- Ground-penetrating radar (GPR) reveals buried structures of ancient Roman city
- GPR lets archaeologists learn about sites without disturbing them
- Computation and machine learning speeds the process of translating raw data into recognizable features
Sifting through sand and digging dirt is just as important to archaeology as reading books and studying artifacts. Manual labor will never totally leave this discipline, but modern technology is taking a lot of the heavy lifting out of our inquiry into the story of humanity.
One such innovation is ground-penetrating radar (GPR). As the name suggests, GPR uses radar pulses to map underground objects or structures. For a field where a strong back is just as important as a strong mind, GPR has been an important tool for decades.
To learn more about what this technology can do and why it’s so important, we sat down with Dr. Lieven Verdonck of Ghent University. He and his colleagues at the University of Cambridge used GPR to study Falerii Novi, a buried Roman city in the current central Italian region Lazio. Their findings help us better understand Roman life, while also showcasing what GPR can do.
Digging deeper into the soil
One of the main goals of the Falerii Novi study was to discover just how different this town was from other Roman settlements like Pompei or Rome. The scientists discovered that, while it wasn’t a completely unique city, it did have certain oddities.
“Some of the city consists of a regular grid (like other Roman cities) and it was probably founded like that,” says Verdonck. “But afterwards all kinds of things happened which gave it the shape it has now, and that shape also has to do with the topography of the city.”
Speaking of the local lay of the land, one of the reasons Falerii Novi was such a good candidate for GPR is the kind of soil that covers the ancient city. Verdonck explains that the dirt they needed to penetrate was dry, allowing the radar to dive down quite deep.
Wet clay soil, on the other hand, would have decreased the chances of finding anything underground. Verdonck also reflects on the fact that GPR doesn’t entirely remove the need for excavations and manual labor.
“Objects in the ground will also generate reflections, and the reflection means that you lose energy because the energy can then no longer travel deeper into the soil,” says Verdonck.
“Stones lying at the surface can also be a problem because in our technique, the instrument needs to be in contact with the soil. If there’s too many stones, we have to clear the field. The preparation of the field sometimes takes even more time than doing the measurements themselves.”
Clearly, GPR doesn’t eliminate all of the work for humans with shovels. But when this technology is combined with machine learning, it becomes an even more vital tool.
Let the computer do the work
Digging up important objects will never go away in archaeology, but GPR is a supplemental instrument that can cut down on unnecessary work. And in terms of working more efficiently, there are few tools better than machine learning.
The data collected by GPR has a very high resolution that can detect details like small pillars in underground heating systems. What’s more, the data collection generally divides a particular space into multiple horizontal sections (‘depth slices’)—sometimes upwards of fifty layers. At the end of the process, Verdonck says it’s not uncommon to have terabytes of data to sift through.
“The traditional way was to draw every single wall or every single archeological feature, digitizing it by hand,” says Verdonck. “That’s no longer feasible, so we need the computer to take a larger role.”
This larger role can take the form of template matching. By defining a template with specific shapes or dimensions, scientists can equip a computer to understand what it’s looking at. For instance, if you defined a template for a wall ten meters long and half a meter wide, the program would be able to match that description to what it sees underground.
“That works really well for simple things such as a rectangular wall, but for much more complicated things, you need to define ever more complicated rules,” says Verdonck. More advanced is a form of machine learning called deep learning. “In that case you have to give the computer a lot of examples by showing it the GPR data, as well as an image of how you as a human interpret the data.”
Despite having access to incredible technological tools, archaeologists will always be in the field lifting rocks and breaking ground. Physical examination of historical objects is a necessary part of discovery. But archaeologists also just enjoy getting their hands dirty.
“I love my job, although it can be really hard sometimes when I'm doing the fieldwork,” says Verdonck. “I'm going back to another site in Italy in July, and I hope it's a bit cooler than a few years ago when it was perhaps 45 degrees Celsius or so at lunchtime. But overall, it's really fun!”