**3. Experimental Setup**

An undistorted model based on the prototype shown in Figure 2 was built with a geometry scale λ*L* = 1:80 in the State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, China. The scaled model was implemented into a water-recirculating system. The bank revetments were built in concrete and had a 0.5 m-deep non-cohesive sediment bed. The flow passed through a triangular weir for flowrate measurement before entering the model, and exited the model through the concrete grade control datum as a free flow which is the same as that of the prototype.

Froude similitude [41] was adopted for the flow motion. Thus, the velocity scale isλ*U* = <sup>λ</sup>*L*0.5 = 8.94, the discharge scale is <sup>λ</sup>*Q* = <sup>λ</sup>*L*2.5 = 1:57243 and the time scale is λ*t* = <sup>λ</sup>*L*0.5 = 8.94. In order to achieve the similitude of sediment motion, *U*/*Uc* in the prototype and model should be the same (*Uc* is the critical velocity of the sediment entrainment). The Shamov formula (Equation (1)) is commonly used for calculating *Uc* in a scale model test, as it is simple to use and can provide reliable estimations [42].

$$\frac{\mathcal{U}\_c}{\sqrt{gd}} = 1.47 \left(\frac{h}{d}\right)^{1/6} \tag{1}$$

where *g* is gravity acceleration, *h* is approach flow depth, *d* is sediment size. Based on Equation (1), the sediment size scale is λ*d* = <sup>λ</sup>2*Uc* = <sup>λ</sup>2*U* = 1:80. The model sediment was scaled down by λ*d* = 1: 80 from the prototype sediment size distribution based on a field survey in the studied river reach (Figure 3). As there is an 18 m-high grade control structure with a stilling basin (for minimizing the downstream local scour) located about 2 km upstream of the 105th Provincial Highway Bridge, the upstream sediment is blocked by the grade control structure from entering the studied river reach. Therefore, the upstream sediment replenish rate was considered as zero and no sediment was fed during the test.

**Figure 3.** Grain size distribution of the prototype and model sediment.

Six flood events (discharge *Q* = 600–4039 m<sup>3</sup>/s, occurrence probability *P* = 1%–50%) were tested for three GCDs (*z* = 527 m, 533 m, 539 m). Among which, *z* = 539 m is the crest of the current GCS, *z* = 533 m is the crest elevation of the new GCS design plan, *z* = 527 m is the bed level downstream of the GCS after the flood event on 9 July 2013.

The tests of each *z* commenced with an initially flattened sediment bed and the smallest discharge *Q*. The gradient of the initial flat bed was set at 4‰, which is the average bed gradient of the studied reach after the flood event in July, 2013. The test stopped and the water dried gradually after the scaled flood duration *t*. Then, the bed profile was measured using a Total Station (Nikon, Japan, DTM-352C). The next test, with a higher discharge, commenced without flattening of the sediment bed.

Before the formal tests, a preliminary test based on the flood event on 9 July 2013 (*Q* = 2710 m<sup>3</sup>/s) was conducted for model calibration. As this model test is aimed to assess the impacts of GCD drop on the upstream bed morphology, only the bed profile upstream of the GCS was measured for calibration. Based on a field survey after the flood event on July 9 2013, the average gradient of the studied river reach was 4‰. The measured model talweg profile of the calibration test has an average gradient of 4.3‰ (multiple correlation coefficient *R*<sup>2</sup> = 0.91), which is close to that of the prototype. The discrepancy of bed profile between the model and prototype is within ± 0.8 m (prototype vaule), which is acceptable for the large scale prototype of this study. As the scour design for instream structures normally adds a safe value greater than 1 m to the estimated scour depth, the model data is reliable for engineering use.
