Authors: Karim Malik*, , Colin Robertson, Wilfrid Laurier University
Topics: Geographic Information Science and Systems
Keywords: Arctic vegetation, map segmentation and classification, object-based image analysis, neural networks
Session Type: Paper
Start / End Time: 10:00 AM / 11:40 AM
Room: Grand Ballroom A, Astor, 2nd Floor
Presentation File: No File Uploaded
The warming of the arctic tundra ecosystems leads to the loss of permafrost, altering vegetation communities and related eco-hydrological regimes. Given the high spatial heterogeneity inherent in Arctic tundra landscapes, monitoring vegetation communities using the state-of-the-art computer vision techniques that mimic human judgment may provide a tool for developing updated maps of vegetation and changes to landscape patterns over large areas undergoing this ecological transformation. However, obtaining high resolution imagery over vast regions is costly and time consuming, while available moderate resolution imagery lacks sufficient spatial detail to discern spatial patterning and vegetation cover classes. In this research, we explore the use of a hierarchical multi-scale image segmentation method to link high resolution field imagery, unmanned aerial vehicle (UAV) image scenes, and moderate resolution imagery for a low Arctic tundra site in Northwest Territories, Canada. By learning and testing a combined segmentation and classification approach based on support vector machines we quantified and characterized the possible dominance of vegetation classes at sites unobserved by high-resolution imagers. Different scales expose diverse composition and configuration of vegetation classes in the landscape. Further research using multi-source/resolution imagery in a machine learning analytical framework will be vital to enhancing understanding on the rate of change in the emergence and colonization activity of disparate vegetation clusters across the Arctic tundra.