WDFW LogoWashington Department of Fish & Wildlife
WDFW LogoPublications

You will need Adobe Reader to view and print publications.

Get Adobe Reader
Get Adobe® Reader

Archived Publications
contain dated information
that do not reflect current
WDFW regulations or policy.
These documents are provided
for archival purposes only.


    Advanced Search
  Search Tips

Download PDF Download Document

Get Adobe® Reader

Shrubsteppe Mapping of Eastern Washington Using Landsat Satellite Thematic Mapper Data

Category: Habitat - Shrubsteppe

Date Published: August 2000

Number of Pages: 42

Author(s): John E. Jacobson and Michelle C. Snyder


Shrubsteppe provides important habitat for many wildlife species in Washington State, such as the sage grouse (Centrocerus urophasianus), sharp-tailed grouse (Tympanuchus phasiannellus), and pygmy rabbit (Brachylagus idahoensis) which are currently listed as threatened or endangered with extinction. Shrubsteppe once extended over nearly all of the non-forested land in Washington east of the Cascade Mountain Range, but now only occupies about 50% of its historical range. The continuous loss of this important habitat makes it imperative the spatial distribution and characteristics of shrubsteppe be mapped for the effective conservation and management of obligate wildlife species. The wide distribution of shrubsteppe landcover throughout eastern Washington made the data obtained from the Thematic Mapper (TM) sensor onboard the Landsat 5 satellite platform a logical and cost-effective choice for this mapping project.

An exhaustive literature review and research effort was conducted to determine an image processing methodology which would optimally discriminate between numerous shrubsteppe habitat conditions, and other landcover in eastern Washington. TM channels 3,4,5, and 7 representing the red visible, near-infrared, and two mid-infrared wavelengths, respectively, were selected because they consistently provided the most effective discrimination among shrubsteppe landcover conditions. Furthermore, the research efforts using these TM channels demonstrated multitemporal TM data provided optimal discriminatory capability over the processing of single date TM data.

An unsupervised clustering technique was applied to each Landsat TM scene to group the enormous amount of variability in the spectral TM data into a set of 175 unique spectral classes. Each of these classes represents subtle landcover variations from components such as vegetative biomass and exposed soil. Field data were collected on the landcover composition of nearly 1300 ground-truth sites which spatially corresponded with the spectral classes. Based on this ground-truth information, the image analyst assigned a spectral class to landcover class of either open water, sand dune, shrubsteppe with less than 10% shrub cover, shrubsteppe with greater than or equal to 10% shrub cover, cropland, forest/shrub, barren, or snow. Additional classes which were not derived from the TM data included palustrine wetland areas obtained from the US Fish and Wildlife Service’s National Wetland Inventory digital database, and Conservation Reserve Program (CRP) land obtained from the Natural Resources Conservation Service (NRCS).

Numerous wildlife species such as sage grouse use CRP land established with grass cover for nesting and other important habitat functions. Therefore, it is important for wildlife managers to know the distribution of CRP land over time, especially in context with the diminishing shrubsteppe habitat. CRP areas were mapped for Okanogan, Douglas, Lincoln, Grant, Adams, Franklin, Benton, Klickitat, Walla Walla, and Yakima counties by compiling and digitizing CRP field boundaries from aerial photographs. Although budget and time constraints prevented CRP mapping of the other eastern Washington counties, the nine mapped counties contained about 80% of the CRP land in eastern Washington.

Substantial spectral confusion between different landcover classes often occurs when processing TM data to obtain landcover information. This spectral confusion results when two different landcover classes have similar spectral characteristics due to similar amounts of live and dead biomass, and exposed soil conditions. Substantially less of this spectral confusion occurred when multitemporal data sets were processed using both an early spring and mid to late summer Landsat scene. Therefore, multitemporal data sets were used to develop landcover information as allowed by budget and availability of scene pairs for the same area.

Considerable effort was required by the image analyst to reduce the spectral confusion by using interpretation experience and ancillary data such as aerial photos. This editing procedure consisted of systematically checking the landcover data as viewed on a computer monitor to determine potential confusion areas. Once the correct landcover was determined for the confusion areas displayed on the screen, the image processing software was used to digitally draw polygons around these areas. Within these polygons, the data elements constituting the confusion areas were edited to a value which would represent the correct landcover class in the final geographic information system (GIS) data file.

The mapping accuracy of the TM-derived landcover classes was determined by an accuracy assessment procedure using data from the NRCS’s National Resources Inventory (NRI). Some of the significant advantages to using the NRI data for the accuracy assessment included resource savings realized by the WDFW, data collection by an independent agency, fairly even distribution of randomly selected sites throughout the study area, and information collected for sites which would have been logistically very difficult to ground reconnaissance.

Overall accuracy based on the TM-derived landcover classes used in the accuracy assessment was nearly 93%. The shrubsteppe, cropland, and forest/shrub landcover classes which comprise about 94% of the study area, had mapping accuracies near or above 90%. The accuracy assessment results suggest this mapping effort would provide effective GIS data products for many natural resource management applications. Shrubsteppe landcover as of 1993 covered only 30% of the eastern Washington landscape compared to approximately 60% historically. The diminishing extent and fragmentation of shrubsteppe makes it imperative this habitat and other interspersed landcover be monitored at least every 5-10 years. Such a mapping effort will assist in the effective management of shrubsteppe and the many wildlife species dependant upon this vital habitat.