Prioritizing Archaeological Inventory and Protection with Predictive Probability Models at Glen Canyon National Recreation Area, USA

Abstract

To prioritize archaeological site inventory and protection at Glen Canyon National Recreation Area (Glen Canyon NRA), predictive probability models were developed using the Random Forests machine learning method. Glen Canyon NRA consists of approximately 5,000 km2 in Arizona and Utah containing evidence of multiple prehistoric cultures spanning at least 10,000 years. Large portions of Glen Canyon NRA have never been inventoried for archaeological resources. Archaeological sites are potentially subject to irreparable damage along roads and receding Lake Powell shorelines accessible to off-road vehicles and boats. The diverse cultural history, highly variable physiographic environments, archaeological site data inconsistencies, and limited available model variable data provided challenges. Model results improved by classifying sites, based on their inferred usage, and distinguishing distinct physiographic regions within the larger NRA landscape. Model results are being used to target ongoing field surveys. Initial validation is positive, with new sites discovered in areas of predicted high probability.