Structurefrommotion with multiview stereo (SfM-MVS) facilitates the acquisition of photos by unmanned aerial vehicle (UAV) to generate the DEM and has been gradually applied to research on surface change processes such as channel erosion and slope erosion. Improving the DEM accuracy of the runoff plot is essential for investigating the interrill and rill erosion process by using SfM-MVS 3D reconstruction technology. The runoff plot 3D reconstruction process using Agisoft PhotoScan was optimized to improve DEM accuracy. The study was carried out in the bare land plot (4.5m wide and 20.0m long) at the Hailun Soil and Water Conservation Monitoring and Research Station of Chinese Academy of Sciences. Before the ploughing in spring of 2019, totally 7 and 30 measured points were evenly distributed around the outside and inside the plot, and the 3D coordinates were measured by using differential global positioning system (dGPS). Using the Phantom 4 Pro UAV, 42 photos covering the runoff plot were taken, and the shooting height was 3.5m. Each photo covered the width of the runoff and the UAV was gradually moved along the runoff to take the next photo. The results proved that potentially DEM accuracy improvement existed for the commercial SfM-MVS 3D reconstruction software by optimizing parameters. Within the PhotoScan, the appropriate processing setting parameters were determined by evaluating the image observation accuracy of the tie point and the ground measurement point, in order to weaken the DEM’s overfitting to the tie points or the ground measurement point. For the default parameter, the check point to control point error ratio was 1.95, and the optimized value was 1.26, which was reduced by 35%. Then, the camera models in PhotoScan were evaluated and the results showed that the 3rd model could reduce the error ratio between the check points and the control points, although the difference among different camera models was small. After the optimization process implementation, the DEM error was reduced by about 40%, both by point verification or point cloud verification method. Compared with the DEM error (20.0mm) under the default parameter setting, the optimized check point error was reduced to 11.0mm, which was similar with the rill erosion depth standard (depth greater than 10mm). Therefore, the optimized SfM-MVS 3D reconstruction process of runoff plot was more suitable for rill erosion research.