Rupasinghe and Chow-Fraser 2018 Final Report to MTO

Abstract

We tested the suitability of high-resolution (80 cm) multi-spectral satellite data from World View 3 (WV 3) to detect small patches of invasive Phragmites within 20-m buffer of the centre-line of the road. We used ENVI 5.5 to classify the image into seven classes: roads, trees, Phragmites, roadsides, ground, grass, andagriculture. We applied the Mixture-Tuned Match Filtering (MTMF) procedure to the image, which is a spectral unmixing method in which the target features could be separated out from the other background features in mixed pixels. The highest confusion with Phragmites were with grasses and agricultural lands. Accuracy of the Phragmites classification was higher for the MTMF image (81.6% producer’s and 75.6% user’s accuracy) than for the reflectance image (73.7% producer’s and 71.4% user’s accuracy), while overall accuracy was 84.4% and 74.6%, for the MTMF and the reflectance image, respectively. We conclude that WV 3 can be used in early-detection programs, as long as the procedure is applied to a relatively small area in wetlands (maximum 100 ha) or roadsides (4-km segment) to increase accuracy and publishing requirements necessary to achieve the best possible product.