Use of High Resolution UAS Imagery to Classify Sub-Arctic Vegetation Types

Abstract:

Sub-arctic permafrost regions are now experiencing annual warming with a resulting thaw that induces changes to the vegetative landscape. This warming trend is directly correlated to increases in annual greenhouse gas emissions including methane (CH₄). Vegetation species and composition are indirect indicators of CH₄ flux, and may serve as a proxy for estimating changes in CH₄emission over time. Three WorldView-2 images (2m² spatial resolution, 8 multispectral bands) were acquired in Jul/Aug of 2012-2014 over the Abisko region in northern Sweden. Color infrared (CIR) sub-meter imagery was also collected over a 4km² area in 2014 using both a multi-rotor helicopter and a fixed wing unmanned aircraft system (UAS). Fifty 1m² ground sample plots were established; these plots cover 5 major ground cover vegetation classes and were used in classification efforts. Texture analysis was conducted on both UAS and WV-2 imagery. Both an unsupervised k-means clustering algorithm to predict vegetation classes and a supervised classification using both random forests and neural networks were conducted; similar texture analysis and clustering were also performed on the UAS imagery. Classifications of the two imagery types were compared with promising results, thus supporting the use of UAS and high resolution satellite image collection to provide landscape level characterization of vegetation.