The Changing Nature of Surveying Infrastructure from Marks in the Ground to Virtual Reference Stations

Surveying infrastructure has always relied on networks of marks in the ground. However, that approach is increasingly giving way to permanently running GPS base stations making data available for real time positioning or for Internet based post processing. The first part of this paper outlines the results of a pilot network established over South East Queensland to investigate the Virtual Reference Station (VRS) concept from Trimble. The system involves permanently running GPS base stations, at spacings up to 70km, feeding GPS data to a processing centre via a computer network. The central processing facility then models spatial errors that limit GPS accuracy. Corrections are then generated for roving receivers to be positioned anywhere inside the network with an accuracy better than a few centimetres in real time. VRS overcomes three main limitations of the current real time kinematic (RTK) technique. Firstly, operators no longer need to establish and run their own base GPS receiver and base radio every time they want to work. Secondly, the use of mobile phone technology overcomes the limitation of the range of radio communications. Thirdly, multiple base stations increase the redundancy and thus the confidence in the resulting rover positions. The second part of this paper examines the implications of the increasingly virtual nature of our surveying infrastructure. The VRS concept lends itself to high use areas such as large urban areas and regions where there are high cost activities like mining or precision agriculture. In remote areas, where it takes a long time to travel to existing geodetic marks and where the cost of maintaining them is high, on-line processing, such as offered by AUSLIG, can be an alternative. Approaches such as the Victorian GPSNet offer a useful approach on a regional basis. One question that arises is whether these approaches can replace ground mark infrastructure? In some cases, these virtual approaches may be less viable and it may be most efficient to run individual GPS surveys that continue to rely on a traditional geodetic network. The most appropriate approach will vary depending on the mix of urban, rural and remote land to be served in a given agency's jurisdiction. This paper opens these and related issues for discussion in an attempt to develop appropriate models for surveying infrastructure in the 21st century. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 1 THE VIRTUAL REFERENCE STATION CONCEPT The Virtual Reference Station (VRS) concept from Trimble is an extension of the real time kinematic (RTK) technique developed for GPS surveying and other forms of high accuracy positioning. With RTK, one can establish a reference station (or base station) at a known point and broadcast the data from the reference station to one or more roving receivers. The computer processor at the roving receivers combines the reference station data with the rover data. With modern equipment, only a few tens of seconds of data are typically required to fix the ambiguities associated with the GPS phase data observable and compute a GPS baseline; the difference in latitude, longitude and height between the reference and rover positions. RTK enables the roving receivers to be positioned with accuracy better than a few centimetres relative to the reference station. Before RTK, GPS baselines had to be post-processed in the office. The ability of RTK to process and display results in real time is further revolutionising the productivity achievable with GPS. VRS takes the productivity increase a step further by overcoming three main limitations of the current RTK technique. Firstly, operators no longer need to establish and run a GPS receiver and radio at their own reference station every time they want to work. Secondly, the use of mobile phone technology overcomes the limitation of the range of radio communications. Thirdly, multiple reference stations increase the redundancy and thus the confidence in the resulting rover positions. The VRS concept involves permanently running GPS reference stations, at spacings up to 70km. They feed their GPS data to a central processing computer via a computer network. The central processing computer can use the reference station data to model spatial errors that limit GPS accuracy and generate appropriate corrections. From the user’s perspective, a roving receiver makes a mobile phone call into the central processing facility, supplying its approximate position (based on a GPS navigation position) and requesting corrections. The central processing computer then generates corrections as though there was a reference station at the coordinates of the rover’s approximate position and the rover is positioned relative to this virtual reference station. For more information on the technical background to VRS see Vollath et al (2000a), Vollath et al (2000b) and Trimble (2000). THE AUSTRALIAN VRS PILOT PROJECT The Department of Natural Resources and Mines (NR&M) of the Queensland Government is responsible for the surveying and geodetic infrastructure of the Australian state of Queensland. Given the potential impact of VRS on future development of that infrastructure, NR&M approached Trimble Australia and its agent in the area, Ultimate Positioning, to establish a pilot project (see Higgins, 2001 and Higgins and Talbot, 2001). The location for the VRS network is shown in Figure 1 and the following points are relevant: The corridor between Brisbane and the Gold Coast is the fastest growing urban area in Australia. That was significant for testing commercial viability of the VRS network. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 2 The station locations are NR&M offices, all of which were linked by high-speed Wide Area Network (WAN) with each building also having a Local Area Network (LAN) enabling all sites to be linked by frame relay. While not directly relevant for this pilot, NR&M has more than 30 such District Offices spread across the state. The stations at Ipswich, Beenleigh and Gold Coast are NR&M District Offices. The Brisbane station was at the Landcentre building, which houses Survey Infrastructure Services (the section in NR&M responsible for the project). The Landcentre is also home to the Information Technology Section for NR&M, which oversee the operation of the WAN and LANs. As such, the Landcentre was the logical site for GPS Network Control Centre. The distance between Ipswich and Gold Coast (76km) was at the limits of station spacing recommended for VRS. This, along with the smaller triangle formed by Ipswich, Brisbane and Beenleigh enabled rover testing under conditions ranging from ideal to limiting. There was also the possibility of running with and without the Beenleigh station to simulate problems with station drop out etc. The Beenleigh receiver could also be run as though it was a rover. That enabled long duration rover data to be gathered. Rover testing was facilitated by the VRS network running over a 3 to 5km density network of existing control stations observed with static and fast static GPS (also known as quick static or rapid static). Figure 1 Location of VRS Base Stations for the Pilot Project 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 3 GOALS OF THE VRS PILOT PROJECT The goals of the pilot project were broken into three main areas as detailed below. GOAL 1: ESTABLISHING AND RUNNING THE VRS NETWORK The project investigated the technical and financial aspects of establishing and running a VRS network, including: Establishing Stations on NR&M Buildings Connections between network base stations using LAN/WAN. Connection of the GPS Network Control Centre to the mobile phone network. Figure 2 shows the architecture for linking the GPS reference stations, the computer network and the mobile phone network. The pilot project uses personal computers running Microsoft Windows NT 4 at each of the reference stations. Their basic purpose is simply to translate the GPS data from the serial port to network protocol (TCP) and send it to the central processing computer as required. As such, personal computers have much more power than is necessary for a production situation but were used for the pilot project to facilitate remote control of reference stations (eg to simulate outages) and to facilitate monitoring of load on the WAN/LAN. Figure 2 Information Technology Architecture for VRS At each site fibre optic cable was used between the GPS PC and the rest of the LAN/WAN. This helped to protect against lightning strike damaging the LAN. For the GPS component, Trimble equipment was used at all sites. At the Landcentre, Ipswich, and Gold Coast sites, 4700 receivers and choke ring antennae, supplied by NR&M, were used. At the Beenleigh site, Trimble supplied the newly released 5700 receiver and Zephyr Geodetic antenna. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 4 GOAL 2: GPS ROVERS WITHIN A VRS NETWORK The project also investigated the technical and financial aspects of running GPS rovers within a VRS network. Testing included matters such as: Establishing that the VRS solution allowed the use of RTK positioning at all suitable locations within the Network with accuracy and initialisation times that were independent of distance from a physical reference station. Establishing how quality measures degraded as rovers moved outside the network coverage. Testing RTK positioning across a range of applications. As mentioned earlier, the VRS network runs over a 3 to 5km density network of existing control stations observed with static and fast static GPS. The homogeneity of the network and its direct connection to the networks forming the backbone of the Geocentric Datum of Australia enable testing against high quality three dimensional coordinates (including consistent ellipsoidal height). A least squares adjustment of the local geodetic network, including GPS baselines between the reference stations and connecting to local marks, ensured that the VRS network maintained a high level of homogeneity with the underlying geodetic network. Stations in this geodetic network were occupied with VRS rovers to quantify the accuracy achievable inside and outside the VRS network. GOAL 3: COMMERCIAL VIABILITY As well as the technical aspects, the project investigated the commercial viability of a VRS network, including potential clients, revenue etc. Sound assessment of commercial viability was facilitated through involvement of key players in government and industry in the fields of surveying, earth moving, mining and agriculture. These included players directly interested in the trial network coverage such as Main Roads Department, local authorities and large surveying firms. There was also exposure to players outside the direct trial network coverage who have an interest in VRS for their areas of operation. These included mining and precision agriculture operators and NR&M's equivalents in other Australian jurisdictions and New Zealand. RESULTS FROM INITIAL VRS TESTING Once the Network was established, it was possible to begin initial testing of roving receivers. The set up for a roving receiver is similar to that for normal RTK except that rather than a radio the GPS receiver is connected to a GSM data module with access to the local mobile phone network. For this project, the rovers mostly used were the recently released Trimble 5700 receivers. In all the testing reported in this paper, a given station was occupied with VRS multiple times and/or with multiple initialisations (ie multiple resolutions of double difference phase ambiguities). Once initialisation was gained, 60 seconds of data was recorded for each occupation at a given station. Often during the testing, an antenna splitter was used to enable a 4700 receiver to operate off the same GPS antenna as a 5700 receiver. On some occasions, this was to enable comparison of the performance of the 4700 against the 5700 receiver. On other 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 5 occasions the 4700 receiver was used to gather fast static data to act as truth for comparing the VRS results from the 5700 receiver. For a particular station, VRS results were compared against the existing latitude, longitude and ellipsoidal height. Another critical point that was noted was the time taken to initialise. As well as such quantitative results, testing also involved noting any anomalies with the performance of the equipment. It should be noted that for this initial testing many aspects of the network and software were still being tested and tuned and there was a deliberate effort to test at the limits of the system. Also during this testing, for more than half the time Beenleigh was being run as though it was a rover station, not a base. Six control stations were used in this initial testing and their locations are shown in Figure 3. Additional characteristics of this testing were: 115 occupations over 2 weeks 6 stations shown in Figure 3 with 3 inside network and 3 outside. Some occupations using 5700 receiver and some using 4700 receiver 52 of the occupations were while Beenleigh was in the network as a base. 63 were while Beenleigh was running as a rover, ie VRS rovers were working from the large surrounding triangle. Taking the above into account, the nearest base to observed stations ranged from 8 to 30km, with an average of 18km. Figure 3 Control for Initial Testing 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 6 For analysis of this initial testing, the residuals between the observed and known positions were expressed as a horizontal (or planimetric) vector and a vertical vector without sign. Of the 115 occupations, 2 had horizontal residuals greater than 1m and were rejected immediately. This level of gross failure is often seen in conventional RTK work and justifies best practice recommendations that ensure multiple occupations or other means of checking for gross error, eg using other techniques. Of the remaining 113 occupations all residuals were less than 0.25m. A further analysis was carried out to reject any occupations with residuals (horizontal or vertical) more than 3 standard deviations from the mean. This iterative process rejected a further 7 occupations before stabilising with all remaining occupations having residuals less than 3 standard deviations from the mean. Table 1 below shows the results for the remaining 106 occupations. Table 1 Results from Initial Testing with 6 Stations Horizontal Distance Absolute Height 3D Vector Mean 0.032m 0.040m 0.054m 0.01 + X ppm 1.4 2.0 2.9 Standard Deviation 0.014m 0.031m 0.029m For this initial testing, the average initialisation time was 2 minutes and all reported occupations were less than 5 minutes. There were other occupations where initialisation time was more than 5 minutes and the initialisation was restarted. Many of those occurrences were while experimenting with various hardware or software configurations. Given that in this stage of the project work focussed on deliberately trying to defeat the system and to test at limits beyond those recommended, these initial results were very encouraging. VRS INITIALISATION TESTING There are many variables that can affect initialisation time during the type of field testing outlined in the previous section. These include time dependent variables such as number of satellites and their geometry as well as site dependent variables such as multipath conditions, RF interference and GSM signal strength. Therefore it was decided to perform a more rigorous test of initialisation time by running the Beenleigh base station as though it was a rover. It is possible to reverse the normal data flow and have the Beenleigh receiver accept VRS corrections and log the results to the local PC. Using special diagnostic software from Trimble the receiver was set to initialise, record for 5 seconds and then re-initialise and repeat this process continually for 15 hours. As an aside, it should be noted that the nature of networked stations is such that it was possible to set up this test remotely from the central PC at the Landcentre site in Brisbane. The approach using the diagnostic software enables statistics about initialisation time to be accumulated. From the 15 hours of initialisation testing, the results were: For the 510 initialisations with 5 or more satellites, the average initialisation time was 1.7 minutes. Of those, 426 initialisations were with 7 or more satellites and the average initialisation time was 1.3 minutes. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 7 It must be remembered that these results were while Beenleigh was out of the base station network, such that the VRS corrections were being generated using the large surrounding triangle. IPSWICH CITY COUNCIL VRS TESTING As well as the diagnostic style of testing described above, the pilot project also included testing in particular applications in real projects. The first such test was conducted in conjunction with Ipswich City Council to observe 36 survey marks to enable upgrade of the spatial accuracy of the Digital Cadastral Data Base (DCDB – Queensland’s digital cadastral map). The project was north east of Ipswich with all stations just outside the triangle, by distances ranging from 3 to 10km (measured orthogonally off the triangle boundary). The direct distance from the stations to the Ipswich base station ranged from 5.7 to 12km. The survey used three parties; two used Ipswich City Council survey staff and the author led the third. Each party observed each of the 36 stations once using VRS, giving a total of 3 independent occupations for each station. As these were new stations without existing coordinates, the author also observed fast static on two thirds of the stations to create some true values for comparison with the VRS results. This was done by running a 4700 receiver split off the same antenna as the 5700, which was observing VRS. Given the time required to observe a fast static occupation (typically 8 minutes), it was possible at those stations for the author’s party to gather multiple initialisations using VRS. This increased the sample size of VRS occupations. The two parties using with Ipswich City Council survey staff simply went to a mark, initialised, recorded 60 seconds of data and moved on. Using that approach, both parties occupied all 36 marks in 8 working hours. This is all the more impressive given that while they had rented GPS equipment for fast static work in the past, none of them had used Trimble GPS equipment before. The only training they had for this VRS work consisted of the author giving 20 minutes of instruction and a brief written procedure to assist them with the use of the 5700 receivers and Survey Controllers and to understand the overall conduct of the project. On those marks where the author observed fast static, a total of 93 occupations were made using VRS. Of those, 10 occupations were rejected as outliers (using 3 standard deviations), with the maximum error being 0.283m in height. Of the remaining 83 occupations, the following statistics apply: The mean longitude residual was 0.001m with a standard deviation of 0.013m The mean latitude residual was 0.001m with a standard deviation of 0.014m The mean ellipsoidal height residual was 0.015m with a standard deviation of 0.036m For all occupations by all parties, 50% of initialisation times were less than 50 seconds and 95% were less than 130 seconds. In terms of productivity, travelling time between the stations (at an average density of 700m) was the limiting factor, rather than initialisation time. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 8 There were 4 stations where initialisation time was significantly worse than others. The impression during the field operation was that this seemed to be due to limited GSM coverage. This was evident through problems using handheld mobile phones for voice. This was subsequently confirmed by the high correlation between the location of these troublesome rover stations with areas where the map of GSM signal coverage showed a change from being suitable for a handheld mobile phone to recommending a vehicle mounted phone with external antenna. VRS BUSINESS VIABILITY As well as technical assessment, work has also been undertaken on assessment of the business viability for a permanent network in South East Queensland. A detailed analysis of business case has been undertaken with assistance from an external accounting and business development firm. That work has identified such things as establishment and running costs, potential users and applications, risk management issues, charging models and potential revenue streams for a VRS correction service. At the time of writing, that business case is about to presented to NR&M management and initial indications are that a VRS network in South East Queensland is viable as a commercial service. FUTURE DIAGNOSTIC TESTING OF VRS If the production network goes ahead, it will be desirable to conduct further diagnostic testing taking into account a new network configuration. That will include more tests of the effect of moving outside the network. The GSM mobile network coverage problems evident in the Ipswich testing also need to be further investigated and quantified. There will also be a continuation of initial investigations into radio-based solutions for users outside mobile phone coverage or for whom the cost of continuous mobile phone use may be prohibitive. There will also be more testing with other potential user organisations and in various application configurations. Testing to date has concentrated on applications coordinating single points. Future testing will need to more fully investigate the use of VRS in dynamic platforms such as rail and road surveys and in earth moving applications. Useful feedback has also resulted from making base station data available to users of post processed kinematic GPS in high dynamic applications, such as aerial photography and airborne laser scanning. GPS BASE STATION APPROACHES OTHER THAN VRS The remainder of this paper investigates the implications for survey infrastructure from GPS bases station approaches like VRS. However, it is first necessary to realise that there are approaches other than VRS that are also useful in certain situations. Two other examples will be used. The first example, that can significantly impact how survey infrastructure should be approached in certain situations, is the on-line processing service offered by AUSLIG 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 9 and known as AUSPOS. The AUSPOS service will be presented at this Congress (see Dawson et al, 2001). As a brief description for the purpose of this paper; the service is free and uses data from the worldwide set of base stations run under the auspices of the International GPS Service (IGS). An AUSPOS user simply needs to gather data with a single GPS receiver (of suitable quality) and submit that data to the AUSLIG web site where the data is post processed, after which results are emailed back to the user. The achievable accuracy depends on the amount of data gathered and submitted by the user. The minimum observation time from which AUSPOS will accept data is 1 hour. The expected accuracy from the minimum recommended observation time of 6 hours is 1cm horizontal and 2cm vertical. Another example of a base station approach is the network of permanent GPS base stations in Victoria known as GPSNet. The network has the capability to support varying approaches. In its basic form, GPSNet supports post processing of GPS rover data to obtain centimetre accuracy. However, it also has the potential to be augmented to enable single station RTK and, in areas with sufficient station density, to enable VRS. Again, this approach will be presented at this Congress and no further details will be given here (see Hale and Mowlam, 2001). IMPLICATIONS OF INCREASINGLY VIRTUAL INFRASTRUCTURE The remaining sections of this paper examine the implications of these base station approaches and what may be generally characterised as the increasingly virtual nature of our surveying infrastructure. The questions that arise are especially relevant for agencies responsible for supplying and maintaining the surveying infrastructure. However, such questions are also of interest to the users of the infrastructure. Some of the questions lead to problems that will need to be addressed while some are opportunities that arise from being able to deliver the infrastructure in new and timelier ways. Some questions that arise include: What is the most appropriate approach for a given set of circumstances? What are the implications of absolute vs relative positioning in terms of accuracy of the resulting positions? What are the implications for maintenance of the geodetic datum in all its dimensions? What new opportunities arise from being able to deliver (and perhaps receive) positions in real time? Each of these questions is now examined in detail. THE MOST APPROPRIATE APPROACH What is the most appropriate approach for a given set of circumstances? In some circumstances, the AUSPOS on-line processing may offer a viable alternative to the traditional approach of connecting to the local geodetic network. From a user’s perspective, the decision on which approach to use will be based on efficiency. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 10 The AUSPOS on-line processing service will deliver the geodetic datum with sufficient accuracy for most types of projects, given sufficient GPS occupation time on the new stations. The alternative is the traditional approach of using GPS to connect to sufficient marks in the local geodetic network. The time required to do that depends on the density of the local geodetic network and is made up of both travel time and the occupation time required for the GPS baseline lengths involved. In remote areas, it is impossible to supply a geodetic network with a high density of ground marks. For example, in remote areas of Queensland, the high quality GPS derived geodetic network is at a nominal density of 75 to 100km. In some states, ground marks are sparser than that. In such areas, connecting to two or three marks will require significant field time for travel and for GPS baseline observation time. The total time may amount to a day or more. There may also be limitations due to the number of GPS receivers available. In such cases a 12 to 15 hour occupation time, say overnight, processed with the AUSPOS on-line approach may give a comparable result and be an attractive logistical alternative to using ground mark infrastructure. At the other end of the scale are high activity areas where users require an infrastructure that can deliver the geodetic datum in real time or with post processing of GPS occupation times measured in minutes rather than hours. Such high activity areas include large urban areas and regions where there are high cost activities like mining or precision agriculture. In such areas this has required establishment of ground mark infrastructure to support RTK with users establishing temporary base stations as required or to support post processing using the fast static technique. For such techniques, geodetic networks with densities ranging from 5 to 25km have been required. Given that even in such areas users tend to be clustered in areas of development, a significant percentage of users can also be serviced by well placed permanent GPS base stations supporting fast static post processing. The current configuration of the Victorian GPSNet is an example of an approach that can offer a viable alternative to ground mark infrastructure for many users. In high activity areas, a permanent RTK service can become viable. This approach can be as simple as servicing a radius around a single base station (for example in a regional city). In other cases the number of users and physical size of the area may justify the establishment costs of a networked RTK approach, such as VRS. Typical areas where VRS is viable are large metropolitan areas such as South East Queensland for the VRS project outlined in the first half of this paper. Another good example is greater Melbourne where parts of the Victorian GPSNet can be transitioned from a network supporting post processing to a full RTK network. Even with all these new options, there will be areas where neither AUSPOS, post processing from local base stations nor VRS are viable. In those cases, it may be most efficient to run individual GPS surveys that continue to rely on a traditional geodetic 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 11 network. These will typically be those areas that are less remote but still rural in nature and where survey activity is only occasional. Even in areas where these new options are viable, suppliers of the infrastructure will still need to consider whether ground mark infrastructure can be totally replaced. There may still be a need for the infrastructure to support users who may be slower to take up these new techniques. Some transition period is likely to be required. ABSOLUTE VS RELATIVE ACCURACY What are the implications of absolute vs relative positioning in terms of accuracy of the resulting positions? Approaches to surveys based on GPS base stations lead to positions that can be thought of as absolute positions within that particular network. Questions then arise about the accuracy of new stations relative to nearby stations. Those nearby stations may be established using the same technique or they may be previously existing geodetic network stations. Existing geodetic stations tend to have good relative accuracy within their network. However, when comparing to a position from AUSPOS one needs to consider the full propagation of error through a hierarchy of geodetic networks back to the IGS stations used in the AUSPOS solution. For example, for the Queensland 100km GPS network, the relative accuracy is typically better than 5cm. However, taking into account its connection to the 500km network and its connection to the IGS stations may give a total absolute accuracy greater than 10cm. In the case of VRS, each initialisation has an accuracy of 1 to 3 cm (horizontal). That needs to be considered when comparing positions derived from different initialisations. On the other hand, it must be remembered that a set of positions derived under the same initialisation may have internal relative accuracy better than the absolute accuracy of the set as a whole. Such issues will vary on a case-by-case basis and need careful consideration to ensure the final positions have sufficient absolute and relative accuracy to meet the requirements of the project. From a broader perspective, this increased mixing of absolute and relative positions will have implications for the management of spatial data generally. For example, this has been recognised in spatial data standards currently emerging that provide better ways of stating accuracy. IMPLICATIONS FOR MAINTENANCE OF THE GEODETIC DATUM IN ALL ITS DIMENSIONS What are the implications for maintenance of the geodetic datum, in all its dimensions? The absolute nature of these base station techniques raises possibilities for differences in how the datum is realised in an area. Such differences in realisation can arise at either 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 12 the reference or rover stations. For horizontal coordinates, the impact of this issue is lessened by the high accuracy and homogeneity of the Geocentric Datum of Australia (GDA). AUSPOS computes in the latest version of the International Terrestrial Reference Frame (ITRF) and at the current absolute position of the tectonic plate. When the user’s position falls in Australia, it is then transformed back to GDA94 (in space and time) based on knowledge of the reference station positions in both frames. That process deals with the transformation at better than the typical noise in the resulting position of the user’s receiver. A similar issue can arise with VRS networks; albeit at a different scale. The coordinates of the physical VRS reference stations need to be consistent at the 1cm level if the VRS software is to model the GPS errors well enough to provide centimetre accuracy corrections to rovers. In the case of the South East Queensland pilot VRS network there was a distortion in the underlying GDA network of 3cm in the 50km of longitude covered by the VRS network. While 3cm is well within the working error of the underlying geodetic network (less than 1 part per million) it is larger than desirable for generating optimal VRS corrections. To get full accuracy and reliability from the VRS network it was necessary to use improved reference station coordinates. At the Ipswich station, that effectively created a variation on the published GDA94 datum in that area, albeit at only the centimetre level. This would be a much more significant issue if the underlying network had larger distortions than is typical of GDA94, as would be the case for establishing VRS in many other areas of the world. As a general comment, these datum definition issues can be minimised if the service provider considers them carefully. The absolute vs relative accuracy issues affecting the roving receivers (outlined in the previous section) are likely to be a greater cause of variations with the horizontal datum in an area. However, it is important to note that all of the above comments about accuracy and about realisation of the datum are much more manageable for horizontal coordinates than for orthometric heights. Variations in the horizontal tend to be geometric in nature and can be managed through a combination of careful selection of reference station coordinates and/or parameters in the processing. On the other hand, variations in orthometric height are more dependent on physical factors that can vary across the network coverage area in a way that is less predictable than for horizontal. At the national scale covered by AUSPOS, the variation between the standard AUSGEOID98 model and the base of the Australian Height Datum (AHD) can amount to more than 1m. Even at the scale of the South East Queensland VRS network, the variation can be greater than 0.2m. The best way to account for this is to create a model of the residual variation surface and add it to the geoid model. In the case of VRS, the roving receivers can apply such a model in real time. The underlying geodetic network in South East Queensland has a large sample of stations that have both orthometric height and ellipsoidal heights derived from static and fast static GPS observations. That facilitates development of a model of the residual variation. 2001 – A Spatial Odyssey : 42 Australian Surveyors Congress 13 For AUSPOS a model of this residual variation is required for the whole country. While some work has begun on this, with coordination through the Intergovernmental Committee on Surveying and Mapping, it will be some time before it will have national coverage. Once such a model is available it could be applied to the user receiver’s position as a step in the AUSPOS processing, similar to the transformation between ITRF and GDA. However, until then, the heights above the geoid that come from AUSPOS may vary from local AHD by more than 1m. In terms of the fourth dimension, time, base station approaches bring an advantage by enabling constant monitoring of the stability of the reference frame. The issue of stability, when viewed internal to Australia (i.e. within the frame of GDA94), relates to the physical stability of the reference station monuments and to their relative positions. When viewed external to Australia, stability relates to issues such as plate tectonics. The fact that this monitoring of stability can be built into the process is an improvement over ground mark based geodetic networks where any movements go unnoticed until the network, or parts of the network are remeasured from time to time.