Modelling and Mapping Fine-Scale Vegetation Biomass in Banff National Park Using Remotely Sensed Data Collected by Unmanned Aerial Systems

• Main Author: Poley, Lucy Gem
• Other Authors: McDermid, Gregory J., Bender, Darren J., Hugenholtz, Chris H.
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: Arts 2020
• Subjects: Geography; Ecology; Remote Sensing; Canada; Bison bison bison; Plains Bison
• Online Access: http://hdl.handle.net/1880/112057; https://doi.org/10.11575/PRISM/37844

Ecological investigations and long-term monitoring programs in grassland ecosystems often require detailed information on vegetation parameters across an area of interest. However, characterization of grassland vegetation is challenging using ground-based measurements or satellite imagery due to the fine-scale heterogeneity present in grasslands. Unmanned Aerial Systems (UASs) provide a way to characterize vegetation at a high spatial resolution, bridging the gap between ground-based measurements and satellite imagery. Motivated by the need for long-term monitoring of fine-scale vegetation parameters following the reintroduction of plains bison (Bison bison bison) to Banff National Park, Canada, this research explored how UAS-derived data could be used to estimate the aboveground biomass of vegetation at remote grassland sites in Banff’s bison reintroduction zone. I assessed the factors affecting quality of UAS-based vegetation estimation in previous research to design a series of analytical experiments. I conducted UAS surveys at study sites in July 2018 using visible-light and multispectral sensors. Concurrent to the aerial surveys, I collected ground-based data on shrub and herbaceous vegetation biomass. I derived spectral and textural variables using one or more wavelengths of light from the UAS imagery and related to ground-measured biomass using linear regression models. I accurately estimated shrub biomass using an area-weighted vegetation index derived from visible-light imagery that fused spectral and structural information into one parsimonious model. For herbaceous vegetation, combining visible-light and multispectral texture information derived from vegetation indices was the best approach to biomass estimation, and I was able to quantify the relative contributions of photosynthetic and non-photosynthetic vegetation within total biomass. I then modelled the distribution of shrub and herbaceous vegetation biomass across the study site and a workflow for collecting and analyzing UAS imagery for ...