A terrain-referenced navigation system works by comparing what the aircraft sees with a stored reference of what the ground should look like. The matching gets most of the attention, but the reference is the half the operator controls. A system with excellent algorithms and a poor map will produce poor positions, and a disappointing flight can often be traced back to a reference dataset that did not cover the right area, was out of date, or was at the wrong resolution for how the aircraft was actually flown. This article covers what to prepare before the aircraft leaves the ground.
The two layers: elevation and appearance
Most terrain references combine two kinds of data.
A digital elevation model (DEM) is a grid of ground heights. It describes the shape of the land: ridgelines, valleys, pit walls and embankments. Strictly, a digital terrain model shows bare earth with vegetation and structures removed, while a digital surface model includes the tops of trees and buildings. The distinction matters, because a downward camera sees the surface, not the ground beneath a forest canopy.
Orthoimagery is aerial or satellite photography that has been corrected so that every pixel sits at its true map position. It describes what the ground looks like: roads, tracks, fence lines, dams, vegetation patterns and buildings.
Elevation changes slowly and is unaffected by lighting or season, but on flat ground it contains very little information. Imagery is rich in detail but changes with season, sun angle and land use. The two complement each other, which is why the combination is common.
Australian sources
Australia is well served for public elevation data. Geoscience Australia publishes national elevation products that cover the whole continent, and finer, lidar-derived elevation data collected by the states and territories is available for many areas. Coverage and resolution vary from place to place, and tend to be better in settled regions than remote ones, so check what exists for your actual operating area rather than assuming.
State and territory spatial agencies publish their own elevation and imagery products and are often the best source of recent aerial photography for their jurisdictions. Freely available satellite imagery covers everywhere at coarser resolution, and commercial aerial and satellite providers offer much finer imagery at a cost. Check each product’s current specifications with its publisher rather than relying on figures quoted second-hand.
Licensing deserves a check before anything is loaded onto an aircraft. Much Australian government data is released under open licences, but not all of it, and commercial imagery comes with its own conditions on use and redistribution.
Coverage margins
The reference must cover everywhere the aircraft might go, not just the planned route. A useful checklist for the extent of the dataset:
- The planned route or survey area.
- The camera footprint either side of the track, which widens with altitude, plus extra for the oblique view during banked turns.
- Wind drift, and the aircraft’s turn radius at the ends of survey lines.
- Loiter areas, holding patterns and the return-to-home route.
- Diversion routes and contingency landing sites.
- The area the aircraft might plausibly be in when GPS is lost, allowing for the position uncertainty at that moment.
It is tempting to trim the dataset tightly to save storage or preparation time. Resist it. An aircraft that drifts off the edge of its reference has lost terrain navigation at the moment it probably needs it most. A generous corridor at modest resolution, with finer data along the core route, is usually a better trade than a tight corridor at high resolution.
Currency: how old is too old
Elevation data ages slowly in most places, but not everywhere. Open-cut mines, quarries, stockpiles, new roads, earthworks and landslips all change the shape of the ground, and in an active mine the relevant surfaces may change week to week. For those areas the reference should come from the operator’s own recent survey rather than a public dataset captured years earlier.
Imagery ages faster. Consider:
- Land use: cleared blocks, new sheds, rotated crops and removed tree lines.
- Seasonal change: green versus dry pasture, crop stages, and water levels in dams and rivers.
- Events: bushfire scars and flood debris can transform the appearance of the ground in a day and take years to fade.
- Time of day at capture: shadows in the reference fall one way, and shadows at flight time may fall another.
Always check the capture date of any imagery before relying on it, and where there is a choice, prefer imagery captured in a similar season to the planned flights.
Resolution trade-offs
Finer is not automatically better. The useful resolution depends on the ground sample distance the aircraft’s camera achieves at its operating altitude, which is set by the altitude, the lens focal length and the sensor’s pixel size. A reference much finer than that adds storage and processing without adding usable information. A reference much coarser than that loses the small features, such as tracks and fence lines, that matter most over otherwise plain country.
Some practical rules:
- Match the reference resolution to the altitude band you actually fly, not the lowest altitude you might conceivably fly.
- If the mission spans very different altitudes, consider preparing coarser and finer layers rather than one compromise.
- Allow for storage and load time on the aircraft’s embedded computer, especially for long corridors.
- Over flat, uniform country, imagery resolution and recency matter more than elevation detail, because the elevation model has little to offer there.
Datums and heights
Two coordinate issues regularly catch people out. The first is the horizontal datum: Australia has used more than one national datum over the years, and data prepared on different datums will not line up exactly. Mixing them silently produces a constant offset that looks like a navigation error.
Heights are the second trap. Elevation products are usually referenced to a height datum that approximates mean sea level, while satellite receivers and many autopilots work in heights above the ellipsoid, and the difference between the two varies from place to place. Every layer in the reference, and every height the autopilot uses, should be on a known and consistent datum, converted with the official model for your area.
A pre-flight data checklist
Before loading the reference:
- Confirm the extent covers the route, camera footprint, contingencies and a margin for uncertainty.
- Record the source and capture date of every layer, and check it against recent change on the ground.
- Confirm the resolution suits the planned altitude band.
- Confirm horizontal and vertical datums are consistent across all layers.
- Spot-check the reference against known points, such as a survey mark or a road intersection.
- Keep a copy of exactly what was loaded, so flight results can be traced back to the data used.
Where TerrainSLAM fits
TerrainSLAM navigates by matching a downward camera’s observations against pre-loaded terrain data, onboard, with no satellite signal, ground station or cloud connection. Because the reference is loaded before flight, preparing it is part of planning a mission, in the same way as the flight plan and the battery check. TerrainSLAM adapts its processing across altitudes and compensates for haze, heat shimmer and changing light, but no matching system can recognise ground that has changed beyond what its reference shows. If you are planning an operating area and want to talk through its data requirements, our engineers are happy to help.