Vous êtes sur la page 1sur 5

Consumer-Grade Global

Positioning System (GPS) Accuracy


and Reliability
Michael G. Wing, Aaron Eklund, and Loren D. Kellogg
Our primary study objective was to test the accuracy and reliability of consumer-grade GPS receivers six consumer-grade GPS receivers collected
in a variety of landscape settings. We established three measurement testing courses in open sky, young 1,800 measurements, with 450 being taken
forest, and closed canopy settings within a conifer-dominated forest in western Oregon and rigorously
ABSTRACT

at each measurement course; these measure-


tested the positional accuracy of six different GPS. All units were produced by established GPS ments were averaged and compared to
manufacturers. We found that performance varied, in some cases considerably, among units and known locations at the measurement courses
appeared to be influenced by canopy cover and satellite availability. Among the top GPS performers, to ascertain positional accuracy and reliabil-
we determined that users could expect positional accuracies within approximately 5 m of true position ity.
in open sky settings, 7 m in young forest conditions, and 10 m under closed canopies.
Basic GPS Concepts
Keywords: GPS, evaluation, geospatial technology, mapping A prime distinction between consumer
GPS and other GPS receiver grades is in the
ability to differentially correct coordinate
data that have been collected. Mapping- and

G
lobal positioning systems (GPS) quality of the equipment and operator skill, survey-grade GPS usually include software
are becoming ubiquitous in society with instrument costs ranging from $2,000 that enables consumers to differentially cor-
for a variety of applications. GPS to $12,000. Although mapping-grade GPS rect data (Van Sickle 2001). Another signif-
hardware and software technology contin- can be somewhat more forgiving than survey icant distinction between consumer-grade
ues to become more accessible and afford- grade in terms of acceptable satellite recep- and other GPS receivers is in the quality
able for consumers, with some GPS receivers tion and required operator skill, the price of control of satellite reception including min-
now available for several hundred dollars or these units is still prohibitive to many poten- imum reception quality standards and the
less. GPS use in forestry has been hindered tial users. In contrast, consumer-grade GPS number of points necessary for coordinate
by challenges to receiving and maintaining receivers are now available at prices under determination. The potential strength of
sufficient satellite reception due to interven- $100, and there are many different manu- satellite arrangement relative to a receiver’s
ing canopy cover and land forms. facturers and styles of consumer-grade GPS position for position determination can be
GPS receivers can generally be classified receivers. Consumer-grade GPS manufac- estimated and reported as a position dilution
into one of three categories: survey, map- turers commonly assert that measurement of precision (PDOP) statistic (Johnson and
ping, and consumer grade. Survey-grade accuracies of this equipment should be Barton 2004). A PDOP value can be gener-
GPS is capable of determining locations to within 15–20 m of true position. ated by GPS mission planning software for a
within 1 cm of true position (Rizos 2002), We tested six consumer-grade GPS re- user-specified position on the earth’s sur-
but requires operator expertise and a sub- ceivers to determine average positional er- face. In general, a higher PDOP value indi-
stantial operating budget because instru- rors and measurement reliability. Our test- cates a satellite geometric arrangement with
ment costs can exceed $25,000. Survey- ing used three measurement courses that reduced reliability for measurement pur-
grade GPS also requires satellite signal represented three distinct measurement set- poses. Mapping- and survey-grade GPS al-
reception that is often unattainable in for- tings (open sky, young forest, and closed low users to set a minimum PDOP level that
ested landscapes. Mapping-grade GPS re- canopy). In addition, we revisited the open is necessary before location coordinates can
ceivers can return accuracies typically within sky course during a predicted period of poor be recorded; consumer-grade GPS do not
2–5 m of true position, depending on the satellite availability. Altogether, each of the typically permit users to specify a minimum

Journal of Forestry • June 2005 169


quality level. In addition, mapping- and sur- platform was glued to the top of staffs for
vey-grade GPS allow for point averaging of a GPS receivers that required horizontal oper-
position. As the earth and surrounding sat- 60° ation.
ellites are constantly in motion, multiple co- At all testing courses, each GPS receiver
ordinate readings and the averaging of these was placed at one of the six measurement
readings typically leads to a more statistically stations to begin recording coordinates. All
reliable location estimate than a single read- GPS receiver operators faced south during
ing alone can produce. The vast majority of data collection to standardize measurement
consumer-grade GPS do not allow for point 7.5m conditions. An audible count at approxi-
averaging, although some exceptions exist. mate 4-second intervals was used to syn-
Hindrances to effective GPS use in for- chronize the collection of 25 measurements
estry include canopy cover (Karsky et al. at each station at the same time by the six
2000), landscape topography (Liu and Bran- Figure 1. Geometry of test courses. GPS operators. This synchronization was
tigan 1996), and other factors (Wing and designed to minimize the effect of satellite
Kellogg, 2005). Key to deriving quality GPS geometry and time on measurements be-
measurements is the reception of at least each stake to mark a center location for mea- tween the six GPS receivers. At the comple-
four satellite signals at sufficient strength surement purposes. tion of 25 measurements, each GPS receiver
and from a robust geometric distribution. The six consumer-grade GPS receivers was moved to the next measurement bench-
Four satellites ensure that a horizontal and we tested were all similar in physical dimen- mark where another 25 measurements were
vertical position can be determined and that sions (Figure 2), although two of the units collected. This process was followed until
GPS timing errors are minimized. Canopy (Etrex Vista and Geko 301) have somewhat each GPS receiver had collected 25 measure-
cover and other solid objects can interfere smaller cases than the rest. All are produced ments from all six measurement bench-
with satellite signals and prevent them from by one of two recognized manufacturers marks. Following an approximate 5-minute
reaching a GPS receiver. In some cases, sat- (Table 1). All receivers had the ability to col- pause, this process was repeated twice so that
ellite signals may reflect off an object, such as lect data in different map coordinate units a total of 450 measurements were recorded
an automobile or other solid surface, and and to store up to 500 coordinate locations. at each measurement course by all six GPS
take a nondirect path to a GPS receiver, We used GPS mission planning soft- receivers. The 450 measurements were
leading to an error known as multipathing ware that was freely available from the Inter- downloaded digitally at the completion of
(Hofmann-Wellenhof et al. 2001). net (ARINC 1999) to identify PDOP times each measurement course. Each of the GPS
of good to poor quality. We scheduled our receivers we tested had a limit of 500 coor-
Methods GPS data collection at each of the test dinate pairs that could be stored, conse-
We established three measurement courses to reflect relatively good GPS data quently collecting 25 measurements at each
courses to test six consumer-grade GPS re- collection potential with one exception: the station (resulting in 450 total at each course)
ceivers. Six measurement stations were es- mission planning software indicated that a approached this limit while not exceeding
tablished in close proximity to each other at time of exceptionally poor PDOP (15) instrument storage capacity.
all three courses. All three courses were would occur on the second day of our All GPS receivers were set to collect
placed on relatively level terrain but environ- planned testing. Consequently, we decided data in a UTM coordinate system. An aver-
mental settings were selected to create open to revisit the open sky course during this age UTM coordinate was calculated from
sky, young forest, and closed canopy condi- time period to examine influences of poor each set of 25 coordinates that was collected
tions. The young forest course was placed in PDOP on GPS accuracy and precision. GPS at all measurement stations. A horizontal
a moderate density stand of predominately receiver testing occurred over 2 days in Mc- distance, or positional error, was calculated
Douglas fir (Pseudotsuga menziesii) with age Donald Forest, a research forest managed by by determining the straight-line distance be-
classes between 5 and 15 years and a canopy Oregon State University and located in tween the average coordinate and the total
closure of approximately 40 –50%. The can- western Oregon. station-derived coordinate. Standard devia-
opy cover in the closed canopy course was A digital total station was used to estab- tions were also calculated to provide an esti-
nearly 100% as provided by a dense stand of lish universal transverse mercator (UTM) mate of GPS reliability. We also determined
35– 45-year-old Douglas fir. coordinates for all measurement stations. a maximum PDOP estimate during each
The six measurement stations that were Wooden staffs were built to standardize the course repetition by having a mapping-grade
established at each of three courses were ini- height of all GPS receivers at 1.2 m above receiver record satellite signals throughout
tially created through the use of a hand com- the ground surface, a height at which oper- the testing processes.
pass and tape measure. A center point was ators could comfortably access GPS con-
established, and measurement stations were trols. We consulted the GPS manuals and, if Results
set 7.5 m from the center point and at 60° necessary, called manufacturers directly to Accuracy and reliability were summa-
angles from each other relative to the center determine whether units should be placed rized for each GPS receiver from three per-
point (Figure 1). Measurement stations were vertically or horizontally for best measure- spectives: by an average positional error and
marked by wooden stakes (4.4 cm2 top sur- ment accuracy. All GPS receivers were af- standard deviation (SD) for each of the three
face) that were driven into the ground until fixed to the wooden staffs using rubber repetitions around the courses, for a sum-
the upper 2 cm remained above ground. A bands (we avoided using metal fasteners to mary of all three repetitions (referred to as
dimpled nail was driven into the head of reduce multipathing risk). A flat wooden course summary), and for all courses com-

170 Journal of Forestry • June 2005


Figure 2. Consumer grade GPS from left to right: SporTrak Map, GPSmap 76S, GPS V, Etrex Vista, Geko 301, Meridian Platinum.

Table 1. Name, manufacturer, and price 301). The Etrex Vista (4.9 m), Geko 301, GPS receiver error averages (Table 2). The
of consumer-grade GPS units that were and GPS V (4.7 m) had measurement error highest positional errors occurred typically
tested. sums in excess of 4.6 m, while all other units in the closed canopy course, followed by the
were 3.2 m or less. The three receivers with young forest course, and open sky courses.
GPS name Manufacturer Price the largest positional errors also had the larg- In the case of the SportTrak Map, however,
est SD (2.8 –5.0 m), with the other units positional errors were improved in the
Etrex Vista Garmin $245
Geko 301 Garmin $220 featuring SD of 1.7 m or less. young forest course in comparison to the ini-
GPS V Garmin $320 We revisited the open sky course during tial visit to the open sky course. Maximum
GPSmap 76S Garmin $310 a period that was predicted by our mission recorded PDOP values also tended to follow
Meridian Platinum Magellan $280
SportTrak Map Magellan $150 planning software to be particularly poor for a similar trend in relation to course cover
GPS data collection. Open sky course posi- types, with the highest PDOPs being re-
tional error averages ranged from 1.0 m corded in the closed canopy course and low-
bined (referred to as GPS summary). Our (Meridian Platinum) to 4.0 m (Geko 301) est PDOPs in the open sky course visits.
results found that measurement accuracy during the second visit. In comparison to In terms of what users should expect in
and reliability varied, in some cases consid- error in the initial visit to the open sky terms of performance, there are several ap-
erably, among the six GPS receivers (Table course, these average errors were within 2–3 proaches to quantifying predicted accuracies
2). m during the second visit with one excep- for GPS receivers based on repeated mea-
In comparing the average positional er- tion: the Meridian Platinum had greatly re- surements (Wilson 2000). However, the
ror for the course summaries of the open sky duced positional errors during the second National Standard for Spatial Data Accu-
setting, positional errors varied from 1.4 m visit. For all other units, the SD were similar racy (NSSDA), published by the Federal
for the GPSmap 76S to 19.6 m for the Me- during both visits to the open sky courses. Geographic Data Committee (1998), is de-
ridian Platinum. Only the Geko 301 and One method to assess the overall per- signed to provide a unified approach to as-
Meridian Platinum had average errors formance of the consumer GPS receivers we sessing the accuracy of digital geospatial data
greater than 4.0 m. The SD was also greatest tested is to average the positional errors and independent of measurement scale. The
for the Geko 301 (2.0 m) and Meridian Plat- SD for all measurement courses to create a NSSDA uses the root mean squared error
inum (3.8 m). For all other GPS units, the GPS summary average error and SD value from measurements and multiplies this by a
SD summary was 1.0 m or less. for each unit. GPS summary average errors constant of 1.7308 to produce a 95% confi-
Within the young forest setting, the ranged from a low of 1.7 m for the Sport- dence level for predicted measurement accu-
course average error ranged from a low of Trak Map to a high of 6.6 m for Geko 301 racies. This constant is based on a normal
1.3 m for the SportTrak Map to 6.8 m for (Table 2). The summary SD ranged from distribution of error values and is designed
the Geko 301. Notably, with the exception 0.9 m for the SportTrak Map to a high of for use when error distances are calculated in
of the SportTrak Map and Meridian Plati- 2.6 m for the Geko 301. In using the GPS two dimensions (e.g., x and y coordinate dif-
num (1.7 m), the course summary average receiver summaries as a benchmark for per- ferences). In contrast, a constant of 1.96
error was greater than 3.5 m for all other formance, a top tier emerged among the would be used for single-dimension error
GPS receivers. The course summary SD was units: the performance of the Etrex Vista, measurements. NSSDA accuracies are given
1.1 m or less for all GPS receivers except the GPS V, GPSmap76S, and SportTrak Map in brackets in Table 2 and can be interpreted
Geko 301 (1.9 m). was superior in comparison to the other two to provide an estimate of expected positional
The average errors for the closed can- GPS. performance among the consumer-grade
opy course summaries varied from 2.7 m In general, course settings and maxi- GPS receivers. For instance, the SportTrak
(SportTrak Map) to a high of 11.4 m (Geko mum PDOP values varied predictably with Map had a NSSDA value of 3.2 m for the

Journal of Forestry • June 2005 171


Table 2. Average positional accuracy, SD, and the National Standard for Spatial Data Accuracy (NSSDA) of GPS measurements.a

Open sky course Young forest course Closed canopy course Open sky revisited GPS summary
Max. Average Max. Average Max. Average Max. Average Average
GPS unit Repetitionb PDOP errorc PDOP errorc PDOP errorc PDOP errorc errorc

Etrex Vista 1 5.0 2.5 (0.9) 5.5 3.1 (1.1) 5.7 4.3 (2.0) 4.3 3.1 (0.8)
2 4.0 2.1 (1.3) 4.6 4.1 (0.7) 5.9 5.5 (4.4) 2.9 3.1 (0.8)
3 4.6 2.1 (0.7) 4.5 4.0 (1.5) 5.7 4.8 (2.3) 3.1 2.2 (0.5)
Course summaryd 2.2 (1.0) 关4.2兴 3.7 (1.1) 关6.7兴 4.9 (2.9) 关9.8兴 2.8 (0.7) 关5.0兴 3.4 (1.4) 关6.4兴
Geko 301 1 5.0 3.1 (1.8) 5.5 6.3 (2.1) 5.7 11.8 (4.9) 4.3 5.1 (1.3)
2 4.0 4.4 (1.7) 4.6 6.4 (1.2) 5.9 10.2 (5.0) 2.9 3.3 (2.2)
3 4.6 5.0 (2.4) 4.5 7.6 (2.5) 5.7 12.3 (5.2) 3.1 3.6 (1.4)
Course summaryd 4.2 (2.0) 关8.0兴 6.8 (1.9) 关12.2兴 11.4 (5.0) 关21.6兴 4.0 (1.6) 关7.5兴 6.6 (2.6) 关12.3兴
GPS V 1 5.0 3.2 (0.8) 5.5 3.8 (0.9) 5.7 3.5 (2.9) 4.3 4.2 (0.6)
2 4.0 2.6 (1.1) 4.6 3.9 (1.1) 5.9 4.9 (3.3) 2.9 3.1 (0.8)
3 4.6 2.0 (0.9) 4.5 3.7 (0.7) 5.7 5.8 (2.1) 3.1 1.6 (0.4)
Course summaryd 2.6 (0.9) 关4.8兴 3.8 (0.9) 关6.8兴 4.7 (2.8) 关9.5兴 3.0 (0.6) 关5.2兴 3.5 (1.3) 关6.5兴
GPSmap 76S 1 5.0 1.1 (0.5) 5.5 3.5 (0.8) 5.7 3.2 (2.0) 4.3 3.7 (0.6)
2 4.0 1.0 (0.4) 4.6 3.7 (1.8) 5.9 3.5 (1.9) 2.9 1.9 (1.0)
3 4.6 2.0 (0.9) 4.5 3.6 (0.7) 5.7 3.0 (1.0) 3.1 1.2 (0.3)
Course summaryd 1.4 (0.6) 关2.6兴 3.6 (1.1) 关6.5兴 3.2 (1.6) 关6.2兴 2.3 (0.6) 关4.1兴 2.6 (1.0) 关4.9兴
Meridian Platinum 1 5.0 13.3 (1.6) 5.5 2.0 (0.9) 5.7 2.9 (1.8) 4.3 0.8 (0.5)
2 4.0 19.6 (8.6) 4.6 1.3 (0.3) 5.9 4.1 (1.9) 2.9 0.8 (0.4)
3 4.6 26.0 (1.2) 4.5 1.9 (1.1) 5.7 1.8 (0.8) 3.1 1.4 (0.7)
Course summaryd 19.6 (3.8) 关34.6兴 1.7 (0.8) 关3.3兴 3.0 (1.5) 关5.7兴 1.0 (0.5) 关2.0兴 6.3 (1.7) 关11.3兴
SportTrak Map 1 5.0 1.4 (0.5) 5.5 1.1 (0.7) 5.7 3.4 (2.2) 4.3 1.4 (0.8)
2 4.0 1.7 (0.4) 4.6 1.2 (0.7) 5.9 1.4 (0.6) 2.9 1.1 (1.0)
3 4.6 2.2 (1.0) 4.5 1.7 (0.5) 5.7 3.4 (2.1) 3.1 0.8 (0.6)
Course summaryd 1.8 (0.6) 关3.2兴 1.3 (0.6) 关2.5兴 2.7 (1.7) 关5.5兴 1.1 (0.8) 关2.4兴 1.7 (0.9) 关3.4兴
a
Units are in meters, SD are in parentheses, and NSSDA values are in brackets.
b
Each course repetition represents 150 measurement values for each GPS.
c
The average error is the distance from GPS coordinates to known coordinates at the test courses.
d
Course summaries are the average of the three course repetitions.

first open sky course. On average, if we were ment accuracy can vary widely depending aging, where a user can log multiple coordi-
to take 100 measurements of a position with on the consumer GPS receiver. nate points and store the average of the
this GPS receiver, we would expect that 95 Our results indicate that at least for the points. None of the other GPS we tested
of them would be within 3.2 m of the actual Etrex Vista, GPS V, GPSmap76S, and featured this capability.
position. For a single coordinate pair taken SportTrak Map (the top four performers as One apparent anomaly was the perfor-
with this GPS receiver, we would be 95% identified by our testing results), positional mance of the Meridian Platinum, which
confident that the point was within 3.2 m of accuracies that are better than the 15–20 m rated poorly during the first visit to the open
the actual position. As the NSSDA is based range often cited as a yardstick should be canopy course but was otherwise a strong
on an assumed statistical distribution and a achieved, with improved accuracies being performer in accuracy and reliability. The
confidence interval (95%), we should not be possible in open sky settings and depending individual coordinates collected by the Me-
surprised to detect measurements outside of on the unit. ridian Platinum were more closely exam-
predicted accuracy ranges. Overall, the Etrex Vista, GPS V, GPS- ined, revealing gradually increasing and de-
In considering the top four GPS receiv- map 76S, and SportTrak Map all performed creasing coordinate positions relative to the
ers in terms of average measurement errors,
similarly in terms of accuracy and reliability. control benchmarks throughout the indi-
expected NSSDA accuracies within open
The SportTrak Map, however, had the low- vidual measurements. The reasons for this
sky settings should be within approximately
est summary average error (1.7 m) and SD performance are unknown but it did not oc-
5 m of true position 95% of the time de-
(0.9 m). The GPSmap 76S had the second cur during the second visit to the open sky
pending on the GPS receiver. In a young
lowest summary average error (2.6 m) and course or at any of the other measurement
forest setting, this figure increases to 7 m,
and in a closed canopy setting approximately SD (1.0 m). courses.
10 m. The Etrex Vista and GPS V were strik- The Geko 301 did not compare well to
ingly similar in performance throughout the the other GPS receivers in any of the mea-
Discussion measurement courses. We expected the GPS surement courses with the exception of the
Consumer-grade GPS receivers now of- V to perform better than all other units, due Meridian Platinum in the initial open sky
fer potential users an inexpensive and man- to the presence of an external antenna, but course measurements. The Geko 301 was
ageable technology for collecting coordinate did not find superior results. The GPS V, the smallest of the GPS receivers we tested
locations of landscape features. Users will however, was the most full-featured of the and would be convenient in this regard.
need to determine whether the potential ac- GPS receivers we tested and included a CD Several considerations should be made
curacies of collected coordinates are suffi- that featured additional basemap data that in assessing the results of this study. The
cient for the intended purposes of the data could be downloaded onto the receiver. The mission planning software we initially used
and must always keep in mind that measure- GPS V also allows for automatic point aver- was available free over the Internet and was

172 Journal of Forestry • June 2005


not a reliable resource for predicting satellite GPS receivers. Mission planning software KARSKY, D., K. CHAMBERLAIN, S. MANCEBO, D.
availability PDOP levels. For instance, our can identify the data collection times that are PATTERSON, AND T. JASUMBACK. 2000. Com-
return visit to open sky course was supposed best in terms of the number of satellites parison of GPS receivers under a forest canopy
with selective availability off. USDA Forest Ser-
to occur during a time that was poor for GPS available and the physical geometry of satel-
vice Project Report 7100. 21 p.
data collection, but average error sums and lites relative to a ground location. GPS users LIU, C.J., AND R. BRANTIGAN. 1996. Using differ-
recorded PDOP values indicated otherwise. may also want to consider keeping the GPS ential GPS for forest traverse surveys. Can. J.
Our expectation was that we would either receiver antenna located at operator head For. Res. 25:1795–1805.
not be able to collect data or that measure- height or above any features that might im- RIZOS, C. 2002. Introducing the global position-
ment errors would be much greater than pede satellite reception. Utilizing an external ing system. P. 77–94 in Manual of geospatial
those that resulted during the projected poor antenna may also increase the accuracy of science and technology, Bossler, J., J. Jensen, R.
PDOP period. As we examined actual McMaster, and C. Rizos, eds. Taylor and
measurements. Unfortunately, few consum-
PDOP values following data collection, Francis, London and New York.
er-grade GPS receivers will accept external VAN SICKLE, J. 2001. GPS for land surveyors. Ann
PDOP values during the second visit were at antennas. Arbor Press, Chelsea, MI.
least marginally better than those recorded WILSON, D.L. 2000. GPS horizontal position ac-
during the initial visit. Literature Cited curacy. Available online at users.erols.com/
Potential GPS users should keep in ARINC. 1999. The system effectiveness model, ver- dwilson/gpsacc.htm; last accessed January
mind that results reported in this study were sion 3.6.4. Annapolis, MD. Available online at 2005.
based on 25 measurements collected at each www.arinc.com/gps/gpsapps/sem.html; last WING, M.G., AND L.D. KELLOGG. 2005. Locat-
accessed January 2005. ing and mobile mapping techniques for for-
of six stations. This process was repeated
FEDERAL GEOGRAPHIC DATA COMMITTEE. 1998. estry applications. Geographic Information Sci-
three times during each measurement Geospatial positioning accuracy standards. Part ences 11(1):87–95.
course, and the entire process was repeated 3: National standard for spatial data accuracy,
at the open sky course. This process of aver- FGDC-STD-007.3. Federal Geographic Data
aging should, statistically, result in coordi- Committee, Washington, DC. 25 p. Michael G. Wing (michael.wing@oregonstate.edu)
nates that are more accurate than that col- HOFMANN-WELLENHOF, B., H. LICHTENEGGER, is assistant professor, Aaron Eklund (aaron.
AND J. COLLINS. 2001. Global positioning sys- eklund@oregonstate.edu) is graduate student, and
lected by a single measurement, or by fewer
tem: Theory and practice. Springer-Verlag, Loren D. Kellogg (loren.kellogg@oregonstate.edu) is
than 25 measurements. Wien, Germany. 382 p.
There are some strategies, however, for JOHNSON, C.E., AND C.C. BARTON. 2004. Where Lematta professor of forest engineering, Forest Engi-
those wishing to maximize the accuracy of in the world are my field plots? Frontiers Ecol. neering Department, Oregon State University,
collected coordinates with consumer-grade Environ. 2(9):475– 482. Corvallis, OR 97331.

Journal of Forestry • June 2005 173

Vous aimerez peut-être aussi