An Evaluation of the Accuracy and Performance of Lightweight GPS Collars in a Suburban Environment
van Heezik Y (2013) An Evaluation of the Accuracy and Performance of Lightweight GPS Collars in a Suburban
Environment. PLoS ONE 8(7): e68496. doi:10.1371/journal.pone.0068496
An Evaluation of the Accuracy and Performance of Lightweight GPS Collars in a Suburban Environment
Amy L. Adams 0 1
Katharine J. M. Dickinson 0 1
Bruce C. Robertson 0 1
Yolanda van Heezik 0 1
Z. Daniel Deng, Pacific Northwest National Laboratory, United States of America
0 Funding: This study was funded by the University of Otago Zoology Department and AA was supported by a University of Otago Postgraduate Scholarship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript
1 1 Department of Zoology, University of Otago , Dunedin , New Zealand , 2 Department of Botany, University of Otago , Dunedin , New Zealand
The recent development of lightweight GPS collars has enabled medium-to-small sized animals to be tracked via GPS telemetry. Evaluation of the performance and accuracy of GPS collars is largely confined to devices designed for large animals for deployment in natural environments. This study aimed to assess the performance of lightweight GPS collars within a suburban environment, which may be different from natural environments in a way that is relevant to satellite signal acquisition. We assessed the effects of vegetation complexity, sky availability (percentage of clear sky not obstructed by natural or artificial features of the environment), proximity to buildings, and satellite geometry on fix success rate (FSR) and location error (LE) for lightweight GPS collars within a suburban environment. Sky availability had the largest affect on FSR, while LE was influenced by sky availability, vegetation complexity, and HDOP (Horizontal Dilution of Precision). Despite the complexity and modified nature of suburban areas, values for FSR (x = 90.6%) and LE (x = 30.1 m) obtained within the suburban environment are comparable to those from previous evaluations of GPS collars designed for larger animals and within less built-up environments. Due to fine-scale patchiness of habitat within urban environments, it is recommended that resource selection methods that are not reliant on buffer sizes be utilised for selection studies.
The development of Global Positioning System (GPS)
technologies in the mid-1990s has enabled the use of GPS telemetry to
investigate habitat and resource selection, space use, and
movement patterns of wildlife [1,2]. GPS telemetry overcomes
many of the disadvantages of traditional VHF (Very High
Frequency) radio-tracking, as more accurate locations can be
continuously collected regardless of season, time of day, weather
conditions, and terrain without the need for fieldworkers. It also
avoids the problem of animals modifying their behaviour due to
proximity to humans [3,4].
Despite the clear advantages, the GPS receiver is subject to two
types of error: receivers may fail to acquire the necessary satellite
signals over a pre-defined time schedule (missed data; termed fix
success rate (FSR)), and locations acquired may be spatially
inaccurate (referred to as location error (LE)). LE can result in
misclassification of habitats and/or resources in selection studies,
leading to poor management decisions regarding species and/or
habitat management [5,6,7]. These two types of error are
influenced by numerous environmental and technological factors
that can affect signal transmission from satellites to receivers .
The main environmental factors affect FSR and LE of GPS
collars by obstructing or reflecting the transmission of satellite
signals and include topography [6,9,10] and vegetation
characteristics, particularly those associated with species composition and
structural complexity (principally stem density, life form, canopy
height and cover) [4,6,10,11,12,13,14,15,16,17,18,19].
Technological factors influencing FSR and LE include the
number of satellites present and their geometric configuration
[11,20]. A three-dimensional (3-D) positional fix is obtained
when signals from four or more satellites are used, while
twodimensional (2-D) positional fixes are acquired from three
satellites [11,16]. Generally, 3-D fixes are more accurate than
2D fixes due to the higher number of satellites used to acquire
the fix [11,16]. The geometric configuration of available
satellites is represented by the Dilution of Precision (DOP,
e.g. horizontal (HDOP) or vertical (VDOP)). Low DOP values
are acquired when satellites are spaced widely apart, resulting in
smaller triangulation errors and more accurate positional fixes
(i.e. low LEs) than fixes associated with high DOP values
(representing poor satellite geometry) [7,16,21]. Other technical
factors influencing FSR and LE include weakening GPS
batteries , differences in collar brands, malfunctioning
electronics, errors in satellite clocks, and multipath signals.
The latter occurs when the GPS collar receiver acquires
multiple satellite signals due to reflection off nearby surfaces
The magnitude of error surrounding a GPS receiver in any
given environment can be estimated and related to
environmental and technical variables . Therefore, researchers
using GPS radio-telemetry on wildlife should first evaluate the
error associated with their GPS device within their specific
study habitat . The performance of GPS collars designed
for large animals has been widely evaluated in both field trials
and stationary tests performed at known locations [9,11,16,20].
However, only a handful of studies have examined the
performance of lightweight GPS collars for use on
small-tomedium sized animals [14,27,28,29], which require their own
testing, as the smaller technological GPS components may result
in differences in how these GPS collars operate compared to
larger collars .
Most research evaluating the performance of GPS collars of
varying size has been conducted in non-urban environments
[7,9,11,14,16,20,23,28,30,31]. Only one study has investigated
the performance of a small GPS receiver (2936 g for
deployment on feral pigeons) within an urban environment in
the central part of a major industrial city (Basel, Switzerland;
). Urban environments are highly heterogeneous: on a
broad scale landscape components include industrial,
commercial, and residential areas as well as parks, reserves, and waste
land, while on a fine scale there can be significant patchiness
within these landscape components, with buildings and
vegetation varying in height and density. Vegetative surfaces have
already been shown to affect GPS error by blocking or
reflecting satellite signals within natural habitats , and this
multipath error is likely to be greater in urban environments
due to the presence of buildings. A large proportion of the
urban landscape is residential (between 20% and 26% of large
cities  and 36% in a smaller city ), and these suburban
habitats have been shown to support significant populations of
wildlife . Evaluation of the performance and accuracy of
GPS collars within suburban areas is therefore necessary to
provide information about the error associated with collected
fixes to enable data correction, or the implementation of
appropriate buffers in selection studies that occur in suburban
habitat . Incorporating error information within a study site
can help minimise the potential of habitat/resource
misclassification, and reduce incorrect conclusions regarding resource
selection, and subsequent poor management decisions.
This study aimed to evaluate the main environmental and
technical factors causing error (FSR and LE) in lightweight GPS
collars using stationary tests across a typical suburban environment
as a precursor to research on resource selection of the common
brushtail possum (Trichosurus vulpecula; hereafter referred to as
possum). We aimed to produce an error estimate specifically for
suburban habitats for use in selection studies. Three main
environmental variables were predicted a priori to influence GPS
collar performance in this environment: suburban areas which
differ in the complexity of vegetation present within a garden; the
degree of canopy closure (referred to here as sky availability); and
proximity to buildings. Technical factors predicted a priori to
influence collar performance included the number of satellites and
satellite geometry (HDOP).
This study was conducted with permission from householders to
conduct stationary GPS collar tests within their gardens.
Collar performance was tested in suburban gardens of Dunedin,
New Zealand (45u529S, 170u309E), with properties typically
consisting of detached single or double storey houses, and property
sizes ranging between 0.018 ha and 0.282 ha (median = 0.061;
mean = 0.07560.01 SE; van Heezik, unpublished data). Houses
are typically surrounded by vegetation, usually on all sides,
covering between 15% and 95% of the total area of the property
(median = 62%; van Heezik, unpublished data). To gain an
understanding of factors influencing collar performance in the
suburban environment, sample sites were randomly selected from
properties where subsequent GPS collaring of possums was to
occur. Suburban habitats fall into three distinguishable categories
(Res 1, Res 2, Res 3; Table 1; [35,36]) according to variations in
housing densities, garden structures, and vegetation complexity.
The properties selected represented this spectrum of suburban
residential development and would be typical of suburbs in most
Stationary Collar Tests
Three 120 g Wildlife GPS data-logger collars (Sirtrack
Electronics, Havelock North, New Zealand, http://www.sirtrack.com)
equipped with a 12-channel GPS receiver Trimble iQ to be later
deployed on possums within a suburban environment were tested.
At each site, collars were configured to acquire a location every 15
minutes in one 24 hour period, which encompasses two complete
satellite constellation cycles incorporating all possible satellite
configurations (97 possible fixes) [17,21]. Data were stored in a
built-in memory capable of storing 40,000 fixes until collar
retrieval, and included information such as the date, time,
longitude, latitude, number of satellites present, and HDOP for
each successful fix. Altitude was not directly measured by these
Firstly, to verify that all test collars were operating similarly, the
three collars were deployed simultaneously at a known survey
mark under open sky to assess the performance and accuracy of
each collar. Differences between the FSR and LE values between
the three collars were quantified using one-way analysis of
The collars were then evaluated under different environmental
conditions within the three suburban habitat types. Three gardens
were randomly chosen in each of the three suburban habitats
(Figure 1); vegetation in these gardens was evaluated to confirm
they fell within the categories defined in Table 1. In each garden,
sites representing four categories of sky availability were tested: 0
25%, 2650%, 5175%, and 76100% sky availability (n = 36
sites; 4 sites per garden). Sky availability was assessed using a
convex spherical densiometer, which provides relative estimates of
percent coverage , with coverage including both vegetation
and built structures. To determine the impact of buildings on FSR
and LE, the distance to the nearest house from the stationary
collar was measured at each of the four sites within each garden:
distances ranged from 4 m to 34 m and were equally distributed
throughout the four sky availability categories. The collars were
left at each of the four sites within a garden for 24 hours on a
30 cm high block representing the height of possums when on the
ground. For optimal reception of signals from satellites, the GPS
collar was placed with the antennae pointing directly upright
[5,12]. The true geographic co-ordinates of each site were
determined using the average of five locations recorded from a
Trimble R7 GNSS using only co-ordinates that were obtained
with more than seven satellites present, with precision criteria of
,0.015 m horizontally.
The FSR for each site was calculated by dividing the number of
collected fixes for each site by the maximum number of fixes
expected for a 24 hour period (97 fixes). LE was calculated for
each collected positional fix by calculating the Euclidean distance
between each of the collected positional fixes and the
corresponding true collar location determined using the Trimble R7 GNSS
Table 1. Descriptions of the three suburban habitat types in which lightweight GPS collars were evaluated in, Dunedin, New
Zealand, as defined by Freeman and Buck .
Residential areas with greater than one third of the property size comprised of mature, structurally-complex gardens containing an assortment of
lawns, hedges, shrubs, and large established trees. Green cover totals 70% with a mean housing density of 11.6/ha (SD = 1.98, n = 14) .
Residential areas with greater than one third of the property size comprised of structurally-less complex gardens dominated by lawns. Green cover
ranges between 4250% with a mean housing density of 12.52/ha (SD = 2.27, n = 20 suburbs) .
Residential areas with no garden or where less than one third of the property is garden dominated by flowerbeds or lawn. Green cover totals 30%
with a mean housing density of 28.6/ha (SD = 3.14, n = 6 suburbs) .
where Dx and Dy are the differences between the collected and the
true x- and y co-ordinates respectively .
Outlier LE values, which occur in all GPS receivers , were
identified as fixes falling outside of three times the standard
deviation of the mean LE value for each site, and were
subsequently removed from modelling. The root mean square
(RMS) of the LE (LERMS) was then calculated within each
vegetation and sky availability categories, and globally for all sites
as a measure of the average location error :
The LERMS estimate can assist in the selection of buffer sizes
around collected positional fixes in resulting spatial analyses for
data collected from GPS collars deployed in field trials on animals.
Additionally, the arithmetic mean (mLE) and median (m1/2LE) of
the LE under each sky availability class were calculated for
comparison with previous studies. One-way ANOVAs were
performed to determine if the FSR and LE differed between
gardens within the three suburban habitats. A two-way ANOVA
was performed to determine if LE differed between type of fix
(2D, which is calculated from three satellite signals, or 3-D, which is
calculated from four or more signals) due to the possibility that
environmental conditions might affect the proportion of 3D fixes,
which could have consequences for precision) and sky availability.
A model selection approach was used to identify the predictor
variables hypothesised a priori that have the greatest impact on
FSR and LE separately. For both analyses, data were screened
using a Spearman pairwise correlation coefficient (|r| .0.6 cut-off
value ) to avoid correlated predictor variables within the same
model. This test revealed that HDOP and the number of satellites
were correlated (Spearmans pairwise correlation test; r = 20.6,
p,0.0001), thus these two variables were not placed together in
the same model.
The influence of sky availability, vegetation complexity (as a
function of suburban habitat type), and distance to the nearest
building on FSR was assessed using fixed effects logistic regression
to model the probability of successful locations per site . Eight
models representing alternative hypotheses, including the null and
global models, were fitted to the standardised data.
LE was modelled using the same standardised environmental
predictors as FSR (vegetation complexity, sky availability, and
distance to the nearest building), as well as technical predictors
including HDOP and the number of satellites present for each
positional fix. Linear Mixed Models (LMMs) were used  to
model the global and null models and all combinations of
predictor variables, with models including a random intercept for
site to control for the non-independence of the collected positional
fixes at each site [41,42]. Additionally, to evaluate the differences
in the generated location errors (which were logarithm
transformed to meet the assumption of normality) between 2-D and
3D positional fixes, linear regression analysis was performed.
For both analyses, model selection was based on the relative
difference in Akaikes second-order corrected Information
Criterion (AICc) values, corrected for small sample sizes . The
information-theoretic approach involves the development of a set
of working hypotheses or models, from which the best model is
selected using Akaikes Information Criterion (AIC), in this case
corrected for small samples sizes (AICc). This approach is effective
in achieving strong inferences from data analyses . For the LE
analysis, we model averaged the coefficients for the fixed effects
predictor variables (sky availability, vegetation complexity,
HDOP) for the top model set comprising the models with DAIC
#2 . All modelling was performed in the statistical software
programme R .
All three collars had a FSR of 1 (100% of the positional fixes
were collected) under clear sky at the selected survey mark. There
were no significant differences in the accuracy (LE) of the three
Sirtrack lightweight GPS collars when simultaneously deployed at
the survey mark (F2,288 = 1.2, p = 0.27), with the LE ranging
between ,1.0 m and 106.3 m (median = 16.9 m).
FSRs did not differ between gardens with differing vegetative
complexity independent of sky availability (x = 90.6%; F2,33 = 0.1,
p = 0.93). The FSR decreased as the amount of clear sky
decreased, ranging from 97% when there was high sky availability,
to 81% when there was low sky availability (Table 2). Across all
gardens, 64% of the positional fixes obtained were 3-D fixes. The
top-ranked model included sky availability only (Table 3); with
FSR decreasing with decreased sky availability (increasing canopy
cover; coefficient = 20.12, SE = 0.05).
LE values were significantly different between gardens of
differing vegetation complexity (F2,3243 = 59.5, p,0.001), with
larger LEs obtained in gardens with complex, mature vegetation
(Res 1; x = 33 m) than properties characterised by lawn and flower
beds (Res 3; x = 28 m). This produced an overall average of
30.1 m for all suburban areas. Additionally, LE tended to decrease
with increasing sky availability (Table 2). The calculated mLE also
increased with increasing HDOP values, with maximum values
being reached for HDOP.10, indicating large, inaccurate values
(Figure 2). The large variation in mLE for each HDOP value and
Figure 1. Sampling locations within the suburban environment. Map of the main urban area of Dunedin depicting the sampling locations
(orange circles) of stationary GPS collar tests in relation to suburban habitats: Res 1 (light grey); Res 2 (mid-grey); Res 3 (black); and other (light green).
associated large standard deviations shows that while LE decreases
with increasing HDOP, the ranges also include some positional
fixes with similar accuracy to those associated with lower HDOP
values (Figure 2).
The magnitude of LE differed depending on whether three
satellites (2-D) or four or more satellites (3-D) were available to
generate the positional fix, with a significant difference occurring
between the type of positional fix obtained (i.e. 2-D or 3-D) and
sky availability (F1,3243 = 15.0, p,0.001): a higher proportion of
2D fixes occurred in the 025% sky availability class compared to
the other three sky availability classes. The LE associated with 3-D
positional fixes (LERMS = 26.2 m), which accounted for 64% of all
positional fixes obtained, were significantly smaller than those
associated with 2-D positional fixes (LERMS = 35.5 m;
This is the first study evaluating the performance and accuracy
of lightweight collars in suburban environments. Sky availability
had the largest effect on FSR, while LE was affected by sky
availability, vegetation complexity within different suburban
habitats, and HDOP. Despite the complexity of structures and
vegetation within suburban areas, values for FSR and LE were
comparable to those obtained in less built-up environments. This
produced a mean LE estimate of 30.1 m for suburban habitat
Within a suburban environment, FSR increased with increasing
sky availability, which is probably due to fewer objects blocking or
reflecting satellite signals [5,45]. For example, high canopy closure
can result in a 50% reduction in the FSR  due to poor satellite
visibility, which also generates a higher proportion of 2-D fixes
[10,31,47]. We also found evidence that variation in vegetation
complexity characteristic of the differing suburban habitat types,
and distance to buildings influenced FSR. However, these two
habitat variables were relatively less important compared with sky
availability. The overall average FSR of 90.6% in this stationary
suburban study was slightly inferior to overall average FSRs of
stationary studies investigating GPS performance in less built-up
habitats with variations in sky availability; e.g. 92.1% in New
Zealand farmland habitat , 93% and 99% for two brands of
collars in forest habitats , 92.8% in mountainous habitat ,
and an average FSR of 94.8% from a review of 35 articles . By
testing the collars in identical conditions, our study provided
evidence that manufacturing differences did not exist between the
F1,3243 = 39.3, p,0.001). After outlier removal, LERMS decreased
to 21.5 m for 3-D fixes, and 30.7 m for 2-D fixes, indicating that
outliers occur in both fix types. A similar number of outliers in
relation to LE were obtained within all four sky availability classes
(Table 1). Filtering the dataset to remove these outliers improved
the LERMS values by several metres for each sky availability class
(Table 1). HDOP values for 2-D fixes ranged from 1.212.7
(median = 3.7; 95% of fixes ,12.4), while the HDOP of 3-D fixes
ranged from 1.012.7 (median = 2.3; 95% of fixes ,5.0).
Additionally, linear regression analysis of the
logarithm-transformed LE and associated HDOP values verified that LE values
increased with increasing HDOP values (coefficient = 0.05,
SE = 0.002, p,2.0216). However, HDOP only explained 14% of
the variation within LE (R2 = 0.14).
Among the top six models explaining variation in LE, the two
that were within D AIC#2 values of each other included the
predictors sky availability, vegetation complexity, and HDOP
(Table 4). Model averaging of the fixed effects within these top two
models showed that HDOP had the strongest effect on LE
(coefficient = 0.13, SE = 0.001), followed by sky availability
(coefficient = 0.05, SE = 0.03), and vegetation complexity
(coefficient = 0.03, SE = 0.03). Therefore, LE increased with increasing
HDOP, canopy cover (i.e. reduced sky availability), and vegetation
complexity. Distance to buildings, which are characteristic of
urban environments, was not present in the top models indicating
that this factor was not important in influencing the variation in
LE (Table 4).
Sky Availability+Vegetation complexity
Sky Availability+Distance to buildings
Sky availability+Vegetation complexity+Distance to buildings
Vegetation complexity+Distance to buildings
K = number of parameters; DAIC = change in AIC; wi = Akaike weight.
Models were ranked based on the Akaikes second-order corrected Information Criterion (AICc).
Figure 2. Mean location error (mLE 6 SD) for each HDOP value. Mean location error (mLE 6 SD) for each HDOP value for lightweight GPS
collars across all suburban habitats and sky availability classes, Dunedin.
individual Sirtrack lightweight GPS collars (also see Blackie 
and Recio et al. ).
The accuracy of positional fixes within the suburban
environment was dependent on a combination of technical (satellite
geometry (HDOP)) and environmental (sky availability and
vegetation complexity) variables. Location error increased with
decreasing sky availability and increasing HDOP, with LE varying
between suburban habitat types of differing vegetation complexity
as predicted. However, distance to buildings did not significantly
influence LE despite the potential for satellite signals to be
reflected off building surfaces.
After filtering outliers, the LERMS decreased for all sky
availability classes, with an overall trend of smaller error values
as sky availability increased. It is therefore important to apply a
preliminary filter to collected GPS datasets to remove these fixes
[8,14]. The range of average LE values (mLE = 16.0 m to 23.8 m)
that were obtained in our study are comparable to those reported
in non-urban habitats [9,10,11,14,31,46]. This could be a
reflection of the suburban habitats containing only low buildings
which may not be significantly affecting satellite signals.
LE increased when the number of satellites used to obtain a
positional fix decreased. Other studies in a variety of habitats have
also documented an increase in error with higher HDOP values
(e.g. [7,15]); higher HDOP values are usually associated with poor
satellite configurations, such as when only three satellites are
available or when satellites are clustered . Additionally, 2-D
fixes, which made up 36% of all fixes in our study, were less
accurate than 3-D fixes, and were largely associated with
decreased sky availability. This result is slightly higher than
proportions obtained in other habitats (30.2% in farmland ;
31% in boreal forests , and 31.4% in mountainous habitat
), and is a reflection of satellite geometry and number of
satellites available (e.g. [16,31,48]). In less built-up habitats, poor
satellite configurations are associated with reduced sky availability
caused by dense canopy cover, high terrain, or physical objects
masking the sky [13,15,18,19,20]. Differences in these
environmental conditions, which influence the number of satellites
available to the GPS device, will also result in differences in the
precision of acquired 3-D fixes [14,20]. In the residential
environment, poor satellite configurations due to reduced sky
Sky availability+Distance to buildings+HDOP
Vegetation complexity+Distance to buildings+HDOP
Vegetation complexity+Sky availability+HDOP
Vegetation complexity+Sky availability+Distance to buildings+HDOP
K = number of parameters; DAIC = change in AIC; wi = Akaike weight.
Models were ranked based on the Akaike Information Criterion (AIC).
availability are likely to be associated with vegetation cover within
individual gardens, particularly in areas containing large
proportions of complex, mature vegetation, especially tall trees (i.e. Res
1). However, the results also indicate that accurate positional fixes
can be obtained for large HDOP values and for 2-D fixes.
Therefore, by using traditional filtering methods, such as
discarding 2-D fixes  or fixes with an associated HDOP$9
(e.g. ), accurate locations will also be discarded. Additionally,
the removal of all 2-D fixes can reduce the dataset significantly.
From these results, it is recommended to utilise more recent
filtering approaches, which incorporate species-specific
assumptions regarding unrealistic speeds and distances travelled between
consecutive locations  to filter collected datasets from
Distance to buildings was not an important predictor variable
affecting the FSR or LE in this study. Our finding is similar to
results from Rose et al.  who documented that LE decreased
significantly in an area with high storage buildings, but was similar
in all other urban locations tested. They also found that FSR
decreased as the amount of open sky decreased , although
there was no mention of whether sky obstruction was due to
building presence or vegetation. Therefore, the non-significant
result reported in this study may be a reflection of the density and
height of the surrounding buildings. Our research was performed
in private gardens of low-rise suburban areas of the city (i.e. one to
two storey buildings) as these areas typically make up a large
proportion of urban landscapes: 36% in Dunedin , and can
support significant populations of wildlife [34,36,50,51]. The low
heights of buildings within these suburbs may not be high enough
to reflect or block satellite signals. The opposite may be true for
high-rise areas of the city, for example the Central Business
District (CBD) and/or industrial sectors, which contain a greater
proportion of tall buildings.
Our study only evaluated stationary lightweight GPS collars and
did not incorporate the impact that animal behaviour and
movement might have on FSR and LE values. Signal acquisition
of collars has been shown to be affected by animal behaviour and
body position in that different activities, particularly denning and
foraging, can result in the orientation of the antennae being
horizontal relative to the sky, leading to lower FSRs and number
of 3-D fixes [5,52]. For example, DEon  reported that the
main source of lost data (i.e. a low FSR) is associated with animal
activity, while Lewis et al.  reported a reduction of 11% in the
FSR when collars were deployed on live animals. However, in
stationary tests the antenna is always directly vertical to the sky,
maximising signal reception, as antennae on collars are designed
in an upright position to optimise signal acquisition
[5,7,12,20,23,53]. Researchers should therefore consider the
behaviour of their study species as well as environmental and
technical factors when evaluating the performance of their collars
in the field. Possums forage both vertically and horizontally, which
may affect the FSR and LEs which is speculated to improve with
increased positional height in trees due to less blockages interfering
with the satellite signals. By evaluating collar performance close to
the ground, we took a conservative approach in calculating collar
error, as error is expected to be greater at ground-level due to
canopy cover. Additionally, by developing an error estimate
independent of animal behaviour for suburban environments, our
results have a general applicability for use in other studies
conducted in similar environments.
Determining the performance and accuracy of GPS collars in a
study location is important for making accurate conclusions from
the collected spatial data and appropriate decisions regarding
species and habitat management. We found a relatively large LE
(x = 30.1 m) was associated with lightweight GPS collars within
suburban habitats. Location error of this size is less important for
determining home range sizes, but is likely to have significant
impacts on habitat and resource selection analyses, and should be
accommodated through the use of buffers reflecting
habitatspecific LEs around each positional fix. However, because urban
environments are highly heterogeneous at a fine scale, large
buffers based on LEs can include multiple habitats or resources,
and it can be difficult to accurately identify which habitats and/or
resources animals are using. Additionally, overlapping buffers, due
to their large size, can be problematic when trying to differentiate
the predictive landscape features of available and used areas .
Unless the error of GPS devices can be reduced through better
technology, our results suggest that conclusions about resource and
habitat selection in heterogeneous suburban environments should
be made with caution, and other techniques that are not reliant on
buffer sizes, such as Brownian bridges [51,52] that incorporate
location error directly into the analysis, should be considered.
We gratefully acknowledge Mike Denham for loaning us the GPS
equipment and subsequent training in their use and Paul Denys for advice
with methodology and data analysis. For statistical assistance, we
acknowledge Sheena Townsend and Aviva Stein. We would also like to
thank Sabrina Hock for comments of the manuscript and the urban home
owners for providing access to their gardens throughout the study period.
Thank-you also to the New Zealand Federation of Women for publication
Conceived and designed the experiments: AA BR KD YvH. Performed the
experiments: AA. Analyzed the data: AA. Contributed reagents/materials/
analysis tools: KD YvH. Wrote the paper: AA KD BR YvH.
1. Hebblewhite M , Haydon DT ( 2010 ) Distinguishing technology from biology: A critical review of the use of GPS telemetry data in ecology . Philos. Trans. R. Soc. Lond ., B, Biol. Sci. 365 : 2303 - 2312 .
2. Rodgers AR ( 2001 ) Tracking animals with GPS: The first 10 years . In: Sibbald AM, Gordon IJ, editors. Tracking animals with GPS. Aberdeen, Scotland: The Macaulay Land Use Research Institute. 1-10.
3. Rodgers AR , Rempel RS , Abraham KF ( 1996 ) A GPS-based telemetry system . Wildl. Soc. Bull. 24 : 559 - 566 .
4. Rumble MA , Lindzey F ( 1997 ) Effects of forest vegetation and topography on global positioning system collars for elk . Resource Technol. Inst. Symp . 4 : 492 - 501 .
5. D'Eon RG , Delparte D ( 2005 ) Effects of radio-collar position and orientation on GPS radio-collar performance, and the implications of PDOP in data screening . J. Appl. Ecol. 42 : 383 - 388 .
6. Frair JL , Nielsen SE , Merrill EH , Lele SR , Boyce MS , et al. ( 2004 ) Removing GPS collar bias in habitat selection studies . J. Appl. Ecol. 41 : 201 - 212 .
7. Moen R , Pastor J , Cohen Y ( 1997 ) Accuracy of GPS telemetry collar locations with differential correction . J. Wildl. Manage . 61 : 530 - 539 .
8. Frair JL , Fieberg J , Hebblewhite M , Cagnacci F , DeCesare NJ , et al. ( 2010 ) Resolving issues of imprecise and habitat-biased locations in ecological analyses using GPS telemetry data . Philos. Trans. R. Soc. Lond ., B, Biol. Sci. 365 : 2187 - 2200 .
9. Cain JW , Krausman PR , Jansen BD , Morgart JR ( 2005 ) Influence of topography and GPS fix interval on GPS collar performance . Wildl. Soc. Bull. 33 : 926 - 934 .
10. D'Eon RG , Serrouya R , Smith G , Kochanny CO ( 2002 ) GPS radiotelemetry error and bias in mountainous terrain . Wildl. Soc. Bull. 30 : 430 - 439 .
11. Di Orio AP , Callas R , Schaefer RJ ( 2003 ) Performance of two GPS telemetry collars under different habitat conditions . Wildl. Soc. Bull. 31 : 372 - 379 .
12. Dussault C , Courtois R , Ouellet JP , Huot J ( 1999 ) Evaluation of GPS telemetry collar performance for habitat studies in the boreal forest . Wildl. Soc. Bull. 27 : 965 - 972 .
13. Hulbert IAR , French J ( 2001 ) The accuracy of GPS for wildlife telemetry and habitat mapping . J. Appl. Ecol. 38 : 869 - 878 .
14. Recio MR , Mathieu R , Denys P , Sirguey P , Seddon PJ ( 2011 ) Lightweight GPStags, one giant leap for wildlife tracking? An assessment approach. PLoS ONE 6: e28225 . doi:28210.21371/journal.pone.0028225.
15. Rempel RS , Rodgers AR ( 1997 ) Effects of differential correction on accuracy of a GPS animal location system . J. Wildl. Manage . 61 : 525 - 530 .
16. Rempel RS , Rodgers AR , Abraham KF ( 1995 ) Performance of a GPS animal location system under boreal forest canopy . J. Wildl. Manage . 59 : 543 - 551 .
17. Samama N ( 2008 ) Global positioning: Technologies and performance . Hoboken , New Jersey: Wiley-Interscience.
18. Stokely JM ( 2005 ) The feasibility of utilizing the cellular infrastructure for urban wildlife telemetry . Virgina: Virginia Polytechnic Institute and State University.
19. Zweifel-Schielly B , Suter W ( 2007 ) Performance of GPS telemetry collars for red deer Cervus elaphus in rugged Alpine terrain under controlled and free-living conditions . Wildlife Biol . 13 : 299 - 312 .
20. Moen R , Pastor J , Cohen Y , Schwartz CC ( 1996 ) Effects of moose movement and habitat use on GPS collar performance . J. Wildl. Manage . 60 : 659 - 668 .
21. El-Rabbany A ( 2006 ) Introduction to GPS: The Global Positioning System . Norwood, MA: Artech House Inc.
22. Gau RJ , Mulders R , Ciarniello LJ , Heard DC , Chetkiewicz CLB , et al. ( 2004 ) Uncontrolled field performance of Televilt GPS-SimplexTM collars on grizzly bears in western and northern Canada . Wildl. Soc. Bull. 32 : 693 - 701 .
23. Graves TA , Waller JS ( 2006 ) Understanding the causes of missed global positioning system telemetry fixes . J. Wildl. Manage . 70 : 844 - 851 .
24. Hofmann-Wellenhof B , Lichtenegger H , Collins J ( 2001 ) Global positioning system: Theory and practice . Wien, Germany: Springer-Velag.
25. Cargnelutti B , Coulon A , Hewison AJM , Goulard M , Angibault JM , et al. ( 2007 ) Testing global positioning system performance for wildlife monitoring using mobile collars and known reference points . J. Wildl. Manage . 71 : 1380 - 1387 .
26. Cagnacci F , Boitani L , Powell RA , Boyce MS ( 2010 ) Animal ecology meets GPS-based radiotelemetry: A perfect storm of opportunities and challenges . Philos. Trans. R. Soc. Lond ., B, Biol. Sci. 365 : 2157 - 2162 .
27. Blackie HM ( 2010 ) Comparative performance of three brands of lightweight global positioning system collars . J. Wildl. Manage . 74 : 1911 - 1916 .
28. Dennis TE , Chen WC , Koefoed IM , Lacoursiere CJ , Walker MM , et al. ( 2010 ) Performance characteristics of small global-positioning-system tracking collars for terrestrial animals . Wildl. Biol. Pract . 6 : 14 - 31 .
29. Rose E , Nagel P , Haag-Wackernagel D ( 2005 ) Suitability of using the global positioning system (GPS) for studying feral pigeons Columba livia in the urban habitat . Bird Study 52 : 145 - 152 .
30. D'Eon RG ( 2003 ) Effects of a stationary GPS fix-rate bias on habitat selection analyses . J. Wildl. Manage . 67 : 858 - 863 .
31. Lewis JS , Rachlow JL , Garton EO , Vierling LA ( 2007 ) Effects of habitat on GPS collar performance: Using data screening to reduce location error . J. Appl. Ecol. 44 : 663 - 671 .
32. Loram A , Tratalos J , Warren PH , Gaston KJ ( 2007 ) Urban domestic gardens (X): The extent & structure of the resource in five major cities . Landsc. Ecol . 22 : 601 - 615 .
33. Mathieu R , Freeman C , Aryal J ( 2007 ) Mapping private gardens in urban areas using object-oriented techniques and very high-resolution satellite imagery . Landsc. Urban Plan . 81 : 179 - 192 .
34. Goddard MA , Dougill AJ , Benton TG ( 2010 ) Scaling up from gardens: Biodiversity conservation in urban environments . Trends Ecol. Evol . 25 : 90 - 98 .
35. Freeman C , Buck O ( 2003 ) Development of an ecological mapping methodology for urban areas in New Zealand . Landsc. Urban Plan . 63 : 161 - 173 .
36. van Heezik Y , Smyth A , Mathieu R ( 2008 ) Diversity of native and exotic birds across an urban gradient in a New Zealand city . Landsc. Urban Plan . 87 : 223 - 232 .
37. Lemmon PE ( 1956 ) A spherical densiometer for estimating forest overstory density . Forest Science 2 : 314 - 320 .
38. Denys P ( 2010 ) A note on the accuracy of consumer grade GPS recievers . Survey Quarterly 62 : 20 - 21 .
39. Hosmer DW , Lemeshow S ( 2000 ) Applied logistic regression . New York : Wiley.
40. Gelman A , Hill J ( 2007 ) Data analysis using regression and multilevel/ hierarchical models . New York : Cambridge University Press . 625 p.
41. Hebblewhite M , Percy M , Merrill EH ( 2007 ) Are all global positioning system collars created equal? Correcting habitat-induced bias using three brands in the Central Canadian Rockies . J. Wildl. Manage . 71 : 2026 - 2033 .
42. Skrondal A , Rabe-Hesketh S ( 2004 ) Generalized latent variable modeling: Multilevel, longitudinal, and structural equation models . New York: Chapman and Hall . 508 p.
43. Burnham KP , Anderson DR ( 2002 ) Model selection and multimodel inference: A practical information-theoretic approach . New York : Springer.
44. R Development Core Team ( 2011 ) R: A language and environment for statistical computing (version 14.1) . Vienna, Austria. R Foundation for Statistical Computing. Available: http://www.R-project.org.
45. Sager-Fradkin KA , Jenkins KJ , Hoffman RA , Happe PJ , Beecham JJ , et al. ( 2007 ) Fix success and accuracy of global positioning system collars in oldgrowth temperate coniferous forests . J. Wildl. Manage . 71 : 1298 - 1308 .
46. Jiang Z , Sugita M , Kitahara M , Takatsuki S , Goto T , et al. ( 2008 ) Effects of habitat feature, antenna position, movement, and fix interval on GPS radio collar performance in Mount Fuji, central Japan . Ecol. Res . 23 : 581 - 588 .
47. Hansen MC , Riggs RA ( 2008 ) Accuracy, precision, and observation rates of global positioning system telemetry collars . J. Wildl. Manage . 72 : 518 - 526 .
48. Edenius L ( 1997 ) Field test of a GPS location system for moose Alces alces under Scandinavian boreal conditions . Wildl. Biol . 3 : 39 - 43 .
49. Bjorneraas K , Van Moorter B , Rolandsen CM , Herfindal I ( 2010 ) Screening global positioning system location data for errors using animal movement characteristics . J. Wildl. Manage . 74 : 1361 - 1366 .
50. McKinney ML ( 2002 ) Urbanization, biodiversity, and conservation. Bioscience 52 : 883 - 890 .
51. Pickett STA , Cadenasso ML , Grove JM , Groffman PM , Band LE , et al. ( 2008 ) Beyond urban legends: An emerging framework of urban ecology, as illustrated by the Baltimore Ecosystem Study . Bioscience 58 : 139 - 150 .
52. Augustine BC , Crowley PH , Cox JJ ( 2011 ) A mechanistic model of GPS collar location data: Implications for analysis and bias mitigation . Ecol. Modell . 222 : 3616 - 3625 .
53. Bowman JL , Kochanny CO , Demarais S , Leopold BD ( 2000 ) Evaluation of a GPS collar for white-tailed deer . Wildl. Soc. Bull. 28 : 141 - 145 .
54. White CG , Schweitzer SH , Moore CT , Parnell IB , Lewis-Weis LA ( 2005 ) Evaluation of the landscape surrounding northern bobwhite nest sites: A multiscale analysis . J. Wildl. Manage . 69 : 1528 - 1537 .