A national model for estimating United States public land visitation

Public land management relies on accurate visitor counts in order to understand and mitigate environmental impacts and to quantify the value of ecosystem services provided by natural areas. We build and test predictive visitation models suitable for publicly-managed parks, open space, and other protected lands based on multiple sources of digital mobility data including geotagged posts to social media platforms, community science observations, and a mobile device location dataset from a commercial vendor. Using observational visitation data series from the United States' National Park Service, Forest Service, and Fish and Wildlife Service, we quantify the accuracy of statistical models to predict on-the-ground visitation using individual and combined sources of mobility data alongside other covariates. We find the predictive models performed best in settings where some on-site visitation data can be integrated into the models. On-site visitation data helps to account for meaningful differences in modeled relationships both within and across the three agencies considered. We find variation in the usefulness of the digital mobility data sources, with models combining multiple data sources outperforming those using a single source, including those based solely on mobile device locations. We discuss the practical implications of these findings as well as paths forward to improve visitation estimation on public lands.