See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274679082

Accuracy comparison between GPS and DGPS: A field study at METU campus

Article  in  European Journal of Remote Sensing · October 2010

DOI: 10.5721/ItJRS20104231

CITATIONS READS

13 6,746

2 authors:

Nazan Kuter Semih Kuter

Çankırı Karatekin Üniversitesi Çankırı Karatekin Üniversitesi
15 PUBLICATIONS   156 CITATIONS    34 PUBLICATIONS   184 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Special Issue "Remote Sensing of Snow and Its Applications" View project

Validation of EUMETSAT H12 Effective Snow Cover Product by Sentinel 2 MSI Data View project

All content following this page was uploaded by Nazan Kuter on 31 March 2015.

The user has requested enhancement of the downloaded file.

 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

Accuracy comparison between GPS and DGPS:

A field study at METU campus
Nazan Kuter1 and Semih Kuter2

1
Department of Landscape Architecture, Çankırı Karatekin University, 18200 - Çankırı, Turkey.
E-mail: nkuter@karatekin.edu.tr.
2
Department of Forest Engineering, Çankırı Karatekin University, 18200 - Çankırı, Turkey

Abstract
Global Positioning System (GPS) is a highly demanded tool for positioning and synchronization
of measurements; therefore, assessing the accuracy of a GPS receiver is an essential phase of
any field survey. The main goal of this study is to briefly summarize different commonly used
GPS accuracy measures and then compare the horizontal and vertical accuracies of GPS and
differential GPS (DGPS) by taking instant measurements, instead of making measurements
for long periods of time at sampling locations. To achieve this, a field study was carried out at
Middle East Technical University (METU) campus in Ankara, Turkey. As a preliminary step
before the field study, a basic sampling strategy was developed. The horizontal and vertical
accuracies of both receivers were calculated and given in terms of root-mean-square error
(RMSE) and arithmetic mean of error. The results revealed out that DGPS has significantly
better performance than GPS, when horizontal and vertical accuracies are considered.
Keywords: Global positioning system, Horizontal accuracy, Vertical accuracy, Wilcoxon
Signed-Rank test.

Confronto di accuratezza tra ricevitori GPS e DGPS: un caso di studio al campus METU

Riassunto
I sistemi di posizionamento satellitare sono strumenti sempre più richiesti per effettuare
misure di posizionamento e sincronizzazione; per questo motivo la stima dell’accuratezza
di un ricevitore GPS è una fase essenziale in qualsiasi rilievo di campagna. L’obiettivo
di questo lavoro è fare una breve sintesi dei sistemi utilizzati per valutare l’accuratezza
dei GPS e confrontare l’accuratezza orizzontale e verticale di ricevitori GPS e GPS
differenziali (DGPS). A questo scopo è stato effettuato uno studio presso il Middle East
Technical University (METU) campus di Ankara, Turchia. Nella prima fase del lavoro è
stata definita la strategia di campionamento. In seguito sono state calcolate le accuratezze
verticali e orizzontali dei due ricevitori in termini di errore quadratico medio (RMSE) e
media aritmetica degli errori. I risultati ottenuti mostrano che le prestazioni del ricevitore
DGPS sono significativamente migliori di quelle del ricevitore GPS in termini di accuratezza
orizzontale e verticale.
Parole chiave: Sistema di posizionamento satellitare, accuratezza orizzontale, accuratezza
verticale, Wilcoxon Signed-Rank test.


Kuter and Kuter GPS and DGPS accuracy comparison

Introduction
Global positioning System (GPS), satellite-based navigation system operated by United
States Department of Defense, has been widely used since it became fully operational in
April 1995 (also known as NAVSTAR). Its area of usage covers a wide range of different
applications by different disciplines from military to civilian, including navigation, target
tracking, search and rescue, geology, mining, farming, forestry, inventory control, land
management and many others.
GPS receiver accuracy and its assessment play a crucial role when collecting field data,
and various methods and statistics have been employed so far in order to investigate and
compare accuracy of GPS. According to Misra and Enge [2001], the error statistics can be
calculated empirically on the basis of a sample of position estimates by letting Δxi, Δyi, and
Δzi (i=1,2,…n) be the errors in the east, north, and vertical components of the ith position
estimate sample. Then root-mean square (RMS) vertical, horizontal (2D) and 3D errors are
given by:
1 n Dz 2 61@
n!
RMS_vertical_error = i
i=1

1 n _ Dx 2 Dy 2i 62@
n!
2D_RMS_error = i + i
i=1

1 n _ Dx 2 Dy 2 Dz 2i 63@
n!
3D_RMS_error = i + i + i
i=1

It was also pointed out that for GPS position estimates, mean error equals to zero if averaged
over a long time interval, and positions estimates averaged over seconds or minutes are
generally biased because satellite clock, orbit, and propagation effects change slowly. If the
position estimates are modeled as unbiased, the above equations can be written in terms of
standard deviations of the east, north and, vertical components of position:
RMS_vertical_error = v U 64@

2D_RMS_error = v 2E + v 2N 65@

3D_RMS_error = v 2E + v 2N + v 2U 66@

where σE, σN, and σU are the standard deviations of the east, north, and vertical components
of the position error, respectively. The 50th and 95th percentiles for the horizontal, vertical,
and 3D errors are frequently used and calculated easily. The circular error probable (CEP),
50th percentile horizontal error, defines the radius of a circle centered at the true position
which would contain the position estimate with probability of 0.5. Spherical error probable


 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

(SEP) is 3D median error, and the 95th percentiles are often used as horizontal error (95%)
and vertical error (95%) as shorthand. It is important to emphasize that sometimes they
are also used as error values at 95% confidence level; whereas confidence level has a well-
defined and different meaning in statistics, and this kind of usage is not recommended.
RMS in 2D given in Eq. [5], also described as distance root-mean-square (DRMS) error by
Kaplan and Hegarty [2006], was used by Rodriguez-Perez et al. [2007], and Yoshimura and
Hasegawa [2003] in order to estimate GPS positional error in terms of precision. There is
another accuracy measure for position estimates, 2DRMS (twice the RMS in 2D), and it is
often used in incorrect manner as equivalent to the 95th percentile of the horizontal error
[Misra and Enge, 2001].
Yoshimura and Hasegawa [2003] conducted a field test to assess and compare horizontal and
vertical positional errors of GPS and DGPS in forested areas. According to their findings,
DGPS improves horizontal accuracy but not horizontal precision, whereas autonomous
GPS is sufficiently useful where maximum horizontal error of 10 m is tolerable, but it is
also noted that DGPS should be used for surveying and mapping where higher accuracy is
needed. On the other hand, by Rodriguez-Perez et al. [2007], four low-cost GPS receivers
under different forest conditions, which were defined by several variables such as stand
density (SDe), Hart-Becking index (HBI) and Assman Dominant Height (H0), were
compared in order to determine ease of use, accuracy and precision. In all of the above
mentioned studies, the following equations were used to calculate the standard deviations
in east and north components of the position error:

! ^ x - xh
n
2

6 7@
i

vx = i=1
n-1

! ^ y - yh
n
2

68 @
i

vy = i=1
n-1

where σx and σy are standard deviations of the east (x) and north (y) components of the
position error, respectively; x and y are the sample mean of positional error along the x
and y directions. These studies estimated the horizontal accuracy with the help of following
equation:

v H_accuracy = ^ x - xtrueh + ^ y - ytrueh 69@

2 2

where σH_accuracy denotes the horizontal accuracy; xtrue and ytrue are the true location along the x
and y directions, respectively. Both Yoshimura and Hasegawa [2003] and Rodriguez-Perez
et al. [2007] employed the following equations to obtain vertical precision and vertical
accuracy:


Kuter and Kuter GPS and DGPS accuracy comparison

! ^ z - zh
n
2

610@
i

vV_precision = i=1
n-1

vV_accuracy = z - ztrue 611@

where z and ztrue are the sample mean of positional error and true location along z direction,
respectively.
Matosevic et al. [2006] made an accuracy comparison between a GPS and a low-cost DGPS,
and developed a simple statistical model to estimate positioning error, which requires only
position information without satellite constellation, and results were introduced in terms of
CEP and SEP values.
As it has been briefly summarized so far, there are numerous ways of dealing with GPS
accuracy, and therefore, the main goal of this study is to demonstrate a simple horizontal
and vertical accuracy comparison between a GPS and a differential GPS (DGPS) by a field
survey based on a basic sampling design strategy, and finally to compare and discuss the
results.

Materials and Methods

Study Area
The study area was Middle East Technical University campus (METU) in Ankara, TURKEY
(Fig. 1), which is 4500 hectares, about 20 kilometers from the center of Ankara at latitude
of 390 53’ 31.93’’N and longitude of 320 46’ 59.09’’E.

Materials
IKONOS imagery of METU campus (multispectral (RGB), 1meter/pixel resolution,
acquisition date: 06.08.2004) was used as base map. Then, major traffic and pedestrian roads
in the study area were digitized and saved as a polyline shape file in ArcGIS v.9.2 (Fig. 1).
As a GPS receiver, Magellan eXplorist 500 manufactured by Thales S.A. was used.
According to the manufacturer’s technical specs, its horizontal positional accuracy is 7 m
at 95% confidence level. MiniMAX, produced by CSI Wireless Inc. [2004] with nominal
horizontal positional accuracy 1 m at 95% confidence level, was chosen as DGPS.
ProMark 500 model multi-constellation GNSS receiver manufactured by Magellan
Professional was used in order to get the exact coordinates at sampling locations with real-
time kinematic position accuracy of 10 mm and 20 mm in fine mode, in horizontal and
vertical respectively.

Methods
The aim of this study was to assess and compare horizontal and vertical positional errors
of both GPS and DGPS receivers. The true values of x, y and z coordinates of sampling
locations were acquired by ProMark 500. Field survey to collect the necessary data was
conducted on March 9 and 10, 2010. All instruments were switched twenty minutes before
data collection and ED50 was chosen as datum and UTM as projection.


 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

Figure 1 – Study area and the sampling distribution.

According to Congalton and Green [2009], clearly identifiable and suitable well-defined
points represent right-angle intersections of roads, or other linear mapped features like
canals, ditches, trails, fence lines, and pipelines. Therefore, as sampling strategy, all road
intersections (646 as total) on the digitized campus roads layer were extracted and saved as
a separate point layer in ArcGIS v.9.2 as the sampling frame of the study (Fig. 1).
The next step was to determine the required sample size, and its calculation is probably
one of the most important steps in statistical analysis. In order to determine the appropriate
sample size, Bartlett et al. [2001] proposed “seven-point scale” to estimate the standard
deviation in the population. For example, if a researcher used a seven-point scale and given
that six standard deviations (three to each side of the mean) would capture 98% of all
responses, the calculations would be as follows:


Kuter and Kuter GPS and DGPS accuracy comparison

7 (number of points on the scale) 6 @

s= 12
6 (number of standart deviations)

Then the required sample size can easily be calculated by the following equation:
^ 2h ^ 2h
n0 = t )2 s 613@
^d h

α = 0.05, two tailed), s is the estimate of standard deviation, and d is the acceptable margin
of error for mean being estimated (number of points on primary scale times acceptable
margin of error; points on primary scale = 7; acceptable margin of error = 0.05). Then the
necessary sample size turned out to be n0 ≈ 43.
If the calculated sample size had exceeded 5% of the population (646*0.05 ≈ 32), the
following correction formula of Cochran [1977] would have been employed to calculate
the final sample size:
n0 614@
^1 + n0 /populationh
n=

Total 60 road intersection points (50% more than the sample size found according to
Eq.[13]) were selected by simple random sampling tool of ArcGIS 9.2 (Fig. 1), and, x, y,
and z coordinates of each sample point was measured and recorded with GPS, DGPS and
ProMark 500 simultaneously.
Horizontal error was calculated in two different ways: Firstly, horizontal root-mean-square
error (RMSEH) was calculated by:

! `^ x - xmih + ^ yti - ymih j /n 615@

n
2 2
RMSEH = ti
i=1

where xt and yt are true coordinates, and xm and ym are measured coordinates.
To calculate the standard deviation of horizontal errors (sH), the following equation was
used [Congalton and Green, 2009]:
sH = ^sx + syh /2 616@

sx and sy are the standard deviations of errors in x and y directions, respectively and given by:

! _^ x - xmih - RMSEx i / ^n - 1h 617@

n
2
sx = ti
i=1

! _^ y - ymih - RMSEy i / ^n - 1h 618@

n
2
sy = ti
i=1


 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

where RMSEx and RMSEy are root-mean square errors in x and y directions, and calculated
with the same manner as in RMSEH. In Eq. [16], the circular distribution of horizontal
errors was assumed based on the bivariate standard normal distribution. Test for circularity
was done by calculating the ratio of the smin to smax, where smin is the lower of sx or sy, and
smax is the larger of sx or sy, and if the ratio of smin to smax is 0.2 or greater, distribution can
be assumed as circular [Congalton and Green, 2009]. Then standard error of RMSEH was
calculated by:
sH = sH / n 619@

Secondly, as an alternative way, the arithmetic mean of horizontal errors ( r ) was calculated
by using the following equation:
n
!r
620@
i

r = i=1
n

where ri is the magnitude of the ith horizontal error vector which is defined as:

ri = ^ xti - xmih + ^ yti - ymih 621@

2 2

Regarding Eq. [20], standard deviation of horizontal errors was calculated by:

! ^r - r h / ^n - 1h 622@
n
2
sr = i
i=1

Then the standard error of r is given by:

sr = sr / n 623@

For calculation of errors in vertical direction, again the same two methods, which were
previously employed in horizontal error calculations, were used: First, vertical root-mean-
square error (RMSEV) was found by:

!^z - zmih /n 624@

n
2
RMSEV = ti
i=1

where zt is true coordinate, and zm is the measured coordinate. The standard deviation of
vertical errors was estimated by:

! (^ z - zmih - RMSEV ) 2 / (n - 1) 625@

n
sv = ti
i=1

And then, the standard error of RMSEV was calculated by:


Kuter and Kuter GPS and DGPS accuracy comparison

sV = sV / n 626@

Alternatively, the mean error in vertical direction was also calculated as arithmetic mean of
absolute errors by using:
n
! zti - zmi
z= i=1
627@
n

Then, the standard deviation of z was calculated by:

- zmi) - z) 2 / ^n - 1h 628@
n
sz = ! ((z ti
i=1

And standard error of z was obtained by the equation:

sz = sz / n 629@

Standard deviation and all other intermediate values used in horizontal and vertical error
calculations are summarized in Table 1.

Table 1 – Standard deviation, RMSE and arithmetic mean

values used in horizontal and vertical error calculations.
GPS (m) DGPS (m)
Horizontal
RMSEH 10.305 2.299
RMSEx 7.215 1.229
RMSEy 7.358 1.943
sx 9.079 1.520
sy 11.344 3.333
sH 10.211 2.427
rmean 8.299 1.419
sr 6.160 1.824
Vertical
RMSEV 6.413 2.587
sv 10.226 4.323
zmean 5.190 2.140
sz 9.282 3.958

10
 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

Most statistical analysis often assume normality without any empirical evidence or test;
however, it is critical in many statistical methods, and when violated, interpretation and
inference may not be reliable or valid [Park, 2008]. Therefore, to test the normality of
horizontal and vertical error distributions of GPS and DGPS, Kolmogorov-Smirnov and
Shapiro –Wilk normality tests were carried out at 95% confidence level (Tab. 2). According
to the result of Kolmogorov-Smirnov test, GPS vertical errors were normally distributed;
however, the result of Shapiro-Wilk test did not indicate normality. Test results revealed that
none of the error distributions were normal. Then, the following two null hypotheses were
tested at 95% confidence level: (1) There is no statistically significant difference between
GPS and DGPS receivers in terms of horizontal accuracy; (2) There is no statistically
significant difference between GPS and DGPS receivers in terms of vertical accuracy.

Table 2 - Normality tests for horizontal and vertical errors of GPS and DGPS (α = 0.05).
Normality Test Kolmogorov-Smirnov Shapiro-Wilk
  stat. df sig. stat. df sig.
GPS horizontal 0.239 60 0.000 0.819 60 0.000
GPS vertical 0.099 60 0.200 0.949 60 0.015
DGPS horizontal 0.301 60 0.000 0.528 60 0.000
DGPS vertical 0.144 60 0.003 0.936 60 0.004

Since the error distributions in horizontal and vertical directions were not normal and the
measurements with two receivers were simultaneously taken in pairs at sampling locations,
Wilcoxon signed-rank test for paired data, which is the non-parametric version of paired
t-test, was applied to the data in order to test the null hypothesis (1) and (2) [Montgomery
and Runger, 2003]. Then, the error bars were plotted with 95% confidence intervals and the
results were interpreted.

Results and Conclusion

The results of Kolmogorov-Smirnov and Shapiro-Wilk tests indicated that horizontal and
vertical errors of GPS and DGPS were not normally distributed (Tab. 2). To test the null
hypotheses (1) and (2), Wilcoxon signed-rank test with alpha level 0.05 was applied. The
test result of hypothesis (1) showed that the horizontal accuracy of DGPS was significantly
better than GPS (p < 0.000), (Tab. 3).

Table 3 - Wilcoxon Signed Ranks test for horizontal and vertical errors.
Statistical Test Wilcoxon signed-rank
  Z Total sig. (2-tailed)
Horizontal errors -6.581 60 0.000
Vertical errors -5.110 60 0.000

11
Kuter and Kuter GPS and DGPS accuracy comparison

According to their manufacturers’ specifications, horizontal accuracy of GPS and DGPS

receivers are 7 m and 1 m respectively, and when these values are compared with the results
shown in Figure 2, the obtained intervals 10.305 ± 2.584 m and 2.299 ± 0.614 m in RMSEH,
for GPS and DGPS receivers respectively, are a bit larger than the specified values (Fig. 2
a). Horizontal accuracies of both GPS and DGPS receivers in terms of r are 8.299 ± 1.559
m and 1.419 ± 0.462 m respectively, and can be seen with error bars in Figure 2 b.
The reason for these differences in accuracies of both GPS and DGPS receivers may arise
from the fact that the measurements at sampling locations were not taken for over a long
time interval (2-4 hours) and then averaged; instead, two coordinate triplets (one for GPS
and one for DGPS) were measured instantly at each sampling location.

Figure 2 - Horizontal accuracy of DGPS and GPS in terms of a) RMSEH, and b) r , with error
bars at 95% confidence interval.

After carrying out Wilcoxon signed-rank test for hypothesis (2), (Tab. 3), it was observed
that again DGPS evidently performed better than GPS in terms of vertical accuracy (p <
0.000). As observed in Figure 3 a, interval obtained for GPS is 6.413 ± 2.588 m in RMSEV,
whereas 2.587 ± 1.094 m for DGPS (Fig. 3 a). Vertical accuracies in terms of z are 5.190
± 2.349 m and 2.140 ± 1.002 m for GPS and DGPS receivers respectively, and given with
error bars in Figure 3 b.
GPS receivers are indispensible instruments in today’s field surveys and necessary
accuracy tests prior to a survey in new environment are essential, and in this study, a
practical method was introduced in order to compare the vertical and horizontal accuracies
of GPS and DGPS receivers and to achieve this, a basic sampling strategy was designed
and applied. As a common convention, RMSE values for horizontal and vertical accuracies
have usually been used, as it was done in this study; however, as an alternative, accuracies
were also calculated and given in terms of arithmetical means (horizontal: r , vertical:
z ). And the results clearly show that, there is a noticeable difference between GPS and
DGPS when horizontal and vertical accuracies are considered. Although the removal of SA

12
 Rivista Italiana di Telerilevamento - 2010, 42 (3): 3-14
Italian Journal of Remote Sensing - 2010, 42 (3): 3-14

(selective availability) has a greater importance for GPS than DGPS, it results in significant
improvement in accuracy of both types of receivers. However, DGPS receivers are still
superior than GPS receivers in terms of accuracy, since sources of error such as, satellite
clock, ionospheric and tropospheric delays can be removed by differential correction.

Figure 3 - Vertical accuracy of DGPS and GPS in terms of a) RMSEV, and b) z , with error bars
at 95% confidence interval.

References
Ashtech (2010). ProMark 500 reference manual.
Bartlett J. E., Kotrlik J. W., Higgins C. C. (2001) – Organizational research: Determining
sample size in survey research. Information Technology, Learning and Performance
Journal, 19 (1): 43-50.
Cochran W. G. (1977) – Sampling Techniques. 3rd Ed., John Wiley & Sons, New York.
Congalton R. G., Green K. (2009) – Assessing the Accuracy of Remotely Sensed Data. 2nd
Ed., CRC Press, Taylor & Francis Group, Boca Raton.
CSI Wireless Inc. (2004). Mini MAX reference manual.
Kaplan E. D., Hegarty C. J. (2006) – Understanding GPS: Principles and Applications. 2nd
Ed., Artech House Inc., Norwood, MA.
Matosevic M., Salcic Z., Berber S. (2006) - A comparison of accuracy using a GPS and a
low-cost DGPS. IEEE Transactions on Instrumentation and Measurement, 55 (5): 1677-
1683.
Montgomery D. C., Runger G. C. (2003) – Applied Probability and Statistics for Engineers.
3rd Ed., John Wiley & Sons, New York.
Misra P., Enge P. (2001) – Global Positioning System: Signals, Measurements and
Performance. Ganga-Jamuna Press, Massachusetts.
Park H. M. (2008) – Univariate analysis and normality test using SAS, Stata, and SPSS:
Technical working paper. The University Information Technology Services (UITS) Center
for Statistical and Mathematical Computing, Indiana University, http://www.indiana.edu/
~statmath/stat/all/normality/ normality.pdf. (May 14, 2009).

13
Kuter and Kuter GPS and DGPS accuracy comparison

Rodriguez-Perez J. R., Alvarez M. F., Sanz-Ablanedo E. (2007) – Assessment of low-

cost GPS receiver accuracy and precision in forest environments. Journal of Surveying
Engineering, 133 (4): 159-167.
Thales S.A. (2005). Magellan eXplorist 500 GPS receiver reference manual.
Yoshimura T., Hasegawa H. (2003) – Comparing the precision and accuracy of GPS
positioning in forested areas. Journal of Forest Research 8 (3): 147-152.

Received 28/07/2009, accepted 09/07/2010

14

View publication stats