Multiclass relevance vector machine classification to explore annual and seasonal dynamics of an Invasive reed

Abstract:

Phragmites Australis forms dense stands which shade native vegetation and alter the ecosystem. Information on annual and seasonal dynamics of this plant contributes to the decision support system of wetland management. The study area is the Bear River Migratory bird refuge (BRMBR) which encompasses the Bear river and its delta where it flows into the northern part of theGreat Salt Lake, Utah. Seasonal change detection was carried out between the months of June 2010 and September 2010. The imagery from June 2011 and July 2011 were used for annual change detection. The remote sensing data was acquired by AggieAir, an unmanned aerial vehicle (UAV) platform, flown autonomously via pre-programmed flight plans at low altitudes to limit atmospheric effects. This UAV acquires high resolution multispectral images in the visible, near-infrared and thermal bands and has a flight interval of about 30 minutes. The reflectance values of the classes in wavebands 550, 650 and 850 nm were used to train the Multiclass relevance vector machine (MCRVM) model developed to classify the imagery of study area. There were a total of 5 classes: water, phragmites australis, marshy land, mixed vegetation and salt flats and three attributes. The multiclass classification accuracy achieved for June 2010, September 2010 and July 2011 were 95.2%, 95% and 98.7% respectively. The seasonal change detection indicated an average increase of 17% in area of phragmites and annual change detection results indicated an average increase of 110% from June 2010 to July 2011. It's astonishing rate of increase in distribution and abundance was alarming.