Keywords: Time Series Analysis, Singular Spectrum Analysis, Flickr, Social Media, Park Visitation
Estimating park visitation counts from social media can greatly reduce the cost of obtaining visitation data. Based on the premise that the recreational behaviors of individuals have a relationship with the social media they post, national park visitation is modeled using data from Flickr, a popular social media platform. The end goal is to forecast percent changes in national park visitation on a month-to-month basis for twenty parks in the United States by using Photo-User-Days (PUD) data from Flickr. The PUD data, which are obtained by counting the number of unique daily users posting photos in a park, are adjusted to account for the popularity of Flickr by counting the number of unique daily users posting on Flickr in the United States. Then, the trend and seasonality components of the adjusted PUD are extracted with singular spectrum analysis (SSA), and are appropriately modified to approximate the corresponding components of the actual national park visitation counts. The resulting model is evaluated by comparing one- to twelve-month ahead forecasts to the actual percent changes.