The Effect of Satellite Imagery's Spatial Resolution on Shrub Land Cover Classification and Habitat Metrics in the Eastern United States

Authors: Lee Ann Nolan*, West Virginia University, Aaron Maxwell, West Virginia University, Tim Warner, West Virginia Unversity
Topics: Remote Sensing, Land Use, Biogeography
Keywords: shrub, shrubland, classifcation, spatial resolution, land cover classification, habitat metrics
Session Type: Paper
Day: 4/3/2019
Start / End Time: 4:30 PM / 6:10 PM
Room: Stones Throw 3 - Mica, Marriott, Lobby Level
Presentation File: No File Uploaded


Accurate classification of shrubland in the eastern United States has become increasingly important due to population declines of shrubland-dependent species. While shrubland has been classified in land cover products like the National Land Cover Database (NLCD) and the NLCD’s Shrubland Products, shrubland in the eastern United States has consistently displayed significantly lower user’s and producer’s accuracy than other land cover types due to its ephemeral nature as opposed to shrubland in the western U.S. Eastern shrubland is difficult to classify because it follows a gradient of change and is not homogenous. This research focuses on how spatial resolution affects shrubland classification.

Shrubland in the eastern United States typically results from disturbance rather than environmental factors. Disturbances including fire, wind, and utility right-of-ways are often smaller than 1 ha in size, which is equivalent to just 11 Landsat pixels, and thus shrublands are also dominated by mixed pixels.

To determine the effect of spatial resolution on shrubland classification, this research resampled 2011 Geoeye (2m) imagery was resampled to coarser resolutions. The imagery at each resolution was classified using both Random Forests and SVM into 4 classes: shrubland, forests, grassland, and other. Vegetation plot data recording the presence of grasses, shrubs, and trees at over 200 plots was used to train using 70% of the data and 30% was used to validate the model. Shrubland classifications were compared among the resolutions. Finally, these classified images were used to determine the impact of spatial resolution on shrubland habitat metrics using Fragstats.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login