OBJECT-ORIENTED CLASSIFICATION OF REPEAT AERIAL PHOTOGRAPHY FOR QUANTIFYING WOODLAND EXPANSION IN CENTRAL NEVADA

摘要

Pinyon-juniper woodland expansion in the Great Basin has been widely documented,but with little quantitative information concerning rates and spatial patterns of expansion.Our study quantified overall rates of woodland expansion over a 30-year period for a 25km2 area in central Nevada.Aerial photos from 1966 and 1995 were used to quantify changing woodland structure.Following orthorectification and pre-processing,an object-oriented multi- resolution segmentation and classification scheme was adopted to classify tree cover at two spatial scales:single-tree and patch level.Trees were differentiated from non-trees on the basis of member functions utilizing information on brightness,patch shape,patch area,distance,textural homogeneity,and local neighborhood relationships.Random validation points were used for georeferencing and accuracy assessment.A pixel-based,unsupervised ISODATA classification was also performed and compared with the object-oriented classification,using the same set of random validation points.Results show substantial woodland expansion,although the rate of expansion depends upon the spatial resolution.Most expansion consists of single trees and small clusters filling in former openings.Similar overall accuracy and Kappa values were obtained for object-oriented and pixel-based classifications,although the object-oriented classification provides some distinct advantages.This study provides an understanding of the usefulness of object-oriented approaches for classifying tree cover patterns in xeric woodland and savanna ecosystems,which should help managers develop strategic plans for rangeland restoration in the future.