DETERMINING THE EFFECT OF GCP QUANTITY ON DEM ACCURACY IN SUAS-BASED STRUCTURE-FROM-MOTION PHOTOGRAMMETRY

The use of small Unmanned Aerial Systems (sUAS) in combination with structure from motion photogrammetry (SfM) is an increasingly important tool for geologists. SfM allows rapid acquisition of the data required to create orthophotos and digital elevation models (DEMs) for small field areas. SfM techniques, in combination with the ability to systematically acquire the requisite sUAS imagery, is an emerging discipline and many questions remain open regarding the effect of methodological techniques on the quality and accuracy of the resulting DEMs. This study focuses on assessing the relationship between DEM accuracy and the number of Ground Control Points (GCPs) used to georeference the imagery. Deploying and surveying GCPs is often the most time consuming aspect of missions. Determining the minimum number of GCPs necessary for a specific level of accuracy and precision in the final DEM product will save time and money during data collection and processing.

We deployed 26 GCPs over a 0.60 ha field site that was chosen for its variation in relief, topography, and natural and human-made features. Of the 26 GCPs, 9 were placed in a profile down the main slope of the site and the other 17 were distributed across the remainder of the site. We used centimeter accuracy RTK GPS to locate the GCPs. After deployment and surveying of the GCPs, a flight mission with 80% side and forward image overlap at an altitude of 35 m above ground level was flown with a DJI Phantom 4 Professional sUAS with a 20 megapixel camera capturing orthogonal images. The SfM processing was completed with Agisoft Photoscan (v. 1.4). We processed the data with varying numbers of randomly selected GCPs (i.e., 2, 6, 11, and 25) to generate a set of 1 cm resolution DEMs for comparison of accuracy. Unused GCPs were treated as assessment points to evaluate the accuracy of the DEMs. The resulting DEMs were exported to ArcMap version 10.5 for analysis.

We found a minimal difference in the accuracy of the DEMs derived from 2, 6, or 11 GCPs. These DEMs yielded elevations for the assessment points that only varied from one another by an average of 1 mm (range 0-3 mm). The DEM created with 25 GCPs was slightly more accurate than the other DEMs, but the absolute difference was only about 1 cm. Our results indicate that more GCPs can result in more accurate DEMs but that the overall gains are relatively minor.