In this month’s ‘Science Notes’, we look at new research that could change the way in which archaeological survey is carried out in the future, exploring an article published in the Journal of Archaeological Science (https://doi.org/ 10.1016/j.jas.2019.10501) that offers the first proof of concept for a method of automating the recording of material culture, such as potsherds, across large areas.

A picture taken by the drone (top) and the results of the automatic potsherd identification (bottom). [Image: H Orengo and A Garcia-Molsosa]

Traditionally, the dispersal of objects across an area is recorded through pedestrian survey, with groups of people fieldwalking, collecting, and recording the location of any finds of archaeological significance. This form of survey has been practised since archaeology was in its earliest phases in the 19th century, but it has several shortcomings. One of the main issues is that, in order to adequately survey a relatively large area, you need large numbers of people working for long periods of time. The post-processing work also takes a significant investment of time and effort, often many years, and demands large amounts of space to store the material collected, which can be expensive and inconvenient.

The drone in flight
The drone in flight. [Image: Anna Karligkioti]

The new method presented in the JAS paper offers a new approach, involving a semi-automated workflow that relies on several recent technological developments. The method was tested in two fields that were being surveyed during the Archaeological Project at Abdera and Xanthi (APAX) in Greece. First, a drone was chosen that was able to follow a planned route for 30 minutes, while taking continuous overlapping high-resolution photographs, flying at a height of 3m above ground level (which was determined to be sufficient to identify ceramic fragments on the surface of the field). This was then immediately followed by pedestrian survey, conducted by an experienced team leader and three students with two weeks of field experience, in order to compare the results of traditional methods and automatic detection.

The photographs collected by the drone were processed using photogrammetry and joined together into an orthomosaic, an aerial photograph that has been corrected so that the scale is uniform. The orthoimages were then uploaded to Google Earth Engine, which offers machine-learning algorithms that can increase the reliability of potsherd identification. The algorithm can be reproduced and altered by anyone with their own orthoimages, and does not require specialist computing resources or coding skills, making this the most accessible way to automatically identify potsherds in the images. Once a vector layer has been generated with features representing each of the potsherds identified, this is incorporated into Geographic Information System (GIS) software, where it can be visualised and analysed in other ways.

When compared to the results of the pedestrian survey, the automated recording method was found to be able to document more potsherds in a shorter time than was possible using the standard method. However, more fragments were recorded during the more intensive ‘total recording’ pedestrian survey. Nonetheless, the time required for pedestrian survey makes it less practical than automated recording in terms of time-cost value.

The automatic recording method focuses on the quantification of surface material, which is the most mechanical part of the process, and allows for the collection of large amounts of information at a much faster speed and with higher analytical capabilities than pedestrian survey. However, it does not attempt to replace human resources in areas such as the analysis of site chronology or landscape use, instead making it possible to direct the efforts of specialised archaeologists towards those areas with higher potential. The new approach does not eliminate the need for material collection, but it can define areas with higher concentrations of material, so that samples can be collected from significant points, speeding up the process and removing the need for the collection and storage of huge amounts of material that exists in traditional systematic collection.

In general, it appears that, although there are some limitations to the automated recording method, it does have the potential to save large amounts of time and resources when applied in favourable conditions. It can also increase traditional survey’s accuracy and consistency, and even identify individual finds and small fragments that would not usually have been found. The presence of a standard methodology could also help with comparability of data, both within and between studies, solving an problem created by the variable strategies used in traditional fieldwalking. The approach is still in its early stages, but the paper does provide a methodological basis on which future applications could be built, and it is possible that, as technological advances continue, it could influence the way that archaeological survey is carried out in the future.


This article appears in issue 359 of Current Archaeology. To find out more about subscribing to CA magazine, click here.


Leave a Reply