SERDP
CSTARS is working with the Strategic Environmental Research and Development Program (SERDP) to use remote sensing to detect invasive species on military grounds. The project involves two military bases, Camp Pendleton and Vandenberg Air Force Base.
The rapid spread of non-native invasive plant species, including noxious weeds, is causing irreparable damage to the natural resources on military installations. This research moves beyond the current generation of remotely sensed vegetation indices and classification algorithms to take advantage of information generated from high-resolution spectra. This project aims to develop and demonstrate a new remote sensing methodology using hyperspectral imaging (HSI) for mapping invasive weeds. The expected outcome is a predictive model of potential weed invasion integrating HSI information with a geographic information system (GIS). This methodology will have broad applicability to military installations.
Seven bases from the Southeastern, Southwestern, and Northwestern
ecoregions of the United States, each with different weed types and intensities
and patterns of environmental disturbances, were selected to demonstrate, refine,
and validate the proposed methodology. These case studies will demonstrate the
portability of the methods under various types of military activities. Airborne
flightlines will be identified to provide data for mapping various species of
weeds under the diverse conditions existing at these military bases. These data
will provide a basis for demonstrating and assessing the benefits of HSI data.
The combination of HSI tools and images will provide a robust protocol for monitoring
weed invasions. The HSI information will be integrated into a GIS database of
other site characteristics to develop a predictive model of potential for weed
invasion. New support vector machine learning tools will be utilized to characterize
the habitats and identify weeds using the HSI imagery. The Hierarchical Foreground
Background Analysis (HFBA) is one example of a multi-scale resolution analysis
that is used to link the spectral variation for each pixel with variation in
the spatial domain. The HFBA decomposition is coupled with a wavelet-based multi-scale
resolution in the spatial domain. This method addresses the following three
issues regarding spectral features which are not addressed by standard methods
of image analysis that focus on each pixel separately: (1) spectral redundancy;
(2) the span and completeness of a supervised classification; and (3) a mechanism
for producing an automatic classification.
The immediate benefit of this project will be a better understanding of the
distribution of major invasive weeds on military bases and the environmental
conditions associated with their distributions and spread. The long-term benefit
will be in developing a cost-effective method for mapping weeds that can be
used to monitor the spread of weeds to new locations.
Currently, the project is being run by five members of the CSTARS laboratory:
Note: PDF Files will require the Adobe Acrobat Reader to view, and Presentations may require Microsoft Powerpoint or the free Microsoft Powerpoint Viewer for Windows or Macintosh.
Publications
Ustin, S., E. Underwood, M. Andrews, G. Scheer, J. Greenberg, J. F. Pierce, and M. O'Niell. 2000. Application of hyperspectral techniques for monitoring and management of invasive weed infestation. In-Progress Review Meeting SERDP, Davis. PPT Format.