Sample designs for multi-objective environmental surveys

Abstract

Large-scale forest and range surveys have been collecting ground data for the better part of the last century. From their inception, these surveys were designed for the estimation of population level means and totals. While these survey programs have been effectively meeting this objective, users of these data are often interested in analytical issues other than population level estimation. One of the most widely requested products are maps displaying the spatial extent of the resource in question. The production of these maps requires the prediction of the attributes at unobserved locations. The concern for a number of surveys programs is that data collected for estimation purposes is often poorly suited for spatial prediction. While it is always possible to collect additional data for the purpose of spatial prediction, this solution is often too costly to be practical. The purpose of this study to determine effective methods for simultaneously meeting both estimation and prediction objectives. This is achieved by first determining the most appropriate sample design for the purpose of estimating population level means and totals and then by adapting this design for the purpose of spatial prediction.