Authors: Gustavo Lopes Queiroz*, University of Calgary, Gregory J McDermid, University of Calgary, Mustafizur Rahman, University of Calgary, Julia Linke, University of Calgary, Guillermo Castilla, Canadian Forest Service, Man Fai Wu, University of Calgary
Topics: Remote Sensing, Physical Geography, Geographic Information Science and Systems
Keywords: remote sensing, machine learning, random forest, classification, GEOBIA, woody debris, vegetation recovery, reclamation, human footprint, boreal forest, lidar, point cloud
Session Type: Paper
Start / End Time: 1:10 PM / 2:50 PM
Room: Buchanan, Marriott, Mezzanine Level
Presentation File: No File Uploaded
The boreal forest in Alberta is home to the woodland caribou, a species that is listed as threatened by the federal government. Seismic lines (i.e. petroleum exploration corridors) have been shown to have negative effects on caribou in many direct and indirect ways. Coarse woody debris (CWD) is an important forest-restoration target on boreal seismic lines and other industrial-disturbance features. When present at optimal densities, CWD can be effective at limiting human and predator movement on the landscape, and creates valuable microsites for seedling growth. Monitoring restoration programs and assessing CWD quantity and quality on seismic lines is important to both government and oil and gas companies. This study aims to develop a novel approach to automated CWD detection and measurement using high-resolution aerial images via geographic object based image analysis (GEOBIA) in conjunction with machine learning classification, while assessing the influence of different input variables and environmental conditions on classification accuracy. Our results show high levels of accuracy (95% completeness and 93% correctness) when training objects are present in the application area, and good accuracy (89% completeness and 96% correctness) when the application area is geographically detached from the training dataset.