INVESTIGATION OF ACCURATE METHOD IN 3-D POSITION USING CORS-NET IN ISTANBUL

In this study, for Istanbul, there are two Cors Networks (Cors-TR, Iski Cors) providing Virtual Reference Station (VRS), and Flachen Korrektur Parameter (FKP), corrections to rover receiver for determining 3-D positions in real time by Global Positioning System (GPS). To determine which method (or technique) provides accurate method for position fixing, a test network consisting of 49 stations was set up in Yildiz Technical University Davudpasa Campus. The coordinates of the stations in the test network were determined by conventional geodetic, classical RTK, VRS and FKP methods serviced by both Cors-TR and Iski Cors. The results were compared to the coordinates by the conventional method by using total station. The results showed a complex structure as the accuracy differs from one component to another such as in horizontal coordinates, Y components by CorsTR_VRS and Cors_TR_ FKP showed ‘best’ results while the same technique provided X components consistent accuracy with the Y component but less accurate than by real time kinematic (RTK). In vertical components, of all the techniques used for the h components, CorsTR_VRS showed ‘best’ accuracy with three outliers.


INTRODUCTION
Global Positioning System (GPS) can provide position fixing in cm level when used in differential mode (Seeber, 2003).This requires at least two receivers, sophisticated software and precise ephemerides, etc.This demands time and does not provide real time positioning.Alternatively, to provide cm level accuracy in position fixing using GPS, countries establish country-wide-cors (Continuously Operating Reference Stations) network (Sunantyo, 2009).It provides a stable and precise positioning in real time.Turkey has established its own network called CORS-TR (TUSAGA-active) (Eren et al., 2009).Along with this even there exists a private CORS system to serve only in a confined area such as Istanbul Municipality Cors Network called Iski-Cors (www.iski.gov.tr).In general, cors network consists of a number of continuously operating stations whose positions are accurately known.Therefore, correction parameters due to ionosphere, troposphere, time, etc. can be calculated and sent to a user that requests corrections ( Öcalan & Tunalıoğlu, 2010).Then the user uses these corrections to its observations to estimate its position in cm level.A number of methods or techniques for calculating correction parameters exists namely; virtual reference station (VRS) (Wanninger, 2003), linear area corrections (Flachen Korrektur Parameter =FKP) (Wübbena & Bagge, 1998) and Master Auxiliary Concept (MAC) Methods (Brown et al., 2005).They have advantageous and disadvantageous over one another.However, which method provides accurate service is still continuing debate and draws scientist attention.Few studies done to investigate the accurate methods include Eren et al. (2009) and Butun Baybura (2010); so more studies need carrying out on the subject to make a clear understanding.
This paper aims at determining accurate method among CorsTR_VRS, CorsTR_FKP, IskiCors_VRS, IskiCors_FKP and classical RTK compared to the coordinates obtained by conventional geodetic position fixing method using total stations.A test network was established in Yildiz Technical University Davutpaşa Campus.The network consists of 49 stations whose positions were fixed by the methods mentioned above and the results were presented.

MATERIALS AND METHODS
To find out accurate method of determining point coordinates by GPS, there are two cors networks involved in this study namely; Cors-TR covered countrywide Turkey and Iski-Cors consists of only 8 stations serving only in Istanbul city, Turkey.A brief description of the networks and the method used to calculate corrections virtual reference station (VRS) and Flachen Korrektur Parameter (FKP) are described.Then acquired data was introduced.

Continuous Operating Reference Stations Networks (Cors-NET)
Differential Global Positioning System (DGPS) based on pseudo-range observations provides real time positioning in meter level provided that there is/are satellite broadcasting correction parameters a roving receiver whose coordinates are in question (Lapucha & Maynard, 1992).However, for applications demanding high precision, phase observations are essential.In this case, integer ambiguity of the number of wavelength makes it more complicated.Current advances in GPS technology enabled us to determine real time positions using phase observations.This is called Real Time Kinematic (RTK) method.However, the distance between reference station and roving station affects the accuracy of position fixing.
To overcome this distance dependence problem, WAAS, WADGPS, etc. have been developed based on a number of control stations located in large areas but they also provide accuracy in decimeter level due to code observations ( Alves et al., 2011).Alternatively, countries established country-wide continuous operating stations network (Cors-Net) (Kahveci, 2009).To make Cors-Net clearly understood, first classical RTK will be given here.

Classical Real Time Kinematic (RTK)
This technique requires a station whose coordinates are precisely known and a roving receiver whose coordinates are in question.RTK is based on phase measurements and communication to roving station.Either raw observations or calculated corrections obtained at the reference station are sent to the roving receiver.The corrections, which are calculated at either the reference station or roving receiver, including position, atmosphere, pseudorange, etc. are possible due to known reference station coordinates.The atmospheric corrections are valid only within the limited area (~15-20 km) from the reference station location.

DATA ANALYSIS
Accuracy is 'closeness' of quantities to their true values while precision is 'closeness' of quantities to their mean values.The strategy followed in this study is as follows.In a sample of coordinates representing the different positions (49 points) with different methods (five methods namely; classical RTK, CorsTR_FKP, CorsTR_VRS, IskiCors_FKP, and IskiCors_VRS), accuracy is the standard deviation of the coordinates differences from the true values.However, it is important to have data free from outlier.It is a well-known fact that an outlier contained in the data influence the mean of the data set dramatically.Therefore, a robust statistic, median, normality, shape of the data are necessary to comment on the data.A box plot may be drawn to see some of the above information in one place.A box plot is a graphic which can be interpreted in terms of spread, centrality, shape and unusual features.
The value of median is known to be the measure of centrality, and simultaneous examination of inter quartile range and the median value can reveal the shape of data.If there is any data outside the limit of whiskers, which means the data possess unusual features.

RESULTS AND DISCUSSION
Point coordinates determined by total station are assumed to be free from errors and 'most' accurate method among the ones used in this study.Therefore, the coordinates obtained by this method were taken as 'true' coordinates.To investigate accurate methods we have used for 3-D positioning by GPS technology, accuracy estimation were performed by taking coordinate differences between true coordinates and the coordinates by classical RTK, CorsTR_FKP, CorsTR_VRS, IskiCors_FKP, and IskiCors_VRS have been taken correspondingly.The results of the differences organized in Y, X and ellipsoidal height h are given in Figures 4, 5,  and 6.
It is clear from Figure 4 that Y components (green line) of CorsTR_FKP are 'most' deviated among others, while X component (Figure 5) partially good and bad and h component (Figure 6) is the 'best' of all when one or two points are excluded from the set.Not in the Y and h components but in X component of the points by IskiCors_FKP and IskiCors_VRS have presented systematical shift from the other three methods, which need further consideration ( Figure 5).Below boxplots were created for the component differences to make sure the data are free from outliers.Figure 7 shows boxplots of Y component differences of the stations for the methods to be tested.It is clear from Figure 7 that Y components ranges are in order from 'best' to 'worst' as CorsTR_VRS, RTK, IskiCors_VRS, IskiCors_FKP and CorsTR_FKP.CorsTR_FKP range is the largest of all.An explanation to this might be CorsTR_FKP parameters were calculated from entire network and then were sent to the user via the nearest Cors_TR station, which may not be convenient parameters as compared to IskiCors_FKP, which were calculated from only 8 stations installed in Istanbul where this test data was collected.For this reason, IskiCors_FKP seems better than CorsTR_FKP.
It is expected that the centers of data from the methods tested tent to zero, because they are differences of Y components with respect to true coordinates.From the Figure 7, CorsTR_FKP and VRS are the closest to zero while the rest were approximately equally apart from the zero line.It is interesting that CorsTR_FKP technique produced zero median although its spread is the largest.
Moreover, IskiCors_FKP and VRS sound like Y components underwent some amount of shift among others.CorsTR_VRS coordinates showed symmetric distribution, which means the data is normally distributed while CorsTR_FKP slightly left skewed.It is clear from the figure that IskiCors_FKP and VRS data were left skewed while RTK slightly right skewed with no unusual features gap or outlier.Of all the methods used in this study for the test, standard deviations of h components are half order lower than horizontal components.
The results in Table 1 may be compared to those of Gordini et al. (2006) which represented the evaluations of Cors Network Technologies included two cors network namely VICpos and MELBpos in Australia.They collected samples of 173795 epochs for only one test point from a sparse network, and produced the standard deviations in cm level and one order of higher magnitude for altimetric components.The results in the Table 1 are found to be slightly worse than that of Gordini et al. (2006) The reason for this might be that we have calculated the differences between the assumed true coordinates (obtained by total stations) and those of the method (five different methods) to be compared to as opposed to one method.

CONCLUSION
This study involves 49 observations site locations in Istanbul whose coordinates were determined by CorsTR-VRS, CorsTR_FKP, IskiCors_VRS, IskiCors_FKP, classical RTK and conventional geodetic position fixed method with total station.The coordinates components were related to the conventional method and their simple differences were taken to compare one technique to another.
The results showed a complex structure as the accuracy differs from one component to another such as Y components showed 'best' results by CorsTR_VRS and Cors_TR_FKP while the same technique provided X components consistent accuracy with the Y component but less accurate than RTK.Of all the techniques used for the h components, CorsTR_VRS showed 'best' accuracy with three outliers.
Another important conclusion one can draw from the test is that IskiCors_VRS and FKP showed some shift in X and Y component which need further research.
This result is compatible with Eren et al. (2009).Based on the results we obtained, it may be recommended that both CorsTR_VRS and FKP technique may be considered before final coordinate determinations.

Figure 4 -
Figure 4 -Comparison of Differences in Y Components (in meters).

Figure 5 -Figure 6 -
Figure 5 -Comparison of Differences in X Components (in meters).
X components are in order from best to worst classical RTK, CorsTR_FKP, IskiCors_VRS, and IskiCors_FKP and CorsTR_VRS.Here the explanation made earlier for IskiCors_FKP components is not supported, this may be because IskiCors_FKP and IskiCors_VRS showed shift pattern from the rest. .