Open Collections

Chadwick, Andrew Joel

University of British Columbia

Surveys to monitor forest regeneration are vital for ensuring sustainable forest management and informing long-term timber supply but are often challenged by high costs, safety, and availability of qualified field crews. Advances in fine spatial resolution imagery and digital aerial photogrammetry have enabled detailed 3D mapping of individual regenerating trees. Deep learning, utilizing convolutional neural networks like Mask R-CNN, offers high accuracy and automation for deriving regeneration survey data. This dissertation explores, in Alberta, Canada, automating the acquisition and analysis of regeneration survey data using these technologies. To do so, a Mask R-CNN model was trained to accurately delineate 85% of coniferous tree crowns across different growth conditions. I then derived photogrammetric height measurements that were strongly correlated with field measurements (r2 = 0.93, RMSE = 0.34 m). I then assessed the potential of near-infrared versus conventional optical imagery for the species classification of delineated lodgepole pine (Pinus contorta) and white spruce (Picea glauca) crowns, concluding that conventional optical imagery was suitable for this task. Using this optical imagery, I then applied transfer learning to extend the crown delineation model to instead delineate and classify crown species simultaneously, achieving a mAP of 0.72 and class F1 scores of 0.69 for lodgepole pine and 0.78 for white spruce. Additional fine-tuning on new data showed a potential for further improvement. To support long-term planning, I then developed a methodology to integrate these derived attributes with Alberta's existing growth and yield model. Despite some discrepancies between initial conditions estimated by these methods and conventional field-based surveys, the trajectories and end points of the growth and yield predictions of these datasets were comparable. This research demonstrates the feasibility of these methods and provides guidelines and recommendations for operational implementation, contributing towards more efficient and sustainable forest management in Alberta.

Text

2023-11-23

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/86656

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/