Several airlines that service the Australasian region have ordered Airbus Industrie's new A380 aircraft; the first is due to fly in to Australia in late 2006. The A380 is the world's largest airliner, boasting a capacity of about 555 passengers. A cargo version can carry a payload of 150 tonnes. With a wingspan of 79.8 metres, a length of 73 metres and a wheelbase of 30.4 metres, the Airbus A380 is considerably larger than the Boeing 747.
The Airbus A380
There are good economic reasons for introducing the A380, but a significant downside of the new aeroplane is that it will require modifications to existing facilities at airports. These modifications will include increasing the width of runways, increasing the separation between taxiways and runways and strengthening bridges. The A380 does not require a longer runway; it requires a wider one.
Plans are currently being prepared to upgrade Tullamarine, Melbourne's international airport, so that it can take the A380. In order to confirm the as-built specifications of their airport, management decided to obtain a survey, and instructed Connell Wagner in February 2003.
Cross-sections were required every 15 metres along the length of the runway. Likewise, the redesign of the taxiways required the production of detail and contour plans of the taxiways and surrounding areas.
To minimise the effect of our work on the operation of the airport, work was strictly confined to a five-hour period (from midnight to 5 am) on three consecutive Sundays. During this time, air traffic is at a minimum, with a maximum of four scheduled flights.
Airport authorities originally requested AHD heights on the cross-sections of the main runway to an accuracy of 3 mm, but this was not possible given the time we were allowed on the runway. After discussions, these criteria were relaxed to allow 10 mm of error. This opened up the possibility of using GPS.
GPS was attractive because of its ability to achieve high productivity under both daytime and nighttime conditions. In fact it can be better to conduct GPS surveys at night because there is less activity in the lower ionosphere.
Even so, the accuracy requirement was at the limit of kinematic GPS techniques, so stringent data collection and data processing procedures were established to ensure that the required accuracy criterion was satisfied.
A first order levelling survey was carried out at Melbourne Airport in the early 1980s by what was then the Australian Survey Office. As a result of this, there were a few good-quality benchmarks with heights based on AHD in the vicinity of the main runway and taxiways. These points could be confidently used to provide vertical control for the purposes of the present project.
A number of horizontal control points were also available within the survey area, but their quality was not well defined. Two issues had to be resolved in defining the horizontal component of the datum:
• the quality of the control point co-ordinates; and
• the relationship between the local planar datum and GDA94.
A rapid-static GPS survey was conducted over the existing control points, using dual-occupation where possible. Dual-frequency Leica System 500 GPS receivers were used. The observed network was connected to the Melbourne GPSnet base station (about 19 km to the south) to provide a link to GDA94, so we could compute GDA94 co-ordinates for all the points. 2D transformation parameters were also derived so that we could relate the observed GDA94 co-ordinates to the local datum and vice versa.
The Melbourne Airport GPS Control Network. The points labelled 'Pillar', 'ARP' and 'PSM 167' were subsequently used as base stations in the execution of the kinematic GPS surveys of the main runway.
The next stage was the survey of the main runway. One way to meet the accuracy requirements would have been to collect several epochs of kinematic GPS data at a suitable number of points on each cross-section.
However, given that the desired accuracy for the project was close to the limit of what is attainable with kinematic GPS, such an approach was not considered sufficiently robust or definitive.
Instead of occupying discrete points along predetermined cross-sections, it was decided to saturate the runway with thousands of points collected by a process of continuous kinematic GPS surveying. By fitting a 3D surface model to the data after it was collected, it would be possible to detect and remove outliers, as well as assess the quality of the measured data.
Three trolleys were constructed. They were designed so that a receiver and antenna could be securely mounted on each one. A person could then pull them over the runway surface.
The design of the trolleys was important in ensuring that good-quality data would be collected. The trolleys were fitted with fixed axles (no suspension), so that the ride height would be constant. The height of the trolley was measured by placing the trolley on a flat surface and measuring from that surface to the base of the antenna. The wheelbase of the trolleys was kept quite narrow so that the height difference between any irregularities in the runway surface and the plane from which the height was measured would be insignificant.
While the runway falls about 30 metres over its 3.6 km length, the grade is fairly constant. The runway also has a slightly convex shape from side to side for the purposes of drainage. As reliable measurement of the antenna height was crucial, a number of repeat measurements were taken to avoid gross errors. The trolleys were made so that they would sit low to the ground, providing better stability.
Measuring the antenna height on the trolleys
Because the antenna would be close to the surface, there were concerns about errors from multipath. To help mitigate this problem, choke ring antennas were used on two of the trolleys.
In addition to the three roving receivers mounted on the trolleys, three other receivers were used as reference stations. They were placed so that the baseline lengths were kept to a minimum and there was good intersection geometry. Baseline length varied between 400 metres and 4.1 km. All sites had an unobstructed view of the sky and there were no obvious sources of multipath. The chosen reference station sites are labelled on page 49 as Pillar, ARP and PSM 167. The accuracy demands of the project meant that we had to monitor data quality in real time. Our field crews needed assurance in the field that the survey would meet its goals. We intended to post-process the data, but the roving receivers were set to operate in real time and thereby provide an indication of performance in the field. When the real-time solution indicated poorer than acceptable vertical accuracy, due to satellite geometry, the survey was halted until the accuracy once again reached a satisfactory level.
We decided that a 3D surface model should be fitted to the kinematic GPS data. Residuals from the computed model would indicate the smoothness of the collected data and allow outliers or problematic data to be identified and investigated. The fitted surface would also provide engineers with a 3D model of the runway from which the required cross-sections could be derived, along with other information relative to the shape of the runway. Statistical analysis of the residuals - the difference between the measured and the modelled runway elevations - would allow the quality of the data to be determined. A total of eight trolley runs on the main runway were carried out over two nights, with data being collected at one second intervals (equating to an observation about every 1.5 metres).
A total of 21,077 points were collected using the trolleys and a further 728 points were collected by conventional kinematic GPS techniques, giving 21,805 points in all.
Two trolley runs, one either side of the centreline, were collected on 23 February. The cross-section data shown on page 52 was collected on the following weekend. On the basis of rejecting one faulty centreline run, the dataset used for the development of the surface model was reduced by 2688 points to 19,117 points.
The surface-fitting computations were carried out using Surfer 8.02, a generalised surface-fitting package developed by Golden Software Inc. in the US. Various modelling options were evaluated before coming to the conclusion that kriging was the technique best suited to the particular requirements of this project.
Kriging is a way of interpolating randomly distributed data onto a regular grid. The contribution of any data point to the interpolation process is controlled through a weighting process, where the weight is determined according to the distance of separation between the data point and the interpolation point.
A particularly useful feature of kriging is the ability to control how rigidly the surface model will follow the measured data.
This is done through careful selection of the variogram - a formula which relates how much weight to give to a reading, its closeness to the point being interpolated - and the refinement of the so-called Nugget Effect. The Nugget Effect adds a smoothing operator to the kriging process, causing the fitted surface to less rigidly follow the behaviour of the measured data.
Not only was it important to select an appropriate surface modelling technique, but experimentation also had to be carried out to choose the most appropriate grid interval. Large grid intervals tended to yield large residuals, so we decided to use small spacing. The only disadvantage of small grid spacing is that it requires significantly more computation. For the purposes of modelling the runway surface, a metre grid was used, with 49 grid lines across the runway and 3662 grid lines along the runway, making 179,438 grid nodes in total.
Since the requirement for this project was to supply runway cross-sections accurate to 10 mm in height, an initial filter of post-surface-fitting residuals of 20 mm was applied. Any point with a residual that exceeded this limit was rejected and the surface fitting was re-computed. A number of iterations were required to reach a result which was deemed acceptable, with no points having residuals greater than the 20 mm limit.
The Tullamarine survey showed that it is possible to get extremely good results in an extremely short space of time using GPS. One needs to be a little inventive sometimes, however.
Geoid Modelling
The height information had to be supplied relative to AHD. It follows that the heights extracted from the processing of the kinematic GPS data had to be converted from the ellipsoidal reference frame to AHD. To achieve this, a geoid model was required.
Past experience indicated that it would not be a good idea to ignore the relative slope between the geoid and the ellipsoid. Under normal circumstances, it would have been possible to develop a localised site geoid model for the purposes of the project, but the sparsity and poor geometric distribution of existing benchmarks made this option unfeasible. So it was decided that the national AUSGeoid98 solution would be used (Geoscience Australia 2003).
Combining AUSGeoid98 with GPS for relative height determination generally allows third order levelling standards to be satisfied. In 1998, Johnston and Featherstone estimated the absolute accuracy of the AUSGeoid98 solution at about 360 mm. The relative accuracy was deemed adequate for the purposes of this project, but the limited absolute accuracy needed to be taken into account to ensure that the orthometric heights derived by combining ellipsoidal heights from GPS with AUSGeoid98 geoid undulations matched known AHD levels at the reference stations.
To this end, an index correction was computed and applied to all geoid undulations.
Connecting the airport control network to the Melbourne GPSNet base station gave an ellipsoidal height at Pillar of 128.397 metres. The geoid undulation at this point was computed from AUSGeoid98 to be 5.437 metres, yielding an orthometric height of 122.960 metres. Precise levelling, on the other hand, gave an AHD height for Pillar of 122.929 metres. The difference between (GPS + AusGeoid98) and AHD was thus 31 mm.
Performing this same computation at a number of other points within the vicinity of the main runway gave similar differences, with a mean of -3 mm and a standard deviation of 2 mm.
The smallness of these differences confirmed the relative accuracy of AUSGeoid98. To harmonise (GPS + AusGeoid98) with AHD for the purposes of this project, an index correction of 31 mm was subsequently subtracted from all (GPS + AUSGeoid98) derived heights.
Philip Collier
