Drone mapping

The August 2012 Te Maari eruption left hundreds of ballistic impact craters across part of the Tongariro landscape. New Zealand Aerial Mapping flew LiDAR and aerial imagery over the area in November that year, producing a broad record of the disturbed surface. Several years later, University of Canterbury research returned to a much smaller crater field with a very different acquisition system. A Draganfly X4P unmanned aircraft carried a digital camera above the site while ground survey points provided positional control. The images were processed in Agisoft PhotoScan Pro to reconstruct the surface in three dimensions.

The 2012 eruption and the later UAV survey were separate events. Ravitej Pitchika’s 2017 MGIS thesis describes field work conducted roughly four years after the eruption. Remembering it as a 2012 drone survey would collapse the event and the later research into one date. Early UAV mapping in New Zealand also appeared through several routes, including research, surveying, engineering and environmental projects that gradually showed what small aircraft and digital photogrammetry could do.

Pitchika's project focused on a site only about 100 square metres in area. The Draganfly was flown at different heights and the photographs were combined with ground-truth survey points. The resulting digital elevation models were used to identify volcanic craters and test how image resolution and flight altitude affected what could be detected. At a 40 metre flight height the model identified 135 craters within the selected site. The thesis compared those results with the much larger LiDAR and orthophoto survey undertaken soon after the eruption and concluded that the UAV approach was feasible and relatively economical for tightly localised work.

The comparison shows the change in acquisition scale. A conventional aircraft carrying a specialist mapping sensor could cover large areas efficiently. A small remotely piloted aircraft could be taken to one slope, one beach, one farm block or one construction site and flown when a particular project required it. The drone did not replace the larger system. It filled a different part of the acquisition spectrum.

The aircraft themselves made the change visible. Earlier aerial mapping usually involved a crewed aircraft, specialist operator, formal flight planning and a contract large enough to justify mobilisation. By the mid-2010s a professional UAV system could travel in a vehicle and be launched close to the site. That reduced the area that had to be worth flying before aerial capture made sense. The practical unit of mapping could shrink from a district or catchment to a landslide, quarry face or work site. A mapping aircraft remained a specialist tool, yet by then it could be small enough to arrive in the back of a vehicle rather than at an airport.

The reduction in scale changed timing as much as cost. A regional orthophoto programme might be repeated every few years. A UAV could return after heavy rain, after a construction stage, after erosion or before and after an engineering intervention. This made aerial data collection more closely resemble field survey. The team controlling the project could choose the day, the flight pattern and the area to be observed, subject to weather, access, aviation rules and the limitations of the aircraft.

The aircraft was only one component. Batteries, camera, GNSS, flight-planning software, ground-control targets, survey equipment, storage, photogrammetric processing and GIS all had to work together. A small machine hovering over a site could look deceptively simple from the ground. Producing a defensible map from it required rather more than pressing the take-off button and admiring the photographs afterwards.

Structure from motion made the new workflow possible for relatively small teams. The basic idea is that the same features appear in many overlapping photographs taken from different positions. Software identifies matching points, estimates the relative location and orientation of the cameras, and reconstructs the three-dimensional arrangement of the visible surface. Multi-view stereo processing can then densify that reconstruction into a much larger cloud of points. The process borrows from photogrammetry and computer vision but became increasingly automated in software that ordinary research and consulting teams could run.

The software calculated three-dimensional geometry from overlapping photographs. A stitched panorama may look continuous while preserving perspective distortions. A photogrammetric model attempts to solve camera geometry and produce spatial measurements. The workflow moves through several different products: original photographs, image metadata, matched features or tie points, camera solutions, sparse and dense point clouds, surface models, orthorectified imagery and later derivatives such as contours, profiles or volumes. Those products should not be treated as interchangeable just because they can all be displayed in a GIS.

The orthomosaic became especially useful because it looked familiar. Photographs corrected for camera perspective and terrain displacement could be combined into one georeferenced image layer. Engineers, planners or environmental staff could use that output in much the same way they used conventional orthophotography, but over a site captured for their own project. At suitable flight heights the ground sample distance could be measured in centimetres rather than metres. That level of visual detail was attractive, but it was not the same thing as centimetre positional accuracy.

The dense three-dimensional point cloud was another reason UAV photogrammetry spread into geospatial work. LiDAR point clouds were created by laser ranging, while SfM could produce a visually similar cloud from photographs alone. To a GIS user, both could support surfaces, sections, profiles, volumetric calculations and three-dimensional display. The similarity of the output, however, should not hide the difference in how it was measured.

Airborne LiDAR sends laser pulses towards the ground and measures their return. Image-based photogrammetry reconstructs geometry from visible texture appearing in multiple photographs. A laser pulse may produce returns through gaps in vegetation and support classification of ground beneath parts of a canopy. A camera usually reconstructs the surfaces it can see. Dense scrub, forest, water, reflective material, uniform surfaces and moving vegetation can all cause difficulties for image matching.

This distinction became increasingly relevant as practitioners gained access to several methods capable of producing high-density terrain data. A point cloud ceased to identify the sensor that created it. Metadata and method became necessary to understand what the points represented. Calling every dense three-dimensional dataset "LiDAR" would be as misleading as calling every raster an aerial photograph.

The 2017 New Zealand Geotechnical Society paper by D. J. Bevan, Martin Brook, Jon Tunnicliffe, Nick Richards and W. M. Prebble provides a clear professional example. The team mapped the Kepa Road landslide in Auckland and the Ohuka landslide south of Port Waikato using a DJI Inspire 1 fitted with a Zenmuse X3 camera. Roughly 200 to 300 overlapping photographs were collected at each site. Up to 20 ground-control points, including targets and identifiable features, were positioned using real-time kinematic GPS.

The images were processed in Agisoft PhotoScan 1.2.6. Blurred images were removed before modelling, camera positions were loaded from image metadata, and the project was aligned within NZTM2000. The workflow produced full-colour orthophotos, digital elevation models and three-dimensional models of the landslide surfaces. The paper presented the method as a practical way to collect high-resolution topographic information in places where other survey methods could be expensive or difficult to deploy.

Ground control is an important part of that story because it separates an impressive image from a controlled spatial product. A consumer camera and onboard GNSS can provide approximate positions, but survey-quality claims require evidence about how the model was tied to the coordinate system and how well it performed against independent measurements. Ground sample distance describes the size represented by an image pixel at the surface. It does not, by itself, say where that pixel is located to the same accuracy.

Later UAVs with RTK or PPK positioning improved the location of camera exposures and could reduce the amount of ground control required in some workflows. They did not abolish the need for quality assurance. Camera calibration, flight geometry, surface texture, coordinate transformations and processing choices still influence the result. A project can have very small pixels and still contain systematic positional error.

The Auckland and Port Waikato landslide work shows another attraction of UAV mapping. Unstable slopes are inconvenient places to carry survey equipment around for long periods. Remote image capture can reduce the amount of time staff need to spend directly on hazardous ground while still providing a detailed three-dimensional record. The 2017 paper argued that UAV/SfM could support engineering-geological mapping of scarps, irregular surfaces and other features at fine scale.

The method became more useful when repeated. A 2019 paper by Martin Brook and Jesse Merkle described UAV/SfM photogrammetry for monitoring active landslides in the Auckland region. High-resolution photomosaics and terrain models could be used to assess deformation and to compare surfaces through time or against LiDAR. A related University of Auckland project funded through EQC, later the Natural Hazards Commission, extended the work into ten monitoring investigations and reported in 2020.

Repeated capture changed the question. One flight could describe a slope. A sequence could show whether parts of it moved, where scarps extended or how the surface responded after rainfall. That still required control and careful comparison. If two surveys were aligned differently, or if vegetation changed between flights, apparent movement might come from the method rather than the landslide.

The same logic applies outside hazards. Construction sites, quarries, stockpiles, river channels, beaches and restoration sites are valuable precisely because they change. A method that can be brought back at project-defined intervals creates a local time series. UAV mapping lowered the threshold at which repeat aerial capture became practical.

Environmental research provides a useful contrast to engineering terrain. In 2015 Grant Lawrence's Auckland University of Technology thesis used low-altitude unmanned aerial imagery at Whatipu Scientific Reserve as part of a project classifying coastal wetland vegetation and investigating change. The study combined high-resolution satellite imagery with UAS imagery whose ground detail was about six centimetres. The small aircraft supplied local observations that could complement rather than replace the broader satellite view.

This difference of scale mirrors the relationship between UAVs and the other sensors covered in the previous chapters. Satellite systems repeatedly observe large areas and provide multispectral consistency. Crewed aerial photography supplies highly detailed regional imagery. LiDAR measures dense three-dimensional points with characteristics particularly useful for terrain. UAVs make it possible to acquire very detailed local imagery at a time chosen by the project. The useful method depends on the question rather than a simple ranking of which technology is newest.

Small aircraft could also carry sensors other than normal visible-light cameras. By the later 2010s New Zealand research included multispectral UAV imagery in horticulture and other environmental work. Those applications also sit within the wider earth-observation lineage, but the UAV contribution is the local scale and control over timing. The same caution still applies to vegetation indices and spectral products. A colourful index is not automatically a direct measurement of plant health or productivity.

Surveyors tested UAV mapping against established accuracy requirements. A high-resolution image is visually persuasive, but a survey workflow needs to know how well coordinates and elevations agree with independent control. The 2017 Auckland and Port Waikato landslide work is useful because it records both the aerial imagery and the RTK-positioned ground targets used to constrain the model. That makes the method comparable with other controlled spatial measurement rather than treating the onboard navigation solution as sufficient.

Ground-control targets serve several purposes. They connect the photogrammetric model to a real coordinate system, help remove scale and orientation ambiguity, and provide a way to detect systematic distortion. Separate checkpoints can be withheld from the model and used afterwards to test how well the reconstructed surface predicts known positions. The distinction between control and check points is easy to lose in casual descriptions, and it becomes central when accuracy is being claimed. A model can fit the control used to build it and still perform less well elsewhere.

The national coordinate framework therefore sits quietly underneath drone mapping just as it does underneath LiDAR. A site model in NZTM2000 can be compared with parcels, roads, earlier surveys and other spatial datasets because it is tied to the same horizontal framework. Heights need an equally clear reference if change in elevation is being measured. The aircraft may be new, but the requirement for coordinate discipline is not.

This also limits what a UAV product can prove. A fence visible in a centimetre-scale orthomosaic may look like an obvious property line. It is still only a visible feature. Legal cadastral boundaries come from the survey record and cannot be relocated by tracing whatever line appears sharpest in the image. Greater visual detail can make old distinctions easier to forget, not less necessary.

Returning to the same site is easy to describe and harder to do consistently. A monitoring programme has to consider flight height, camera, lens, overlap, lighting, vegetation, ground control and the coordinate solution from one campaign to the next. If these change substantially, differences between two surfaces can reflect the survey method as well as the landscape. A time series therefore depends on repeatable acquisition and processing as well as repeated flights.

Vegetation creates a particular problem. Grass, shrubs and trees move in wind and change with season. Photogrammetry reconstructs what the camera sees, so a taller or denser canopy can appear as surface change even when the ground beneath it has not moved. Bare slopes and engineering sites are often easier to compare than forested terrain. Analysts may need to mask vegetation, select stable features or use another sensor where the ground itself is the required measurement.

Lighting and water can be equally troublesome. Strong shadows may hide texture needed for image matching, while low-texture surfaces offer few reliable features. Water moves, reflects the sky and changes appearance between photographs, so SfM often struggles to reconstruct it as a stable surface. Coastal and river projects may still use UAV imagery effectively, but the method does not automatically turn every visible scene into precise geometry.

These limits are part of the reason UAV mapping settled into a toolkit rather than becoming a universal replacement. Practitioners learned which surfaces and conditions suited image-based reconstruction. A technique can become ordinary without becoming suitable for everything.

UAV products often arrive with very small numbers attached to them. Ground sample distance may be two centimetres, five centimetres or another impressively fine figure. That number describes the approximate size represented by an image pixel at the ground surface under the flight geometry. It says nothing by itself about whether a feature is located within two centimetres of its true coordinate.

Several sources of error sit between the camera and the map. The onboard GNSS may locate the camera imperfectly. Lens distortion has to be modelled. Ground-control coordinates have their own uncertainty. Poor flight geometry can weaken the camera solution. Surfaces with little texture may reconstruct badly. Coordinate transformations and elevation references can introduce further problems if they are handled carelessly.

This is why professional UAV reports often include error statistics from checkpoints rather than relying on resolution alone. The map may be extraordinarily detailed and still need a more modest statement about absolute accuracy. Conversely, a model with larger pixels can be well controlled for the purpose at hand. Resolution and accuracy answer different questions.

The distinction also affects volume calculations. Stockpile, cut-and-fill or landslide volumes are derived from differences between surfaces. Small vertical biases across a large area can produce substantial volume errors even when the model looks excellent. Repeated surveys therefore need stable control and consistent processing if the derived change is to be trusted.

UAV mapping created a new archival problem because local projects could generate data far faster than earlier aerial programmes. A single flight might produce hundreds or thousands of original photographs, logs, calibration information, ground-control files, project databases, point clouds, surface models, orthomosaics and reports. Saving only the final JPEG or map image destroys much of the evidence needed to reproduce or assess the work later.

A sensible archive needs to retain enough of the production chain to explain the result. At minimum that includes acquisition date, site, coordinate reference system, control information, key processing details and the authoritative final products. Some projects also need the original imagery and complete processing project, especially where legal, engineering or scientific traceability is important. Storage is cheap compared with reflying an event that cannot be repeated.

The same issue becomes more obvious in monitoring programmes. If each flight is named differently, control files are separated from imagery or surfaces are overwritten by the next campaign, the apparent convenience of frequent acquisition becomes an information-management liability. The useful time series depends on preserving each observation as a dated, comparable record.

This is another way in which drones followed the wider GIS trajectory. The technology reduced friction at the point of capture and shifted some of the effort downstream. Once detailed spatial data could be created by many project teams, organisations had to decide which products were authoritative, how long to retain them and how other users would know what they were looking at.

Professional mapping also required the aviation system to catch up with a rapidly expanding class of small aircraft. On 1 August 2015, a modernised Part 101 and the original Part 102 remotely piloted aircraft framework took effect in New Zealand. Part 101 provided the operating rules for activities that met its conditions. Part 102 created a certification framework for operations that could not be conducted within those Part 101 conditions.

Aviation rules governed how UAV operators could carry out the work. The broader effect is enough. A surveyor, university, engineering consultant or council using a remotely piloted aircraft was operating an aircraft as well as collecting spatial data. Flight planning therefore involved airspace, people, property, consent, weather and safety alongside overlap, camera angle and ground control.

That regulatory setting is one reason UAV mapping matured into a professional activity rather than remaining a hobbyist extension of photography. Organisations needed procedures, competent operators and documented operating limits. The mapping workflow had to fit inside an aviation workflow. A perfect orthomosaic produced from an unlawful or unsafe flight would be a poor professional result.

Privacy and public acceptance sat beside aviation safety. A low-flying camera over a small property can feel more intrusive than a satellite passing hundreds of kilometres overhead or an aircraft conducting a regional imagery programme. Professional projects therefore had to consider access, communication and what the camera might record beyond the immediate mapping target. These concerns formed part of using UAV technology responsibly.

By the late 2010s the New Zealand literature had changed tone. The early question had been whether a small UAV and SfM could produce useful spatial data. Later papers assumed that they could and concentrated on how to apply the method, test it, repeat it and integrate it with other measurements. The 2017 engineering-geology work used RTK control and standard coordinates. The Auckland monitoring programme used repeated terrain models. Environmental projects combined low-altitude imagery with satellite or field data.

Professional drone mapping spread across sectors on different timelines, under different accuracy requirements and for different site sizes. Surveying, hazards, environmental monitoring, agriculture, construction and infrastructure each had their own reasons to adopt or reject the method. That pattern is a better indicator of adoption than finding a company that advertised the first drone service.

The software changed with the practice. Agisoft PhotoScan, later renamed Metashape, became one widely used option, while Pix4D and other packages also appeared in professional workflows. Increasing automation made camera alignment, dense matching and orthomosaic generation accessible to teams that were not specialist photogrammetric bureaux. That accessibility could conceal complexity. A model can process successfully while still being poorly controlled or unsuitable for the measurement being claimed.

Computing also mattered. Hundreds or thousands of high-resolution photographs consume storage and processing time, while dense point clouds and textured meshes can become large very quickly. Faster workstations, GPUs and later cloud processing made larger projects easier. The practical point is simpler: the small aircraft could generate a surprisingly large digital workload after it landed.

Many users needed a map, elevation surface or measurement derived from the model. An orthomosaic could enter the GIS as a basemap. A surface model could support contours, slope calculations or volumes. A profile could be extracted for an engineer. A geologist could map scarps against the reconstructed terrain. The sophisticated photogrammetry often disappeared behind products that fitted familiar GIS practice.

This helped UAV mapping spread. Organisations did not need to redesign every downstream workflow around the aircraft. They could receive a georeferenced raster, a surface or a point cloud and combine it with parcels, assets, geology, field observations or earlier surveys. The acquisition method was new, while much of the analytical environment was already familiar.

At the same time, the products carried new provenance requirements. Users needed the flight date, aircraft and sensor where relevant, coordinate system, control method, processing settings and accuracy information. Repeat projects needed consistent naming, storage and archiving so that one flight could be compared with the next. A folder full of attractive orthomosaics without dates or control information is not much of a monitoring system.

The same issue appears throughout New Zealand's digital mapping history. Making data easier to create increases the need to manage it properly. Drones lowered some barriers to acquisition, but they could also produce duplicate, poorly documented or unnecessarily large datasets with impressive speed. The aircraft made capture easier. It did not make information management optional.

By 31 December 2025, UAV photogrammetry had become one ordinary option for detailed local capture. Drones joined the range of available survey and observation methods. Each method occupies a different combination of area, accuracy, timing, sensor capability, cost, risk and terrain. A regional council still has reasons to commission broad aerial imagery. A national elevation programme still has reasons to use airborne LiDAR. A surveyor still has reasons to occupy points on the ground.

The mature choice is therefore methodological rather than fashionable. A drone is useful when the required area is small enough, access and aviation conditions are manageable, the surface can be reconstructed from imagery, and the project benefits from very high local detail or repeat timing. It is less useful where dense vegetation hides the ground, weather prevents reliable flying, the site is too large, the regulatory environment is difficult or another sensor provides a better measurement.

Te Maari demonstrates the distinction neatly. The November 2012 LiDAR and aerial survey mapped a broad eruption landscape soon after the event. The later Draganfly project worked over a much smaller crater field and examined what could be resolved from local UAV imagery. One did not invalidate the other. They answered different acquisition problems at different scales.

UAV photogrammetry also pushed GIS further away from the assumption that geographic information would be primarily flat. The camera produced photographs, but the processing increasingly produced dense clouds, surfaces and textured three-dimensional models. Engineers and earth scientists could rotate a landslide, inspect a crater from different angles, measure a volume or compare two reconstructed surfaces. Those outputs overlapped with LiDAR and other 3D technologies even though their acquisition methods differed.

By the end of 2025, point clouds and three-dimensional models had become familiar enough that attention was shifting from how they were captured to how organisations used them as general spatial information. Terrain, buildings, infrastructure and other objects could increasingly be combined in explicitly three-dimensional GIS environments, carrying the map beyond a flat representation of place.