2. MATERIALS AND METHODS

2.1 Study area

This study was conducted at 102°54′12″ E and 23°37′13″ N and an altitude of 1300–2500 m (Figure 1), in Jianshui County, Honghe Hani and Yi Autonomous Prefecture, Yunnan Province, China. This region is a karst graben basin with strong stratigraphic erosion since the Tertiary Period (Cao, 2021). The red weathering crust of carbonate rocks, known as carbonate laterite, is formed in tropical and subtropical environments through a long weathering process that is unique to the karst graben basin. The long-term average annual temperature of this area is 19.8 °C, with most of the 805 mm mean annual precipitation occurring between May and October. The summer/autumn wet season lasts from April through September and provides approximately 70%–90% of the total annual rainfall.
The E. robusta forests in the experimental area are approximately 1.5 km2, which cover the Nanpan River water system in the upper reaches of the Pearl River and the middle and upper reaches of the cross-border Red River Basin. Three study sites (2.8 m × 2.8 m each) were selected for field tests and marked as Sites 1, 2, and 3. The bulk density of the soil was approximately 1.2 g/cm3 and the pH was 6.5. Table 1 details the physical and chemical properties of the tested soil.

2.2 Experiment design

2.2.1 GRP surveys and ground truthing

This study was divided into field survey and infiltration experiment stages. The soil cracks and their properties were characterised in accordance with the GPR lines and excavation of the soil pedons using an embedded Linux GPR system (CAS-S500, IECAS, China) with a shielded antenna of 500 MHz. Horizontal and vertical GPR lines were deployed in accordance with crossed grid principles and with horizontal lines parallel to contour lines. The GPR grid (2.8 m × 2.8 m) consisted of 15 horizontal (H1–H15) and vertical (V1–V15) parallel survey lines with 20 cm between them (Figure S1). The horizontal lines were oriented downslope with a local 5° slope. 90 radargrams (3 sites × 30 survey lines) were collected over the study area, with data acquired from the distance travelled by a survey wheel equipped with a position sensor. The sampling frequency of the system was 25.6 Hz, and the wheel moved at a uniform speed (scanning rate of 390.625 traces/s) in conformity with the survey line. The time window was set at 40 ns and, for each 3 m long survey transect, 258–276 traces and 38 time windows were recorded. Based on the signal fluctuations appearing in the radargrams, six pedons (three sites × two random pedons) were stochastically excavated for verification (Figure S1). The length and depth of each pit were 120 cm and 60–70 cm, respectively. Various soil crack shapes were screened through the radargrams and pedons of the soil profile (Figure 2b).

2.2.2 Infiltration experiment design

Preferential flow paths were simulated to account for the natural cracks and pore structures of soils detected by GPR. Based on the identification of crack configuration in the field which was defined as “I”, “V”, and “Λ” morphologies, the infiltration experiments were conducted to explore the effect of isolated cracks on water infiltration. In addition, we found that soil cracks were filled with sand grains or rock fragments of different grain sizes when excavating the soil profile. It has been shown that inclusions (Yang et al., 2016; Liu and She, 2020), width (Ou Yang, 2020), and configuration (Liu and She, 2020) influence the infiltration process. For simplicity, soil cracks are always assumed to resemble smooth parallel plates (Tsakiroglou et al., 2012; Rouchier et al., 2012). Therefore, three properties of different inclusions (sand grains and rock fragments), crack widths (1, 1.5, and 2 cm), and configurations (I-shape, V-shape, and Λ-shape) for infiltration were simulated (Figure 2d and Table 2). A soil column without cracks (CK) was used as the control. Based on these properties, all experiments were divided into six groups (Figure 2d) in which four soil columns (including CK) were conducted simultaneously, with each experiment repeated three times.
Disturbed soil samples were collected in September 2021 from the upper 70 cm of pedons in layers. The tested soil was passed through a 2 mm sieve, air-dried, thoroughly mixed, and hierarchically backfilled into Plexiglas columns (20 cm × 5cm × 60 cm, Figure 2c) to ensure that the particle size composition of the backfilled soil was as consistent as possible with the field site. For homogeneous soil columns, the backfilled soil needed to be compacted, and the surface was roughened every 5 cm before the next layer was added. The top layer was covered with a metal mesh after backfilling soils of 50 cm to prevent water flow impact on top soils. Different crack inclusions were replaced by sand grains and rock fragments. The diameter of all sand grains was approximately 2 mm, whereas the rock fragments ranged 2–5 mm. All these debris (i.e. sand grains and rock fragments) was wrapped with 500-mesh nylon netting to prevent soil particles from passing through and blocking the paths (Figure 2c). From the field excavation of representative sites and GPR large-scale detection, cracks are mainly scattered approximately 10 cm from the surface. Therefore, the lab-based experiments of cylindrical “channel” was placed in column when backfilled to 30 cm from the upper edge of the column to simulate various cracks.
Brilliant Blue FCF allows clear visualisation of the infiltration process of non-uniform flow in the soil (Lipsius et al., 2006); as such, a Brilliant Blue FCF dye volume of 5 L (4 g/L) was added to the Marriott bottles to visualise the preferential flow paths. The volume of the dye solution used corresponded to the amount of water collected in a 20 mm rainfall event, considering more than 65% of regional rainfall events were for 20 mm during 2009–2018. Therefore, a constant pressure head of 20 mm was supplied using Marriott bottles (Figure 2c). For each experimental group, four images (three variances and a control) were taken with a digital camera (Canon EOS 60D, Canon, Japan) directly in front of the soil column at 3, 5, 10, 20, 30, 50, 70, and 90 minutes after the Brilliant Blue FCF dye solution was applied. The infiltration continued until the wetting front penetrated the bottom of the column. The dye infiltration depth and wetting front trajectory of each soil column were recorded simultaneously using a ruler with millimetre precision attached to the edge of the soil column. The infiltration rate was determined by measuring change in water level in Marriott bottles over time.

2.3 Data analysis

2.3.1 GPR data processing

GPR data were processed following a conventional routine (Dal et al., 2019; Guo et al., 2014) via Radar View V18.08 (Institute of Electronics, Chinese Academy of Science, China) to enhance the signal-to-noise ratio, which follows the sequence: 1) background filter wiped off antenna reverberation; 2) time-zero correction calibrated the radar antenna start time; 3) energy gained compensated energy attenuation with propagation depth; 4) Hilbert filter removed high- and low-frequency; and 5) DC-drift suppressed low-frequency noise and remedied average amplitude. Finally, the high signal areas float compared to their surroundings in the radargram, and were converted to average envelope curves (Figure 2a). The envelopes can estimate the configuration of soil cracks, and the abnormal signals, defined as cracks on the radargram, were used to calculate the soil crack densities at the study site. Total amplitude area (TotAra) is the area enclosed by the envelope curve and the coordinate axis. Accumulative amplitude (Acuma) is the accumulation of the amplitude at every 1 cm soil depth on the envelope curve (The crack signal illustration method is shown in Figure S2).

2.3.2 Image processing and data analysis of preferential flow

The captured images were cropped, and the threshold was adjusted using Photoshop 2020 (Adobe Systems Inc., San Jose, CA) for further analysis. The stained area was replaced with black (0) and the non-stained area was replaced with white (255) via the colour replacement function. Separation was then conducted using Image Pro Plus 6.0 (Media Cybernetics Inc., Rockville, MD) to obtain only 0 and 255 dual-value matrices after bitmap analyses were done.
Dye coverage (DC ) is defined as the ratio of the stained area to the total soil vertical profile, and the area was measured using a matrix converted from pixel grids and was calculated as follows:
, (1)
where D is the number of matrix values of 0 andND is the number of matrix values of 255.
Many historians have argued that the matrix flow depth (UF ) is defined as the depth at which the stained area ratio is greater than 80% (Schaik, 2009; Bargues Tobella et al., 2014). In contrast to preferential flow, water infiltration in the matrix flow region is relatively uniform and continuous. Matrix flow often leads to a delay in the preferential flow.
The preferential flow ratio (PF ) is defined as the ratio of the stained area of the preferential flow area to the entire vertical profile; a high PF value leads to a high degree of preferential flow. This can be calculated as follows:
, (2)
where W is the width of the Plexiglas column (W = 20 cm) and DT is the total stained area of the soil profile (cm2).
The length index of preferential flow (Li ) describes the heterogeneity of the dye penetration, which is higher in the preferential flow region than in the matrix flow region. A high Li value leads to a high degree of preferential flow. This can be calculated as follows:
, (3)
where n is the number of vertical soil layers in the soil profile (one pixel down along the vertical profile named one layer, the size of which is determined by the number of pixels contained in the maximum infiltration depth), and Dc(i+1) andDci are the stained area ratios corresponding to soil profile layer i +1 and i , respectively.