The procedure has eight steps, shown in sequence in Figure 1. The first three are preparation, the fourth is the field capture, and the last four turn photographs into a record that others can use. A return loop marks the repeat surveys that document the advance of a quarry.
Sequence of the proposed recording procedure, with repeat survey as quarrying advances
Work starts with what is already known: earlier survey reports, district gazetteers, topographic sheets and free satellite imagery, on which the larger circles and cairns can often be seen. The limits of quarry leases and the direction of working are established from the mining authority or by observation, since they decide which part of the landscape is surveyed first. Permission is needed from the landowner, the archaeological authority and the aviation regulator. Rules differ between countries and change often. In India, for example, operations are governed by the Drone Rules (Ministry of Civil Aviation, 2021), which classify aircraft by weight and restrict flight in designated zones. An active quarry adds its own hazards, and blasting times must be known.
Targets that will be visible from the air, such as painted crosses on bedrock or printed boards pinned to the ground, are set out before flying and their positions measured. Their number and distribution matter more than any other single factor for the accuracy of the result (Sanz-Ablanedo et al., 2018). They should surround the survey area, include points in the interior and span its range of heights. A further set, the checkpoints, is measured in the same way but kept out of the processing so that it can test the result.
Survey-grade satellite positioning receivers are the ideal instrument and can often be borrowed from an engineering or geography department. Where none is available, a local grid can be established with a total station, or with careful taped distances between targets, which gives a model that is correct in scale and internally consistent though only approximately placed on the map. That is an acceptable compromise for a rescue record, provided it is stated.
Flying height sets the ground sampling distance, the size of one pixel on the ground. For megaliths the aim is that a slab edge or a kerb stone is several pixels wide, which with a typical consumer camera means flying low. Flight lines are planned with high forward and side overlap, so that every point on the ground appears in many photographs. A second set of lines flown at right angles, with the camera tilted away from vertical, is planned as well. The oblique images strengthen the geometry of the block, counter the doming to which self-calibrated vertical blocks are prone (James and Robson, 2014), and show the sides of upright stones.
Flights are made in even light, under thin cloud or within a short interval around midday, so that shadows do not move between flight lines. Exposure and focus are fixed, the shutter speed is kept fast and the aircraft slow. Images are saved at full resolution. Each battery covers a limited area, and the block is divided into overlapping sections accordingly.
The aircraft cannot see inside a chamber or under a capstone. Selected monuments, including every type present and any with port-holes, passages or carvings, are therefore photographed on foot with the same or a better camera, circling each at two or three heights with a scale bar in view. If these pictures overlap with the low oblique aerial images, the software can join the two sets in one model.
Photographs are first sorted and blurred frames removed. Alignment produces camera positions and a sparse cloud. The control targets are marked in the images, and the bundle adjustment is run again with them as constraints. Dense matching then produces the point cloud from which a surface model and orthomosaic are generated. Whatever software is used, the settings, the version and the number of images are noted. General accounts of the workflow and its pitfalls are given by Smith et al. (2016).
The coordinates of the checkpoints are read from the finished model and compared with their surveyed values. The differences are summarised as root mean square error in plan and in height and reported with the product. A check on the residuals at control points alone is not enough, because the adjustment has been fitted to them. If the checkpoint errors show a pattern, such as height error growing towards the edges, the block is reprocessed with different calibration settings or more images before any map is drawn.
The orthomosaic and elevation model are loaded into a geographic information system. Low cairns and robbed circles that are nearly invisible in a photograph often show clearly in the elevation model once it is displayed with a relief visualisation such as hillshading from several directions or the sky-view factor (Kokalj et al., 2011). Each candidate monument is digitised, given an identifier and described by type, dimensions and condition. Because natural boulder clusters mimic cairns, the inventory is verified on the ground with the map in hand before it is treated as final.
The record consists of the original photographs, the control and checkpoint coordinates, the processing report, the derived products and the inventory, together with a plain statement of how each was made. Professional guidance on photogrammetry for heritage stresses that the raw images are the primary archive, since they can be reprocessed as software improves (Historic England, 2017). Copies are lodged with the responsible archaeological authority and at least one other repository.
Where quarrying continues, the survey is repeated at intervals using the same control. Comparing successive elevation models shows where rock and monuments have been removed, and the dated sequence of orthophotographs provides evidence that can be put before a planning authority or a court.