The height of the concrete plate-insert sand fence (CPISF) is 2 m. Each section is made of seven horizontal straight-installed plates (SIPs) (Figures 2A,B). The length, thickness, and height of the SIPs are 1.50, 0.05, and 0.25 m, respectively. There are eight rectangular holes uniformly distributed across each SIP, with each hole being 0.25 × 0.05 m in size. The porosity of the SIPs is 25%; however, as there are gaps between the SIPs after it is assembled as a fence, the overall porosity of the CPISF is increased to 30%. In each wall section, six gaps exist between the seven SIPs, with each gap being 1.50 × 0.05 m in size. The row spacing between the double-row CPISFs placed at the forefront of the wind-blown sand protection systems along the Lanzhou-Xinjiang High-Speed Railway is 74 m (equivalent to 37 H, where H is the fence height).

Experiments were conducted in a wind tunnel at the State Key Laboratory of the Desertification and Aeolian Sand Disaster Combating, Gansu Desert Control Research Institute. The non-circulating blow-type wind tunnel has a total length of 38.9 m and consists of air inflow, power, stabilization, shrink, test, adjustable, and diffusion sections. The test section is 16 m long and has a cross-sectional area of 1.2 × 1.2 m. The expected wind speed in the wind tunnel ranges from 4 to 35 m s⁻¹, and the boundary layer in the test section has a maximum thickness of 0.5 m (Wang et al., 2018).

To simulate the CPISF more realistically, we used the SIPs (Figure 3A) as the models in our wind tunnel experiments. Two groups of models were tested: 1) a single row of SIPs and 2) a double row of SIPs with a row spacing of 9.25 m (37 H). To simulate the Gobi surface, gravels sampled from the field were spread out on the floor of the tunnel test section. Free-stream wind velocities (U) of 10, 15, 20, and 25 m s⁻¹ were selected for the measurements of wind velocity profiles in this study. The wind profiles were measured using an array of pitot-static tubes at nine heights, from 1 to 40 cm, above the wind tunnel floor; the pitot-static tubes array was moved throughout the profile along the direction of inflow, and 27 measurement points were selected around the sand fences (Figures 3B,C). For each point where wind velocity was measured, the duration of wind data acquisition was 30.0 s, with a data acquisition interval of 0.5 s. The wind data were time-averaged and interpolated via bilinear interpolation to reconstruct the flow field around the sand fences using Origin 9.6 drawing software. The relative wind velocity (u/u ₀) at each measurement point is the ratio of the wind velocity u to the reference velocity u ₀, which is the inlet velocity at the corresponding height. The following equation represents the wind velocity reduction efficiency (R _w ) around the sand fences: R w = ( 1 − u u 0 ) × 100 % (1)

We observed the wind profiles, sand transport profiles, and shapes of the sand deposition on the windward and leeward sides of the double rows of CPISFs in the “Yandun” windy area, along the Lanzhou-Xinjiang High-Speed Railway.

The wind profiles were measured using 3-D ultrasonic anemometers (Wind Master Pro, Gill Instruments) at height, z = 1.0, 1.5, and 3.0 m above the Gobi surface. The ultrasonic anemometer array was moved throughout the profile perpendicular to the sand fences, and around the sand fences, 28 measurement points, x = −40, −30, −20, −10, −6, −4, −2, −1, 1, 2, 4, 6, 10, 20, 30, 40, 53, 63, 67, 69, 71, 72, 74, 76, 78, 80, 84, and 94 m, were selected (Figure 4A). Because of the placement requirements of the ultrasonic anemometer and the principle of velocity measurement, all ultrasonic anemometers were aligned in the direction of magnetic north. Therefore, the positive direction of the horizontal velocity (u) is due north, the positive direction of the vertical velocity (w) is vertically upwards, and the positive direction of the transverse velocity (v) is due west. For each measurement point, the duration of wind data acquisition was 5 min, with an acquisition interval of 1 s. A 3-D ultrasonic anemometer, installed at a height of 1 m, was used to establish a reference point at 60 m (30 H), windward of the first sand fence row. The wind velocity profiles around the sand fences were measured from 11:00 to 17:50 on 11 November 2016. The 10-min average wind velocity at a height of 2 m during the observation period varied from 9.32 to 12.15 m s⁻¹, with a mean of 10.66 m s⁻¹. The wind direction during the whole measurement period was relatively steady, and the 10-min average angles ranged from 59.33° to 76.48°, with a mean of 67.43° (ENE winds). Post-processing of data acquired from the ultrasonic anemometer conducted as described by Walker (2005). The higher acquisition frequency of 3-D ultrasonic anemometer allows the analysis of airflow turbulence, and hence, we calculated the turbulence intensity (Tu) using the following formula (Bennett and Best, 1995; Venditti and Bauer, 2005): u i ′ = u ¯ − u i (2) T u = [ 1 n ∑ i = 1 n u ' i 2 ] 0.5 u ¯ = [ 1 n ∑ i = 1 n ( u ¯ − u i ) 2 ] 0.5 u ¯ (3) where u _i ’ is the horizontal fluctuating wind velocity, u _i is the horizontal instantaneous wind velocity, and u ¯ is the average horizontal wind velocity during the observation period.

Sand transport profiles were measured using three vertically segmented sand samplers. The height and width of the sand samplers were 2.8 m and 0.03 m, respectively, with segments at 5 cm and 10 cm intervals in the lower 1.3 m and upper 1.5 m of the samplers, respectively. The first sand sampler was located 60 m (30 H) windward of the first sand fence row (Figure 4B) and was used to measure the quantity of sand transported over the gravel surface without sand fences. The second and third sand samplers were located 20 m (10 H) leeward of the first and second rows of sand fences and were used to measure the quantity of sand transported over the gravel surface after interaction with the sand fences. All sand samplers were situated perpendicular to the prevailing wind direction. The sand transport profiles were measured from January to March 2016. During the observation period, there were six intense sandstorms, and the duration of each sandstorm exceeded 18 h (Table 1). Profiles of the rate of sand transport (Q) on the windward and leeward sides of the fences were obtained by converting the masses of sand collected at each sampling height of the sand samplers into rates (kg m⁻¹). The sand trapping efficiency (T _s ) of the sand fences can be represented by the following equation: T s = ( 1 − Q t l Q t w ) × 100   % (4) where Q _tl and Q _tw are the total rates of sand transport on the leeward and windward sides of the fences, respectively.

A 3-D laser scanner (Leica Scan Station C10; detailed information on this device is described in An et al., 2018) was used to obtain point-cloud data of the shapes of sand deposits on both sides of the sand fences (Figure 4C). Multistation scanning was performed at the surveyed sites. Sphere targets were used to join and mosaic the overlapping data from multiple stations, with four sphere targets set between every two stations, limiting alignment errors to no more than 6 mm. Cyclone software was used to splice the data from different sites and eliminate the interference from point cloud data. ArcGIS software was used to establish TIN and DEM models (Figure 4D), and accurate data for the dynamic deposition of sand were obtained. Field operations were carried out on 15 September 2015 and 14 September 2016.

The airflow field in the vicinity of a sand fence determines the sand transport and deposition regime around the fence (Li and Sherman, 2015). A deceleration zone containing a small eddy field appeared in the windward airflow around the SIPs of the sand fence, while an acceleration zone emerged over the top of the fence. Additionally, a main deceleration zone with a larger eddy field appeared leeward of the fence, and a restoration zone was present behind the main deceleration zone. This was consistent with the airflow regimes identified near sand fences in previous studies (Dyunin, 1964; Plate, 1971; Judd et al., 1996; Dong et al., 2007). In theory, transported sand is deposited in deceleration zones. Our field measurements demonstrated that sand was indeed deposited in the windward and leeward deceleration zones of the double-row CPISFs and that sand dunes developed on both sides of the fences, indicating that the fences sufficiently reduced wind velocity. Eddy fields only appeared when the porosity of the sand fence was lower than 40% (Gloyne, 1954; Lee and Kim, 1999). Eddy fields appeared on both the windward and leeward sides of the SIP, which indicates that the fence induced a strong disturbance in the airflow field and that turbulence around the fence was intensified (Lee and Lee, 2012). The field observations demonstrated that turbulence was greatly intensified by the presence of the CPISFs and increased 2.7 and 4.0 times on the leeward sides of the first and second rows of the sand fence, respectively. Owing to the intense blocking effect of the low-porosity SIP, a vertical displacement flowfield was present, which formed an acceleration zone over the top of the fence. This is consistent with field observations that the upward vertical velocity increased to approximately 20 w _ref over the top of the CPISF. In addition, the direction of horizontal wind gradually became parallel to the sand fences as the wind approached the CPISFs, which indicates that a transverse displacement flow also appeared windward of the fences (Huang et al., 2013).

Traditionally, the wind-shelter effects of sand fences have been evaluated based on the wind velocity reduction efficiency (R _w ) and effective shelter distance (D _s ). The results of 3D-CFD simulations have indicated that maximum R _w decreases with increasing fence porosity (Fang et al., 2018). Our wind tunnel experiments demonstrated a limit in the maximum R _w of individual SIPs of 91%, but when they were assembled as a CPISF in the field, their maximum R _w was reduced to 78%, owing to the higher porosity of the sand fences in the final assembly. The optimal spacing between multiple rows of sand fences is determined by the D _s of the fence (Dong et al., 2006; Liu et al., 2018). Previous studies suggested that the peak value of D _s occurs with a fence porosity of 30–40%, and the optimal spacing of double-row sand fences with 30–40% porosity is approximately 5–7 H (Boldes et al., 2001; Wu et al., 2013; Fang et al., 2018). Our wind tunnel experiments demonstrated that the D _s of the SIP decreased as free-stream wind velocities increased, with an average value of 10 H. However, the spacing between the double-row CPISFs of the system installed as protection against wind-blown sand along the Lanzhou-Xinjiang High-Speed Railway is 37 H, which is much larger than the D _s of the sand fence. Consequently, the distance between the first and second sand fence is wide enough for the wind velocity to recover to velocities above the saltation threshold velocity. The impact of the first sand fence on the second sand fence is dissipated, and the two fences function independently.

In field observations, T _s values on the leeward side of the first and second CPISF fences were 72.43 and 62.97%, respectively. These observations raise two questions. First, why was the T _s on the leeward side of the second CPISF fence lower than that of the first? On the one hand, sand dunes developed in front of and behind the fences provided ramps for sand to blow over each fence. Furthermore, as discussed above, the spacing between the two fences is much larger than the D _s of the sand fence, allowing the wind velocity between the two fences to recover above the saltation threshold velocity. Thus, sand on the ground surface between the two fences becomes entrained by the wind and transported downwind. Because of these two effects, the rate of sand transport downwind of the second fence exceeded that downwind of the first fence. Second, why was the R _w of the CPISF high but its T _s low? Most previous studies indicate that the maximum T _s of a sand fence is obtained with a porosity of 40–50% (Manohar and Bruun, 1970; Bofah and Al-Hinai, 1986; Zhang et al., 2004), and fences that are spaced too closely will change the wind direction, causing relatively low T _s (Savage and Woodhouse, 1968). In the current study, wind velocities were adequately reduced because of the low porosities of the CPISF; however, the large size of the openings in the fences allowed more sand particles to be transported through them and reduced the T _s of the fences (Savage, 1963; Savage and Woodhouse, 1968; Manohar and Bruun, 1970; Lee and kim 1999).

Wind tunnel experiments by Hotta and Horikawa (1990) demonstrated that sand is deposited on both sides of a sand fence when its porosity exceeds 10% and that dune deposition leeward of the fence occurs more intensely than that on the windward side. In addition, sand fences with 40% porosity trap larger volumes of sand, and lower porosity sand fences are more easily buried by dunes. Our field measurements showed that dunes developed on both sides of the double-row CPISFs and that the leeward dunes were larger than the windward dunes. The leeward dunes at the first and second fences grew to 75 and 50% of the fence height in 2016, respectively, and dune deposition continued at rapid rates each year. The growth of the sand dunes around the CPISFs is expected to decrease the sand trapping efficiencies of the sand fences (Ning et al., 2020), and they may gradually lose their sand-blocking capacity once the deposited dunes grow to 80% of the height of the fence (Hotta and Horikawa, 1990).