Hyperspectral Image Analysis Algorithms for Improved Invasive Species Decision Support Tools
Lori
Mann
Bruce, Mississippi State University, bruce@ece.msstate.edu
(Presenting)
John
E.
Ball, Naval Surface Warfare Center, john.e.ball@navy.mil
Matthew
Lee, Mississippi State University, mal12@gri.msstate.edu
Paul
Evangelista, Colorado State University, paulevan@nrel.colostate.edu
Tom
Stohlgren, Colorado State University, toms@nrel.colostate.edu
Jeffrey
T.
Morisette, NASA, jeffrey.t.morisette@nasa.gov
The invasive species known as tamarisk (Tamarix ramosissima) is a major problem in the U.S.&rsquos desert southwest, where it is aggressively consuming the water supply and displacing cottonwood, willow, and other native plants. Since tamarisk can re-grow from root crown buds, even after burning, the current management practices for tamarisk involve combinations of chemical, mechanical, and biological techniques. Detection of tamarisk when it is in its earliest growing stages, through the use of remote sensing, could greatly reduce the cost associated with this invasive species.
In this NASA-funded project, researchers developed hyperspectral-based models for detecting invasives in remotely sensing imagery. These models incorporate image processing and pattern recognition techniques, including advanced hyperspectral feature extraction (discrete wavelet transforms, linear discriminant analysis, etc) and classification (maximum likelihood, nearest neighbor classifiers, etc). Quantitative verification and validation of these algorithms, in terms of target detection accuracies and false alarm rates, were conducted via a testbed of hyperspectral signatures, where the ground truth, in terms of tamarisk presence or absence, is known.
The project was part of an on-going collaboration between Mississippi State University, Colorado State University, USGS, and NASA researchers. The ground truth (field survey data) and hyperspectral data (both handheld data and HYPERION imagery) were collected in the Grand Staircase-Escalante National Monument, in southern Utah. Experiment results were very promising. Results showed that classification accuracies of 75%-95% are possible for discriminating tamarisk from native plants when target abundances were greater than 50%, i.e. mixed pixels had at least 50% ground coverage of tamarisk.
NASA Carbon Cycle & Ecosystems Active Awards Represented by this Poster: