Using machine learning to predict merchantable volume, piece size of central interior tree species in British Columbia, Canada

Abstract

The ability to produce forest inventory product estimates that are cost effective, accurate and can be available in a timely manner has been a struggle in most native forests in the world, but especially in British Columbia, Canada due to its size, species diversity, and disturbance history. LiDAR and optical imagery provides an opportunity to fill this void.  The goals of this study are to demonstrate how piece size and volume per hectare can be imputed from variable radius plots with similar accuracy to fixed radius plots, use object-based image analysis to produce more precise estimate of LiDAR and imagery metrics at the plot level, and show the cost- benefits of using remote sensing metrics at different resolutions plot size and plot sample sizes.

Advisory Team

Principal Advisor - Dr David Pullar

Venue

Room 314/315, Steele Building (03), St Lucia Campus