Content uploaded by D. Terrance Booth
Author content
All content in this area was uploaded by D. Terrance Booth on Dec 23, 2013
Content may be subject to copyright.
NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
327
R E F E R E E D R E S E A R C H
L E A F Y S P U RG E
AER I AL ASS E SSM ENT OF
D Terrance Booth, Samuel E Cox,
and Deena Teel
High-resolution aerial surveys are a highly effective means for monitoring new and
dispersed invasive-species infestations across extensive areas of wildland. Invasive
species constitute a leading threat to native vegetation in wildland settings.
Monitoring and controlling these species are essential actions to preserving native-
plant community integrity and historic wildland character. Infestations continue to
increase. Monitoring with conventional ground or lower-resolution aerial data may
be problematic as these tools are of questionable value for detecting small or dis-
persed weed populations. In July 2006, we conducted a dual-camera aerial survey
acquiring 1- and 10-mm ground sample distance (GSD) imagery in the Medicine
Lodge watershed in eastern Idaho. The survey included most of the 2003 Deep Fire
Burn. Survey data were used to determine leafy spurge (
Euphorbia esula
L.
[Euphorbiaceae]) distribution on burned and unburned lands and to relate spurge
distribution to ecosystem structure, associated vegetation, and control efforts. Leafy
spurge was detected in 10% of 10-mm GSD samples versus 8% of 1-mm GSD sam-
ples. We conclude that 10-mm GSD is best for detecting leafy spurge because it opti-
mizes the balance between resolution and field-of-view. Litter was about 4% greater
where spurge was present than where it was not, and spurge occurrence was asso-
ciated with significant decreases in cover of native grasses, forbs, and sagebrush
(
Artemisia
spp. L
.
[Asteraceae]). Leafy spurge proximity to water was higher than a
random distribution would predict.
Booth DT, Cox SE, Teel D. 2010. Aerial assessment of leafy spurge (
Euphorbia esula
L.) on Idaho’s Deep Fire
Burn. Native Plants Journal 11(3):327–339.
KEY WORDS
digital images, invasive species, light sport airplane, sampling, weed control,
Euphorbiaceae
N O M E N C L A T U R E
Plants: USDA NRCS (2009)
Animals: ITIS (2009)
A B S T R A C T
ON I DAH O’S DEE P F I RE BUR N
(Euphorbia esula L.)
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
328
structure, native vegetation, and human
activities, including weed-control
efforts (see Hobbs and Humphries
1995).
METH O D S
Study Site
Our survey area, which we refer to as
the Medicine Lodge Project Area
(MLPA), was centered at lat
44°20'52''N, long 112°33'45''W, north-
west of Dubois, Idaho, over 22130 ha
(54 683 ac) of public land extending
north from Idaho’s Upper Snake River
Plain to the Continental Divide on the
Montana-Idaho border (Figure 1). The
area is administered by the Upper Snake
Field Office, Idaho Falls District, US
Bureau of Land Management (BLM). It
has an elevation mean of 2030 m (6660
ft), ranging from 1710 to 2650 m (5610
to 8695 ft) and a mean, generally south-
facing slope, of 10.9 ± 8.1%. Nine
perennial streams run through the sur-
vey area, making the area especially pro-
ductive relative to the adjacent plains.
Major land uses are livestock grazing,
recreation, and wildlife habitat. The
native Yellowstone cutthroat trout
(Oncorhynchus clarki bouvieri [Salmo-
nidae]) is present within the watershed
(IDEQ 2003) and is classified as a type-
2 sensitive species (high likelihood of
being listed under the Endangered
Species Act in the foreseeable future) by
the BLM and the Idaho Department of
Fish and Game. In August 2003, a wild-
fire known as the Deep Fire Burn (DFB)
consumed 15 378 ha (38 000 ac) of
mountain big sagebrush / Idaho fescue
(Artemisia tridentata Nutt. ssp. vase-
yana (Rydb.) Beetle [Asteraceae] / Fes-
tuca idahoensis Elmer [Poaceae]) with
dispersed infestations of leafy spurge
(Figures 1 and 2). Average annual pre-
cipitation ranges from 305 to 508 mm
(12 to 20 in) and was 112% and 136%
of normal in 2004 and 2005, respective-
ly (Hankins 2006a).
Leafy-Spurge Control Efforts
Chemical (2,4-D; picloram) and
biological (black dot spurge flea beetle
[Aphthona nigriscutis (Chrysomeli-
dae)], brown legged spurge flea beetle
[A. lacertosa (Chrysomelidae)], red-
headed spurge stem borer [Oberea ery-
throcephala (Cerambycidae)]), control
agents were used pre- and post-burn.
Some pre-burn control was sporadic
and poorly documented; but, beginning
five years before the burn (1998), con-
trol efforts were documented using geo-
graphic information systems (GIS)—
the GIS information is presented here.
Approximately 297000 beneficial insects
were released at 32 sites prior to the
DFB. The fire’s effect on the introduced
insect population was not assessed.
Post-burn, approximately 30 ha (74 ac)
were treated annually with herbicide
within the burn, yet leafy spurge
expanded in drainages and up canyon
slopes (Hankins 2006b). Biological-
control agents were released in batches
of 10 000 insects at 29 and 20 sites in
2004 and 2005, respectively, for a total
release of approximately 490 000 insects.
Aerial sagebrush seeding occurred in
2004 across 2816 ha (6958 ac). These
independent but overlapping treat-
ments resulted in the following man-
agement conditions (Figure 1): 1) no
spurge-control treatments (untreated);
2) sagebrush seeded; 3) picloram 0.56
kg/ha (0.5 lb/ac) + 2,4-D 0.8 kg/ ha; 4)
picloram + 2,4-D + sagebrush seeded. A
surfactant was added to both picloram
and 2,4-D at a rate of 1:400 solution (1
qt/100 gal). All treatments overlapped
with biological-control releases, but the
specific area of potential influence by
the released populations is not known.
Control treatments were targeted at
known leafy spurge infestation areas
(Figure 1) and were not replicated;
therefore, we used GIS analysis to iden-
tify covariant environmental factors
that differed among the 4 management
conditions. The management condi-
tions will be referenced throughout this
report.
nvasive species, taken together, are
one of the most pressing challenges
of managing the vast landscapes of
the western US. DiTomaso (2000)
reported the presence of more than 300
rangeland weeds in the US, and listed 21
as invasive alien species that cause sig-
nificant problems. He cites cheatgrass
(Bromus tectorum L.[Poaceae]), leafy
spurge (Euphorbia esula L. [Euphor-
biaceae]), and knapweed (Centaurea
spp. [Asteraceae]—particularly C. dif-
fusa Lam. and C. stoebe L. ssp. micran-
thos (Gugler) Hayek) as among the
especially noxious weeds that infest
more than 1 million ha (2 470 000 ac)
across the western US. Public wilder-
ness areas and national parks have been
judged at particular risk for new inva-
sions and most lower-elevation public
lands have established weed popula-
tions (see Parks and others 2005). Land
managers of many public and private
organizations are faced with more inva-
sive species producing more infesta-
tions but they are also often seeing
declining monitoring budgets. Yet, the
importance of monitoring as part of an
effective invasive-species management
program is well established. This was
further emphasized by the recent report
that unmanaged weed populations can
be expected to increase exponentially
within management units (Maxwell and
others 2009). Efficiency of invasive-
species monitoring for wildlands needs
improvement but conventional ground,
and lower-resolution (>1 m [3.28 ft]
ground sample distance or GSD, a
measure of image pixel coverage) aerial
monitoring, are of questionable value
for detecting small or dispersed weed
populations in the interior of large
areas. Here we report the utility of very-
large scale aerial (VLSA) imagery for
detecting small, dispersed populations
of leafy spurge across the breadth of a
large wildland area. The specific objec-
tives were to assess the information
gained from the aerial survey relative to
ground information and to relate leafy
spurge distribution to the ecosystem
I
D TE R RA N CE BO O TH AN D O T HE R S NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
329
Aerial Imagery
We conducted the VLSA survey in
2006 as a general resource inventory for
management purposes to determine the
feasibility of using VLSA imagery for
assessing ecological condition and for
extrapolating the acquired information
over the geographic area of interest.
Given known infestations of leafy
spurge, the survey was scheduled and
executed in mid-July when leafy spurge
flower bracts aided identification of the
plant in the imagery (Everitt and others
1995; Hunt and others 2004). The
methods for conducting a VLSA survey
have been previously described, includ-
ing the 2-camera system as used here
(Booth and Cox 2006, 2008, 2009).
Briefly, we acquired color digital aerial
images from a light sport airplane
equipped with: 1) a navigation system
interfaced with a Wide Area
Augmentation System-enabled Global
Positioning System (GPS) device; 2)
11- and 16-megapixel, single lens reflex
digital cameras (RGB) fitted with 100
mm f/2.0 and 840 mm f/5.6 lenses,
respectively; 3) a laser rangefinder used
as an altimeter in conjunction with
LaserLOG software (Booth and others
2006a) to continuously read and record
the airplane’s altitude above ground
level (AGL) below 300 m; and 4) asso-
ciated laptop computers and power
supply. For our general survey, the nav-
igation system was programmed to
automatically trigger the camera at
804-m intervals along 25 flight lines,
resulting in images for 1383 locations.
We also acquired images for 110 sta-
tions over 23 BLM permanent tran-
sects.
We simultaneously acquired images
having a 3.5 × 5-m field-of-view (FOV)
and 1.04-mm (0.04 in) GSD (designat-
Figure 1. Medicine Lodge study area in Idaho, near the Montana border, showing burned area, streams, and areas of no leafy spurge treatment
(control), aerial seeding, and herbicide application. Leafy spurge occurrence was qualitatively assessed from 1378 11-megapixel VLSA samples
as none (0%), low (< 30%), moderate (31–60%), or high (> 60%). Leafy spurge occurrence from 1998–2005 BLM ground surveys is shown.
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
330
ed 1-mm GSD) centered within an
image having a 29 × 43-m FOV, and
10.7-mm GSD (designated 10-mm
GSD). Images were acquired from a
flight altitude of 121 m AGL and were
initially saved as raw files and later con-
verted to 24-bit Tagged Image File
Format (TIFF) files for analysis.
Flight altitude was displayed for the
pilot on the screen of one of the laptops
storing the images, while stored data
were saved for later time-correlation
with images. The flight plan was created
by extracting coordinates of user-
defined points drawn on a digital raster
graphic in ArcView 3.3 (Economic and
Social Research Institute [ESRI],
Redlands, California), then using
Track’Air SnapXYZ flight planning soft-
ware to create the flight plan utilized in
flight by Track’Air SnapShot software.
Occurrence and
Cover Measurements
Both 10- and 1-mm GSD images
were examined for the presence of
weeds, riparian areas with and without
leafy spurge, disturbance, or other spe-
cial circumstances. For our purpose,
image samples were classified as ripari-
an leafy spurge if a stream channel was
in the image FOV. Leafy spurge abun-
dance was visually assessed from images
as low (< 30%), moderate (31 to 60%),
and high (> 60%) and, by measuring
ground cover for every 1-mm-GSD sam-
ple from each of the 4 management con-
ditions identified under the Methods sub-
section Leafy-Spurge Control Efforts
(Figure 1). Canopy cover was measured
using SamplePoint software with a 100-
point grid (Booth and others 2006b).
SamplePoint facilitates digital-image
point sampling by using single pixels as
sample points and by allowing dynamic
image magnification (“zoom”) on up to
3 monitors to help users understand the
context of the sample-point pixel. Where
the image GSD is equal-to-or-less than 1
mm, the analysis has a potential accura-
cy of 92% (Booth and others 2006b).
Here, cover was measured using 535,
1-mm GSD images from the general
survey (black dots in Figure 1, cumula-
tively covering about 0.003% of the
management unit area [0.67 ha]) for 22
ground-cover categories: bare ground,
litter, gravel, rock, brown grass, brown
shrub, brown forb, snakeweed (Gutier-
rezia sarothrae (Pursh) Britton & Rusby
[Asteraceae]), sagebrush, rabbitbrush
(Chrysothamnus viscidiflorus (Hook.)
Nutt. [Asteraceae]), bitterbrush (Purshia
tridentata (Pursh) DC. [Rosaceae]),
unknown shrub, forb, leafy spurge,
knapweed, cheatgrass, pricklypear cacti
(Opuntia polyacantha Haw. [Cacta-
ceae]), unknown herbaceous species,
water, willows (Salix spp. L. [Salicaceae]),
and (other) deciduous or coniferous trees.
Additionally, cover was measured
with SamplePoint using 1-mm GSD
images over nineteen 30.5-m fire-reha-
bilitation-plot transects and three 402-m
step-point “trend” transects. All transects
were established within the DFB perime-
ter in 2004 by BLM ground crews for
vegetation-recovery monitoring. Each
transect was covered by one to 8 VLSA
images, with each image analyzed by 2 to
4 users. Multiple SamplePoint users
allowed us to assess user precision
through coefficients of variation (CV). A
similar measure of ground-work varia-
tion was not done because of the high
cost of repeating that effort.
Spatial Analysis
ArcGIS (ESRI, Redlands, California)
was used with a digital-elevation model
and other spatial data layers to describe
image FOV by elevation, aspect, slope,
plant community, distance to road, dis-
tance to open water, and fires of record.
To assess the influence of water on
spurge infestations and on treatment
efficacy, 2 random points per ha (that is,
> 1000 random points for the 4 manage-
ment conditions; Figure 1) were gener-
ated within the burned untreated area
and distance to water measured from
each random point. Mean random-
point distance to water was then com-
pared to the mean distance-to-water of
leafy spurge-containing aerial samples
within the untreated area to determine if
leafy spurge exhibited a preference for
invasion of areas close to water.
Leafy spurge cover from 535 1-mm
GSD general-survey images (all man-
agement conditions plus burned
untreated) was compared with distance
to water, slope, elevation, and aspect to
assess the relationship between leafy
spurge infestations and physiography.
Leafy spurge cover was measured
quantitatively using SamplePoint soft-
ware and qualitatively using a 0 (no
infestation) to 3 (completely infested)
rating system. Relationships were
assessed using correlation coefficients
(r; Microsoft Excel, Redmond,
Washing-ton) and multivariate analy-
sis to determine the quadratic response
surface (proc rsreg; SAS v. 9, SAS
Institute, Nashville, Tennessee)
A second spatial analysis was con-
ducted specifically to assess the effect
of distance from water on cover of
grass, forbs (except leafy spurge), leafy
Image resolution Road Fire scar Stock trail Gravel pit Riparian Leafy spurge Riparian leafy spurge
10-mm GSD 136 11 86 3 91 139 18
1-mm GSD 10 06 0 51 105 —
TABLE 1
Image count by ground feature for both 10-mm ground sample distance (10-mm GSD) and 1-mm GSD image resolutions. There were 1359 1-mm and
1378 10-mm GSD images examined for this analsyis. Feature counts may overlap.
D TE R RA N CE BO O TH AN D O T HE R S NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
331
spurge, and shrubs. Cover data from the 535 general-survey
images analyzed with SamplePoint were stratified into bins of
80- or 160-m increments from water and the values were com-
pared using one-way analysis of variance (proc glm, SAS). T-
tests comparing upland-vegetation cover means with and with-
out leafy spurge were run using Prism 5.0 (GraphPad, San
Diego, California).
A third spatial analysis was conducted to investigate the rela-
tionship between biological control and leafy spurge. We used
COSTools (Chasen 2005) in ArcMap to calculate the distance from
each of the 535 general-survey images analyzed to the closest bio-
logical-control point, and then used this distance as an indicator of
biocontrol influence for selected aerial samples. Correlation coeffi-
cients between distance to closest biological-control point × leafy
spurge cover were calculated.
RESU LTS AND D ISCU S S I ON
Ocurrence Measurements
We obtained VLSA samples (aerial images) at 2 resolutions
(10-mm and 1-mm GSD) from 1383 general-survey sample
locations. Leafy spurge was detected in 139 upland and 18
riparian 10-mm GSD samples (10%) and 105 of the 1-mm
GSD samples (8%) (Table 1). From 1998 to 2005, BLM field
crews detected 214 separate infestations. Though leafy spurge
plants showed up more clearly in the 1-mm-GSD imagery
(Figure 3), detection capability was lower owing to its smaller
FOV (P= 0.03, n = 1383), and was poorly correlated with the
10-mm GSD imagery (r = 0.55). The lack of correlation in
physical-feature detection between resolutions was also true for
roads, fire scars, stock trails, gravel pits, and riparian areas (r <
0.25; Table 1). Thus, we conclude that a qualitative assessment
of physical features or leafy spurge presence/absence is best car-
ried out using 10-mm GSD (wide-FOV) imagery. Eighteen 10-
mm GSD images were classified as riparian with leafy spurge.
Less-conspicuous weeds like spotted knapweed may require the
higher resolution for positive identification. Spotted knapweed
and cheatgrass were not detected from either image resolution,
although they are known to be present within the survey area.
Leafy spurge was scattered throughout the MLPA, with the
heaviest infestation in the southern portion (Figure 1). Of the
1383 VLSA samples collected, 322 samples fell within the 4 man-
agement conditions identified under the Methods subsection
Leafy-Spurge Control Efforts (Figures 1 and 4). Of these, the
untreated and aerial-seeded treatments have >100 samples, while
the remaining treatments have < 42 samples and thus are of
questionable value for deducing a treatment effect. Leafy spurge
occurrence was lowest in the herbicide + seeded area (6.0%) and
highest in the herbicide treatment area (21.4%, Figures 1 and 4).
Figure 2. Deep Creek Bench following the Deep Fire Burn, August
2003 (A). Approximately the same area in August 2005 before the
June 2006 aerial survey (B). Reprinted from Hankins (2006a,b) with permission.
Figure 3. Leafy spurge captured in 10-mm GSD (left) and 1-mm GSD
(right) images. The teal rectangle shows coverage of the nested 1-
mm GSD image.
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
332
Cover Measurements
The cover measurements from the
general survey provide an overall char-
acterization of vegetative conditions.
Upland sites with leafy spurge had
more litter, rock, and bitterbrush with
less brown grass, sagebrush, and forbs
(P< 0.01; Table 2). Other measured
cover types did not differ. Green-forb
cover represented the largest difference:
24% in samples without leafy spurge,
but only 13% in samples with leafy
spurge, a finding also reported by
Kazmer and Marrs (2005). Lesica and
Hanna (2009) reported that a plant-
diversity increase after 14 y of leafy
spurge control was largely due to
increased forb richness. Rock (stones >
13 cm [5.1 in] diameter) accounted for
3% of ground cover where leafy spurge
did not occur and 10% where it did
occur (P< 0.001). This finding is in
agreement with the report of Mitchell
and Glenn (2009) that leafy spurge in
this area is associated with rock out-
crops. We suspect this association is due
to high soil moisture near rock out-
crops as a result of water runoff (Figure
3). The greater litter cover on sites with
leafy spurge may reflect a buildup or
mulching effect of spurge material; or,
it may result from the tendency of leafy
Figure 4. Frequency plot for percentage of images within each management condition showing leafy spurge cover in 10% cover classes. The
inset table details leafy spurge cover and percentage occurrence from 1-mm GSD imagery within each management condition.
TABLE 2
Cover means (percentage) and standard deviations for select categories measured from upland
images taken within the Deep Fire Burn with leafy spurge (LS+) and without leafy spurge (LS–).
Welch’s (to control for heteroscedasticity) t-test mean comparison p-values are shown. LS+ n = 99;
LS– n = 122. Ground cover was obtained by measuring the selected cover categories from every 1-
mm GSD sample in each management combination (Figures 1 and 4) using SamplePoint software
(Booth and others 2006b).
Cover category LS+ LS– p
Bare ground 5.4 ± 6.6 6.6 ± 8.3 0.228
Litter 6.8 ± 7.0 2.4 ± 2.6 <0.001
Brown grass 7.0 ± 8.0 11.4 ± 9.4 <0.001
Brown shrub 1.5 ± 3.4 1.9 ± 2.9 0.402
Brown forb 1.0 ± 5.4 3.2 ± 3.7 <0.001
Green grass 22.8 ± 15.4 23.2 ± 12.4 0.806
Gravel 3.4 ± 5.2 2.4 ± 3.6 0.137
Rock 10.1 ± 11.0 3.1 ± 6.0 <0.001
Snakeweed 0.3 ± 1.0 0.2 ± 0.6 0.108
Sagebrush 5.5 ± 8.3 9.7 ± 12.6 <0.001
Rabbitbrush 0.5 ± 2.6 0.2 ± 1.1 0.275
Bitterbrush 1.2 ± 3.1 0.1 ± 0.3 <0.0001
Green forb 12.9 ± 14.3 24.3 ± 14.0 <0.001
Leafy spurge 16.6 ± 24.6 0 ± 0 na
D TE R RA N CE BO O TH AN D O T HE R S NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
333
spurge to occupy wetter, more productive microsites.
Sagebrush cover was lower in burned areas with leafy spurge
(P< 0.001; Table 2), and while we can’t establish a cause-and-
effect relationship, it may prove important that leafy spurge is
associated with reduced sagebrush seedling cover after a fire.
Image analysis of 22 transects by multiple users resulted in
CV means of 14 for grass, 27 for forb, 59 for shrub, 20 for lit-
ter, 78 for soil, and 63 for rock, indicating that precision across
users is not very high for most categories (Table 3).
Corresponding precision measurements for field-collected
point-intercept data are not available for this project. Data
from other non-image-based methods indicate that ground
methods have their own precision shortcomings (Booth and
others 2005, 2008; Cagney and others, unpublished manu-
script); however, these are rarely reported because of the cost
of assessing precision when using conventional ground meas-
urements. The CVs suggest historical comparisons would best
be made by having evaluators do their own SamplePoint
analyses on historical and current imagery so that the same
interpretations are used for all years.
Spatial Analysis
Using untreated sample locations within the burned area,
we determined that leafy spurge occurrence was significantly
closer to water than a set of random points (P< 0.001), indi-
cating that the threat of leafy spurge invasion increases in
proximity to riparian areas (moist soils; Figure 5). The finding
is consistent with Kazmer’s and Marrs’ (2005) observations
from their 6-y study that, generally, leafy spurge cover was
highest in riparian areas and that spurge is particularly per-
sistent in riparian areas even when treated with biological-
control agents. This finding is also consistent with our
hypothesis that water runoff accounts for the association
between leafy spurge and rock.
Distance to water, slope, aspect, and elevation had little to
no correlation with either leafy spurge cover or level of infes-
tation (r < 0.51, n = 535). Aspect had a particularly weak rela-
Figure 5. Distance to water (mean ± 95% CI) from a set of random
points (2 points/ha; n = 55,390) and from VLSA untreated, burned
sample locations showing leafy spurge (Spurge Plots; n = 18).
tionship (r < 0.04, n = 535). Results from multiple regres-
sions of physiographic variables differed depending on which
leafy spurge metric was compared. Slope and elevation (P<
0.05, n = 535), as well as distance to water (P= 0.09, n = 535),
were significant factors related to leafy spurgecover, although
the correlations indicate that the influence of these factors
was weak. The significance of all 3 factors was much higher
when compared with the qualitative level of infestation (P<
0.001, n = 535). Distance from water had no effect on forb or
shrub cover (P= 0.46 and 0.67); however, grass cover was
lower closer to water (P= 0.025, n = 535; Figure 6). Using
unpaired t-tests we found that grass cover was 8.5% lower 0
to 80 m from water compared to > 80 m from water (means
= 28.2 [n = 121] as compared with 36.6 [n = 415]; P< 0.001).
Leafy spurge cover was essentially divided into 3 groups of 0 to
80 m, 80 to 320 m, and > 320 m (means = 9.06 [n = 121], 3.58
[n = 257], and 1.78 [n = 178]; P= 0.001]). We strongly suspect
that the depression of grass cover in the first increment from
water is due to the abundance of leafy spurge within that dis-
tance. A similar conclusion was supported by the cover meas-
urements reported by Kazmer and Marrs (2005).
Leafy spurge cover was highest nearest areas of biological-
control release (Figure 7), with leafy spurge cover > 10% with-
in 400 m of a release point, > 5% within 600 m of a release
point, and < 1% more than 1400 m from a release point. In the
absence of a suitable control, the effectiveness of the biological
control cannot be determined. Leafy spurge is high in the
vicinity of the biological control, but this is probably more a
function of the BLM releasing biological-control agents in
areas with established, high-density leafy spurge populations.
What is apparent is that leafy spurge populations remain unac-
ceptably high—a finding in agreement with Mitchell and
Glenn (2009). This is particularly true near the release sites;
however, dramatic reductions in, and in some cases eradica-
tion of, leafy spurge is reported to have occurred in some areas
of the MLPA as a result of treatment with biocontrol agents
(Wright 2009).
Management Implications
Maxwell and others (2009) simulated invasion rates of a
hypothetical plant species to make objective comparisons
among 4 monitoring/management strategies and a no man-
agement (control) treatment, for a hypothetical 20-y period
and management area. The management area was created
with initial populations (15 or 30) distributed at random in
the central portion or along a road of a 100 ×100-cell map of
the management area. Simulations resulted in growing the
number of populations by spatially expanding from the ini-
tial source populations. Short-distance dispersal was
assumed. The management strategies were: 1) managing a
fixed number of populations at random each year (early
detection rapid response [EDRR] random); 2) managing an
Random points
Leafy spurge plots
Distance to water (m)
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
334
equivalent number of populations along a road each year
(EDRR road); 3) managing half of the fixed populations that
were determined by monitoring to be sources of new popula-
tions (monitoring every year); and 4) managing an equivalent
set of source populations only on even years, leaving the odd
years for monitoring (monitoring every other year). They
report that simulated metapopulations with no management
increased exponentially from 15 or 30 initial populations in
the management area. Given equal financial resources, devot-
ing funds to annual monitoring at the expense of management
resulted in greater population reduction, long-term, than
applying all resources toward management of random popula-
tions alone. They write further that:
When detection was allowed to occur earlier in the inva-
sion . . . , some management strategies could consistently
drive the metapopulation to extinction or set it on a trend
to extinction. The EDRR that restricted management to
roadsides was the least effective management strategy for
reducing the number of populations, regardless of the ini-
tial conditions or parameter value changes. This result was
disconcerting considering the prevalence of management
restricted to roadsides (emphasis added).
We stress that Maxwell and others (2009) are not arguing
against roadside monitoring—roads are known corridors of
invasion. Rather they, and we, emphasize that roadside moni-
toring must be accompanied by methods that address the
whole area of concern. They note that early detection means
finding the species while it occurs with low frequency—but
that the lower the frequency, the more costly it is to detect.
Figure 6. Leafy spurge and grass cover in 80- or 160-m increments up to 1100 m from water source, mean ± 95% confidence interval (n = num-
ber of aerial samples and is the same for leafy spurge and grass).
Figure 7. Predictive map of leafy spurge cover based on cover analy-
sis of 500 VLSA image samples within the Deep Fire Burn. Biological-
control release points are shown.
% Leafy spurge cover
Leafy spurge Grass
D TE R RA N CE BO O TH AN D O T HE R S NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
335
Systematic sampling of wildland vegetation is recognized as
usually superior to random sampling (Mueller-Dombois and
Ellenberg 1974, 39), but it may require greater sampling inten-
sity for equal accuracy (Wei and Chen 2004). A comparison of
leafy spurge detection points for 8 y of BLM ground monitor-
ing with our 1-time aerial survey (Figure 1) illustrates both the
advantage of a systematic aerial survey (> 30 new populations
identified) and a need for greater aerial-sampling intensity (aer-
ial detection was 73% of the number detected by the 8-y ground
effort). Although an unknown number of ground-detected
populations may have been eliminated before the aerial survey,
we doubt that number is great enough for us to change our rec-
ommendation for higher VLSA sample density in future aerial
surveys. Blumenthal and others (2007) recommended that for
toadflax in the mixed-grass prairie of Wyoming, a VLSA sam-
pling intensity capturing 0.5% of the area of interest in image
fields-of-view would be appropriate. The MLPA survey, con-
ducted a year before the Blumenthal recommendation, had an
order of magnitude lower sampling rate, capturing only 0.05%
of the area of interest (10-mm GSD images).
A further point for consideration can be deduced from
other leafy spurge data collected in 2006 along Medicine
Lodge Creek (Mitchell and Glenn 2009). A ground survey of
leafy spurge was conducted in late July and early August 2006,
using fifty-six, 168.25-m2(1716-ft2) plots, with some upland
overlap of the MLPA. Leafy spurge was detected in 96% of
their ground samples. Our VLSA survey detected leafy spurge
in only 21% of samples. The difference is most likely due to
differences in survey designs and location since their survey
concentrated on high spurge infestations and covered only
16% of the area of our larger, predominantly upland (less
spurge) survey. We concede the possibility that we may have
TABLE 3
Coefficients of variation among users for manual image analysis (SamplePoint) of VLSA images acquired over permanent transects.
Transect Number images Users Grass Forb Shrub Litter Soil Rock
R08 3420 48 9 31 99 48
R11 23617 28 24 43 22
R14 3414 46 49 36 49 41
R17 2318 53 39 22 35 31
R24 2311 29 87 9 26 60
R25 3312 14 135 5 158 115
R26 24914 98 26 100 108
R27 3327 14 173 18 80 53
R29 23919 88 24 89 65
R31 33810 46 16 107 87
R33 3211 7 34 1—141
R35 329335 15 8 20
R38 3222 2 141 4 20 10
R39 23917 22 38 160 125
R40 3214 10 35 15 141 141
R42 2236 62 33 16 47 52
R43 12224 11 14 —0
R45 3219 30 96 25 47 80
R46 32371 47 35 49 28
T7 8324 23 51 29 173 87
T8 8317 21 11 27 62 25
T9 7312 50 20 20 68 57
Mean 14 27 59 20 78 63
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
336
missed some vegetative-stage (no
showy bracts) plants. That possibility
serves to emphasize our position that,
as much as possible, people analyzing
VLSA images should have substantial
ground experience in the survey areas.
The findings of Maxwell and others
(2009), particularly the failure of road-
only based monitoring and the demon-
strated capability of the systematic VLSA
sampling used for the MLPA and in
Wyoming (Blumenthal and others
2007), combine to suggest the impor-
tance of VLSA-type surveys for detecting
low-frequency or dispersed populations
in wildland settings. The importance of
developing aerial methods for rangeland
surveys has long been recognized. West
(1999) observed, in a context of ground-
cover measurements, that:
I see no hope that traditional me-
thods of monitoring . . . on the
ground, will be able to accomplish
those (monitoring) needs . . .
especially when landscape and
regional perspectives are required.
There are simply not enough ade-
quately trained people and that
approach would not be afford-
able, even if the necessary profes-
sionals existed (emphasis added).
West’s observation is true whether
the monitoring objective is ground
cover or weeds. We are unable to pro-
vide a side-by-side comparison of on-
the-ground and VLSA survey monitor-
ing costs; however, available cost infor-
mation for each method is presented in
Table 4. (Note that BLM costs may lump
monitoring and control efforts.)
The aerial survey methods used here
provide information about the resource
and about management results; howev-
er, great care must be exercised in draw-
ing conclusions, as is evident in this
work where pre-existing differences in
the management conditions prevented
conclusions about weed-control treat-
ments. For example, biological-control
agents were released primarily in areas
of highest spurge density near the ripar-
ian areas (Figure 7), so even after treat-
ment, the biological-control area had
the highest spurge presence. Untreated
ground in dryer areas had less spurge. A
careless consideration of the data, and a
lack of knowledge of the resistance of
riparian leafy spurge to biological con-
trol (Kazmer and Marrs 2005), could
be interpreted to indicate biological
control increased spurge. (Similarly, the
herbicide management condition had
the greatest spurge cover [Figure 4].)
Those attempting to glean knowledge
from uncontrolled comparisons must
be fully aware of the area’s land-man-
agement history and adapt to this real-
ity by recognizing confounding situa-
tions and the extent to which valid con-
clusions can be drawn. In this study, we
attempted to normalize the physio-
graphic characteristics between man-
agement conditions to permit direct
comparisons, but we were unable to do
so to our satisfaction. Where agencies
wish to enhance the potential to gain
knowledge while implementing treat-
ments, they should plan for valid com-
parisons (Boyd and Svejcar 2009) and
recognize that the best way to test treat-
ment effect is with before-and-after
measurements to show trend. Although
trend data are most desirable, there are
some conclusions that can be drawn
using one-time-only monitoring data.
CONC L U S IONS
From this study, we conclude:
Ten-mm GSD imagery is better than
1-mm GSD for detecting the pres-
ence of leafy spurge because it opti-
mizes the balance between resolu-
tion and field-of-view.
Our sampling intensity resulted in
0.05% of the MLPA being captured
within the field of view of the 10-
mm GSD imagery. That was too low
a sampling intensity and we believe
repeat surveys should obtain an
order-of-magnitude greater sample
density.
Our data support those of Kazmer
and Marrs (2005) that leafy spurge
generally supplants other forbs, and
to a lesser extent grasses, when it
invades.
BLM costs for 3 y monitoring
Internal resource specialists and technicians ............................................ $85 200
Temporary employee for weed inventory and follow-up control .............. 28 000
Vehicle for temporary employee .............................................................. 5 600
USDA Agricultural Research Service costs
Aerial-image acquisition costs
CloudStreet .......................................................................................... $3 344
Associated costs (flight planning, ground support, travel,
data management).......................................................................... 2 015
Analysis .................................................................................................... 4 420
______________________________________________________________________________
TABLE 4
List of costs (2006 US dollars) associated with aerial and ground monitoring. These costs are not
comparable and are presented here only to provide some cost information for each method. USDI
Bureau of Land Management (BLM) costs are from Hankins (2006a,b).
D TE R RA N CE BO O TH AN D O T HE R S NAT I V E P L A N T S | 1 1 |3 | FA LL 2010
337
Sagebrush cover in the area of the DFB was lower where leafy
spurge was present. We believe this is the first evidence that
leafy spurge is associated with a reduction in sagebrush
seedling cover.
Currently, the leafy spurge infestation (cover) on the DFB is
highest in the biological-control area and lowest in the
untreated area. We draw no conclusion about the efficacy of
the biological-control treatment for the reasons discussed
above. However, the survey data are evidence that leafy
spurge has supplanted native plants on 10% of the 22130-ha
MLPA and that it has significantly reduced forage where it
occurs; thus, it remains a significant threat to the affected
native-plant communities. The survey is a basis for future
comparisons of leafy spurge cover and occurrence to deter-
mine whether control efforts are effective.
Leafy spurge proximity to water is higher than random dis-
tribution would predict. We believe this explains the fre-
quency of leafy spurge supplanting native plants around rock
outcrops away from riparian areas.
Slope, aspect, elevation, and distance-to-water, taken individ-
ually or together, appear to influence cover and distribution
of leafy spurge in the MLPA, but there is not a close associa-
tion (r) between these variables.
VLSA sampling is an effective means for systematically
acquiring high-density sampling over extensive wildland
areas. We recommend it be used with ground-based methods
for early detection of invasive species that might threaten
wildland native-plant populations.
ACKN O W L EDGM E N T S
Project funding was provided by a grant to Dr DT Booth from
the Upper Snake Field Office of the Bureau of Land
Management, US Department of the Interior, Idaho Falls,
Idaho. ARS technicians Carmen Kennedy, Pam Freeman, and
Larry Griffith; and, BLM personnel Lance Brady, Harley
Bateman, and Dawn Loomis provided technical assistance.
Aerial imagery was acquired by Joe Nance of CloudStreet Flying
Services, Fort Collins, Colorado. Dr Mark West, USDA
Agricultural Research Service, Fort Collins, Colorado, provided
statistical consulting. Drs N Glenn, Idaho State University, Boise
Center Aerospace Laboratory, and WE Limbach, Bureau of Land
Management, Pocatello, Idaho; and, Mr G Guenther, Mr J
Wright, and Mrs J Hankins-Smith, all with the Bureau of Land
Management, Idaho Falls, Idaho, provided helpful manuscript
reviews. Mention of a proprietary product does not constitute
an endorsement—to the exclusion of the other products that
may be suitable—by USDA, USDI, or the authors.
L E AF Y S P UR G E
NAT I V E P L A N T S | 1 1 |3 | FA LL 2 0 1 0
338
REFE R E N CES
Blumenthal D, Booth DT, Cox SE, Ferrier C. 2007. Large-scale aerial
images capture details of plant populations. Rangeland Ecology
and Management 60:523–528.
Booth DT, Cox SE. 2006. Very-large scale aerial photography for
rangeland monitoring. Geocarto International 21(3):27–34.
Booth DT, Cox SE. 2008. Image-based monitoring to measure eco-
logical change on rangelands. Frontiers in Ecology and the
Environment 6(4):185–190.
Booth DT, Cox SE. 2009. Dual-camera, high-resolution aerial assess-
ment of pipeline revegetation. Environmental Monitoring and
Assessment 158:23–33.
Booth DT, Cox SE, Johnson DE. 2005. Detection-threshold calibration
and other factors influencing digital measurements of bare
ground. Rangeland Ecology and Management 58:598–604.
Booth DT, Cox SE, Berryman RD. 2006a. Precision measurements
from very large scale aerial digital imagery using Image-
Measurement, Laserlog, and Merge software application.
Environmental Monitoring and Assessment 112:293–307.
Booth DT, Cox SE, Berryman RD. 2006b. Point sampling digital
imagery using ‘SamplePoint’. Environmental Monitoring and
Assessment 123:97–108.
Booth DT, Cox SE, Meikle T, Zuuring HR. 2008. Ground-cover measure-
ments: assessing correlation among aerial and ground-based meth-
ods. Journal of Environmental Management 42(6):1091–1100.
Boyd CS, Svejcar TJ. 2009. Managing complex problems in rangeland
ecosystems. Rangeland Ecology and Management 62:491–499.
Chasen R. 2005. COSTools: Multiple Minimum Distance script for
ArcMap. URL: http://arcscripts.esri.com/details.asp?dbid=13520
(accessed 16 Jun 2009). Redlands (CA): Economic and Social
Research Institute (ESRI).
DiTomaso JM. 2000. Invasive weeds in rangelands: species, impacts,
and management. Weed Science 48:255–265.
Everitt JH, Anderson GL, Escobar DE, Davis MR, Spencer NR, Aandrascik
RJ. 1995. Use of remote sensing for detecting and mapping leafy
spurge (
Euphorbia esula
). Weed Technology 9:599–609.
Hankins J. 2006a. Post-fire regeneration of mountain big sagebrush
rangeland in southeast Idaho: a case study. Poster presentation.
59th annual meeting, Society for Range Management, Vancouver,
British Columbia (abstract #147). Idaho Falls (ID): Bureau of Land
Management, Upper Snake Field Office.
Hankins J. 2006b. Deep Fire F658 year three closeout report. Idaho
Falls (ID): Bureau of Land Management, Upper Snake Field Office.
Hobbs RJ, Humphries SE. 1995. An integrated approach to the ecol-
ogy and management of plant invasions. Conservation Biology
9(4):761–770.
Hunt ER, McMurtrey JE, Parker-Williams AE, Corp LA. 2004. Spectral
characteristics of leafy spurge leaves and flower bracts. Weed
Science 52(4):492–497.
[IDEQ] Idaho Department of Environmental Quality. 2003. Medicine
Lodge subbasin assessment and TMDLs. URL: http://www.deq.state.
id.us/water/data_reports/surface_water/tmdls/medicine_lodge/med
icine_lodge_execsum.pdf (accessed 21 Jul 2008).
[ITIS] Integrated Taxonomic Information System. 2009. URL:
http://www.itis.gov/ (accessed 15 Oct 2009). Washington (DC):
Smithsonian Institution.
Kazmer D, Marrs RW. 2005. Assessing long-term impact of leafy
spurge biological control agents: conclusions from a 6-year study.
Project report to USDA-APHIS 5179 #58-5436-1-221. URL:
http://www.team.ars.usda.gov/v2/publications/WyoReport%20M
arrs2.pdf (accessed 20 Apr 2009).
Lesica P, Hanna D. 2009. Effect of biological control on leafy spurge
(
Euphorbia esula
) and diversity of associated grasslands over 14
years. Invasive Plant Science and Management 2(2):151–157.
Maxwell BD, Lehnhoff E, Rew LJ. 2009. The rationale for monitoring
invasive plant populations as a crucial step for management.
Invasive Plant Science and Management 2:1–9.
Mitchell JJ, Glenn NF. 2009. Leafy spurge (
Euphorbia esula
) classifica-
tion performance using hyperspectral and multispectral sensors.
Rangeland Ecology and Management 62:16–27.
Mueller-Dombois D, Ellenberg H. 1974. Aims and methods of vege-
tation ecology. New York (NY): John Wiley and Sons. 547 p.
Parks CG, Radosevich SR, Endress BA, Naylor BJ, Anzinger D, Rew LJ,
Maxwell BD, Dwire KA. 2005. Natural and land-use history of the
Northwest mountain ecoregions (USA) in relation to patterns of
plant invasions. Perspectives in Plant Ecology, Evolution and
Systematics 7:137–158.
USDA Natural Resources Conservation Service. 2009. The PLANTS
database. URL: http://plants.usda.gov (accessed 30 Apr 2009).
Baton Rouge (LA): National Plant Data Center.
Wei H, Chen D. 2004. The effect of spatial autocorrelation on the
sampling design in accuracy assessment: a case study with simu-
lated data. Environmental Informatics Archives 2:910–919.
West NE. 1999. Accounting for rangeland resources over entire land-
scapes. In: Eldridge D, Freudenberger D, editors. Proceedings VI, peo-
ple and rangelands: building the future; 1999 July; International Range-
land Congress. Aitkenvale, Queensland 4814, Australia. p 726–736.
Wright J. 2009. Personal communication. Idaho Falls (ID): USDA
Bureau of Land Management. Wildlife Biologist.
NAT I V E P L A N T S | 1 1 |3| FA LL 2010
339
A U TH O R I NF O R MAT IO N
D Terrance Booth
Rangeland Scientist
Terry.Booth@ars.usda.gov
Samuel E Cox
Biological Science Technician
Samuel.Cox@ars.usda.gov
USDA ARS Rangeland Resources Research Unit
High Plains Grassland Research Station
8408 Hildreth Road
Cheyenne, WY 82009
Deena Teel
Supervisory Natural Resource Specialist
Bureau of Land Management
Upper Snake Field Office, Idaho Falls District
Idaho Falls, ID 83401
Deena_Teel@blm.gov
D TE R RA N CE BO O TH AN D O T HE R S
NAT I V E P L A N T S | 1 1 |3| FA LL 2010
340
