Simulation studies to examine bias and precision of some estimators that use auxiliary information in design-based sampling in forest inventory

P W West

doi:10.1186/s40490-017-0101-7

Abstract

BackgroundVarious double sampling methods using both target and auxiliary variables have been developed over many years for use in natural resource inventory.MethodsSimulations of inventory were carried out using four different ratio estimators and model-assisted estimation in each of five rather different example forest populations. Estimates of population means and their standard errors from each of these methods were compared with those obtained using simple random sampling.ResultsWith all five double sampling estimators, bias in estimates of means and standard errors (the latter estimated analytically or through bootstrapping) was generally small and consistent with theoretical expectations. Their efficiency increased as either the first- or second-phase sample sizes increased. All were more efficient than estimates obtained using simple random sampling as long as there was some positive level of correlation between the target and auxiliary variable. However, none of the double sampling estimators was more efficient than any of the others.ConclusionsFor many forest inventory tasks, users may well be able to use whichever of the estimators is most convenient to their purpose. However, model-assisted estimation has application in a wider range of circumstances than the other methods, which perhaps recommends it for general use.

Highlights

Various double sampling methods using both target and auxiliary variables have been developed over many years for use in natural resource inventory
Estimators Numerous authors have described in formal detail the various design-based sampling techniques and estimators used in forest inventory (Schreuder et al 1993; Kangas and Maltamo 2006; Gregoire and Valentine 2008; Mandallaz 2008; Ståhl et al 2016); these texts informed the approach and the estimators considered here
The present work compared results obtained when sampling from five simulated forest populations that had rather differently shaped frequency distributions of their target variables

Summary

Introduction

Various double sampling methods using both target and auxiliary variables have been developed over many years for use in natural resource inventory. Forest inventory either maps how some forest characteristic (the target variable) varies across a forested area or estimates the total or mean value of that variable across the area. To assess the level of confidence that one may have in the result, it is usual to estimate the level of precision of Forest target variables can be difficult, time consuming and expensive to measure on the ground. It is common to have available one or more auxiliary variables that are correlated, at least to some extent, with the target variable and can be measured relatively . Stand basal area might be such a variable for the two examples mentioned above. Such variables can be used very effectively

Methods

Results

Discussion

Conclusion