An Accuracy Assessment of Tree Detection
Algorithms in Juniper Woodlands
Aaron J. Poznanovic, Michael J. Falkowski, Ann L. Maclean, Alistair M.S. Smith, and Jeffrey S. Evans
Abstract
This research provides a comprehensive accuracy assessment
of five methods for classifying western juniper (Juniperus occi-
dentalis) canopy cover from 1 m, 4-band National Agriculture
Imagery Program (
NAIP
) imagery. Two object-oriented classifi-
cation approaches (image segmentation and spatial wavelet
analysis, (
SWA
)) are compared to three pixel based classifi-
cation approaches (random forests, Iterative Self-Organizing
Data Analysis (
ISODATA
), and maximum likelihood). Methods
are applied to approximately 250 km
2
in the intermountain
western USA. A robust suite of statistical approaches, which
offer an alternative to traditional kappa-based methods, are
utilized to determine equivalence between methods and over-
all effectiveness. Object-oriented approaches have the highest
overall accuracy among the assessed methods. Each of the
methods varied considerably in cover class accuracy. SWA has
the highest class accuracy when juniper canopy cover is low
(0 to 40 percent cover), ISODATA performs best at moderate
cover (60 to 80 percent) and maximum likelihood performs
best at higher cover (60 to 100 percent cover).
Introduction
Western juniper (
Juniperus occidentalis
Hook.) is a tree
species native to the semi-arid shrub-steppe ecosystem of
the western United States. Low density juniper woodlands
that are subordinate or codominant with native shrubs (e.g.,
mountain big sagebrush (
Artemisia
tridentata
spp. vaseyana
)),
provide excellent habitat for 23 species of mammals and 83
species of birds (Davies
et al
., 2011; Miller
et al
., 2005; Miller
and Wigand, 1994; Poddar and Lederer, 1982). Furthermore,
3 to 5 percent of western juniper woodlands are old-growth
forests, which provide increased biodiversity, structural
diversity, high quality habitat, genetic diversity, and serve
as ecological legacies (Miller
et al
., 2005; Miller
et al
., 1999;
Waichler
et al
., 2001).
Characterizing the magnitude and rate of encroachment
of woody plants into semi-arid lands is important in further
understanding the assessment of ecological dynamics (Hunt
et al
., 2003; Miller
et al
., 2000; Sankey and Germino, 2008;
Sankey
et al
., 2010; Smith
et al
., 2008; Strand
et al
., 2008).
Western juniper historically grew in rocky refugia and other
areas that were protected from relatively frequent fire return
intervals common to the region (Camp
et al
., 1997; Miller and
Wigand, 1994). However, the intensity and frequency of fires
have decreased, allowing for expansion and establishment of
woody species like mesquite and juniper into deeper, more
productive soils (Burkhardt and Tisdale, 1976; Miller and
Rose, 1999; Waichler
et al
., 2001). Additional studies indicate
the historic range and density of western juniper has been in-
creasing since the late 1800s due to what is believed to be the
interacting effects of fire suppression, increased cattle grazing,
and favorable climatic conditions for juniper growth at the
turn of nineteenth century (Archer
et al
., 1995; Blackburn and
Tueller, 1970; Burkhardt and Tisdale, 1976; Crawford
et al
.,
2004; Johnsen, 1962; McPherson
et al
., 1988; Miller and Rose,
1999; Miller and Wigand, 1994). In eastern Oregon, it is esti-
mated that juniper has increased in range from approximately
600,000 to 2.63 million hectares since the turn of the nine-
teenth century (Azuma
et al
., 2005). Given these large extents,
developing and evaluating computer assisted classification
methods to identify and map juniper trees across large spatial
extents are critical to managing a changing landscape and
maintaining the biodiversity and ecological functioning of the
western sage-steppe ecosystem (Campbell
et al
., 2012).
Various methods have been developed to identify woody
plant encroachment over large spatial extents from remotely
sensed data. Common approaches include sub-pixel meth-
ods (Sankey and Glenn, 2011), pixel-based analysis meth-
ods such as maximum likelihood and ISODATA clustering
(Anderson and Cobb, 2004; Ball and Hall, 1965; Pillai
et al
.,
2005), contrast thresholding (Knapp and Soulé, 2002), and
texture-based assessment (Hudak and Wessman, 1998). These
approaches produce assemblies of pixels based on statistically
determined criteria that have similar spectral characteristics
(Jensen, 2005).
Object oriented image analysis (
OBIA
) methods such as
image segmentation (Laliberte
et al
., 2004; Laliberte
et al
. 2007;
Pillai
et al
., 2005), feature extraction, and object identification
(e.g., Spatial Wavelet Analysis (
SWA
)) (Falkowski
et al
., 2006;
Garrity
et al
., 2008, 2012; Smith
et al
., 2008; Strand
et al
.,
2006; Strand
et al
., 2007) are also gaining popularity in the
assessment of woody plant encroachment. Feature extraction
methods for
OBIA
utilize texture, spatial context, and reflec-
Aaron J. Poznanovic and Michael J. Falkowski are with the
Department of Forest Resources, University of Minnesota, St.
Paul, MN 55108 (
).
Ann L. Maclean is with the School of Forest Resources and
Environmental Science, Michigan Technological University,
Houghton, MI 49931.
Alistair M.S. Smith is with the Department of Forest, Range-
land, and Fire Sciences, University of Idaho, Moscow, ID 83844.
Jeffrey S. Evans is with The Nature Conservancy, Fort Collins,
CO 80524, and the Department of Zoology and Physiology,
University of Wyoming, Laramie, WY 82071.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 7, July 2014, pp. 000–000.
0099-1112/14/8007–000
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.7.000
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2014
627
