OOS 68-5
High resolution, high frequency imaging to track the evolution of stress in agro-ecosystems
By allowing us to make high resolution, landscape-scale measurements more frequently than current satellite or manned aerial platforms allow, unmanned aerial vehicles (UAVs) can fill important gaps in our current ecological knowledge. However, realizing this tremendous potential depends as much on developing manageable image processing workflows as it does on flight and sensing technologies. We developed a method to automatically ortho-mosaic and co-register time series of UAV-collected imagery, using a combination of AgiSoft PhotoScan Professional and the open-source Orfeo Toolbox. We apply this procedure to detect the spread of a fungal pathogen and other stress factors in a perennial fruitcrop.
Results/Conclusions
We collected imagery (near-infrared, blue, and green) from 100 m altitude over a ~15 hectare area on ten dates separated by two week intervals over the course of the growing season. We use our automated image processing method to create ortho-rectified scenes for each date in the time series, with a between-date effective resolution of <30 cm. Using an image classifier to detect vegetative death caused by a fungal pathogen in the first set of images, we measure the extent and rate of change in area of individual blight locations over the time series. We then correlate the rate and extent of spread with varying fungicide treatments applied within a randomized complete block design (20 blocks, 5 treatments, 7-8 replicates each). The co-registered time series allows changes in blighted area to be detected when the rate is greater than two-three times the effective resolution during the imaging interval (approximately 0.02-0.03 m/day), as well as significant differences in the extent and rate of blight spread between treatments and controls.
More recent work demonstrates the utility of this methodology for creating time series of ~90 cm resolution 5-band ortho-mosaics of visual, near-infrared, and thermal imagery. We evaluate the usefulness of this higher spectral resolution data for detecting fungal pathogen spread in the same study location, as well as spatial heterogeneity in stress and growth due to other factors, such as uneven nutrient management practices.