Katherine Peck Publishes New Research on Machine Learning in Archaeology

Katherine Peck Publishes New Research on Machine Learning in Archaeology

The New Mexico Consortium (NMC) is pleased to highlight a recent publication by former Junior Research Scientist Katherine Peck, “Using Simulated Training Data to Locate Archaeological Sites with Machine Learning,” co-authored with Claudine Gravel-Miguel, Grant Snitker, and Matthew Helmer.

While working with NMC’s Cultural Resource Sciences program (and completing her PhD in Anthropology at the University of New Mexico) Peck helped analyze LiDAR data from the Kisatchie National Forest in Louisiana. The team noticed unusual features that resembled historic tar kilns but didn’t quite match known examples. Searching for similar features across the landscape by hand would have been slow and difficult.

To speed up the process, the team turned to machine learning. Because there were few real examples of these features to train a model, they created simulated ones by modifying LiDAR data using an approach called procedural generation. This technique adds randomized features to the LiDAR data, inspired by techniques used in video game design. They then trained a computer vision model to look for these patterns.

The model successfully identified the original features and flagged additional areas of interest, though it also produced false positives that required manual review. Field visits ultimately showed that the features were not tar kilns, but likely historic military infrastructure.

While the model required some cleanup and was less precise than similar studies, the project shows how simulated data can help archaeologists find unusual features and guide fieldwork more efficiently.

Peck graduated in fall 2025 and is now a postdoctoral researcher in the School of Computing at the University of Wyoming.

Read the entire article at: https://doi.org/10.1017/aap.2025.10130

Top image caption: Examples of simulated features added to a hillshaded LiDAR tile.