An estimation strategy to protect against over-estimating precision in a LiDAR-based prediction of a stand mean

Abstract

A prediction of a forest stand mean may be biased and its estimated variance seriously underestimated when a model fitted for an ensemble of stands (stratum) does not hold for a specific stand. When the sampling design cannot support a stand-level lack-of-fit analysis, an analyst may opt to seek a protection against a possibly serious over-estimation of precision in a predicted stand mean. This study propose an estimation strategy to counter this risk by an inflation of the standard model-based estimator of variance when model predictions suggest non-trivial random stand effects, a spatial distance-dependent autocorrelation in model predictions, or both. In a simulation study, the strategy performed well when it was most needed, but equally over-inflated variance in settings where less protection was appropriate.