A comparison of change detection measurements using object-based and pixel-based classification methods on western juniper dominated woodlands in eastern Oregon

• Main Authors: Ryan G. Howell, Steven L. Petersen
• Format: Article
• Language: English
• Published: AIMS Press 2017-03-01
• Series: AIMS Environmental Science
• Subjects: western juniper; object-based image classification; pixel-based image classification; remote sensing; juniper encroachment
• Online Access: http://www.aimspress.com/environmental/article/1352/fulltext.html

Encroachment of pinyon (<em>Pinus</em> spp) and juniper (<em>Juniperus spp.</em>) woodlands in western North America is considered detrimental due to its effects on ecohydrology, plant community structure, and soil stability. Management plans at the federal, state, and private level often include juniper removal for improving habitat of sensitive species and maintaining sustainable ecosystem processes. Remote sensing has become a useful tool in determining changes in juniper woodland structure because of its uses in comparing archived historic imagery with newly available multispectral images to provide information on changes that are no longer detectable by field measurements. Change in western juniper (<em>J. occidentalis</em>) cover was detected following juniper removal treatments between 1995 and 2011 using panchromatic 1-meter NAIP and 4-band 1-meter NAIP imagery, respectively. Image classification was conducted using remotely sensed images taken at the Roaring Springs Ranch in southeastern Oregon. Feature Analyst for ArcGIS (object-based extraction) and a supervised classification with ENVI 5.2 (pixel-based extraction) were used to delineate juniper canopy cover. Image classification accuracy was calculated using an Accuracy Assessment and Kappa Statistic. Both methods showed approximately a 76% decrease in western juniper cover, although differing in total canopy cover area, with object-based classification being more accurate. Classification results for the 2011 imagery were much more accurate (0.99 Kappa statistic) because of its low juniper density and the presence of an infrared band. The development of methods for detecting change in juniper cover can lead to more accurate and efficient data acquisition and subsequently improved land management and monitoring practices. These data can subsequently be used to assess and quantify juniper invasion and succession, potential ecological impacts, and plant community resilience.