A remote sensing methodology for sub-stand growth and yield projection of post-harvest forest regeneration

Abstract Sustainable forest management is driven by accurate characterization of current forest conditions and predictions of future stand development. The spatial scale of these data affects the level of detail with which decisions can be made. We demonstrate a methodology to forecast forest regeneration attributes derived from fine spatial resolution image data at a sub-stand resolution. A previously trained deep learning model was applied to 10 m × 10 m image tiles to delineate crowns by species in 13-year-old previously harvested stands in Alberta, Canada. Delineated crowns were used to produce stem counts and derive top height measurements from photogrammetrically derived canopy height models. These data were combined with site index values from ecosite phase classification to forecast future forest attributes with the Growth and Yield Projection System. We compared initial and forecasted image-derived forest attributes with conventional methods utilizing field data inputs. The image-derived and field data projections had similar trajectories and endpoints, affirming the utility of this methodology for conventional use. In addition, by mapping growth and yield projections to the 10 m × 10 m grid, our approach provides a spatially detailed resource that can be used to advance understanding of silviculture dynamics and better inform decision-making throughout stand development.