Mapping Non-native Plants Using Hyperspectral Imagery

One of the significant threats to global biodiversity and ecosystem functioning is the spread of invasive plant species (Mooney and Cleland, 2001). Continuing anthropogenic related disturbances such as land conversion, grazing, and habitat fragmentation, combined with international trade and climate change indicate that these trends are likely to continue (Zedler and Scheid, 1988). In this context, the major challenge for land managers and ecologists is how to effectively manage non-native plants to preserve native biodiversity. Being able to delineate the spatial extent and ascertain the severity or intensity of the invasion is essential for resource management (Byers et al., 2001). This information provides a baseline for monitoring future expansion, the effectiveness of control efforts, and assists in identifying targets for control activities, such as satellite populations and ‘invasion fronts.’ Techniques such as remote sensing offer significant opportunities for providing timely information on invasions of non-native species into native habitats. To date, there have been two divergent approaches (Everitt et al., 1996, Dewey et al., 1991). The first approach uses imagery with a high spatial, but low spectral resolution, such as black and white or color infrared aerial photographs. These photographs have the benefit of being relatively cheap, large amounts of archival data are available for many sites, and photographs are available at hyperspatial resolutions (0.1–2 m). However, the major disadvantage is that they rely on the non-native plant possessing visually detectable unique characteristics, extensive manual labor for processing and, finally, the resolution means that it is only feasible to collect data over a relatively small spatial area. The second approach uses digital images with greater spectral resolution, although coarser spatial resolution have been utilized—predominantly various airborne or spaceborne multispectral instruments. The use of digital multispectral imagery offers the opportunity for automated image processing, access to recent historical data for time series analyses, and large spatial coverage. However, the spatial resolutions of these sensors, such as TM and AVHRR imagery, mean that invasive species populations can often only be detected once, dense and widespread (Carson et al., 1995). In addition, traditional classification techniques like isodata or maximum likelihood, usually identify vulnerable land cover classes, rather than the non-native species itself. Consequently, maps have only a general applicability. The availability of Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) imagery with both increased spatial, but in particular spectral, resolution offers an enhanced potential for mapping invasive species. Because of the large number of wavebands (224) image processing is able to capitalize on both the biochemical and structural properties of the target invader. However, to date, there has been little evaluation of different processing techniques suitable for imaging spectrometry data for identifying invasive plants. The objective of this research is to investigate the use of AVIRIS imagery to detect the invasive species iceplant (Carpobrotus edulis) and jubata grass (Cortaderia jubata). More specifically, to compare three techniques for processing the imagery: Minimum Noise Fraction, Continuum Removal, and Band Ratio Indices, and to critically evaluate the relative ease of processing and repeatability of each method.