Optimising nearest neighbour information—a simple, efficient sampling strategy for forestry plot imputation using remotely sensed data

ABSTRACT

The rapid adoption of LiDAR-assisted forest inventory, together with nearest neighbour imputation, has created new challenges for those involved in sample selection. It is not obvious a priori what type of sample will be the most efficient when used with imputation. In this paper we explore a number of sampling approaches, including conventional methods such as stratification, spatial methods based on tessellations, and balanced sampling. We introduce a new approach called nearest centroid (NC) that optimises the survey design by using the distance properties of the sample in the space defined by the auxiliary variables, and examine this and other methods across a range of survey situations and objectives. The NC method is shown to be highly efficient when compared with other methods and is very flexible in the way it can be implemented.