Land Cover Accuracy Assessment for COGAP

A land cover map for Colorado was completed in 1997 by the Colorado Gap Analysis Project(see Land Cover for more details).
"While this map is a data layer constructed for COGAP habitat modeling, it also can be used for other applications by scientists and land managers. To ensure that all users understand its strengths and limitations, thematic accuracy assessment using aerial videography was initiated in 1997."

Objectives of the Accuracy Assessment

Four main goals were developed using Hay's(1979) "five general questions that should be addressed to understand classification accuracy."
Our goals were:

1. Develop a sampling scheme that maintains statistical rigor and: a) estimates the dominant land cover in each heterogeneous sample unit (polygon), b) represents the collection of individual cover types mapped across Colorado and c) can by analyzed within the constraints of budget and time.

A. Explore the efficiency of video sampling for achieving these sampling goals by modeling sampling scenarios. 2. Use contingency table analyses based on reference data from interpreted air videography to calculate overall map accuracy, producer's accuracy, user's accuracy and the Kappa statistic for the Colorado land cover map. A. Determine the above for the Gap primary land cover type and the Anderson level land cover classes. 3. Modify the fuzzy accuracy measures proposed by Gopal and Woodcock (1994) to apply them to a large project like the assessment of the Colorado land cover map. Calculate and interpret the functions MAX, RIGHT, DIFFERENCE, MEMBERSHIP, CONFUSION and AMBIGUITY. A. Determine the above for the Gap and Anderson level classifications. 4. Evaluate the utility of aerial videography as a tool for compiling reference and training data in Colorado and other analogous Rocky Mountain and northern plains areas. A. Explore the accuracy of the interpretation of the videography for each of the land cover types.
B. Examine relationships between interpreter confidence in interpretation of individual video frames and matches of interpreted frames with mapped polygons.
C. Make recommendations for future use of air video in these areas.

Identification Key
"Interpretation of air video requires interpreter training based on identification of video frames on the ground. A key was constructed that included prints of video frames (both zoom and wide angle) and corresponding photographs and descriptions of the actual sites seen in video frames. "

"When sites were located with confidence, a brief description was recorded in a field notebook, significant features and vegetation visible on the video prints were labeled directly on the prints, and the GPS coordinates of the center of the frame were noted, for comparison with the GPS coordinates from the aircraft. Photographs were taken of each site and of the context of the site. Whenever possible multiple sites were visited for each sampled cover type to capture the range of variation in the types. Some types (e.g. open water) that are easily identified in the video required fewer site examples than more heterogeneous types. "

"The training site information was assembled into a 3-ring binder that included the video prints, prints of the sites taken in the field and written descriptions of the sites. The information was organized by cover type, along with the type definition from the land cover mapping. In some cases photographs of individual plants were also included. This was especially important for forest trees, where canopy shape was an important consideration for the video interpretation."

Photographs used in the identification key can be viewed from a table of land cover types, or from locational maps of major land cover types (bare/water cover & anthropological uses, forest and woodlands, shrublands, or grasslands)

For more detail on the accuracy assessment, refer to Page 20 of Chapter 2 of the:
Colorado Gap Analysis Project: A Geographic Approach to Planning for Biological Diversity,
Schrupp et al., 2001
or
"Statistical Evaluation of the Wyoming and Colorado Landcover Map Thematic Accuracy Using Videography Techniques" (Reiners et al., 2000)

Land Cover - COGap Home