Airborne Laser Scanning-Assisted Sampling for Remote Regions.

Acquisition of probability samples of sufficient size to produce population live tree volume, biomass and carbon estimates for remote regions that satisfy precision criteria can be extremely expensive, even when the estimation procedure is enhanced using remotely sensed auxiliary information.  An alternative is to use non-probability samples with model-based inference whose validity is based on correct model specification, not probability samples.  This approach has been investigated for forestry applications but always with probability samples or existing non-probability samples.  Little attention has been directed toward devising optimized remote sensing-based sampling schemes for use with probability-based estimators, but no attention is known to have been directed toward devising such schemes for use with model-based estimators.  For a study area in northern Minnesota in the United States of America, the objectives focused on comparing combinations of simple random, stratified, and adaptive sampling schemes and both probability- and model-based estimators with respect to estimates of mean volume per unit area and the standard errors of the estimates.  The most important result was that an adaptive sampling scheme in combination with model-based inference greatly increased the precision of the population estimator of mean live tree stem volume per unit area and simultaneously greatly reduced the sampling costs associated with travel to and from sample unit locations.