Spatial-temporal models for improved county-level annual estimatesThis article is part of a larger document. View the larger document here.
The consumers of data derived from extensive forest inventories often seek annual estimates at a finer spatial scale than that which the inventory was designed to provide. This paper discusses a few model-based and model-assisted estimators to consider for county level attributes that can be applied when the sample would otherwise be inadequate for producing low-variance estimates in the smaller counties. I present and demonstrate simple spatial and/or temporal estimators that draw strength from neighboring counties and/or years in order to increase confidence in the county level annual estimates. The spatial estimators are restricted to those that do not require knowledge of exact plot locations in order to enable their use with privacy protected, publicly available data. A series of simulations is used to compare and contrast the performance of these estimators relative to position in the time series of interest under various variance prescriptions. Although none of the estimators is shown to be superior in terms of minimum mean squared error (MSE) overall, a few general conclusions are drawn. The first is that estimators that draw strength through consecutive measurements of the same set of field plots show a significant reduction in MSE under a wider variety of circumstances than those that draw strength from plots in neighboring counties. The second conclusion is that of the estimators that rely on a temporal model, a simple, centralized weight-adjusted moving average (with weights specific to time-series position) often was the most robust.