Date of Award

Spring 5-2016

Document Type

Thesis

Department

Environmental Resources Engineering

Thesis Advisor

Lindi J. Quackenbush

Keywords

Invasive plants, Image processing, Remote sensing

Abstract

The impact invasive species have on native habitats can be devastating as they reduce plant biodiversity, decrease wildlife habitat, and change ecological function of an area. Lythrum salicaria (purple loosestrife), Fallopia japonica (Japanese knotweed), and Phragmites australis (common reed) are three invasive species located in Onondaga County, New York. This study aimed to test the feasibility of identifying the presence of these invasive species using readily available oblique imagery. Imagery used in the study came from Google Earth Street View imagery matched to the coordinates of locations reported to have dense populations of common reed, Japanese knotweed, or purple loosestrife. Images were loaded into ERDAS Imagine 2015 where unsupervised object-based segmentation was performed. ArcGIS was used to determine spectral, spatial, and textural characteristics for each object within the image. MATLAB was used to classify the segmented images using a classification tree based off the spectral bands, length to area ratio of the objects, and average texture of the objects. Results show that Japanese knotweed and common reed classified by these means have an overall accuracy of about 84% and 64%, respectively. Purple loosestrife was not classified during this study due to a lack of imagery corresponding with available data points. To improve this methodology, spatial resolution and capture date of photos should be optimized. Future research should include reference photos with confirmed species identification, higher quality images, and ground validation. This research is an important step towards improving invasive species detection efforts.

Share

COinS