Large-scale habitat inventories are necessary for understanding how human land-use, natural disturbances (e.g., fire) and climate change are influencing ecological processes, species declines and how restoration can address these changes. Within the boreal forest of Alberta, Canada anthropogenic habitat alteration, also referred to as human footprint or the physical disturbance of a landscape as a result of human activity, covers over 19% of the land area. Much of this human footprint is not recovering on its own. As a part of the Alberta Biodiversity Monitoring Institute’s vegetation regeneration mapping projects, we provide up to date and accurate information on the status of vegetation (e.g., structure and composition) on human footprint using remotely sensed data, such as lidar, and machine-learning approaches.

Lidar data were acquired via fixed wing aircraft with a high-end aerial lidar sensor, Teledyne Galaxy T2000, at a density of 12-16 points per square meter. Derivatives were generated using lidR and included canopy height mode, canopy cover estimate and the arithmetic mean intensity of the signal return. We also used PlanetScope Ortho Scene Level 3B 8 band imagery as input data for the machine-learning algorithms. Vegetation classes (coniferous, deciduous, shrub, other vegetation, non-vegetated) were modeled for lidar-based tree top and randomly generated ground point locations. The model was trained and validated with labeled data generated by trained photointerpreters from stereo models generated from imagery collected concurrent with lidar acquisition using two PhaseOne iXM-RS150F digital cameras. XGBoost was the highest-performing algorithm.

In addition to modelling vegetation class, vegetation height and cover were estimated from lidar data on human footprint and in the surrounding undisturbed forest. The dominant vegetation classes on most human footprint features were shrubs. Most features were characterized by low vegetation structure. When compared to surrounding vegetation structure and composition, these data suggest that natural recovery of these features is low. We developed a robust and repeatable workflow using remotely sensed data to assess vegetation recovery across a large portion of boreal forest in Alberta, Canada which will inform regional and sub-regional land use planning and management, including for the Woodland Caribou, a threatened species and pressing conservation issue across Canada.