Core Science Analytics and Synthesis - USGS-NPS Vegetation Characterization Program
The USGS-NPS vegetation map at Theodore Roosevelt National Park (THRO), completed in March 2000, produced via contract with the Bureau of Reclamation Remote Sensing and GIS Applications Group, is being used in support of an ongoing project that is demonstrating the use of NASA AVIRIS (Airborne Visible and Infrared Imaging Spectrometer) imagery for detecting and mapping invasive leafy spurge in the park and its immediate surroundings. Leafy spurge is a highly aggressive and difficult to control herbaceous exotic plant. It was first introduced at the park's western boundary in 1966 and now occupies approximately 10 percent of the total area of the South Unit. It grows in extensive, dense monocultures, often occupying entire sections of stream floodplains or smaller drainages.
Although the plants grow in dense stands, openings in the canopy can expose litter and soil, adding to the challenge of mapping via remote sensing.
A team of scientists from the NPS, USGS, and the University of California, Davis have been analyzing AVIRIS data collected over the park on July 6, 1999 to examine the feasibility of detecting and mapping leafy spurge via imaging spectroscopy. The recently completed USGS-NPS Vegetation Characterization Program map at THRO has proven to be a valuable aid in this research because leafy spurge was one of the vegetation types included in the mapping process. Because the vegetation mapping products were in digital form, the leafy spurge class was rapidly and easily extracted from the overall vegetation database for comparison and overlay with the AVIRIS data.
Although the leafy spurge in the new USGS-NPS vegetation map was interpreted from 1996 aerial photographs, it was recent enough to be of great value to the remote sensing research team for locating probable leafy spurge spectra directly from a georeferenced version of the AVIRIS data. The only other source of information was a 1993 map of leafy spurge, produced by the park in earlier efforts to generate baseline data to be used in control efforts. The more recent USGS-NPS THRO vegetation map represented a significant update to this original baseline map.
The leafy spurge class as represented in the more recent USGS-NPS vegetation map also proved to be a useful resource for the leafy spurge research team because it served as a basis for setting thresholds on AVIRIS classification "rule" images that portray probabilities of leafy spurge occurrence. By comparing the probability image with leafy spurge polygons from the vegetation map, thresholds can be determined and set for the probability images, taking the "guesswork" out of this process.
In the image to the left, the black areas are classified AVIRIS spectra with a threshold applied to approximate the red polygons that represent leafy spurge stands from the new NPS vegetation map.
Direct overlay of the two totally independent data sets have also produced some additional side benefits. Because of an ongoing program at the park to control leafy spurge populations some polygons on the 1996 map do not correlate with the AVIRIS classification, adding an additional variable to the verification process. Assuming that the AVIRIS classification is accurate, areas that do not correlate may represent locations where control efforts were successful. On the other hand, areas on the AVIRIS classification showing leafy spurge that are not represented on the 1996 map probably represent locations of spread during the past 3 years, with allowances for some errors in classification.
Comparing the georeferenced AVIRIS data and the USGS-NPS vegetation map also provided some valuable information on the ground referencing accuracy of each of these products. By checking the vegetation map polygons against recently produced Digital Orthophoto Quarter Quads (DOQQ?s), it was determined that most of the locational error was in the AVIRIS data, and not the USGS-NPS vegetation map. Armed with this knowledge, some allowances can be made when comparing the two data sets during the process of selecting spectra within the AVIRIS data for further classification refinements. This process also has demonstrated the need for using localized "rubber sheeting" algorithms for better georeferencing of the AVIRIS data.
The text and photos were kindly provided by Dr. Ralph Root, USGS, former lead investigator on this project.