Maintaining undisturbed habitat is key to the recovery strategy of the Woodland Caribou boreal population. Seismic lines have been identified as a human footprint feature that requires specific attention for habitat restoration in part because linear features are linked to increases in the predation rates of caribou as they provide lines of sight and easy transportation routes for predators. It is estimated that there are hundreds of thousands of kilometers of seismic lines within Alberta (ABMI & AHFMP 2023). Using the standard 1.5 m resolution SPOT 6 satellite photo interpretation methods employed by the Alberta Human Footprint Inventory, narrow low impact seismic lines are under reported. This makes achieving and maintaining the 65% undisturbed habitat required for herds to have a 60% chance of self-sustainability difficult in some areas (Environment Canada 2012). . Establishing an updated inventory of seismic line features is critical to an evidence-based approach to habitat restoration. 

To create an updated seismic line inventory within caribou ranges as a part of ABMI’s vegetation regeneration mapping projects in northern Alberta, a combination of high-density lidar data, imagery, machine learning, and human verification are utilized. The lidar data is collected via fixed wing aircraft by a high-end aerial lidar sensor Galaxy T2000 with a density of 12 to 16 points per square meter. Using this point cloud data, 0.5 m resolution derivatives are created based with lidR including normalized intensity, canopy height model, and topographic position index (tpi-8m). .  The seismic shift model is a segmentation neural network based on U-net architecture that receives this input data and determines linear features within the environment. The NN was trained and validated on approximately 473 km of seismic lines from four different areas within the boreal forest. To ensure that the linear features detected are located in the centre of the corridors and that seismic lines and not related to other human land-uses, the lines are reviewed by GIS Technicians in one kilometer square tiles. Linear features such as gravel and paved roads, transmission lines, and pipelines are removed and seismic lines missed by the model are manually added.

With the newly developed workflow (a pipeline consisting of neural network predictions, automated post-processing, followed by human correction) the speed and accuracy of seismic line delineation was dramatically increased. For example, in one study area, we identified approximately 90% more linear kilometers of seismic lines compared to the existing seismic line features in the Alberta Human Footprint Inventory (ABMI, 2023). This increase is attributed to enhanced ability to delineate low impact seismic lines by the newly developed method.