2.2 Rainfall test
The simulated rainfall device was composed of a raindrop generator and a splashing raindrop collection device (Figure 1) (Fu et al., 2017). The raindrop generator was a cylinder (10 cm in diameter and 10 cm in height) with a top opening, 21 syringe needles were installed at the bottom of the cylinder, and raindrops of different diameters were generated by changing the needle size. The collection device for splashed raindrops was a stainless-steel pan (110 cm in diameter). To prevent the raindrops from being disturbed by the horizontal airflow, a shield was placed around the experimental device. The duration for testing each raindrop diameter was 10 minutes, and each diameter test was repeated 3 times.
Three rainfall intensity levels (5.76, 68.61, and 217.26 mm h-1) determined by three raindrop diameters were selected in the experiment. The corresponding raindrop diameters were 2.67 mm, 3.39 mm and 4.05 mm, respectively. The rainfall intensity was measured by the direct method. For all needles of each size, continuous rainfall was performed for 10 minutes, and the precipitation volume was determined. The rainfall intensity was calculated according to the rainfall area and rainfall time. Each test was conducted 10 times, the maximum and minimum values were removed, and the average of the remaining 8 replicates was calculated to be the rainfall intensity. When the raindrop diameter was greater than or equal to 1.9 mm, the modified Newton formula (equation [1]) was used to calculate the final velocity of the raindrops. The raindrop velocity under this test condition was calculated using equation [2]. The raindrop energy was calculated by equation [3].
d >1.9 (1)
(2)
(3)
where Vi is the terminal velocity (m s-1), d is the raindrop diameter (mm), V is the raindrop velocity (m s-1), g is the acceleration of gravity (m s-2), H is the height from which the raindrop falls (m), Ers is the raindrop energy (J m-2 s-1), n is the number of raindrops, i=0,…, and m is the individual raindrop mass (g). The main parameters of the simulated raindrops are shown in Table 1.
After the rainfall test, the soil in the cutting ring was air dried naturally. Dry clods (2 cm cubes) from the surface of the impacted soil (a depth of approximately 0.5 cm) were obtained by using a knife before and after each rainfall event. The soil clods were placed in a container with a sponge to keep the structure intact. A total of 54 dry clods were selected for CT scanning.
2.3 SR-μCT scanning and image processing
The soil clods (5 mm in diameter) were scanned using synchrotron-based X-ray microcomputed tomography (SR-μCT) at beamline BL13W1 in Shanghai, China. The scan was performed at an energy level of 24 kV, with an exposure time of 1.8 s, a detection distance of 11 cm, and a resolution of 3.25 μm. The sample stage rotates at a constant speed from 0 to 360°, and the scan is continuously performed at a scan interval of 0.625 mm; 720 original images (tomo images) were obtained for each soil sample. Then, PITRE (Phase-sensitive X-ray Image Processing and Tomography Reconstruction) software was used for phase retrieval and slice reconstruction based on the back-projection algorithm (Chen et al., 2012). A total of 1500-2000 slice images (2048×2048 pixels) were obtained for each soil clod and stored as 32-bit grayscale images in tiff format.
The open-source software ImagePy was used to complete image processing, visualization and quantification of the three-dimensional aggregate structure (Wang et al., 2018). To avoid the influences of soil depth and image boundary, a total of 512 images with serial numbers 512-1024 were selected, and a nonedge section of 512×512×512 pixels (i.e., 1.664 mm×1.664 mm×1.664 mm) in size was selected as the target area for image analysis. The process of dividing the grayscale image into the soil matrix and pores by thresholding, which is called binary segmentation, was key to the quantitative analysis of soil structure. This study used the automatic Otsu algorithm of the global threshold method (Zhou et al., 2012; Garbout et al., 2013) to perform binary segmentation which was subsequently completed by visual observation and manual adjustment. After image binarization, the three-dimensional visualization function of the ImagePy software was used to visualize the three-dimensional structure of soil aggregates (Wang et al., 2018) (Figure 2).