Geospatial segmentation of recreational visits in the Boundary Waters Canoe Area Wilderness, Minnesota: an unsupervised machine learning approach to travel modeling for wilderness management

Keywords: travel modeling, itinerary prototyping, unsupervised learning, k-means, visitor management, wilderness
Session Type: Virtual Paper Abstract
Day: Friday
Session Start / End Time: 2/25/2022 09:40 AM (Eastern Time (US & Canada)) - 2/25/2022 11:00 AM (Eastern Time (US & Canada))
Room: Virtual 64

Abstract

The Boundary Waters Canoe Area Wilderness (BWCAW) in the Superior National Forest (Duluth, Minnesota) is a reference wilderness for lacustrine and fluvial recreation in the USA. With nearly 12,000 individual trips stored in the US Forest Service visitor census, the BWCAW provides a rich source of public data for the modeling of travel patterns in wilderness and a gold standard to develop business intelligence for the management of outdoor recreation. This paper illustrates an innovative adaptation of the k-means unsupervised machine learning algorithm for the recognition and clustering of spatial travel patterns in datasets of trip itineraries. The analysis follows a multicriteria workflow to identify the optimal number of trip clusters looking at cluster distortion, information loss (via the Akaike Information Criterion), predictability of the cluster arrangement, and the spatial separation of trip clusters (using the Jaccard Index). Results describe (1) several clusters of trip events based on the geospatial footprints of their itineraries across the wilderness, and (2) trip profiles relative to the operational and demographic features of the trips that make up each cluster, providing insights into how BWCAW places and resources are differently used by visitors as a function of the nature of the trip. Outcomes of this study aim to inform BWCAW managers in the recognition of travel patterns among recreational visitors and in the building of behavioral models for the forecasting of future travel patterns, enabling an enhanced decision-making pipeline for adaptive planning in this wilderness.