Plant Species Identification via Drone Images in an Arid Shrubland
Dr David Gallacher, Mr Tamer Khafaga, Mr Tamer Mahmoud Ahmed,Mr Hatem A. Shabana
The Arabian Peninsula is famous for its vast areas of desert rangelands and described by its sparse and low vegetation diversity. It is prohibitively expensive to monitor these lands by using satellite or manned aircraft due to its scale; while, ground monitoring is very labour intensive. In the present study, we aimed to assess whether low altitude aerial photography via drones stratified sampling could provide a feasible alternative to ground-based monitoring of vegetation.
Seventeen abundant species were selected and between 30 and 44 individual plants were numbered and identified by the resident botanist, and then photographed from above at 10, 30 and 100m (Ground sampling distances, GSDs, of 2, 6 and 20mm) using DJI s1000 multirotor drone with a Sony NEX7 24MB camera. Images were selected and cropped to include one sample of each labelled plant at each GSD and assigned a randomised filename, each image contained between one and 54 labelled plants. Two botanists who are familiar with the region, but not with the particular location, then classified each image to species, genus and plant group (tree, shrub, herb, grass, sedge).
The result have shown that plant species identification by drone is less accurate than ground-based assessment at any resolution, but it does provide several other benefits; of which, the collection of data from large sample areas in a short period. More data could, therefore, collected during seasons of peak growth or reproduction. Our study indicates that it would be feasible to establish a database of all perennial plants within predefined sample areas and thus aerial collects biomass estimates periodically for georeferenced perennial species.
