Predicting Archeological Sites Ahead of the Southern Pine Beetle

The Preservation Technology Podcast, produced by the National Park Service's National Center for Preservation Technology and Training (NCPTT), features the people and projects bringing innovation to historic preservation. In episode 139, host Sadie Schoeffler Whitehurst and NCPTT's Tad Britt spoke with Zac Selden of the Heritage Research Center at Stephen F. Austin State University about his NCPTT grant: modeling archeological sites in the Sabine National Forest in preparation for the southern pine beetle. Selden's comments are paraphrased here.

The episode: "Southern Pine Beetles and Archeological Site Modeling", from the National Park Service's Preservation Technology Podcast.

Zac Selden (left) with Tad Britt. NPS, NCPTT

How the project began

The work grew out of Selden's meeting with Britt and a predictive modeling seminar at the University of Arkansas, where he also met Forest Service staff interested in modeling the Davy Crockett National Forest. With Britt, he used a bibliometric survey to identify the most competitive tools and techniques, chose software and methods, and moved the archeological collections from the national forests and grasslands in Texas to his lab to be documented.

Why beetles matter to archeology

The southern pine beetle is a native insect that devastates pine trees. Selden described it moving through a forest in bursts before shifting east or west from season to season, and said its range is expanding because of climate change: it had moved west into Mississippi and Louisiana and spread north as far as New York and New Jersey. Forest managers respond to outbreaks in predictable ways, but until now they have largely relied on informed guesswork about where archeological sites might be affected. The goal was to have a model of the Sabine National Forest ready before the beetle arrived — and to make it portable to other forests and properties facing similar threats.

Building a predictive model

A predictive model starts with the locations of known sites. Through discussions with Britt and with national forest staff, the team chose environmental variables that seemed to correlate with where sites are found — proximity to water, elevation, soil and vegetation. In the earlier Davy Crockett model they tested every available variable, using a bootstrap method to find those most closely tied to site locations, while taking care to avoid overfitting, which produces high accuracy that does not generalize. That model served as a stage for design, research and learning as a group.

The team uses a maximum entropy (Maxent) approach, a machine-learning method. They began with a point-and-click interface and moved to the open-source version available in the R programming language, so the code could be shared and reused wherever similar resources face similar risks, with local variables plugged in. The method uses 70% to 80% of known sites to build predictions and holds back the remaining 20% to 30% to test how accurate they are.

Selden has also been exploring side questions, such as where mounds occur in different periods and how land use and population change over time.

Sources

• National Park Service, National Center for Preservation Technology and Training, "Podcast 139: Southern Pine Beetles and Archeological Site Modeling": https://www.nps.gov/articles/000/podcast-139-southern-pine-beetles-and-archeological-site-modeling.htm
• Words on this page from people and organisations outside the federal government are paraphrased; rewritten in hubnx's own words.
• The episode's audio is the National Park Service's own file, from the podcast's feed.