Applications of variable probability sampling using remotely sensed covariates
University of New Brunswick
Variable probability selection methods are regarded as the most efficient sampling design because sample selection is based on parameters of interest. However, a lack of prior knowledge of covariates for study areas restrict the application in practice. This thesis explored the use of covariates derived from airborne light detection and ranging (LiDAR) scanning (ALS) and a consumer-based spherical camera in selecting sample locations with variable probability. Results shows that list sampling with big BAF sample plots is a highly efficient and cost-effective sampling strategy for effectively calibrating ALSderived estimates to local conditions. For the spherical photography study, ratio estimation also showed the capability to calibrate imprecise covariate estimates; however, sampling efficiency under variable probability selection was not improved relative to simple random sampling. The low correlation between the photo-derived covariate and parameter of interest most likely impacted these results. Optimal covariates need further exploration to improve sampling efficiency.