Evaluating drone-based photogrammetry for morphologic mapping of a hydraulic model with a mobile bed

Physical models are valuable tools for investigating changes in river morphology while also allowing the analysis of three-dimensional processes such as scour in bends or the vicinity of hydraulic structures. An important aspect is an accurate assessment of the river bed and morphological structures that occur, which today is often based on optical measurement systems, such as LiDAR, or photogrammetric techniques like Structure for Motion (SfM). These sensors are usually mounted on tripods, requiring multiple look angles to cover the whole model, and therefore need to be manually repositioned several times. Alternatively, they can be mounted on overhead tracks, limiting the possible look angles and requiring expansive installation.

To overcome these limitations, this study utilized a drone equipped with a high-resolution camera to survey morphological bed changes of a 70 m long and up to 6 m wide physical model with a movable bed and fixed rip-rap embankments. The bed material consisted of coarse sand and fine gravel with a mean diameter of 2.1 mm. Several surveys covering a total area of 180 m² were carried out and drone-based SfM results were compared with data obtained using a terrestrial laser scanner (Leica RTC360). The DJI Mavic Mini 3 Pro drone was equipped with a 48 MP camera, featuring a 1/1.3'' CMOS sensor which captured up to 240 camera positions from three vertical angles within 30 minutes. This was a similar acquisition time required by the tripod-mounted laser scanner to cover the whole model with six setups. Post-processing, from ground control point detection and tie point matching to cloud construction and digital elevation model (DEM) generation, was automated in this study to reduce processing time.

By comparing the DEMs produced by SfM and the RTC360, it became obvious that SfM cannot only map morphological structures but also produces denser point clouds, with a mean surface point density of 127 pts/cm² compared to 75 pts/cm² by the laser scan. The mean absolute cloud-to-cloud distance for the model bed is 1.8 mm, with a standard deviation of 1.5 mm. This compares favorably to the accuracy of the RTC360 of 1.9 mm at a distance of 10 meters.

Notably, there are disagreements between the SfM model and the laser scan, especially in areas with coarser materials, e.g. rip-rap at the embankments, or areas with low-feature texture, e.g. plastic structures or smooth concrete faces. The final calculated volume differences from the resulting DEMs before and after an experimental trial also show good agreement, with a 3 % discrepancy in the volume difference.

The results of this study showed that the accuracy of drone-based SfM-generated DEMs is similar to that of an RTC360 with much lower equipment costs. Furthermore, the mobility of drones offers the advantage of achieving a wider range of look angles, which improves the quality of the resulting SfM model. Hence, the application of drone-based SfM for morphological measurements in laboratory experiments is a promising technique for a wide range of measurements of morphological processes.