DETERMINING THE EFFECT OF MISSION DESIGN AND FLIGHT PATTERN ON THE QUALITY AND ACCURACY OF DEMS AND ORTHOPHOTOS DERIVED FROM UAV IMAGERY

Unmanned Aerial Vehicles (UAVs) are a cost-effective means of collecting imagery that can be used to create high-resolution DEMs (Digital Elevation Models) and orthophotos of field sites. UAVs allow the rapid collection of aerial images across field sites. Additionally, flight planning software allows the precise design of missions with a high-degree of control over flight pattern, camera angle, and image overlap.

Because of the ease of image acquisition and a lack of scientific literature exploring best practices in mission design, most UAV projects err on the side of obtaining large amounts of imagery to ensure a suitable final product. However, acquiring more imagery than necessary leads to considerable time and costs involved in processing the data. Understanding the minimum number of images and ideal flight plan(s) required to obtain a specific standard of DEM and Orthophoto quality is an important question.

In 2017 imagery of the effects of the April 2017 flooding in the North Fork of the White River watershed was collected at seven different stream reach corridors ranging from 2.5 to 65 hectares. At each study site, four separate image acquisition missions were flown with varying flight patterns and camera angles. In order to georeference and rectify the imagery sub-decimeter accuracy dGPS positions were collected at 5-10 visible targets distributed across each study site. In addition to the GCPs, ground assessment points were surveyed to provide known coordinates on the ground that could be compared to coordinates extracted from the produced DEMs.

The data were processed into DEMs and Orthophotos using various combinations of the imagery missions from each study site. The relative accuracy and resolutions of the resulting DEMs and Orthophotos are being compared to determine which combinations of image acquisition methods yields the best results for the least investment of time and effort. Thus far we have found a relationship between the overall number of photos used to process the data and the resulting DEM, but only up to a certain point where the rate of improvement rapidly diminished. Overall differences in the resulting DEMs have been small, some averaging around 4mm, despite instances where there were 4 times as many photos used to process the DEMs.