UAS-derived imagery and terrain models for rangeland mapping and monitoring

TitleUAS-derived imagery and terrain models for rangeland mapping and monitoring
Publication TypeConference Paper
Year of Publication2009
AuthorsLaliberte, Andrea S., Rango A
Conference NameUnmanned Vehicle Systems (UVS) Canada Conference
Date Published11/2/2009
Conference LocationVictoria, BC, Canada
ARIS Log Number257843

Images from satellites and piloted aircraft have been used extensively for mapping and monitoring rangelands, which comprise approximately 50% of the world’s land area. Unmanned aircraft systems (UAS) are ideally suited for monitoring these vast and remote areas, and derived data can bridge the gap between coarser resolution imagery and ground-based information. Small UAS (<50 kg) offer the advantage of low operating costs, fast and repeated deployment, and low flying heights resulting in very high resolution imagery. However, image processing approaches require adaptation or customization to handle the large number of small-footprint images, account for low accuracy GPS/IMU data, and classify images from low-cost sensors into quality vegetation maps. We are presenting a proven workflow for orthorectification, mosaicking, and classification of 6 cm resolution UAS imagery acquired with a low-cost digital camera. The geometric accuracies of orthorectified image mosaics comprised of 200-300 images were in the 1-2 m range. Classification accuracies for rangeland vegetation maps ranged from 78%-92%, depending on number of classes. Digital surface models (DSM) extracted from the imagery can be used to estimate parameters for hydrologic and erosion models. Dense DSMs extracted at the pixel level have potential for deriving vegetation heights. A multispectral sensor acquiring 10-bit data in 6 narrow bands ranging from blue to near infrared is currently undergoing testing. Current FAA regulations in the U.S. limit the potential for widespread UAS image acquisition missions, but the image processing and analysis methods developed on smaller areas are scalable to larger areas.