Photogrammetry, lidar and small uncrewed aircraft have changed the economics of mapping. Work that once took a crew a fortnight can produce a surface in an afternoon. What has not changed is that these are measurement systems subject to the same discipline as any other: they need control, they need checks, and they produce results that are wrong in specific, predictable ways if either is skipped.
The most useful way to think about them is by what they can and cannot see. Photogrammetry measures what a camera can see. Lidar measures what a laser pulse can reach. Neither sees through solid ground, and only lidar has any chance of reaching the ground through vegetation. Almost every decision about which to use follows from that.
Photogrammetry: geometry from overlapping images
Photogrammetry recovers three-dimensional positions from two-dimensional images. A point appearing in two or more images taken from different positions can be intersected in space, provided the position and orientation of each camera and the internal geometry of the camera are known. Modern software solves for all of those simultaneously in a bundle adjustment, using thousands of automatically matched features, then produces a dense point cloud, a surface and an orthorectified image.
The scale of the imagery is described by ground sample distance, the ground size of one pixel.
GSD = (flying height × pixel pitch) ÷ focal lengthA sensor of 5472 by 3648 pixels at that GSD covers about 179.8 m by 119.9 m on the ground. Typical overlaps of 80 percent forward and 70 percent sideways put the exposure spacing at roughly 24 m along the line and the flight lines about 53.9 m apart. Those numbers determine flight time, image count and processing time, and they are worth computing before the flight rather than discovering afterwards.
Lidar: ranging with light
A lidar unit measures the range to a surface by timing a laser pulse, and combines that range with the position and orientation of the sensor to compute a coordinate for each return. Because it makes its own light and measures range directly, lidar does not depend on surface texture or on ambient illumination, and it works at night and over featureless surfaces such as fresh snow or open sand where photogrammetric matching fails.
Its decisive advantage is vegetation penetration. A single pulse spreading as it travels may strike leaves, branches and finally the ground, producing multiple returns from one emission. Classification algorithms then separate ground returns from vegetation and structures, yielding a bare-earth surface under a canopy where a camera sees nothing but treetops. No amount of photogrammetric processing recovers ground the camera never saw.
Point density is specified in points per square meter, and it maps directly to feature resolution: 8 points per square meter corresponds to an average spacing of about 0.354 m, while 2 points per square meter gives about 0.707 m. Density in the delivered bare-earth class is far lower than the raw density wherever there is canopy, so specify the density required for ground points rather than total points.
Comparing the methods
| Method | Strengths | Limitations |
|---|---|---|
| Aerial photogrammetry from a small uncrewed aircraft | Low cost, imagery and surface together, very high resolution over small sites | Cannot see ground under vegetation; degraded on uniform or reflective surfaces; needs ground control |
| Uncrewed aircraft lidar | Bare earth under moderate canopy; works in poor light; sharp break definition | Higher equipment cost; no inherent imagery; heavier payload cuts endurance |
| Crewed aircraft lidar | Efficient over large areas; consistent quality; dense canopy penetration | Only economic at scale; mobilisation cost; scheduling and weather windows |
| Terrestrial laser scanning | Extremely dense detail on structures and vertical faces | Occlusion behind objects; registration effort; slow over open ground |
| Satellite imagery | Wide area, repeat coverage, no field mobilisation | Resolution and accuracy generally insufficient for engineering design |
| Conventional field survey | Highest accuracy on discrete points; sees inside structures and under any cover | Slow per point; impractical for full-coverage surfaces |
Ground control is not optional
This is the point at which remote sensing projects most often fail. A photogrammetric block without ground control is internally consistent and externally arbitrary: it can be scaled wrongly, rotated, and warped into a bowl by uncorrected camera calibration, and nothing in the software's own reporting will tell you. Onboard positioning of the aircraft improves matters considerably but does not remove the need for independent verification.
- Set targets before the flight, large enough to be unmistakable at the planned GSD and with a clearly identifiable center.
- Distribute them across the block, including at the perimeter and at least one near the center. Control clustered in one area leaves the rest of the block unconstrained.
- Cover the vertical range of the site. Control all at one elevation constrains the surface poorly where the ground rises or falls.
- Survey the targets to an accuracy comfortably better than the accuracy required of the product, and record how they were surveyed and to which datum.
- Hold back a set of checkpoints that are not used in the adjustment. These are the only honest measure of accuracy; residuals at points used to constrain the solution flatter it by construction.
- Report the checkpoint residuals in the deliverable, as a table of differences with a summary statistic.
Fitting them into a workflow
In practice these methods complement field survey rather than replace it. A typical site mapping project might fly the open areas for the surface and imagery, use conventional methods under the tree line and inside the building, and level the sewer inverts because no remote method can see inside a manhole. The remote data supplies coverage; the field data supplies the points where accuracy actually matters.
The deliverable should therefore be honest about provenance. A surface assembled from three sources with different accuracies should carry a note describing which areas came from which method and what accuracy each supports. A designer who assumes uniform accuracy across a blended surface will eventually place something critical in the part that came from the least accurate source.
Finally, treat the classification of a lidar dataset as a deliverable in its own right, not as an automatic by-product. Automated classification is good and not perfect, and the errors concentrate exactly where they matter: low vegetation misclassified as ground raises the surface, and a bridge deck classified as ground blocks a drainage model.