Identification of invasive vegetation using hyperspectral imagery in the shore of the Kinu River, Japan

Abstract

Weeping love grass (Eragrostis curvula) has become a well-established invasive species along the Kinu River, Japan and is now considered a problematic invasive weed species. The aim of this study was to map the probability of the establishment of this invasive grass in the shore of the Kinu River using airborne hyperspectral imagery. Binary logistic regression analysis was used to model the probable presence/absence of weeping love grass. This study tried entering two types of input variables, original reflectance bands and MNF (Minimum Noise Fraction) transformed bands, into the regression model. No available variable of original reflectance data was selected, but two bands of MNF were selected in the regression analysis. The final classification, using the selected MNF bands, has distinguished weeping love grass from pseudo-absence pixels with user's and producer's accuracies of 100% and 66.7% respectively. The kappa coefficient was 0.74. These results indicate that the MNF transformed hyperspectral bands are more suitable than the original reflectance data to estimate the distribution of invasive weeping love grass in the shore of the Kinu River.