Object-oriented image analysis for mapping shrub encroachment from 1937-2003 in southern New Mexico

TitleObject-oriented image analysis for mapping shrub encroachment from 1937-2003 in southern New Mexico
Publication TypeJournal Article
Year of Publication2004
AuthorsLaliberte, Andrea S., Rango A., Havstad K, Paris J.F., Beck RF, McNeely R.P., Gonzalez A.L
JournalRemote Sensing of Environment
Date PublishedSeptember 20, 20
Accession NumberJRN00414
ARIS Log Number163386
Keywordsaerial photography, desert grassland, mapping, object-based classification, satellite image, segmentation, shrub encroachment

Shrub encroachment into arid and semiarid grasslands in the southwestern United States is of concern because increased shrub cover leads to declines in species diversity, water availability, grazing capacity, and soil organic matter. Although it is well known that shrubs have increased over time, we have little quantitative information related to the nonlinear nature of this vegetation change over a particular time period. On the USDA-ARS Jornada Experimental Range and the adjacent Chihuahuan Desert Rangeland Research Center (CDRRC)(New Mexico State University) in southern New Mexico, shrub increase has been measured with various ground survey techniques extending back to 1858. For this study, we used 11 aerial photos taken between 1937 and 1996 that covered a 150-ha study area and had sufficient resolution for shrub detection. A QuickBird satellite image provided coverage for 2003. We used image segmentation and object-based classification to monitor vegetation changes over time. Shrub cover increased from 0.9% in 1937 to 13.1% in 2003, while grass cover declined from 18.5% to 1.9%. Vegetation dynamics reflected changes in precipitation patterns, in particular effects of the 1951-1956 drought. Accuracy assessment showed that shrub and grass cover was underestimated due to the constraint of the pixel size. Eighty-seven percent of all shrubs >2m2 were detected. The use of object-based classification has advantages over pixel-based classification for the extraction of shrubs from panchromatic aerial and high-resolution satellite imagery. Incorporating both spectral and spatial image information approximates the way humans interpret information visually from aerial photos but has the benefit of an automated classification routine. Combining several scales of analysis in a hierarchical segmentation method is appropriate in an ecological sense and allowed for determining shrub density in coarser level classes. Despite encountering difficulties in analyzing a greatly varying aerial photo dataset, including variability in spectral and spatial resolutions, moisture conditions, time of year of observation, and appearance of grass cover, aerial photos provide an invaluable historic record for monitoring shrub encroachment into a desert grassland.