Spectral unmixing of airborne hyperspectral imagery for mapping giant reed infestations

Abstract

Spectral unmixing techniques applied to hyperspectral imagery were examined for mapping giant reed (Arundo donax L.), an invasive weed that presents a severe threat to agroecosystems and riparian areas throughout the southern United States and northern Mexico. Airborne hyperspectral imagery with 102 usable bands covering a spectral range of 475–845 nm was collected from two giant reed-infested sites along the US-Mexican portion of the Rio Grande. The imagery was transformed with minimum noise fraction (MFN) to reduce the spectral dimensionality and noise. Linear spectral unmixing (LSU) and mixture tuned matched filtering (MTMF) were applied to the transformed MNF imagery based on endmember spectra extracted from the imagery. The abundance images were then converted into classification maps. For comparison, spectral angle mapper (SAM) and support vector machine (SVM) were used to classify the imagery. Accuracy assessment showed that MTMF was slightly better than or similar to LSU and that SVM performed better than the other three methods. The results from this study will be useful for distinguishing giant reed from associate plant species.