Costa, Ana Cristina Marinho daTang, VicenteGranell-Canut, CarlosHubach, Christopher Paul2025-03-132025-02-27http://hdl.handle.net/10362/180558Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial TechnologiesBackcountry navigation in trail-less wilderness areas is infrequently studied. Denali National Park and Preserve (DENA) is a largely trail-less park, containing only 57.1 kilometers of official trails in over 6 million acres (2.45 million hectares). This makes backcountry travel in DENA dependent on environmental and terrain factors, rather than infrastructure like trails. This study utilizes ranger patrol data collected in DENA since 1998, comprising thousands of kilometers of GPS track lines, to model off-trail navigation choices. We optimize tree-based ensemble machine learning models, comparing techniques like ADASYN, RUS, and focal loss to address class imbalance, to generate a probability surface of likely travel routes. This surface was then used with an A* routing algorithm to predict the most probable routes between two points. Sample routes are then compared to line samples from the patrol tracks dataset, as well as visitor samples, to assess generalizability. The selected model achieved a G-Mean of 0.65, indicating low to moderate predictive capability, and additional research is warranted to improve model predictions. Key terrain features, such as slope, elevation, and landform type, show importance in the influence of predicted travel paths. This research demonstrates the potential of using ranger patrol data and machine learning to understand off-trail movement patterns in wilderness areas, informing management decisions related to visitor safety and resource protection.engDenali National Parkwilderness travelbackcountry navigationtrail-less routingmachine learningSDG 12 - Responsible production and consumptionSDG 13 - Climate actionSDG 16 - Peace, justice and strong institutionsmaster thesis203923731