) to the square of pore perimeter (P).Values of A/P 2 greater than 0.04 were designated as round, those less than 0.015 as elongate, others as irregular.For round and irregular, equivalent diameter equals 2(A/π) 0.5 ; and for elongate pores, equivalent diameter equals the pore width, 0.25[P − (P 2 − 16A)] 0.5 (Hu et al., 2015).
The numbers of macropore tubes, macropore volumes, and macropore surface area were calculated with the plugin "3D objects counter."These 3D pores were divided into wide macropores or macropores, depending on whether the volumes exceeded 1 mm 3 or not, respectively.3D visualization of soil macropore networks in the soil cores was reconstructed using the plugin "3D viewer," and then, the skeleton of the 3D soil macropore networks was established using the plugin "Skeletonise 3D."The numbers and lengths of both macropore tubes and their branches were analyzed using the plugin "Analyse skeleton."

Statistical analyses
Differences among measured parameters within-group sites were analyzed using a one-way analysis of variance combined with Fisher's protected least significant difference test.All statistical analyses were conducted using SPSS software (version 19.0,IBM Inc.), applying a p = 0.05 level of confidence.

Soil properties
Across the investigated catena, the soil properties varied greatly from the ridge to the valley (Table 1).The soils were clay loam, loam, and sandy loam at the ridge, slope, and valley, respectively.The sand contents increased from the ridge to the valley, whereas the clay contents followed the opposite trend, and silt contents were highest at the slope.The bulk densities at the ridge were insignificantly higher than at the slope, whereas bulk densities at both sites were significantly higher than at the valley floor for all depths.However, the differences of total porosity contrasted with the bulk density.
The capillary porosities increased from the ridge to the val- ).
Abbreviations: DW, dry weight; K fs , field-saturated soil hydraulic conductivity; MWD, mean weight diameter of aggregates; W, weight.
Vadose Zone Journal ley but only showed significant differences between the ridge and the valley at 20-40 cm.No significant differences were found between non-capillary porosities of the 0-20 cm soil layers at different sites; however, at the 20-60 cm layer, the non-capillary porosities at the valley were significantly higher than both at the ridge and slope.The field water capacity only showed a significant difference at the 20-40 cm layer.The aggregate weight percentages and MWD increased from the ridge to the valley.The K fs of the ridge and slope were 135.7 ± 98.7 and 80.6 ± 56.1 mm h −1 , respectively.Ridge sites had higher K fs than slope sites, but their differences were not significant (p = 0.449).The K fs at the valley could not be measured as it exceeded the maximum measurement range of the DualHead Infiltrometer (1150 mm h −1 ).

Visualization of macropore networks
Visualizations of soil macropore networks in soil columns are shown in Figure 4. Soils at the valley have the greatest macroporosity and more macropores than soils at the ridge; however, the soils at the slope have minimum macroporosity and macropores.At the ridge and slope, soil macropores are mainly distributed at the surface layer (0-15 cm), where they formed networks, whereas at deeper soil layers (>15 cm), soil macropores are fewer and isolated (Figure 4a,b).Exceptionally, the Ridge 3 soil core showed many macropores at 15-35 cm depth, because this sample contained an ant nest at this depth (Figure 4a).The soil columns of the valley showed high macroporosity at all depths (Figure 4c, especially Valley 3).Sometimes, the macropores at the deeper layer (30-40 cm) showed higher numbers than those at the shallower layer 0-15 cm (Figure 4c, Valley 2).The directions of soil macropores at the ridge and slope were mainly vertical with a few horizontal networks at the surface layer; however, the soil macropore directions of the valley were mainly horizontal, which differed from the other two sites (Figure 4).

CT-measured macropores and macroporosity of whole columns
The total number, volume, and surface area of soil macropores, as well as the macroporosity at the valley, were significantly higher than those at the ridge and slope (Table 2).The wide macroporosities (volume >1 mm 3 ) were 3.3% ± 1.9%, 1.4% ± 1.4%, and 10.5% ± 3.3% for soil columns at the ridge, slope, and valley, respectively, and the numbers of wide macropores (volume >1 mm 3 ) were 428 ± 75, 266 ± 188, and 1468 ± 194, respectively.Although the numbers of wide macropores only supplied a small part of the total macropores detected by CT, the large macropore volumes and wide macroporosity accounted for more than 94% of the total macropore volume and 93% of the macroporosity, respectively.The mean Ridge, mid-erosion 1828

F I G U R E 5
The macropore numbers along the soil depth at (a) the ridge, (b) the slope, and (c) the valley.The black, red, and green solid lines are the macropore numbers of three replicates with the blue solid line being their mean for each 3 cm depth.The blue short dash lines are the number of macropores with equivalent pore diameter >1 mm.
volume and mean surface area of macropores did not show significant differences between different sites.The mean connectivity at different sites showed great differences (Table 2); however, the differences were insignificant because there were great variations among each site.

Spatial distribution of macropores
Large variations were found in the macropore number distribution along the depth for different sites (Figure 5).Although the trends were different in different replicates, the average trend of macropore numbers decreased with increasing soil depth.At the top 3 cm of the soil cores, the wide macropore numbers were 11.4,10.4, and 27.5 at the ridge, slope, and valley, respectively.At the ridge, the average wide macropore number decreased to 7.4 at a depth of 3-6 cm and ranged between 4.8 and 3.7 at a depth of 33-45 cm.The average wide macropore number at the slope decreased to 3.8 at a depth of 6-9 cm and remained within the range between 1.9 and 3.2 at a depth of 27-45 cm.The wide macropore numbers at the valley were highest among the three sites with a maximum number of 27.7 at a depth of 3-6 cm.The numbers of total macropores also decreased with increasing depth (Figure 5).The percentages of the wide macropore number to the total macropore number increased with soil depth at both the ridge and the valley.Soil macroporosities differed greatly among the three investigated sites (Table 2) and also varied along depths based on the data of slices (Figure 6).Soil macroporosities mainly decreased with increasing soil depth except for Ridge 3 and Valley 2. At the top 3 cm of the analyzed soil cores, the mean macroporosities were 0.62%, 0.48%, and 2.04% for the ridge, slope, and valley, respectively.The macroporosities at the ridge remained above 0.50% over 12 cm depth and then became lower than their mean macroporosity of 0.34% except for the depths of 15-18 and 30-33 cm, where extended ant nests were accidentally sampled at Ridge 3. Ant nests are common in this region, and it cannot be avoided that these are included in the sample.At the slope site, only macroporosities at soil layers above 9 cm depth exceeded 0.48%, and the macroporosities of the soil layers below this depth did not exceed 0.16%.The macroporosities in the valley were highest among the three sites for each layer, ranging from 0.27% to 1.33% at depths below 18 cm.The trends of the wide macroporosities were similar to those of the macroporosities (Figure 6) because wide macropores represent the majority of total macropore volume (Table 2).

Macropore size distribution
The macropore sizes (φ) mainly ranged between 0.1-1 and 1-2 mm, accounting for more than 44.0%, and 32.3%, respectively (Figure 7).φ > 5 mm were less than 1% of the total macropores at all three sites.At the slope site, the percentage of 0.1-1 mm macropore was highest among the three sites, and the percentages of the wide macropores larger than 2 mm were lowest.The percentages of 1-4 mm wide macropores at the valley were highest among the three sites, whereas the percentages of wide macropores >4 mm were highest at the ridge.

F I G U R E 7
The frequency of equivalent pore diameters of soil columns at different sites.

Characteristics of macropore at different sites
Macropore networks were distinctly different among different positions at the collapsing gullies and different types of macropores were detected by CT.Macropores formed by earthworms were highly continuous, relatively large, and tubular (Luo et al., 2010); however, earthworms were not found at the ridge and slope as the soil was very barren with soil organic carbon levels as low as 1.71 g kg −1 (Zhang et al., 2020).However, a network of ant nests was unexpectedly found in the middle of Ridge 3 (Figure 4).The ant network macropores were characterized by several big chambers connected by tunnels, and these characteristics were similar to those observed by Li et al. (2017).Macropores formed by roots were highly continuous and round in shape, the size of which generally decreased with depth.These macropores could be found in all soil cores, especially within the top 10 cm, which is in accordance with the roots that are mainly distributed within the topsoil layer.Smaller, more randomly distributed, and less continuous macropores were likely inter-aggregate macropores, which were formed by wetting and drying (Hu et al., 2015;Luo et al., 2010;Zhang et al., 2022).These are also the main macropores at soil layers deeper than 20 cm at the ridge and slope sites.Another phenomenon is the presence of many horizontal macropores at the valley site.This is because plants generally have shallow roots and grow horizontally (e.g., bamboo) because of the high soil water level and the absence of water deficits in the valley.
In addition, the erosion and sediment process greatly affected the number, size, macroporosity, and continuity of soil macropores.Soil macropores at the slope were comparatively rare and small (Table 2), similar to previous findings of Zhang et al. (2016).Soil erosion was more intensive at the slope compared to the ridge (Sabzevari & Talebi, 2019), and the soil horizon A at the slope had eroded during observation.Soil horizon A usually has higher porosity and more macropores than horizon B (Hu et al., 2020b).The barren slope soil without horizon A also restricted the growth of plants, so that the P. massoniana growing there were stunted, and the undergrowth was extremely sparse or even absent, which further limited the development of macropore.However, at the valley, the soil was deposited by floods from the slope, and many dead tree branches were buried in different layers, which then formed macropores after decomposition.Moreover, the sediment mainly consisted of sand with an abundance of inter-aggregate macropores.

Implementation of macropores, infiltration, and soil erosion in the Benggang landscape
Macropore characteristics altered at different sites, which affected the soil water movement and the processes of soil erosion and collapse.The slope had the lowest K fs , which may be due to the few macropores limiting the infiltration capacity (Hu et al., 2020a;Iversen et al., 2012).Superimposing the flow from the upslope, the overland flow at the slope was strong and erosive.Severe erosion limited soil water because of few infiltrations, and soil barrenness caused by the loss of horizon A inhibited the establishment of plants.Thus, the interaction between the low number of soil macropores, low vegetation cover, and fierce overflow at the slope was mutually promoting.The development of slope erosion thus stretches into the collapsing gullies.This could be one of the evolutionary mechanisms of collapsing gullies after the destruction of the original vegetation (Figure 8).This agrees with previous literature, which proposed that collapsing gullies developed because of the loss of vegetation (Zhang & Zhong, 1990).
The ridge had more macropores, which were larger in size and volume, and the macroporosity was also higher than that of the slope.This was caused by the relatively weaker water erosion and better vegetation conditions at the ridge compared with those at the slope.Activities of plant roots and soil macrofauna (e.g., ants) are advantageous for the formation of macropores.These conditions promote water infiltration at the ridge.According to former research, the soil water content is an important factor that promotes the development of an existing collapsing gully (Li, 1992;Shi, 1984).More rainfall infiltration decreases the shear resistance of the soil layer while also adding to the weight of the soil column.These developments further increase the probability and risk of soil collapse and the development of a collapsing gully (Figure 8).Macropores decreased with increasing soil depth, indicating that the infiltration ability decreases with increasing soil depth as Duan et al. (2018) reported.Thus, the soil water would remain at the topsoil layer, thus increasing the weight of the soil layer and decreasing the shear resistance of soil (Deng et al., 2018).
The valley site was covered by sediment from the upslope and had an abundance of macropores.Consequently, infiltration would be large, and the underground flow may be strong under these conditions.The high soil water content and subsurface flow increase the risk of subsurface flow erosion (Tebebu et al., 2010).The sediment was sandy, containing only a little clay.It has been reported that alluvial sediment is very erosive when the vegetation cover is insufficient (Jiang et al., 2020).As observed in the field, once the check dam was completely filled, the overland flow could easily erode the inside sediment and thus form a deep gully.Once the sediment was removed, the base of the collapsing wall would be lost, and the risk of collapse would increase without base support (Figure 8).

CONCLUSIONS
The ridge of collapsing gullies had more and larger macropores than the slope, whereas the sediment at the valley

Vadose Zone Journal
had the most macropores.The numbers of macropore and their macroporosities mostly decreased with increasing soil depth.The equivalent pore diameter of macropores mainly remained below 2 mm.These characteristics differed according to the tomography position, soil erosion, plant cover, and soil macrofauna.The high macroporosities of the valley and ridge increase the risks of subsurface flow erosion and soil collapse, which would promote the development of collapsing gullies.The soil macropores at the slope were rare, and infiltration abilities were comparatively small; thus, overland flow and slope erosion can be assumed to be strengthened.These effects contribute to the formation of collapsing gullies.2020GDASYL-20200102013, 2020GDASYL-20200301003, and 2021GDASYL-20210103046).

C O N F L I C T O F I N T E R E S T S T A T E M E N T
The authors declare no conflicts of interest.

R E F E R E N C E S
U R E 1 (a) A photo of a collapsing gully; (b) the location of soil core sampling sites in Guangdong Province, China, and (c) their geomorphic and environment around.Shaded provinces in (b) represent the main distribution areas of Benggang.
Procedures of image analysis in the study.

F
Three-dimensional visualization of soil macropore networks in the soil columns and soil macroporosity along the soil depth at the (a) ridge, (b) slope, and (c) valley.The number after ridge, slope, or valley is the replicated column number.T A B L E 2 Characteristics of macropores of different stages of the slope.
Distribution of macroporosity along the soil column depth at the ridge, slope, and valley.

F
Conceptual model of the interplays among soil macropore, infiltration, and soil erosion in the Benggang landscape.
This work was supported by the National Natural Science Foundation of China (Grant Nos.NSFC 41977010, 41930865, and 42177065), the Guangdong Basic and Applied Basic Research Foundation (Grant Nos.2019A1515010628 and 2018A030313696), the Guangdong Foundation for Program of Science and Technology Research (Grant No. 2020B1111530001, 2019QN01L682, 2019B121205006, and 2019B121201004), the GDAS Special Project of Science and Technology Development, China (Grant Nos.
Soil properties at the ridge, slope, and valley of collapsing gully region.
T A B L E 1※: K fs at the valley could not be measured as it exceeded the maximum measurement range of the DualHead Infiltrometer (1150 mm h −1